A Study Of Stored Food Insect Image Pattern Recognition

Digital image processing and pattern recognition technologies are applied to pattern recognition of stored food insect images, and the automatic recognition of stored food insects is realized in this paper. The following is main research contents and achievements.(1) Pretreatment of stored food insect images. The paper put the emphasis on smoothness and eliminating noise, normalization in position and image segmentation. The result shows middle filter is very efficient. Rotating the image according to the inertial principal axis can normalize it in position. As the background very fair, edge detection and linking method can be used to isolate insect, and a good effect is achieved. And we analysis the effect that the difference color element have on edge detection and image segmentation, which shows blue element is the most effective. Two methods are used to segment the legs and antennae.(2) Feature extraction of stored food insect images. According to the information offered by Expect System, shape, color and texture features of stored food insects are extracted. Shape features include many global shape features, such as invariance moments, geometric shape features and projection statistical features, and many local shape features, such as the ratio of antenna length and body length and antenna statistical projection feature. In addition the auther extracts insect color feature, yellow spot color feature and protergum and proala Crustacea texture feature. The result shows that all this features are typical and effective, and invariable with the image displacement, rotation and flex.(3) Feature selection of stored food insect images. Considering the few samples and the need of real time, the paper doesn't use K-L transform or any other methods commonly used for feature selection, but use RS theory to select features which has the ability of analyzing, reasoning and finding the relationship between data, selecting conditional property combination. The result shows its efficiency.(4) Classification of stored food insects. According to principle of reducing recognition error rate, increasing the ability of real-time processing, making the loss least, the multilevel classification mode is selected to classify the stored food insects. Compared with classical statistical approaches, neural networks approach to pattern recognition have many advantages, such as self-adaptability, parallel processing, robustness and strong classification ability. So BP neural network is used to classify the stored food insects primarily. The result shows our method is effective.