SID 5 Research Project Final Reportrandd.defra.gov.uk/Document.aspx?Document=PH0191_6314_FRP.pdf ·...

SID 5 (Rev. 3/06) of 23

General enquiries on this form should be made to: Defra, Science Directorate, Management Support and Finance Team, Telephone No. 020 7238 1612 E-mail: [email protected]

SID 5 Research Project Final Report

Note In line with the Freedom of Information

Act 2000, Defra aims to place the results of its completed research projects in the public domain wherever possible. The SID 5 (Research Project Final Report) is designed to capture the information on the results and outputs of Defra-funded research in a format that is easily publishable through the Defra website. A SID 5 must be completed for all projects.

• This form is in Word format and the boxes may be expanded or reduced, as appropriate.

ACCESS TO INFORMATION The information collected on this form will

be stored electronically and may be sent to any part of Defra, or to individual researchers or organisations outside Defra for the purposes of reviewing the project. Defra may also disclose the information to any outside organisation acting as an agent authorised by Defra to process final research reports on its behalf. Defra intends to publish this form on its website, unless there are strong reasons not to, which fully comply with exemptions under the Environmental Information Regulations or the Freedom of Information Act 2000.

Defra may be required to release information, including personal data and commercial information, on request under the Environmental Information Regulations or the Freedom of Information Act 2000. However, Defra will not permit any unwarranted breach of confidentiality or act in contravention of its obligations under the Data Protection Act 1998. Defra or its appointed agents may use the name, address or other details on your form to contact you in connection with occasional customer research aimed at improving the processes through which Defra works with its contractors.

Project identification

1. Defra Project code PH0191

2. Project title

Generic system for detection of statutory pests and pathogens

3. Contractor

organisation(s) Central Science Laboraory Sand Hutton York YO41 1LZ

4. Total Defra project costs £ £265,016.00

(agreed fixed price)

5. Project: start date ................ 01 March 2003 end date ................. 28 February 2007

SID 5 (Rev. 3/06) of 23

6. It is Defra’s intention to publish this form. Please confirm your agreement to do so....................................................................................YES NO

(a) When preparing SID 5s contractors should bear in mind that Defra intends that they be made public. They should be written in a clear and concise manner and represent a full account of the research project which someone not closely associated with the project can follow.

Defra recognises that in a small minority of cases there may be information, such as intellectual property or commercially confidential data, used in or generated by the research project, which should not be disclosed. In these cases, such information should be detailed in a separate annex (not to be published) so that the SID 5 can be placed in the public domain. Where it is impossible to complete the Final Report without including references to any sensitive or confidential data, the information should be included and section (b) completed. NB: only in exceptional circumstances will Defra expect contractors to give a "No" answer.

In all cases, reasons for withholding information must be fully in line with exemptions under the Environmental Information Regulations or the Freedom of Information Act 2000.

(b) If you have answered NO, please explain why the Final report should not be released into public domain

Executive Summary

7. The executive summary must not exceed 2 sides in total of A4 and should be understandable to the intelligent non-scientist. It should cover the main objectives, methods and findings of the research, together with any other significant events and options for new work.The main objective was to determine whether volatile and acoustic technologies have the potential to be used as effective on site detection methods for interception of quarantine pests and diseases by the Plant Health and Seeds Inspectorate. A. Volatile detection of potato diseases and invertebrate pests

It is now well established that some diseased plants eg potato tubers with soft rots, produce characteristic volatile compounds that can be detected by chemosensors such as electronic noses. Our aim here was to determine whether the latest developments in biosensors and chemosensors could be applied to portable instrumentation for PHSI to use in improved detection of quarantine pathogens and pests. The selected targets were potato ring rot and potato brown rot caused by the bacteria Clavibacter michiganensis subsp sepedonicus and Ralstonia solanacearum respectively and the insect pest Thrips palmi.

The objectives and summary of progress were:

A1. Obtain/prepare samples of chosen representative commodities with and without the selected pest and diseases. Thrips palmi is the selected pest. Potato ring rot and potato brown rot are the selected diseases. It was not possible to produce ring rot or brown rot tubers by direct inoculation due to secondary invasion by soft rot bacteria and fungi. Instead we had to infect seed tubers, plant them in quarantine conditions and harvest the naturally infected progeny for use in volatile analysis. For brown rot we cultured potato plants under quarantine glasshouse conditions but for ring rot we had to use relatively low temperature quarantine growth cabinets at 18°C and as a result space was limited. Although we managed to produce reasonable quantities of brown rot infected tubers throughout the project it proved difficult to provide a regular supply of ring rot infected tubers. A colony of Thrips palmi was produced. A2. Determine unique volatiles or combinations of volatiles by analysis of headspace and liquid phase run-offs to establish chemical information representative of problem. Early work identified the volatile compounds using gas chromatographic mass spectroscopy (GCMS). No volatile signals were produced by Thrips palmi and it was agreed to do no further work with insect pests. For potato ring rot and brown rot there were potentially unique volatile compounds produced but later work showed that changes in the total volatile profile offered greater potential for detection than individual compounds. Few if any of the compounds were only found in diseased tubers. Nevertheless the profiles of diseased tubers were clearly differentiated from those of healthy tubers. A3. Investigate the potential of biosensors especially array based biosensors to detect the identified

SID 5 (Rev. 3/06) of 23

changes and markers in headspace at realistic levels of problem. The absence of specific diagnostic volatile compounds in diseased tubers resulted in a change of direction from a biosensor to a chemosensor approach and we selected an Electronic Nose using an array of 8 individual chemo sensors to facilitate detection of the range of volatile compounds found in diseased tubers. The volatile molecules were captured in the headspace by using solid phase microextraction (SPME) fibres. The sampled odours interact with the sensor array and change the physical properties of each sensor, which are measured, converted from analogue to digital output and analysed by pattern recognition. A4. Develop and test demonstrator instrument to confirm realism of approach and establish generality of application in specific situations eg for a selected pest in a glasshouse or a disease in a containerised crop commodity An initial prototype Enose was produced and profiles of brown rot and ring rot tubers screened and compared with healthy controls. Tubers from more than 10 different potato varieties were included to show the natural variation that may occur. Tubers with different levels of infection were also included to represent early to advanced symptoms. However it was difficult to produce guaranteed latently infected tubers without recourse to bacteriological analysis which wounded them and altered the volatile profile. We also used naturally infected tubers from outbreaks and interceptions of both diseases. Identification of diseased tubers was made through training of an artificial neural network using the Enose. A second improved prototype was produced at the end of the project but this does not yet have a database of volatile profiles for brown rot or ring rot. A5. Make recommendations for further development and use. We now have a prototype Enose capable of identifying tubers with visual brown rot symptoms. The databases for this need to be developed using profiles from a range of brown rot infected tubers. It was demonstrated at the PHSI Technical Conference in 2007. CSL staff have been trained in its use and once the database has been developed it could be used by PHSI. The equipment is also likely to be able to detect ring rot infected tubers but this would probably require more processing and validation than that required for brown rot detection. B. Acoustic detection of invertebrate pests

There have been recent advances in cost effective acoustic detection systems and software that facilitates acoustic pattern recognition. Our aim was to determine whether these advances were appropriate for portable systems that could be used by PHSI for specific and sensitive detection of selected quarantine insect pests. The selected targets were the Asian Longhorn Beetle, Anoplophora glabripennis, Thrips palmi and the moth, Spodoptera littoralis.

