Research And Application Of Diagnosis Technologies For Crop Pests Based On Image Recognition

In view of crop pests diagnosis in our country at present mostly remains in the artificial stage, there are such problems as poor objectivity, low efficiency, huge labor intensity, also, the current crop pests diagnosis based on image recognition exists myriad deficiencies. The key theory, algorithm and the practical application based on image recognition of crop pests diagnosis are researched in this paper. The summary of the main work as follows:The fundamental of crop pests diagnosis based on image recognition were researched systematically, including the classification, basic frame of image recognitions and traditional image data set. In this part, two kinds of insect pest image acquisition modes were introduced, and a suitable way of crop pests image preprocessing was proposed and verified. Also, a construction principle of crop pests image data set was putted forward, guiding by this principle and utilizing the existing resources of crop pests image, four common field crop pest image sets of rice, rapeseed, corn, soybean were built respectively.The existing crop pests image segmentation algorithms focused on simple background, or the segmentation of gray level image which mostly use the threshold segmentation algorithm. But in the real scene, under the influence of environmental factors such as plant leaves and weeds, soil and illumination, pests images generally has the complicated background of farmland, so the existing automatic pest segmentation algorithms are often not ideal. To solve above problems, this article put the semi-automatic image segmentation algorithm, such as GrabCut region merging algorithm and region merging algorithm by maximal similarity into the crop pests image segmentation, and proposes a maximum similarity region merging algorithm combined with texture-color histogram. Through the complex segmentation experiments of the crop pests image with complicated background proves the feasibility and effectiveness of the algorithm.Current crop pests image feature extraction is mostly focused on the extraction of single feature, such as the color feature, shape feature or texture feature, because of its recognition object shall focus on one or several pests recognition, so using a single feature can obtain a good recognition rate, but spreading to the recognition of a variety of pests, the result is often bad. To solve these problems, this paper puts forward the crop pests recognition based on multi-feature fusion. The method uses Fisher line discrimination method to calculate weights of all the features, combined with Euclidean distance classifier, the34kinds of insect pests in the four kinds of plants as rice, rapeseed, corn and soybeans for experimental data, through a variety of features combination experiment respectively, the experimental results prove that the algorithm has a high recognition rate.Aiming at the practical application, a friendly user interface, easy extension of crop pests diagnosis system based on image recognition has been developed. The system not only can read the local pest image recognition, but also you can get the external image acquisition equipment to capture real-time image recognition, the users only need a simple interaction can obtain pest species and the method of prevention and cure, thus effectively guide the users to rationally use pesticides to prevent and control pests.