Research On Crop Leaf Disease Recognition Based On Multiple Classifiers Selection Ensemble

In recent years, research on the identification of crop diseases caused people’s attention due to the urgent need for agricultural information. For problems of the common crop diseases recognition, many scholars resolved them use a variety of methods and from multiple angles. But to apply to actual production, there are still lots of problems to solve. To identify crop leaves collected from the natural environment, the accuracy of lesion features, the adequacy of features description and the performance of the classifiers will have a great impact on the recognition results. Therefore, this thesis focuses on how to improve the recognition performance and enhance the practicality of crop diseases recognition in the natural scene. The main work is as follows:(1) In order to enhance the recognition effect and practicability of diseases in cucumber leaf, heighten the capacity of character description, a novel method which segmenting the cucumber disease images and then extracting color features and texture features of these images is put forward. Firstly, preprocess and segment the disease images collected in natural environment in different color spaces. After that, fuse the images that are segmented. Secondly, fuzzy quantization histogram and polymerization degree of color of lesion area are extracted as the color features, then the texture features of lesion are extracted via the color co-occurrence matrix calculated by color similarity measurement function. Finally, utilize kernel principal component analysis method to fuse the color features and texture features so that the redundant components can be removed and support vector machine is used for classification. Comparison of the experimental results shows that the proposed method can fully describe leaf lesion features, and improve the recognition rate of leaf disease effectively.(2) For the limitations of single classifier on disease recognition, the thesis put forward a disease classification method based on multiple classifier selection ensembles. First, in order to guarantee the recognition effect and the otherness of the base classifiers, using Simba algorithm to sort the characteristics, build feature subset by targeted sample features, train some candidate classifiers; Second, using the classifiers trained by the kernel principal component analysis algorithm and Simba algorithm to do complementary calculation, choose classifiers which recognition rate is high and the complementarity is large to integrate; Finally, weighted voting for multiple classifiers to obtain the final result according to the corresponding weights. The experimental results show that, the proposed method can effectively use the prediction results of multiple classifiers, improve the recognition effect and robustness of crop leaf disease identification.