The specific objectives were:

B1. Develop and utilise novel acoustic technology to detect and differentiate between key selected quarantine and indigenous pests. These will include Asian Longhorn beetle, Thrips palmi and Spodoptera littoralis. These were selected from the PHSI list because they represent different types of target for acoustic detection. Few of the target species were available consistently throughout the project and it was agreed to work on locally available native species that fed on similar plant materials to the target species. A total of 14 insect species were studied, of which the most studied were beetles with wood boring larval stages. A range of recording devices was compared as well as a range of acoustic detectors. Recording devices included mini disc CD recorders and digital tape recorders. Detection systems included biomorphs, piezoelectric sensors, electret microphones and an accelerometer. The first 3 are cheap devices for which many can be strung together to give good coverage of the volume to be screened. Accelerometers are large expensive single detectors. They were all tested in various combinations over a range of different plant materials representing the food plants of the selected insects. The next phase looked at the analysis of the acoustic signals to determine whether biting signals could be differentiated at species level. Different signal process technologies were developed to enable detection and identification of individual insect sounds, largely as a result of biting activity. A bite is characterised by a short duration impulse–like signal. Detection of the biting sounds was carried out by both vertical and horizontal threshholding. Once the acoustic event had been located and isolated, the next step was to extract the appropriate features of the signal for input to a classifier. We used Time Domain Signal Coding (TDSC) because it provides a good feature set with minimal computational complexity. Classification of the TDSC signal was made using two different forms of artificial neural network (ANN). The early work was on detection of beetle larvae (Hylotrupes bajalus) using the different combinations of recording devices and sensors with varying types of neural network. The best combination for detection of wood boring insects was found to be piezoelectric transducers with the LVQ ANN.

SID 5 (Rev. 3/06) of 23

Signals from a total of 14 different insects were tested, mostly comparing signals of H. bajalus in ash twigs with Prionus coriarius in small oak logs Again the combination of piezoelectric transducers and LVQ network proved best and gave highly accurate identification of the sounds of these insects. Most of the 14 insects species tested gave unique diagnostic signals and apart from discrimination between insects there was good discrimination between these and background sounds. Final work moved towards real time rather than recorded analysis to avoid the problems of searching through the recorded data. It was also clear that hardwood signals were more easily detected than soft wood signals. B2. Investigate the effects of climatic conditions and cultural practices on acoustic recognition of the selected quarantine pests. Supplies of insect pests were not good enough for these studies to be satisfactorily completed and the milestones were adjusted to take account of this. B3. Validate the use of the acoustic detection technology against available outbreaks of the selected quarantine pests. For each of the selected pests we will aim to determine thresholds of detection eg for Asian Longhorn beetle, could a single insect be detected in a consignment of Bonsai ? No outbreaks were available. A possible trial with Acers infected with Anoplophora chinensis failed to materialise at the last minute when the imported consignment was re-exported to China. B4. Develop a prototype acoustic recognition apparatus and produce and deliver appropriate guidance to PHSI on its use. Prototype equipment has been produced and demonstrated to PHSI. No outbreaks were available for further development with PHSI. Thus for both methods, prototypes have been developed and are available for use by PHSI and proof of principle has been shown. For both, some further development is necessary before they can be handed over to PHSI. For the Volatile detection this requires populating the database with more brown rot and ring profiles from individual infected tubers together with assessments of thresholds of infection. For the acoustic detection, the database needs to be populated with signals from the pests that PHSI wish to detect and then be validated. For both, the major advantages are generic platform, field portability, ability to screen non infected samples rapidly, remote detection over long periods of time and avoidance of common sampling issues. Both have great potential for use as tools by the PHSI in their inspection capacity.

Project Report to Defra

A. As a guide this report should be no longer than 20 sides of A4. This report is to provide Defra with details of the outputs of the research project for internal purposes; to meet the terms of the contract; and to allow Defra to publish details of the outputs to meet Environmental Information Regulation or Freedom of Information obligations. This short report to Defra does not preclude contractors from also seeking to publish a full, formal scientific report/paper in an appropriate scientific or other journal/publication. Indeed, Defra actively encourages such publications as part of the contract terms. The report to Defra should include:

the scientific objectives as set out in the contract; the extent to which the objectives set out in the contract have been met; details of methods used and the results obtained, including statistical analysis (if appropriate); a discussion of the results and their reliability; the main implications of the findings; possible future work; and any action resulting from the research (e.g. IP, Knowledge Transfer).

OBJECTIVES The main objective is to determine whether volatile and acoustic technologies have the potential to be used as effective on site detection methods for interception of quarantine pests and diseases by the PHSI. A. Volatile detection of potato diseases and invertebrate pests The overall objective of this part of the proposal is to conduct strategic research to develop a generic solution to aid plant health inspectors detect a range of pests and pathogens of statutory importance. The output of this research will be proof of principle and a demonstration apparatus with some validation using both laboratory test

SID 5 (Rev. 3/06) of 23

and field samples. For each of the selected pests we will determine the probability of missing it under defined circumstances eg for brown rot what is the chance of not detecting a defined number of latently infected or symptomatic tubers in a lot of given size in a closed refrigerated container. Our studies will start with potato brown rot and potato ring rot detection and if successful will move onto detection of Thrips palmi. In the event of problems with Thrips palmi, one pest will be studied instead from among Spodoptera littoralis, Anoplophora glabripennis, Liriomyza huidobrensis, L. sativa and L. trifolii. A1. Obtain/prepare samples of chosen representative commodities with and without the selected pest and

diseases. Thrips palmi is the selected pest. Potato ring rot and potato brown rot are the selected diseases. A2. Determine unique volatiles or combinations of volatiles by analysis of headspace and liquid phase run-offs to

establish chemical information representative of problem. A3. Investigate the potential of biosensors especially array based biosensors to detect the identified changes and

markers in headspace at realistic levels of problem. A4. Develop and test demonstrator instrument to confirm realism of approach and establish generality of

application in specific situations eg for a selected pest in a glasshouse or a disease in a containerised crop commodity.

A5. Make recommendations for further development and use. B. Acoustic detection of invertebrate pests The overall objective of the acoustic detection component is to undertake strategic research to investigate the viability of utilising novel acoustic signal processing technologies to detect insect pests of quarantine importance. Specific objectives are: B1. Develop and utilise novel acoustic technology to detect and differentiate between key selected quarantine

and indigenous pests. These will include Asian Longhorn beetle, Thrips palmi and Spodoptera littoralis. These were selected from the PHSI list because they represent different types of target for acoustic detection.

B2. Investigate the effects of climatic conditions and cultural practices on acoustic recognition of the selected quarantine pests.

B3. Validate the use of the acoustic detection technology against available outbreaks of the selected quarantine pests. For each of the selected pests we will aim to determine thresholds of detection eg for Asian Longhorn beetle, could a single insect be detected in a consignment of Bonsai ?

B4. Develop a prototype acoustic recognition apparatus and produce and deliver appropriate guidance to PHSI on its use.

