Research On Greenhouse Tomato Disease Diagnosis Based On GA-BP Network

Diseases of tomato in greenhouse were a serious threat to tomato production and agricultural economy, most common way of inhibiting the disease was spraying chemical pesticides. However, the use of pesticides frequently caused a serious damage to the ecological balance of the greenhouse, leaded to pest resistance higher and higher,the work of resistant to insect pests became more difficult. Therefore, detecting the tomato disease timely and accurately was necessary. Traditional detection methods of greenhouse tomato was observation appearance by visual or weighing fruit quality, then determined tomato diseases category and divided levels by using knowledge. The methods that determined tomato diseases category and divided levels by using knowledge was influenced by human factors, did not have the reliability, stability and scientific.In the study, tomato leaves that characteristics of plane was good and had a longer life cycle, was used as the research object, by machine vision technology and Matlab mathematical software and optimized neural network model based on genetic algorithm, after image acquisition, image preprocessing, feature extraction, the establishment and training of model processes, make the precise and quantitative research for early blight, late blight, leaf mold of tomato diseases, and estimated the severity of each disease.In the process of image acquisition, achieved early detection and accurate collection of tomato disease in greenhouse using the P2P monitoring and digital camera collecting;In the preprocessing of image, the traditional median filtering algorithm is improved by using divide and conquer, the filtering average rate of per image improved 9.8%;In the segmentation process of image, achieved the separation of the blades and background,and the segmentation of lesion through the improved watershed algorithm;constructed the BP neural network, get training results of tomato diseases in greenhouse by learning and training of the neural network repeatedly; Based on the problems that BP neural network had some local minimum, network did not converge, fell into local minima easily. Using the genetic algorithm to optimize the initial weights and thresholds between neurons, then optimized BP neural network, the accuracy of new network model was higher, speed of convergence was faster.By the tests of GA-BP model, the recognition rate of the model for tomato early blight, late blight, leaf mold in greenhouse reached 100%,98%,96%. In terms of the estimation of disease severity, the disease severity was divided into five grades by the method of ratio product based on shape parameter.