Study On Recognition Method Of Rice Disease Based On Image

The rice is one of the most important grain crops in our country. The improvement of rice yield and quality is an important goal for rice production. However, the loss caused by rice diseases annually is so astonishing that the control of rice diseases in rice production and national economy development holds the extremely important status. Now the usual method for recognizing rice diseases still relys on artificial subjective judgment, which is an inefficient way and could be result in unconstant outcomes. In this study, the digital image processing and pattern recognition technology were used to recognize rice diseases. The database of rice diseases was constructed and the intelligence recognition system was developed that allowed us segment rice disease spot, extract features of rice disease spot, carry on indoor disease spot appraisal and automatic recognize the rice plant disease. In addition, with the networking technology, rice plant disease remote recognition system was also developed. This research complied with the modern accurate agriculture and is a developing direction of automatic detection technology in modern agriculture. Outcomes of this study can be used over a large range. The main content and results are as follows:(1) A rice disease database was established. It realized the storage and reappearance of the information which included rice plant disease picture, disease features and other description information. This database provided the support for the recognition system as well as the network inquiry system. It also provides the data for later expanding based on the content retrieval.(2) A new segmentation method was proposed for segmenting disease spots based on color features and the outline of disease spots. The results showed that this method can effectively avoid the empty holes phenomenon appearing in disease spot. In addition, the method has a stronger anti-noise nature.(3) The inhomogeneous quantification histogram was used to extract color characteristic and it enhanced the extraction efficiency and recognition robustness. The gray level co-occurrence matrix was used to extract texture characteristic and it compressed the grayscale of spot image. The computation was reduced by 3/4 and the complexity of feature extraction was also reduced.(4) The method of stepwise discriminant analysis was used to select effective identification features in different feature sets. The results showed that the recognition accuracy of hue texture features was higher but the color parameter is the lower. The used parameters can be reduced to 57.2% without affecting the recognition rate. The redundant parameters were also ejected effectively. It can reduce the burden of computer memory and operation and improve the disease recognition efficacy.(5) Three methods including Bayes discrimination, neural network and support vector machine were used to recognize six kinds of rice diseases by four feature sets. The results showed that the recognition accuracies of color set and shape set were similar by the three methods. Regarding the texture feature set, the neural network sorter shows a weak ability, its accuracy is lower than 50%.However, the recognition accuracies were above 98% by using the three methods when all features were operated. Because the support vector machine displays some unique advantages for solving small sample, non-linear and high dimension problems and its algorithm is fixed and avoids non-convergence and random of output result, it was selected for recognizing rice diseases.(6) With IE, ActiveX, and unified the Sockets, the rice plant disease remote recognition and inquiry system based on the B/S structure was realized. It makes remote recognition, thus enhances the real time of plant disease recognition. The system is very useful to popularize the modern technology and makes information exchangeing between technical departments and users much easier.(7) A method of twice segmentent strategies was proposed to segment disease spot and healthy leaf in the indoor disease spot appraisal based on computer vision. This method improved the recognition accuracy of rice diseases. A method using the four vertices mean of rectangular was proposed to remove sclerotium. This method was feasible and didn't affect the segmentment accuracy.(8) A system of rice disease automatic recognition was developed. Its functions included image de-noising, disease spot segmentment, feature extraction, indoor disease spot appraisal, rice disease recognition and remote recognition, inquiry and browse. Moreover, the techniques developed by this research could be extended to other crops disease recognition.