EXTENT TO WHICH OBJECTIVES HAVE BEEN MET In the volatiles work, the nature of the volatile patterns were different to those anticipated and as a result, there was a switch from development of biosensor to an electronic nose. This is discussed below. We studied the 3 diseases and pest listed but since there were no volatile signals from the insect pest we concentrated on the 2 potato diseases. No unique volatiles were found. Instead there were significant differences between the volatile patterns from diseased and healthy potato tubers, more so for brown rot than for ring rot. The first prototype carried a database of brown rot, ring rot and other potato rot disease profiles. This was replaced with an improved demonstration model late in the project. None of the previous data was collected under the new conditions and there was insufficient time to develop the database further. CSL now have this demonstration Enose together with the operating software and a manual for its use either by CSL staff or by PHSI. All that remains is to produce a database within the software for analysis of field samples. In the acoustics work, we struggled to obtain cultures of the Asian Longhorn beetle (Anoplophora glabripennis) due to lack of availability in the UK and elsewhere. However, its close relative Anoplophora chinensis was available and made a good substitute. The other 2 target insects were studied but due to the nature of the signals, more emphasis was placed on a series of beetle species that were readily available throughout the project but had similar biting acoustic signals. Good use was made of the available outbreaks of quarantine pests through liaison with PHSI but these were not numerous and not always easy to work with due to the nature of the quarantine status. For example, a consignment of Anoplophora chinensis infected Acers were re exported before work could begin. A prototype instrument has been constructed and is the focus of a follow up bid to DEFRA for further development together with PHSI. TECHNICAL REPORT The partners

SID 5 (Rev. 3/06) of 23

CSL’s role was to lead the project (D. Stead), to provide supplies of brown rot and ring rot infected potato tubers (N. Parkinson) and insect pests (D. Morgan, J. North, L. Blackburn) and to carry out GC Mass Spectrometric analyses of volatiles (J. Chambers, G. Bryning). The University of Manchester’s role was to determine the biosensor approach and later develop the electronic nose (Prof. K. Persaud, J. Stinson (PhD student)) The University of York’s role was to develop the acoustic detection system (Dr D. Chesmore, I. Farr (PhD student)) A. DEVELOPMENT OF AN ELECTRONIC NOSE BASED ON VOLATILE SIGNALS 1.0 INTRODUCTION There is a risk that pathogens present in imported potatoes may infect indigenous potato crops. Both brown rot (Ralstonia solanacearum) and ring rot (Clavibacter michiganensis) in potatoes are quarantine diseases in the EC Plant Health Directive and are notifiable in the UK. It is very difficult to identify infections rapidly at the ports of entry and a need was identified by Defra for portable instrumentation that could be used by a Plant Health and Seeds Inspector (PHSI) that would be capable of detecting these diseases. The objective of this research was to develop non-invasive techniques for detection of quarantine pathogens in potatoes, with the aim of developing an instrument that could potentially be used by PHSI inspectors at ports of entry for screening imports of potatoes. A PhD research project at the University of Manchester in collaboration with Central Science laboratories focussed on detection of potato ring rot and brown rot. The project commenced with a review of the literature that encompassed the nature of potato infections involving Ralstonia solanacearum (RS) and Clavibacter michiganensis subsp. sependonicus (Cms). Biological, biochemical and chemical changes/markers associated with the two pathogens that were to be focussed on within the project, and existing sensor technologies that might have some impact in terms of devising a strategy for the project. Both biochemical and chemical sensing approaches were considered, and problems of developing an instrument that could be used for rapid screening were considered. From the available information, a system capable of detecting volatiles was considered to be a viable approach. The question asked was - can volatile organic compounds (VOCs) emitted during shipment of plant tissues by microorganisms be utilised as early disease indicators? Accordingly a programme of chemical analysis of volatiles emitted from potatoes – both infected and non-infected was commenced to determine whether suitable chemical markers could be identified for the pathogens to be investigated. We identified that there would be problems associated with detection of low concentrations of volatiles, and so a method of pre-concentration of volatiles from the headspace was required. Solid-phase microextraction (SPME) was selected over other methods of volatile pre-concentration methods, and was investigated as a way forward. METHODS AND RESULTS 2.0 PROVISION OF BIOLOGICAL MATERIALS Producing a regular supply of potato tubers infected with ring rot and brown proved to be a challenge. For brown rot, a method was devised that allowed production of reasonable numbers of tubers but for ring rot infected tubers success was very variable, often with no tuber initiation after 4 months of culture. Initial studies with inoculated potato tubers at various populations by placing standard volumes in wounds through the tuber vascular system did not produce any tubers with typical systemic symptoms of either brown rot (Ralstonia solanacearum) (Rsol) or ring rot (Clavibacter michiganensis subsp sepedonicus) (Cms). Most wounds developed typical Pectobacterium induced soft rots. These tend to produce much greater volatile signals than Rsol or Cms induced rots. This approach was abandoned. All further studies used naturally infected tubers produced by planting chitted seed that had been stabbed lightly with Cms cultures or watered at tuber initiation time with Rsol suspensions. In each case inocula were c 106-107 cfu/ml. Rots were produced in a wide range of potato cultivars. Plants were maintained in a quarantine glasshouse for up to 4 months before harvesting tubers for use in GC-MS or Enose studies. Cultivars included Cara, Desiree, Estima, Jersey Royal, Maris Peer, Maris Piper, Marfona, Pentland Crown, Pentland Dell, Rocket and Wilja. All these were successfully infected with Rsol. Samples of infected potato tubers were produced throughout the project, although not necessarily always in the quantities required at the time. Not all were successfully infected with Cms. For most ring rot trials, few infected tubers were found either as latent or as visual symptoms. This was assumed to be due to high temperature peaks greater than 25°C. Further production of infected tubers was carried out in quarantine growth cabinets at 18°C, from which small numbers of infected tubers were obtained for initial studies but not in sufficient quantities to develop an Enose able to detect ring rot infected potatoes. A major problem was production of guaranteed latently infected tubers. Since these show no

SID 5 (Rev. 3/06) of 23

external symptoms, they need to be tested for the presence of pathogen. This required wounding of the tubers which affected the volatile profile. Thus, confirmation of presence of latent infections had to be done after screening for volatiles and volatile profiles for any pathogen - free tubers abandoned. Milestone changes were agreed to account for these difficulties. Volatile patterns for other potato pathogens, commonly encountered in stores were included. Tubers were thus also infected with strains of soft rot bacteria including Pectobacterium carotovorum and Pectobacterium atrosepticum, together with fungal pathogens including Fusarium caeruleum. These infections were much simpler to produce and involved introduction of the pathogen into a wound in the tuber, sealing and incubation in humid conditions for several days at ambient temperature, after which rotting ensued. Confirmation of tuber infection was made from heel end cores or visual symptomatic infections. Rsol and Cms were confirmed by a range of methods including isolation onto semi selective, semi diagnostic media eg SMSA for Rsol, by PCR based molecular and Immunofluorescence, all in accordance with EU Directives for control of potato ring rot and brown rot. On site diagnosis, eg at the University of Manchester, was made using a CSL produced Lateral Flow device for Rsol marketed under the trade name Pocket Diagnostics. Occasional samples of ring rot and brown rot infected tubers were obtained following import interceptions by PHSI (brown rot) or from UK outbreaks (ring rot). In addition, an EU wide request for samples of ring rot infected tubers produced samples of ring rot infected tubers from Dr J.Tegel, of the Finnish Plant Protection Service, which is gratefully acknowledged. A culture of the insect pest, Thrips palmi was provided by entomologists at CSL. This produced no volatile signals and as a result, it was agreed that no further testing of insects would be carried out within the project. 3.0 KEY VOLATILE IDENTIFICATION Work carried out at CSL and the University of Manchester, utilised SPME using Supelco 75µm Carboxen/Polydimethylsiloxane fibres to monitor production of volatiles from potatoes infected with Cms and RS. Analyses of headspace volatiles from infected potatoes were performed utilising Headspace Solid Phase Microextraction Gas Chromatography Mass Spectrometry (HS-SPME-GCMS). For ring rot infected tubers a specific marker of disease; 3-methyl-2-pentanone was identified as unique, while for brown rot 1-undecene was tentatively identified, but concentrations relative to the overall composition of the headspace were very low and the relative intensities of compounds varied considerably. For brown rot infections it was found that secondary alcohols and ketones increase markedly in concentration. The total-ion chromatograms for 9 infected and corresponding healthy control samples were analysed using Principal Component Analysis (PCA) (Figure 1), which demonstrated discrimination of each infected cultivar from the controls which is not cultivar dependent.

