A Study Of Forest Insect Pest Classify Method Based On Image Pattern Recognition

In this paper, dung beetle, which is common in inspecting and reporting forest insect pest, was studied with modern information technologies, such as neural network and digital imageprocessing. The technology of image-processing is made use of extracting features of some images. The packages of MATLAB7.1 neutral network are made use of identifying insect pests. In this paper, not only an overall neural network design but also the solution has been provided.The following are main research contents:(1) Treatment of forest insect pest digital image The paper put the emphasis on the theory expounding and methods research of grayscale equilibration, noise removal by filtering, image segmentation and so on, because of the essential and importance of pretreatment. Though comparing the two filtering modes, at last using average filtering to remove noise, and then doing threshold segmentation on the filtered images, transforming the grayscale to binary, then detecting edges by many edge detectors, after comparing, selecting Canny detector to detect the edges of forest insect pest digital image.(2) Feature extraction of forest insect pest digital image The segmented binary images can't be used to design neural network, first, because the dimension is to high, it doesn't fit the input of neural network, second, which is the most important, these direct describing of the binary images can't reflect the essence of insect pest, so, should find some characters, this is what is called feature extraction. Shape is one of the features, in this paper, including perimeter, area, rotundity, elongate etc geometry feature, and seven moments invariant which don't change with displacement, circumrotate, flex.(3) The technique of neural network First, saying the basic content of neural network, including summary, classify, study, training, mathematic model, transfer function and BP neural network.(4) The design of neural network This part simply introduces MATLAB neural network toolbox, put the emphasis on the design of neural network, and neutral network are made use of primary identifying insect pests.