CNN based on Overlapping Pooling Method and Multi-layered Learning with SVM & KNN for American Cotton Leaf Disease Recognition

The cotton agriculture industry faces the economic losses due to the pest infections and bacterial or viral contagions, the farmers lose nearly 10-20% of the total profit on an average annually in India. In this paper, a solution to the problem related to agriculture has been suggested, which involves the cotton leaf disease recognition by using the visual features. This paper implements the expert system for the recognition of crop diseases in the given set of images. The overlapping pooling has been utilized with the multi-layered perceptrons (MLP) with flexible layering mechanism to classify the plant leaves for detection of infected and healthy leaves. The graph based MLP network model has been employed to dynamically transform the feature orientation to discover the similar components in the given database. The k-nearest neighbor (kNN) and support vector machine (SVM) are utilized for the overlapping layer of classification to minimize the errors by double layered modeling. Various techniques like morphological segmentation, pattern matching, hue matching are combined in this model to accurately localize the disease region with more than 96% accuracy.