Pc1/Pc2 ( all chemicals)

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

-0.8 -0.6 -0.4 -0.2 0 0.2 0.4

% R.S Infected Tubers Uninfected pooled tubers

Uninfected Tubers

Desiree 37.5% +ve

P.Dell 82.6% +ve

Marfona 51.9% +veMaris Peer 30.3% +veWilja 76.2% +ve

Maris Piper 53.3% +ve

Rocket 53.5% +ve

Estima 78.3% +ve

Cara 43.5% +ve

Figure 1. PCA chart of PC1 versus PC2 of pooled (between 5-7 tubers where the % corresponds to the

actual number of infected tubers) RS infected tubers and their healthy controls. Due to the complexity of detecting unique markers in the field, a strategy of profiling the entire headspace emitted from potatoes was investigated. 4.0 ELECTRONIC NOSE INSTRUMENTATION

PC1

PC2

SID 5 (Rev. 3/06) of 23

An instrument (Figure 2) was developed comprising an automated SPME sampler for preconcentration of volatile chemicals from potatoes, together with an 8 metal oxide sensor array for detection of volatiles desorbed from the SPME fibre.

Figure 2. Demonstration electronic nose

After sampling the SPME fibre is placed into an automated loader in the Enose system that is controlled using an interfaced PC and specifically designed software. The headspace is isolated for passive thermal desorption of the volatile on fibre directly onto the sensor array see figure 2. The automated system is controlled by integrated electronics, after the sample volatiles have been desorbed from the fibre clean/dry air is drawn into the sensor chamber to clean the fibre. Followed by the retraction of the coated section of the fibre and further flushing of the headspace volume with clean/dry air. The fibre is then fully retracted to the home position and can be immediately reused. The sensors remain at a constant controlled temperature in clean/dry air ready for the insertion of another fibre. 5.0 ELECTRONIC NOSE INVESTIGATIONS The key objectives for electronic nose investigations were as follows;

• To determine the effect of humidity, age, damage and storage temperature/material on the measurements of healthy potato tubers

• To determine if healthy and RS and Cms disease state potato tubers can be discriminated using the electronic nose system

• To compare measurements from soft rot and dry rot diseased potato tubers Sampling was performed by exposing the adapted SPME fibres to the headspace of pooled tubers (between 5-7) (Figure 3) and followed by direct exposure of the collected volatiles onto the sensor array using the automated system.

Automated Sampler

Sensor array

Electronics

SID 5 (Rev. 3/06) of 23

Figure 3. Sampling procedure

Tested samples included; Two cultivar healthy control baseline study RS infected potato tubers of different cultivars/levels of infection and corresponding healthy controls Cms infected potato tubers of different cultivars/levels of infection and corresponding healthy controls Dry rot and soft rot infected potato tubers and corresponding healthy controls The instrument was linked to data processing software and a radial basis neural network for real time identification of incoming data. The aim of the control experimentation was to determine whether the shifts of the healthy profiles with time makes the data non-discriminating to RS infected tubers. Where experimentation investigated both cv. Jersey Royals and cv. Pentland Crown the collected data is presented together. RS data for cv. Maris Peer recorded during the time of testing is also presented, comparisons to this data will indicate the level of discrimination. The presented data shows that changes in potato tuber age, damage, storage temperature, humidity and material do affect the volatile profiles when compared to the benchmark (young undamaged tubers stored at 4oC in paper at low humidity). Figure 4 shows all the data from both cv. Jersey Royal and Pentland Crown when exposed to changes in these conditions. The Sammon map shows the actual distances between each sample, it is evident there is a spread of the control data in Euclidian space although there are a few samples that overlap with the RS infected samples. This statistical method is low-dimensional and the neural network would be able to discriminate between the data sets in high dimensional space.

Figure 4. Sammon map of healthy cv. Jersey Royals and cv. Pentland crown potato tubers tested under

various conditions and time intervals and brown rot infected cv. Maris Peer.

Adapted fibre holder 500ml Glass jar SPME fibre exposed Potato sample

SID 5 (Rev. 3/06) of 23

Various samples of RS infected potato were supplied by CSL. The tubers which were produced from infected plants exhibited various symptoms at the time of testing. The pooled samples were tested using the electronic nose system and then tested for presence of RS using TaqMan PCR or lateral flow devices. The PCA plot (figure 5) shows data containing 3 control batches and 3 RS infected batches exhibiting small, medium and rotten symptoms at the time of testing. The data shows discrimination of the RS infected samples regardless of the level of symptoms from the healthy controls, which is independent of cultivar.

Figure 5. PCA plot of RS infected cv. Desiree, Estima, and Marfona (Small, Medium and Rotten symptoms) and their healthy controls.

The PCA plot shows differences in the level of symptoms of the RS infected potato tubers cannot be discriminated. This is however, an advantage as the potato tubers exhibiting small symptoms can still be discriminated from the healthy controls. The plot also shows drift of the samples with time, which indicates the samples change with time. One batch supplied by CSL contained cv. Rocket samples containing asymptomatic tubers (from infected plants) at the time of testing. These potato tubers are presented as Rocket NES in figure 6.

Figure 6. PCA plot of RS infected cv. Rocket, Rocket no external symptoms and their healthy controls.

The highlight of the data is the ability of the Enose system to discriminate the NES samples from healthy controls and infected samples. This suggests that the Enose can differentiate between healthy controls and controls which were harvested from infected plants. Further tests of NES samples are needed to prove whether this is true for different cultivars. However, these findings could prove to be very useful in the screening of potato tubers to determine the origin of infected plants.

Sample D ift

SID 5 (Rev. 3/06) of 23

For the in-field application the measurement system has to be capable of detecting a single infected tuber within a large lot. It is difficult to simulate the field conditions within the laboratory, this combined with the limited provisions of infected material and corresponding healthy controls meant only a few samples could be investigated. This investigation involved placing a small single RS infected potato tubers into a sampling vessel containing 25 healthy potato tubers. The headspace was sampled and compared to the profiles for the healthy controls. Three replicates were recorded for two different RS infected tubers. The PCA plot (figure 7) shows that samples containing the RS infected tuber can be discriminated from the healthy controls. Due to contamination of the samples when the RS infected tuber is added to the healthy controls only a limited number of samples could be recorded. Further work is needed to determine the LOD for the measurement system.

Figure 7. Sammon map of one single RS infected cv. Maris Peer among 25 healthy tubers and the

25 healthy controls. Cms infected samples tubers included; 135 potato tubers from a Cms infected consignment from Finland of an unconfirmed cultivar. The 135 tubers were split in smaller batches consisting of 5 tubers sampled as a batch or sub-batch (A/B). Samples were tested prior to confirmation of Cms infection. Due to the deterioration of the samples with time only confirmation of presence or absence of infection could be confirmed for certain batches. It must be noted the majority of the samples which tested positive for the presence of Cms were also presenting secondary infection due to soft rots. The PCA plot (figure 8) shows data containing 50 batches of suspected Cms infected potato tubers. The confirmed Cms infected data for RS infected tubers shows the samples drift with time. Although the samples were presenting Cms at the time of testing it cannot be confirmed whether the volatiles measured from the Cms infected tuber were masked by the presence of soft rots.

