CLASSIFICATION OF IMAGE PATTERNS USING SVM FOR WEB-BASED DIAGNOSIS OF RICE DISEASES
Recently, Support Vector Machine (SVM) is being used increasingly as an efficient pattern recognition tool also in the field of bioengineering. A classification method based on SVM pattern discrimination model was investigated aiming to apply it in an online identification of rice disease and their stages. The experiment was carried out with 425 images of 3 types of rice diseases named Leaf blast (divided into 2 classes of A-type and B-type), Sheath blight, and Brown spot. As for the image features, 5-shape variables and 2-color variables were used. Pattern discrimination model was assessed by 3-fold cross validation. In case of using only the shape features, discrimination accuracy of each classes was 40 to 82％. However, in case of using shape and color features simultaneously, the accuracy was raised to satisfactory level, 93％ for one class and 80％ for the rest of the classes. Furthermore, method of discrimination in two steps, first discriminating by using shape features to determine the possible/similar diseases, and use both shape and color features in next step to discriminate the possible diseases presumed in the first step, showed the best result with average recognition rate of 86％. 2 of 4 classes, Leaf blast A-type and Sheath blight showed 94％ and 87％ respectively. Also the stage discrimination for Leaf blast and Sheath blight showed high accuracy. Study on increasing effective variables and parameter adjustments are still necessary to improve the recognition rate. In comparison with the other non-linear discrimination models, SVM outperformed the neural network and quadratic discrimination analysis. With this result, recognition method with SVM classifier was considered to be effective in pattern recognition of the disease spots/lesions of rice. Moreover, it is expected that integrating additional visible and non-visible features sent by user to this method will provide better performance in the case of online identification of diseases.
Image features web-diagnosis disease identification SVM pattern discrimination
Gauri MAHARJAN Teruo TAKAHASHI Shu-Huai ZHANG
The United Graduate School of Agriculture Science,Iwate University,Japan Faculty of Agriculture and Life Science,Hirosaki University,Japan
国际会议
北京
英文
1-6
2009-10-14（万方平台首次上网日期，不代表论文的发表时间）