Study On Soybean Leaf Disease Recognition Based On Image Processing

Soybean is one of the important food crops in China.In recent years, due to the rapid development of economy in our country, accompanied by more and more serious environmental pollution. Water pollution,fog weather and so on various factors induced the occurrence of soybean plant disease, which have bean seriously influenced to the yield and quality of soybean.The crop diseases automation implementation can be real-time and accurately to judge diseases.Agricultural producers could timely adopt corresponding prevention measures.This not only improved the yield and quality of soybean, but also could avoid the abuse of pesticide. Above all, it could protect environment. This paper was written by based on fully understand the current research in this yield at home and abroad,which was in-depth research of segmentation the lesion of soybean leaf disease image.According to the characteristics of the disease lesion,this paper discussed the question how to extract the characteristic parameters. Finally at the end of this paper, some diseases identification methods were studies and compared.In this paper, the main works and achievements of this paper are as follows:(1)In terms of segmentation research on soybean leaf disease images.We present two method to intelligently detect soybean leaf disease. Firstly, Unsupervised Segmentation method for diseases of soybean color image based on Fuzzy Clustering. This method is an improvement on fuzzy mean clustering, which was by gradually increased the value c and changed the distance function iteration,on the basis of effectiveness evaluation to measure, unsupervised to search the optimal classification number c, so as to realize the segmentation of disease spot;Second, we will use the significant region extraction method to segment soybean diseases,then combine the threshold segmentation and morphological processing to get the complete disease lesion.Compared with the traditional methods,our method reduced two steps of the pretreatment and background separation.(2)In terms of feature extraction, we choose to first order moment and Second order moment of every channel of RGB and HSV models as color features, texture features mainly adopts a operators described the local texture feature of image uniform LBP (homogeneous binary mode), according to the uniform LBP histogram distribution we can get the texture feature.(3)On the disease recognition.Four identification method were used to implement soybean leaf disease classification,which were neural network, support vector machine, support vector machine improved by Particle Swarm Optimization algorithm and neural networks improved by dropout. Aim at parameters for support vector machine (SVM) is difficult to determine the problem,the paper put forward support vector machine (SVM) improved by particle swarm optimization, which mainly used particle swarm optimization(PSO) algorithm to search the optimal kernel function parameter and punish coefficient,so improved the accuracy of support vector machine (SVM), the result of the experiment data shows that the recognition accuracy than support vector machine (SVM) was improved13.25%. Finally based on the thesis with less training samples this problem, at the same time in order to prevent model fitting and improve the effect of soybean disease classification identification, the paper put forward to using the dropout to improve the neural network. The experimental results data showed that more than the recognition accuracy of improved support vector machines (SVM) is increased by2.273%.