SID 5 (Rev. 3/06) of 23

Figure 8. PCA plot of the batches infected with Cms and the confirmed healthy batches and unconfirmed batches.

The Sammon map (figure 9) shows data containing cv. Maris Peer RS infected, soft and dry rot infected potato tubers. The data illustrates discrimination of the healthy controls and the infected samples irrespective of infection.

Figure 9. Sammon map of RS, Fusarium caeruleum and Pectobacterium atrosepticum infected cv. Maris

Peer and their corresponding healthy controls. Figure 9 shows the dry rot infected samples can be discriminated from the brown rot and soft rot samples. This is most likely because the dry rot disease is cause by the fungi (Fusarium caeruleum). Both RS and Pectobacterium atrosepticum are bacterial diseases and it is difficult to determine whether these samples can be separated. The electronic nose system was demonstrated to be capable of identifying infected material in a controlled laboratory environment. Tests demonstrate that the system can discriminate each disease from the controls and is not cultivar dependent. Alternative pre-concentration techniques were investigated as a method to optimise the technique. 6.0 DISCUSSION AND CONCLUSIONS The volatile investigations of pooled potato tubers tentatively identified unique markers. However, their detection among the entire profile of volatiles using sensor technology would be very difficult. The most important finding was marked increases in 6 key compounds; 2-propanone, 2-propanol, 2-butanone, 2-butanol, 2-pentanone and 2-pentanol from potato tubers infected with RS when compared to healthy controls. Statistical analysis demonstrated the discrimination of RS infected tubers from healthy controls to be independent of cultivar. When compared to other publications this work describes the VOC profiles of brown/ring rot on intact potato tubers, which were harvested from infected plants. The second aim was to fabricate instrumentation capable of discriminating infected tubers from healthy uninfected ones. During the project two Enose instruments were produced for this purpose. The Enose systems’ were tested using a variety of infected material. Laboratory based testing demonstrated that the measurement systems could confidently discriminate between pooled RS infected potato tubers and their healthy controls irrespective of cultivar. Limited supply of Cms samples makes it difficult to state whether the systems are capable of discriminating between Cms infected potato tubers and their healthy controls. The initial objective of this research was to develop techniques for measurement of key potato volatiles of control and disease state tubers. Due to the low concentrations of volatiles emitted from both infected and healthy tubers a pre-concentration step was required. For this purpose the use of commercially available SPME fibres for headspace analysis was adopted. HS-SPME-GC-MS was used to identify the bacterial metabolites of RS and Cms. The Supelco™ 75µm CAR/PDMS fibres were successful in pre-concentrating the volatiles from infected and uninfected potato tubers. This method was adopted and coupled to the measurement system.

SID 5 (Rev. 3/06) of 23

The next objective was to develop a non-invasive measurement system which can be applied to the problem. Both instruments incorporate automated SPME samplers for direct desorption of volatiles onto the sensor array with minimal sample loss. The Enose II measurement system incorporates an interchangeable guide for use with SPME fibres and a new HSSE device. After sampling the adapted fibres can be inserted into the instruments. The instruments have robust designs and are portable for in-field analysis. Interfacing of the measurement systems to custom designed software enables fully automated operation via RS232 serial port to a PC/laptop. The software performs; data acquisition, data processing, data analysis and finally online recognition of incoming data. The last objective was to develop suitable odour classifications and discrimination based on the measured data coupled with identification and recognition of complex odours using artificial neural networks. A range of samples were tested using the systems. The RBF neural network was trained using various datasets and tested using validation samples. For the majority of the RS and corresponding healthy controls samples the RBF was successful at recognising the validation samples. The influence of secondary infections decreased the confidence of the RBF to discriminate RS infected samples from Pectobacterium atrosepticum infected samples. In summary, we have developed a prototype Enose potentially capable of detecting lots of potatoes infected with potato brown rot. The Enose is easy to use, is fully portable, requiring only a laptop and is primed ready to be populated with a database of brown rot infected tubers. Since it is difficult to obtain potato tubers carrying pure Rsol infections in the total absence of other pathogens eg the soft rot Pectobacterium species, there is the possibility that our Enose has been trained to detect mixed infections and this may influence the possibility of false positive results with other rots. In addition we have no data on the sensitivity of the detection although our data indicate that detection of symptomatic tubers is likely to be as sensitive as current methods for detection of single symptomatic tubers in the usual 200 tuber samples. We have no data to indicate whether this holds for low numbers of latently infected tubers. Before the Enose can be used by PHSI further work would be necessary to fill these gaps, develop a working database of volatile profiles and validate the Enose against standard EU Directive detection methods. Development of an Enose for ring rot is likely to be a greater challenge due to the problems of producing an adequate supply of infected tubers as well as the apparent paucity of volatile signals from ring rot infected tubers. Volatile detection of insect pests such as Thrips palmi would appear to represent an even greater challenge 7.0 MAIN POLICY IMPLICATIONS Both potato ring rot and brown rot are significant quarantine pathogens in most countries. Current detection methods all rely on taking a sample and testing that sample for the presence of the pathogen. These methods are laid down in the EU Control Directives for these bacterial pathogens. All rely on transport of the sample back to the laboratory. The sample is usually 200 and such sampling gives an 87% chance that a single infected tuber will be present in the sample, assuming a 1% level of infection. In reality, infection levels are much lower than this but the problems of increasing the sample to several thousand tubers and transporting them back to the laboratory become significant and costly. In all these tests a core of tissue is removed from each tuber, which is labour intensive, whether it is done at the site of inspection by PHSI or back at the laboratory. The value of a method which does not rely on such sampling is thus obvious. The SPME fibre could theoretically be left in contact with the tubers for several days, either in stores or in pallets during import or unloading at dockside. The fibre is then simply placed in the Enose and a result is obtained in 2 minutes, with a report as to the likelihood of infection expressed as a percentage. It is feasible that several diseases could be included in the same Enose. Many samples can be processed in a day. Not only is sampling efficiency increased since the volatile signal will represent the whole consignment rather than a sample, but the result is obtained at the point of inspection. This approach has been successful with lateral flow devices and portable real time PCR but the Enose has the advantage of extremely little preparatory work prior to detection. Until fully validated, the Enose would screen only negative samples and lots with positive results would require confirmation back in the laboratory. Additional Enoses trained to detect brown rot would cost less than £5000 each. Assuming that the sensor array for brown rot is appropriate for other diseases the Enose could be adapted and trained to detect other diseases. The costs for this would be based on approximately 3 staff months per disease but this would not include the validation. There are other statutory diseases for which Enose detection may have value. Technology transfer is underway and CSL staff have been trained by University of Manchester in the use of the Enose. A training and operating manual has been produced by University of Manchester and resides with CSL.

