Combination linear feature detector for effective location of insects in grain images

This paper reports work that has been orientated at the design of a cost-effective system capable of inspecting harvested cereal grains for insects. Earlier work showed the potential of the 'linear feature detector' concept, and employed a two-mask vectorial type of operator, subsequently replacing it with a one-mask isotropic operator in order to save computation. In this paper ways of combining the best features of the vectorial and isotropic operators are discussed and a highly effective combination operator is demonstrated. The importance of the combined operator is that it preserves the shapes of insects better than the other two operators: hence subsequent shape analysis has more opportunity for discriminating between insects and artefacts such as small seeds or chaff. The new operator is also more sensitive than the component operators as it allows a greater proportion of the insect to contribute to each point in the output signal. Thus the combined operator is both sensitive and discriminating: it also retains a low computational overhead, in keeping with the demands of real-time inspection.