Research Of Ramie Diseases Recognition System Using Neural Network And Support Vector Machine

With the rapid development of agricultural modernization and information technology,our country put forwarded the intention of agricultural informatization,its purpose to using computer technology to centralizing analysis the agricultural data,it can improve the efficiency and reduce the cost.In the process of agricultural informatization,machine learning is widely used.The main applications are agricultural production,price calculate,agricultural information recommendation and crop disease recognition which in this paper.Use computer to recognize crop diseases can reduce the subjectivity,get the reference results to help experts to recognize diseases.This paper focus the recognition of ramie’s brown spot disease,and analysis the performance of the popular machine learning method in the diseased leaf image recognition,and made the following work:(1)Using the method of FindMaxima: setting a tolerance to pick up the local maximal value,to divide the disease lesions in diseased leaf images.(2)Using convolution neural network(CNN)to train the brown spot lesion’s pixel gray level data,and use the CNN’s output features to train support vector machine(SVM)model for recognize and classify the disease.(3)Use the combination of multiple features and multiple models to classify disease,it can improve the accuracy of the system and reduce the deviation,better than one model and one type of feature.After a series of research and testing,and considered the recall in lesions dividing,the accuracy rate of ramie leaf’s brown spot disease recognition perhaps around by 90%.If all the lesions are success to divided by FindMaxima method,the accuracy will up to 96%.After test,the FindMaxima method is better than the threshold method in used to divide ramie leaf’s brown spot lesions,especially for the lesion in yellow area of disease leaf.During the research,we are confirmed the advantages of CNN in the image recognition task,and the feasibility of using CNN’s output to train SVM model to recognize the crop diseases.Mining features by CNN can simplify the manual feature mining procedure by digital image process,improve the automation level of the image feature mining,CNN’s output features are more stable and reliable,strong with adaptability,will help us convenient to expand more disease types to recognition.