SID 5 (Rev. 3/06) of 23

8.0 POSSIBLE FUTURE WORK Further work is required prior to in-field online recognition of RS and Cms diseased potato tubers using the demonstration electronic nose system. The sensitivity of the instrument needs further work to determine if the system is capable of detecting a single infected potato tuber in a large lot and indeed a single latently infected potato in the lot. So far the system has been operated in laboratory conditions. When sampling in the field there are various factors to consider. Field sampling conditions need to be addressed to assess the impact of dockside shipping condition and influence of background on the sampling method and instrument performance. For maximum efficiency when sampling large imported consignments a suitable sampling methodology needs to be adopted. 25 tonne lot consignments are split into 1 tonne sub-lots, as such, a protocol needs to be devised whereby the 1 tonne lots can be probed and individually sampled. Another approach could involve strategically placing adapted SPME and HSSE devices within the hold of the ship. This would sample the cargo for the entire time it takes to reach the point of entry. This would involve strict monitoring to control sampling time, minimise sample losses and prevent tampering with the devices. B. DEVELOPMENT OF AN ACOUSTIC DETECTION SYSTEM FOR INVERTEBRATE PESTS 1.0 OBJECTIVES The original objective of the project was to investigate the feasibility of automatic detection of quarantine insects species using a combination of acoustics, time domain signal coding and artificial neural networks. This objective has been successfully attained with reliable detection and in some cases identification of species particularly within hard plant materials. Due to problems with obtaining insect interceptions the acoustic detection system has not been tested at field sites. The only occasion within the project period when it may have been possible to test the system, the shipment was turned around at dockside without docking. There were also problems in obtaining target insect pests such as the Asian Longhorn beetle during the project which led to a search for native species; this is discussed in section 2 of the report. 2.0 METHODS USED AND RESULTS OBTAINED Sensing of insect movement or feeding activity from within woody plants has previously been carried out for a variety of insect taxa including adult beetles, beetle larvae, moth larvae and termites using a variety of sensors, concentrating mainly on accelerometers. However, these are very expensive and one of the objectives of this project was to investigate the application of low cost sensors including electret microphones, piezoelectric sensors and bimorph elements. All previous research has been for idetection only of insects and not identification; the major aim of this work is to provide both detection and identification. In the early stages of the project the Central Science Laboratory was having difficulties in obtaining cultures of target quarantine insects. It was therefore decided to concentrate on collecting locally available native species that fed on similar plant materials to the target species. Table 1 gives a complete list of the species investigated during the project and their source. It was also decided at an early stage during the project that Thrips palmi was considered to be physically too small to produce acoustically significant signals. 2.1 Recording Methods and Sensor Selection A variety of recording methods have been employed including the following: a) PC soundcard. It is possible to directly record from sensors to a soundcard using the microphone input or

the line input if signals have been amplified. It is important to note that laptop computers in particular are electrically very noisy and can produce high levels of interference in an audio recording. It is therefore recommended that an external soundcard is employed.

b) Minidisc (MD). Minidisc recorders such as the Sony MZ series (e.g. MZ-R909) provide a good, low cost

recording method. It has been suggested that the compression methods used in the minidisc can cause significant signal degradation for non-human or non-musical signals. We have not found this to be the case.

SID 5 (Rev. 3/06) of 23

c) Digital Audio Tape (DAT). DAT is an industry standard direct PCM recording method which preserves the audio signal with high fidelity.

For this project the sampling rate was 44.1kHz at 16 bits/sample (standard audio CD quality).

Table 1. Species Used in the Project

Insect Species Plant Species Source Anoplophora chinensis (Coleoptera: Cerambycidae)

Ficus sp Ash (Fraxinus sp) Twigs

CSL

Prionus coriarius (Coleoptera: Cerambycidae)

Oak (Quercus sp) Royal Holloway

Rhagium bifasciatum (Coleoptera: Cerambycidae)

Pine (Pinus sp) Oak (Quercus sp)

Local source collected by PI & researcher

Trichoferus griseus (Coleoptera: Cerambycidae)

Fig (Ficus sp) Brach (≈45vm)

CSL

Hylotrupes bajulus (Coleoptera: Cerambycidae)

Pine (Natural) Pine (Kiln Dried) Oak (Quercus sp) Ash (Fraxinus sp) Twigs

CSL, recordings from Australia

Lucanus cervus (Coleoptera: Lucanidae)

Oak (Quercus sp) Ash (Fraxinus sp) Sycamore (Acer pseudoplantanus)

Royal Holloway

Dorcus parallelopipedus (Coleoptera: Lucanidae)

Oak (Quercus sp) Pine (Pinus sp)

Royal Holloway

Hylobius abietis (Coleoptera: Curculionidae)

Pine (Pinus sp) Bark Wafers Forest Research

Leptinotarsa decemlineata (Coleoptera: Chrysomelidae)

Potato Leaf CSL

Agrilus planipennis (Coleoptera: Buprestidae)

Fraxinus sp CSL

Anobium punctatum (Coleoptera: Anobiidae)

Meal Cookies CSL

Acherontia atropos (Lepidoptera: Sphingidae)

Privet Larva provided by PI

Spodoptera exigua (Lepidoptera: Noctuidae)

Chinese Leaf CSL

Spodoptera littoralis (Lepidoptera: Noctuidae)

Tomato Plant Cotton Plant Chinese Leaf

CSL

It is important to note that MP3 recording is not suitable as it severely affects the reproduced signal spectrum. The recording equipment described above has two input channels (stereo) and this has been utilised for simultaneous recording from two sensors, either two different sensors or two identical sensors. In the latter case, the second sensor has been placed away from the source to provide a reference for background noise levels. One of the objectives (B02/02) of this project was to determine the most appropriate sensors for hardwood and softwood materials. A variety of sensors were tested as indicated in Table 2 and their suitability for various materials are shown in Table 3. A comparison between the three types of sensor investigated indicates that: a) Bimorphs, whilst very sensitive, are fragile, difficult to attach and are therefore not practical. b) Piezoelectric sensors are also very sensitive but narrowband and highly resonant, and are thus ideal for

detection; identification is possible to a certain degree. c) Bimorphs and piezoelectric sensors are contact devices and ideal for non-invasive substrate-based

vibrational signals. d) Electret microphones are broadband but suffer from a lack of coupling to the substrate with airborne signals

only being detected.

SID 5 (Rev. 3/06) of 23

e) The accelerometer is physically large which makes attachment difficult and is very expensive. Newer, physically small and low cost integrated accelerometers are now available which should be considered as sensors in a future project. One potential disadvantage of accelerometers is that they can have low bandwidths.

In general, problems of attachment to soft plant materials and low signal levels effectively excludes applications involving such material. Figure 10 shows a typical set-up with an enclosed, waterproof piezoelectric sensor and amplifier.

Table 2. Sensors Evaluated during the Project

Sensor Type Comments Brüel & Kjaer 2513 portable vibration meter

accelerometer Very expensive and physically large sensing element. Also has a low frequency response

Microphone Electret (various sizes) Low cost, simple to use. Good for airborne signals but not for substrate-based vibration. Requires amplification. Wide frequency response.

Bimorph linear piezoelectric element

Low cost, very sensitive but fragile and easily damaged. Narrow frequency response.

Piezoelectric transducer planar piezoelectric elements of various sizes (circular and square)

Low cost, very sensitive and robust. Can be encapsulated to improve robustness. Highly resonant (around 2kHz).

Table 3. Suitability of Sensors for Different Plant Materials

Sensor Attachment Suitability for Soft Plant Materials

Suitability for Hard Plant Materials

Accelerometer Magnetic via nail or screw Not suitable – difficulty of attachment

Suitable if nail/screw can be attached.

Microphone No physical attachment Cannot be directly attached but can pick up airborne signals if high enough intensity

Not generally suitable unless inserted into a hole and sealed.

