Biosecurity insect trapping programs in Australia use low-cost or disposable units which are inspected on a weekly or fortnightly basis, and re-lured or replaced monthly or quarterly. Inspectors visit trap sites and collect specimens, remove debris and non-target species from traps. Many inspectors never encounter target species. While that is a good thing, it also means many costly trips to trap sites.

Research outcomes:

 A dataset of over 400GB of images of three pairs of insect species from the Orders Diptera, Coleoptera and Lepidoptera, was acquired using hyperspectral cameras in the near infrared, visible and UV. A consistent protocol was used across species and could easily be applied to other insect pests. State of the art recognition methods used on the dataset revealed that digital shape and pattern can successfully differentiate between the model insect species. Recognition rates of up to 95% were obtained during experiments, confirming the feasibility of developing an automatic detection system to deploy in insect traps. Additional benchmarking and improvements are required before the recognition system can be efficiently deployed in traps. Expertise in trap design, wireless technology and sensors networks will be required in the future to develop a prototype trap with an auto-reporting semi-automated system for field testing.

Research implications:

This scoping study has shown the potential of recognition techniques to identify harmful plant pests. This opens-up the possibility and opportunity to develop an insect trap with auto-reporting capabilities. It also provides a proof of concept of the capability of spectral imaging, pattern recognition and computer vision to aid solve biosecurity problems, in partner with the wider community.

Acknowledgements:

We are grateful to John Lester of CSIRO Entomology for rearing model insects when required and preparing specimens for further processing and to Cong Huyhn of NICTA for acquiring images used in the analyses. We are grateful to our collaborators from DAFWA, I&I NSW (including Ann Mooney and Andrew Jessup) and Vic DPI for sourcing the insects and stimulating discussions. We also wish to thank CRC Program Leader Darryl Hardie and David Everitt, NICTA CRL Lab Director for their continued support.

The main goal of this project was to undertake research that will develop technically sound sample/survey methodologies and systems to enhance the ability to capture a wide range of plant health information in an accurate and cost-effective manner both domestically and internationally.

What is the biosecurity problem?

I aimed to create the necessary tools to significantly reduce the amount of human intervention, as required in present systems. The computational techniques will be required to recognise and identified EPPs in real-time on-board automatic insect traps. The use of imaging technologies based upon hyperspectral and UV ranges to develop a statistical and computational framework for the classification and identification of selected EPPs are challenges.

The main outputs of this project were to:

1. devlop new, better understanding of shape/image descriptors suitable for spectral imagery specially designed for biosecurity surveillance systems
2. create a set of kernel-based, statistical methods that can be used to perform classification of emergency plant pests making use of the descriptors developed in item 1, and
3. develop a framework that will permit the extension of the classification methods developed in item 2 to include domain knowledge. This will involve the development of semi-supervised pattern recognition and interactive computer vision methods based upon statistics that are scalable and capable of real-time analysis.

Who are the end-users of this research?

Automatic and continuous monitoring capabilities of ‘smart traps' have a high potential for commercialisation, both nationally and internationally. Such technology will most likely be adopted by state agencies and plant-based industries involved in early warning networks for emergency plant pests to increase efficiency and reduce cost of monitoring early warning insect traps. Outcomes from this project may also be applicable to a range of biosecurity issues such as semi-automated surveillance systems in quarantine facilities, index databases, building of libraries for future reference, etc.