Automatic Identification And Classification System For Rice Pests

The identification of rice pests is the premise and foundation of the prevention and management of pests.The automatic identification and classification system of pests can reduce the labor intensity of worker and improve prediction accuracy,thus reducing the loss of rice yield.This paper build a identification system of feature extraction of rice pests by using digital image processing and pattern recognition technologies.It can effectively identify 15 kinds of rice pests and 5 common pests.This system mainly contains pests image acquisition,image preprocessing,feature extraction of pests and classification of pests.More details are as follows:(1)High resolution and multi-focus image acquisition method was proposed.Harris point and image entropy were used as image quality inspection standard and the best image acquisition and image fusion strategy were obtained based on wavelet transform,this application is a reliable one to acquire an extended-depth-of-field image for stereoscopic pests.Finally,an image library of different backgrounds,gestures and equipment were established.(2)The preprocessing of insect images collected from different backgrounds,gestures and equipment include image adjustment,motion blur image filtering and image segmentation.The image filter changes the brightness,contrast and gamma value of the image through the lookup table(LUT)to reduce the influence of the external environment on the image;An image restoration based on constrained least square method and wiener filtering were used to reduce the confusion caused by motion jitter;Image segmentation includes color and grayscale image segmentation,and the influence of background can be eliminated by means of K-means clustering method and Otsu method,which makes the characteristics of pests more obvious.(3)The insect color feature,texture feature and Hu moment invariants were extracted.The first,second and third order moment of color are extracted as a color feature;The texture characteristics of pests were extracted using the gray level cooccurence matrix;The invariant characteristics of Hu moment with rotation,translation and scale are extracted.Finally,the pest characteristic library was established.(4)A recognition model based on support vector machine is established,the optimal parameters are found through the grid search method,also the identification rate of pest images and the overall system from different equipment were tested.Automatic identification system of rice pests was established which includes image opening,image preprocessing,feature extraction and pest identification.The experimental results show that the image can be well recognized for different equipment.The recognition rate for training set is 100%.And the insect recognition rate of multi-device test set was achieved more than 90% except Oxya chinensis and Mythimna separata(Walker).