Study On Web-Based Intelligent Diagnosis Techology Of Wheat Disease

The traditional agricultural expert systems are mainly stand-alone ones, which are restricted by time and space, lack of agility and can not adapt to the increasing development of internet period. Nowadays, with the advancement of internet technology, web has becoming a main approach of information transmission. Establishing the web-based crop disease and pest diagnosis system has become an inevitable trend of agricultural informatization. In order to keep up with the trend, this dissertation takes wheat disease diagnosis as research object, and studies web-based expert system model, knowledge base construction, knowledge acquisition and reasoning methods in wheat disease diagnosis. Moreover, an algorithm testing software and a three layer B/S-pattern prototype system is designed to validate the proposed methods.The main contributions of the research include:(1) For making knowledge base meet the network requirement, this paper proposes a new method of wheat disease and pest knowledge presentation based on object-oriented and XML technology, and constructs the XML knowledge base which can be easily extended and do not depend on software and hardware environment. In addition, discussing the advantages of web-base expert system compared to the traditional stand-alone expert system, a web-based expert system model based on J2EE and XML is prompted, in which the reasoning machine is designed by EJB technology and data processing can be separated with reasoning processing.(2) Aiming at the limitations of knowledge acquisition and inaccurate reasoning in traditional expert system, a knowledge acquisition and reasoning method based on BP neural network is proposed. Wheat disease feature parameters are coded and the causality between symptoms and diseases is got by neural network. Experiment results show that the average diagnosis precision rate of this method is 80%.(3) In order to make the diagnosis process of artificial neural network transparent and comprehensible, a cross entropy function with penalty term is applied to replace the usual squared error function, and if-then rules are extracted from neural net work by three steps of...