Application of neural networks in the recognition of stored-grain pests

A study of pest classification in the stored-grain pests automatic detection system based on image recognition was conducted. From the binary pest images, we extracted over ten shape features and finally select five effective features. By combining genetic algorithm with BP algorithm, we trained a multi-layer forward neural network, and overcame the shortcomings of the conventional BP algorithm, such as the slow convergence and its tendency to fall into local minimum. By use of this neural network model, an experiment for recognizing twenty samples of four kinds of stored-grain pests was performed, and the accurate recognition ratio was up to 100 percent. This lays a foundation for the practical use of the system.