Bimorph Must be placed in contact with the surface

Only suitable if a good contact can be made

Only suitable a good contact can be made

Piezoelectric transducer

Must be placed in contact with the surface

Only suitable if a good contact can be made

Suitable for all hardwoods

Figure 10. Prototype detection system with enclosed piezoelectric transducers and amplifier

2.2 Signal Processing and Signal Identification A number of signal processing techniques were developed during the project to enable the detection and identification of individual bites. These included optimal event detection (an event being defined as an insect bite

SID 5 (Rev. 3/06) of 23

or short event such as interference), a new enhancement to time domain signal coding to eliminate the problems associated with codebook generation and a software package written in Matlab to perform event detection and identification. An example screen shot of the analysis screen of the WASP (Waveform Analyser and Sound Profiler) package is shown in Figure 2.

Figure 11. Screen Shot of the Waveform Analysis Page of WASP 2.2.1 Event Detection An insect bite is characterised by a short duration impulse-like signal as indicated in Figure 3, caused by the insect’s jaws breaking wood fibres. Reliable detection of bites and other sounds including stridulation and interference is very important and is carried out using two types of thresholding – vertical and horizontal. a) Vertical threshold A vertical threshold is basically a level check based on excursions above the mean value within a signal segment. This threshold can be user-based or automatic, being a percentage of the peak signal within the segment. Vertical thresholding returns a time series corresponding to samples in the signal which exceed the threshold level.

SID 5 (Rev. 3/06) of 23

Figure 12. Example bite waveforms for Hylotrupes bajulus. Lower trace is a single bite in more detail. Horizontal axis is time in seconds. Signal amplitude is normalised.

b) Horizontal threshold Horizontal thresholding is applied to the time series generated from the vertical threshold. It is essentially a time duration check performed in two stages. Firstly, samples that occur within a specified time of each other are grouped together to form an acoustic event; this is termed the Event Separation Time (EST). Secondly, grouped events are checked to ensure they exceed a specified Minimum Event Time (MET); events which are below the MET are then discarded. Both the EST and MET can again be user-based or automated. 2.2.2 Feature Extraction using Time Domain Signal Coding (TDSC) Once an event has been located and isolated, the next step is to extract appropriate features of the signal for input to a classifier. Feature extraction is the most important part of any signal classification system and here TDSC is employed since it has been shown to provide a good feature set with little computational complexity. TDSC is based on time encoded speech (TES) originally developed by King and Gosling in 1978 as a method for the transmission of speech at very low data rates. It operated by encoding the shape of a waveform (as the number of minima) between successive zero-crossings (termed an epoch) as indicated in Figure 4. TDSC extends TES by incorporating amplitude scaling to overcome the loss of amplitude information in the encoding process. Each epoch is described by the duration (D) in number of samples and the shape (S) as the number of positive minima or negative maxima. The number of possible combinations of (D, S) can be very large and TDSC has in the past employed a non-linear mapping to create a single codeword for each epoch. The histogram of the frequency of occurrence of code words over a given time interval is termed the S-matrix and a second matrix, the A-matrix, describes the number of occurrences of codeword i followed by codeword j as given by:

( ) ( )∑=−

=N

2nijij nx

1N1a (1)

where aij = element (i,j) of matrix A xij(n) = 1 if t(n) = i and t(n-L) = j (0 otherwise) L = lag (L=1 in this case) and t(n) = nth codeword

SID 5 (Rev. 3/06) of 23

The A-matrix is a fixed size histogram (currently 28x28 code words) with time-invariant dimensions representing the conditional probability of finding pairs of code words and has been used extensively for a variety of acoustic applications ranging from fault detection in gearboxes to heart defect classification and insect identification.

Figure 13. Principle of Time Domain Signal Coding

The codebook is application specific and, although a general codebook is often sufficient, a new codebook has to be produced by manually examining all D-S combinations for a given set of signals. The approach used here removed the requirement for a codebook by mapping D-S combinations onto a 1-dimensional matrix, termed the D-matrix. The mapping employed here is: ( )( )DSSC F +×= (2) where S = shape D = duration SF = scaling factor The purpose of the mapping is to separate duration based on shape. The scaling factor is determined by the maximum duration in the signal; SF must be sufficiently large to avoid any overlaps. In the current system, the sampling rate for all signal is 44.1kHz and the signals are band-limited to 500Hz-8kHz, thus limiting the duration to between 125�s (5 samples) and 2ms (88 samples). SF must therefore be greater than 88; a value of 100 has been selected. Figure 5 shows D-matrices for two different species and sound types. The D-matrix forms a feature vector for input to the classification stage which takes the form of an artificial neural network (ANN). Two forms of ANN have been tested – a multilayer perceptron (MLP) and learning vector quantisation (LVQ) network. Each network is trained with a representative set of D-matrices from known species and then tested with unseen D-matrices from the species used in the training sets.

Positive Minimum

Negative Maximum

Duration

Epoch (n) Epoch (n +1)

Time

Amplitude

10 20 30 40

SID 5 (Rev. 3/06) of 23

a) Distributed Code Matrix (D-matrix) for Hylobius abietis Bite Event

b) Distributed Code Matrix (D-matrix) for Lucanus cervus Stridulation Event

Figure 14. Example D-matrices for two different species

2.3 Results for Larval Detection using MLP & LVQ Networks The results given here are a summary of many tests carried out; please refer to Ian Farr's thesis for a complete set of results. Tables 4 to 7 show results from testing the trained MLP and LVQ networks with samples of H. bajulus bites and noise. These tables indicate the success of each network combination and sensor type at detecting the presence of insect bites from background noise. Each network was trained with the 20% “best” (highest SNR) files and tested with the remaining 80%. The best performing device in each test is indicated by bold type and the worst performing device is highlighted in red (bold border).

Table 4. MLP Network Tested With Bite Sounds from H. bajulus Recording

Sensor Bite as Noise Bite as Bite No of Files Microphone 9.09% 90.01% 66

Bimorph 29.94% 70.06% 157 Piezo Transducer 11.04% 88.96% 154

Table 5. MLP Network Tested with Noise Sounds from H. bajulus Recording

Sensor Noise as Noise Noise as Bite No of Files Microphone 40.91% 59.09% 66


SID 5 (Rev. 3/06) of 23

Table 6. LVQ Network Tested with Bite Sounds from H. bajulus Recording

Sensor Bite as Noise Bite as Bite No of Files Microphone 3.03% 96.96% 66


Table 7. LVQ Network Tested with Noise Sounds from H. bajulus Recording

Sensor Noise as Noise Noise as Bite No of Files Microphone 42.42% 57.58% 66


Tables 4 and 6 indicate that correct classification of a bite sound was achieved with high degrees of confidence for all sensor and network combinations, with electret microphones providing the best performance closely followed by the piezoelectric transducers. Classification of noise sounds shown in Tables 5 and 7 were not as good in comparison; only the bimorph/LVQ network combination provided a high level of confidence for correct identification of noise sounds. The overall indication would suggest that the best combination for detection of wood-boring insects would be to use data from recordings made using piezoelectric transducers and to test this data with the LVQ network. Whilst the results presented show the electret microphones have slightly higher confidence levels for classifying bites as bites they have a lower rate of detecting bites within a recording. This is shown by the number of files in the tables; generally the bimorph and piezoelectric transducers will detect around 60% more events from a recording. 2.4 Species Classification Results using MLP & LVQ Networks Tables 8 to 10 show classification results for bites of H. bajulus and P. coriarius. Recordings were made using piezoelectric transducers, H. bajulus was inside a bundle of ash twigs and P. coriarius was inside a small oak log. Again, these results are only representative examples, and the reader should refer to Ian Farr's thesis for more results.

