Stored-grain Pests Image Classification Research Based On Multi-label Learning

The food is closely related with the beneficial to the people’s livelihood issues, but atharvest and during storage, the harm will inevitably be influenced stored grain insects. So theidentification of these insects is a very important research in microbiology research. With theinformation technology and the application of computer pattern recognition rapiddevelopment, people began to classify the insects form the image recognition field. By usingthe large storage capacity and high speed of computers, the problems of personnel shortageand hard classification can be perfectly solved.At present, many methods have been proposed to classify the insects in imageclassification methods. But it is still hard classify the insects perfectly, because an insectimage contains not only many kinds of insects but also some cereal grains and impurities.This paper the introduction use the multi-label learning on grain insect image classificationfield, we give the each complexity grain pests image multiple labels, and use multi-labelclassifier to classify grain pests images.First, this paper describes and analyzes the basic theory of neighborhood average andmedian filtering and their respective advantages and disadvantages, which are the commonmethods of the image smoothing, puts forward the improved wavelet threshold method forgrain pests image smoothing, then researches threshold segmentation, edge detection andsegmentation method, and analyzes the respective advantages and disadvantages, choosesregion the region growing method to segment the grain pests image, and then researchesseveral commonly used kinds of image feature extraction methods, which include colorTexture, geometry features, On this basic this paper select color moment,texture featuresbased on GLCM and geometry and moment invariant feature as the feature extraction methodof grain pests image. After detecting the feature of grain pests, this paper transforms eachsegmented grain pests image to30d feature vectors, each grain pest is regarded as an instance,this all instance of grain pests image make up a package, and then gets the feature vector ofeach image package, thus transforms the Multi-Instance Multi-Label learning of grain pests toMulti-Label learning by Sparse Coding. Finally this paper adopt Discrete AdaBoost.MHwhich is the extend method of AdaBoost method to build the multi-label classifier. This paperexperiment100grain pests image, the results show that stored-grain pests image classificationresearch based on Multi-label learning is simple, efficient and feasible.