Research Of Diagnostic Technologies For Rice Leaf Diseases Based On Image

With the computer vision and image processing technology carried out in agriculture application field extensively, How to diagnose crop disease quickly and effectively has become a hot research topic. This paper focuses on four kinds of rice leaf diseases, crop science and information technology were combined to carry out a series of studies. a variety of experiments and further comparative analysis were conducted after understood current situation in the field widely,method,work processes and technical means were proposed from the rice leaf diseases image acquisition to identifying diseases,and a prototype system of rice leaf diseases diagnosis was developed based on image pattern recognition, Steadily pushing the laboratory basic research to the field practical applications was realized. new ideas and methods were provided for rapid diagnosis research of other crops diseases. other plants diseases diagnostic.In this study the works were completed as follows:First of all, in the image acquisition stage, Imaging environmental impact factors and the influence of their value level were considered fully for image acquisition effect. Using the orthogonal test method to design photo experiment, established the optimal combination of various imaging environmental factors by the results of image quality evaluation, and built computer vision system indoor through the combination of digital cameras and portable computer. on the foundation of image acquisition in light box, carried on image acquisition under simple background and complex background in agriculture rice fields simultaneously, and rice leaf disease image acquisition standard under different conditions was presented through a large number of experiments.According to images which were captured under different environments, an image evaluation method was proposed based on the features of single image to choose images for pre-processing stage.Then in the image preprocessing and image segmentation stage, For the problem of the image size is too large captured by digital camera, Rice leaf diseases image zoom-out method was put forword to ensure image fidelity has little change based on visual attention model and human visual system; In order to push laboratory studies to the practical application of field steadily,with the images captured under complex background in agriculture rice fields, using the phase coherence testing and area mathematical morphology method to propose an extracting method of rice leaf under complex environment ,which can transform the complex background into the simple background, and make the complex question become simplification; Diseased spot area always has higher brightness than other regions, a mixed space image enhancement method based on gray-scale transformation was put forward, improve the brightness and contrast of disease spot area further; Analyzing the color features of diseased spot area and obtain gray image using R-G,best iterative threshold method, Otsu method,fuzzy c-means clustering method were utilized for gray image segmentation, the results show that fuzzy c-means clustering segmentation method is better, but the speed is unsatisfactory, Otsu segmentation method is slightly worse, but the speed is quick, which method can be selected according to the actual situation.After that, in the feature extraction and selection stages, extracting feature parameters from three aspects such as color, texture and shape to represent diseased spot, after the massive analyses and confirmation, 22 features were extracted effectively for disease spot classification, with more features and high computational complexity, PCA was used to realize feature dimensionality reduction, 3 main components were formed, the accumulation technical progress factor achieved 86%, and 8 features were selected whose classification contribution degree are more than 90%.Finally, in recognition and classification stage, nearest neighbor, BP neural network, and support vector machine (SVM) classification algorithm were used to identify diseases.From the result of recognition rate, the 8 features whose classification contribution degree are more than 90% based on PCA being found is the optimal parameter combination, the optimum classification method is SVM. Simultaneously,on the basis of single disease spot identification, mixed diseases were realized effectively combined with regional characteristics, the original idea of a single spot identification were extended.Based on the above research contents and gradualness research results, this paper has established rapid recognition system of rice leaf diseases, its recognition accuracy rate achieved more than 83% for 4 rice leaf diseases, and it can satisfy the production request.at the same time,the study in this paper can be referenced in other rice leaf diseases and other crop diseases classification.