Feature Selection using Modified PSO with Mutual Information for Tomato Leaf Disease Identification

Automated disease detection in agriculture is one of the primary focus areas of researchers recently. The captured images are used as input to diagnose the diseases in plants. However, during automated diagnosis, the large number of features with respect to each image is one of the major challenges. The optimal feature selection techniques applied to these input images are often stuck into the local optimum. These features often share a large amount of mutual information, which can be used to reduce the dimensionality of feature set at the primary level before applying any standard feature selection approach. In this work, we have exploited the mutual information to reduce the size of feature set and then applied one of the most popular optimization techniques, PSO, to obtain the optimal feature set. We have compared the results with the optimal features in the absence of any prior reduction in the dimensionality of the feature set.