Research And Application Of Intelligent System Of Field Wheat Leaf Disease Detection

The study on wheat disease detection using image processing technology is significant to recognize wheat disease accurately, reduce the impact of disease on wheat yield and ensure national food security. However, as a new field of application, the technique of wheat disease detection based on image processing still has no systematic study in our country. Reviewing of relative research at home and abroad, an algorithm for segmenting spots in wheat leaf disease image with complicated background was studied, color feature, texture feature and shape feature were extracted, after feature selection central features for multi-class disease detection were received, a multi-class classification algorithm based on support vector machine and decision tree, and the intelligent system of wheat leaf disease detection based on image processing was designed and developed, which provided the basic research to study and develop the intelligent system of yield wheat diseases. The contents of the study could be briefly summarized as follows:1. The segmenting algorithm of wheat leaf disease image was studied. For the case of the background was complicated in image acquisition of wheat leaf disease and was difficult to segment, the image segmenting technology based on mathematical morphology was studied. And combined with many algorithms such as watershed algorithm and threshold segmenting algorithm, a new image segmenting algorithm which is suitable for yield environment was received.2. Color feature extraction algorithm for wheat leaf disease image was studied. Consideration of the research results in various crop disease identification, the new color features were defined, and the feature extraction algorithms of wheat leaf disease in RGB color space and HSI color space were studied.3. Shape feature extraction algorithm for wheat leaf disease image was studied. After studied the research results in shape feature extraction, comprehensive consideration of the difference of various diseases of wheat leaf in shape, the new shape features were defined, and the shape feature extraction algorithms of which was suitable for wheat leaf disease detection were studied.4. Texture feature extraction algorithm for wheat leaf disease image was studied. Consideration of the research results in texture feature extraction, texture feature extraction algorithms of which was suitable for wheat leaf disease detection were studied, and the texture features were extracted.5. By using the extracted color, shape and texture feature, through the appropriate feature selection method, the suitable features for wheat leaf multi-class disease identification were found.6. The identification algorithm of wheat leaf disease was studied. In order to decrease the sample training time effectively, improve the identification rate, and make the model has good generalization ability, a new class partition project based on samples is proposed. This project makes a comprehensive consideration of the number of waiting classification samples and the capability of class partition, and takes a compromise between the "first classifying the classes with a large number of samples" and the "first classifying the classes that can be partitioned easily". And a new decision-tree-based support vector machines multi-class classification algorithm is proposed, which adopts the balance decision tree structure. The experimental results by using Statlog database indicate that the algorithm can significantly reduce system training time at the condition of not reducing identification rate, and is an effective multi-class classification algorithm. At the same time, in order to avoid the disadvantages of treating the differences between different attributes of the samples equally and taking no account of the correlativity of different variables in computing the inter-class separability measure in European space, we proposed a method of computing the inter-class separability measure based on Mahalanobis distance, and gained a multi-class classification algorithm based on SVM and decision tree utilizing the advantages that the Mahalanobis distance has dimensionless impact and has nothing to do with the unit of measurements with the original data. Experimental results show that the classifying project we obtained by this algorithm is a better one and this algorithm could have a higher recognition rate, and the algorithm is an effective multi-class classifying algorithm.7. The intelligent system of wheat leaf disease identification was designed and developed.