Table 8. MLP Network – 15 Hidden Nodes

H. bajulus P. coriarius H. bajulus 64.94% 35.06% P. coriarius 69.70% 30.30%

Table 9. MLP Network – 7 Hidden Nodes


Table 10. LVQ Network


Species classification results from the MLP and LVQ networks again indicate that LVQ network outperforms the MLP network. Results shown in Table 8 and 9 for the MLP network show an inability to correctly identify between the species. Adjusting the number of nodes in the hidden layer shifts the emphasis to the other species. This may indicate that the optimum number of nodes could lie between 7 and 15. Table 10 shows very good classification of the species and supports the findings in the detection tests that piezoelectric transducer and LVQ networks provide the best chance of detection and classification.

SID 5 (Rev. 3/06) of 23

3.0 DISCUSSION This project was a feasibility study into the automated detection and identification of insect species within various plant materials. The objectives of the project included investigation into low-cost sensors, optimal signal analysis techniques for the extraction of bites and an investigation into whether it is feasible to determine species on the basis of individual bites. The project has been a success in all areas. Difficulties in obtaining quarantine species led to a search for native species that could be used as analogues; this led to collaboration with Royal Holloway University of London and the Department of Agriculture in Australia. Problems encountered with acoustic transmission in soft plant materials, notably the requirement for airborne sensors such as microphones, led to the project concentrating on hard plant materials. The detection of wood-boring insect larvae in hard plant materials has been very successful with piezoelectric sensors being the most suitable for this purpose. The identification of species based on individual bites has proven to be successful, however, more work needs to be done in this area to expand the number of species the system is capable of identifying. The identification part of this work has also shown that it is possible to discriminate between insects and other sounds such as the ubiquitous mobile phone interference. Many hours of recordings were made and the time taken to analyse this manually to locate bites and other sounds was extremely time-consuming. A decision was made to develop a software package that was capable of automatically locating sounds. In order to reduce the search time substantially. In a “real system", the signals would not be recorded, but analysed in real-time thus removing the need for recording equipment. 4.0 MAIN POLICY IMPLICATIONS This project has shown that it is possible to detect and identify wood boring insect larvae using low-cost sensors and in the presence of interfering signals. The major implication of this is that it is entirely possible for a real-time fields deployable system to be produced, thus providing PHSI inspectors with a new tool for locating quarantine species. It should also be noted that this work is not limited to quarantine species but can also be used in forestry applications and ecological studies, for example, the non-invasive detection of stag beetle larvae in tree stumps. The results obtained so far appear to have potential for the detection of insects in many situations and particularly in small confined spaces such as for young trees imported in containers. Further work is necessary before the method could be used by PHSI. 5.0 FUTURE WORK Future work is as follows: a) Further investigation into optimal and robust detection algorithms. Also, further work is required on species

identification using different types of neural networks together with temporal processing, e.g. determining whether inter-bite interval and feeding rtes can be used.

b) Investigation into the use of 2- and 3-axis miniature accelerometers for substrate-based vibration detection.

If successful, they will exhibit both characteristics necessary for detection and identification – high sensitivity and less resonant than piezoelectric sensors.

c) Implementation of the algorithms for detection and identification on real-time platforms. Two types of

platform can be developed (i) a hand-held device using a PDA or hand-held computer and (ii) a datalogging device capable of being left unattended for extended periods of time. Both systems will optimally detect the presence of insects and, if required, determine the species.

SID 5 (Rev. 3/06) of 23

References to published material

9. This section should be used to record links (hypertext links where possible) or references to other published material generated by, or relating to this project.VOLATILES Peer reviewed publications and thesis • Stinson J. A., Persaud K.C. and Bryning G.2006 “Generic system for the detection of statutory potato

pathogens”. Sensors and Actuators B. 100-106. • Stinson JA, Persaud KC, Bryning G, and Stead D. 2006. Detection of potato pathogens using an

SPME-Enose. Chemical Senses 31: E36. • Stinson, J.A 2007 Sensing potato pathogens:an automated approach. PhD thesis University of

Manchester Conference Presentations And Proceedings • Jade Stinson was awarded the Duncan Davies Prize for 2003/4. This is awarded by the University of

Manchester to the student who has shown the most outstanding contribution in terms of bringing together the instrumentation and analytical science element of the work of the Department of Instrumentation and Analytical Science.

• CSL Student Poster day presentation, April 2005. Won first prize in the poster competition. • Britain’s Young Engineers at the House of Commons - Dec 2005. Poster presentation. • British Society of Plant Pathologist Annual Conference – Dec 2005 Oral and poster presentation. • Defra Plant Health Seed Inspectorate Technical Conference – Jan 2006 Poster presentation. • CSL Student Poster day presentation, April 2006. Won first prize in the poster competition for final year

students. • Stinson, JA. Persaud, KC. Stead, DE, Bryning, G And Parkinson, N. (2006) Using an SPME Enose to

identify quarantine pathogens of potatoes. In Elphinstone, J.G., Weller, S.A., Thwaites, R., Parkinson, N., Stead, D.E. and Saddler, G. (2006) Proceedings of the 11th International Conference on Plant Pathogenic Bacteria. Pp-29-30.

• Defra project review meeting, University of Manchester, October 2006. • Defra Plant Health Seed Inspectorate Technical Conference – Jan 2007. • Poster presentation and demonstration of Enose. • Defra project review meeting, CSL, Feb 2007. ACOUSTICS Thesis • Farr, I.J., Automated Bioacoustic Identification of Statutory Quarantined Insect Pests, PhD Thesis,

University of York, 2007. Conference Presentations And Proceedings • Farr, I.J., Chesmore, E.D. & Morgan, D., Pursuing pests – acoustic recognition of quarantine insects,

Entomology 2004, 21-23 July 2004, University of York. • Farr, I., Chesmore, D., Harvey, D., Hawes, C & Gange, A. Bioacoustic detection and recognition of

Stag Beetle (Lucanus cervus) larvae underground using vibration sensors. Entomology 2005, 12-14 September 2005, University of Sussex.

• Farr, I.J. & Chesmore, E.D. Acoustic detection and recognition of wood-boring insects, Entomology 2005, 12-14 September 2005, University of Sussex.

• Farr, I.J. & Chesmore, E.D., Automated bio-acoustic recognition of statutory qurantined insect pests: Defra funded project PH0191. In: PHSI Technical Refresher Course, course notebook 2006, Central Science Laboratory, pp 81-86.

• Farr, I & Chesmore, D. Automated bioacoustic detection and identification of wood-boring insects for quarantine screening and insect ecology. 4th International Conference on Bioacoustics 2007, Institute of Acoustics, Loughborough, 29(3), 10-12 April, 2007, University of Loughborough.

Other presentations: • Royal Entomological Society Postgraduate Forum, Northern Region Meeting (Leeds University) 2003

Presentation, inc. question and answer session. • Royal Entomological Society Technology Special interest group meeting 2003 Presentation, inc.

question and answer session. • Research Seminar (York University) Feb 2005. • Defra project review meeting, University of Manchester, October 2006. • Defra project review meeting, CSL, Feb 2007.

SID 5 Research Project Final Reportrandd.defra.gov.uk/Document.aspx?Document=PH0191_6314_FRP.pdf ·...

Documents

Transcript of SID 5 Research Project Final Reportrandd.defra.gov.uk/Document.aspx?Document=PH0191_6314_FRP.pdf ·...