Detection of Leaf Disease Using Hybrid Feature Extraction Techniques and CNN Classifier

Identification of leaf disease using multiple descriptors is presented. Initially the images are resized to 256 x 256 to maintain the uniformity throughout the experiment. The Histogram Equalization (HE) technique is employed on resized leaf images to improve their quality. Segmentation is performed using k means clustering. The contour tracing technique is applied on leaf images to trace the boundary of affected areas. Prominent features of leaf image are extracted using DWT, PCA and GLCM techniques. The performance of the system is evaluated using three different classifiers viz., SVM, KNN and CNN using Matlab on bean leaf database. The performance of proposed model on CNN classifier is better when compared with other existing methodologies.