Study On Detection Method Of Pests And Plant Disease Based On Low-rank Representation

As the major agricultural disaster, to a certain extent, crop diseases and pest threaten the safety of food production and the quality of agricultural products. In our country, the overall situation of crop diseases and pest is multiple, retransmission and frequent, it has brought huge economic losses to farmers. At present, the detection of crop diseases and pest still uses the traditional artificial observation method, and the artificial detection faces a great work intensity, poor timeliness and many other shortcomings. At the same time, the artificial method will cause the flying of pests, affect the accuracy of the observations, and bring a great deal of inconvenience to the early warning and disaster control of crop diseases and pest. Therefore, aiming at the characteristics of crop pest, according to the actual production situation, and with the application of computer vision technology, the research of crop diseases and pest based on automatic detection technology can realize real-time diagnosis to the crop diseases and pest condition, it can effectively solve the shortcomings of artificial detection with great timeliness and usability.In this thesis, we firstly study the area proportion of crop disease and the counting of pest automatic detection. According to the low-rank characteristics of natural images and combined with the color characteristics of crop pests, an algorithm based on the low-rank representation is proposed. And we also effectively extract pests in crop leaf images with pests and process for further handling. Thus the problem that agricultural workers encountered in the detection and counting of diseases and pests in the practical agricultural production is solved, and it can reduce the workload and improve the accuracy of statistics. Specifically, the work of this thesis can be summarized as follows:(1) In order to measure the ratio of the disease area on a single leaf to the entire leaf or realize the detection and counting of pests, we propose an automatic blade extracting algorithm based on super pixel segmentation. Firstly, we compare the differences and similarities between various super pixel segmentation algorithms. Then based on the SLIC algorithm, we design a method for super pixel combination. Finally, we achieve the re-clustering of the original image that is segmented by super pixel, and the automatic extraction for image in plant leaves.(2) This thesis proposes a pest detection model. This model based on low-rank representation algorithm, and through it the original image is decomposed intoa low-rank matrix and a noise matrix. The characteristics of natural images determine the pests will be fully included in the noise matrix. Next, through the further processing of the noise part with pests on it, we can achieve the detection of pests.(3) We show a pest detection and counting system in this thesis. According to the above method, we present a system with the functions of the calculation of crop disease area ratio, automatic detection and counting of insects, it can identify the crop leaves of the images conveniently, and do detection and statistics of the number of pests on the leaf. This can help agricultural researchers accomplish the statistical calculation and judge the severity of pests quickly.For the purpose of automatic detection to crop diseases and pest, we design a detection system based on the further research on image characteristics of crop diseases and pest, and propose some innovation works of automatic extraction of crop leaves and pest detection. Seen from the results, our method can not only effectively segment crop leaves from complex background, but also accurately extract and count the number of pests of a single leaf, calculate the disease percentage of a leaf.