Rapid automated detection of insects and certain other contaminants in cereals

Abstract

The research reported here provides a basis on which to develop the first automated system for detecting contaminants in post-harvest grain which will meet trade requirements in terms of price, reliability and speed. This has been done by developing analytical methods for /images of wheat samples containing various contaminants.

A suitable system to capture /images has been developed. The most promising methods for analysis of the /images were found to be thresholding and linear feature detection but each needed substantial innovative modification to deliver the required performance. Application of these approaches to /images recorded in the visible region has shown that it will be possible to detect and distinguish contaminants into the following categories: (a) adult and larval insects external to grain kernels, (b) ergot and droppings from rats and mice, (c) rapeseeds, (d) large dark objects, and (e) other contaminants. The system would not report as contaminants permitted admixture such as hulls and damaged grains. Reliability of detection and recognition varied with type of contaminant from complete success with sets of /images containing a total of 60 items of either ergot or rapeseed to 6 missed items out of 150 larval O. surinamensis (saw-toothed grain beetle). To minimise false alarms, it is recommended that each image does not exceed 25 grains. Painstaking improvements to the software have brought a substantial reduction in time needed to analyse each image and the application of faster computers soon to be available should allow the system to handle samples at the required rate of 3 kg in 3 minutes. The cost of the computing hardware, excluding that required for sample handling, is likely to be around £4000.

No existing device has a performance comparable to that expected from the system to be developed from the work presented here and a suggested specification for it has been written. The performance figures it contains should be regarded as starting points for further improvement, not the best ultimately achievable. Attention is now being given to protection of the intellectual property upon which this system depends and the identification of a suitable vendor to bring the system to market.

Internally infested grains cannot be detected by imaging in the visible region. Incorporation of an X-ray camera into the system was investigated but this would increase cost, complexity, and time needed, so its inclusion is not justified. However, such infested kernels were detectable from /images recorded at carefully chosen wavelengths in the near-infra red (NIR) region and the incorporation of an NIR capability into the inspection system should permit the reliable and rapid detection of internal infestation.

HGCA Project Number: 0048/1/94