A Classification Approach for Predicting Plant Leaf Diseases in Digital Image Processing

In agriculture, detection and diagnosis of plant disease using digital image processing techniques focused on accurate segmentation of healthy and diseased tissue. Among various segmentation methods, the most widely used semiautomatic segmentation is based on gray scale histogram. In a novel semi-automatic segmentation process, the edges were removed along with pixels and then color conversion was done. After color conversion, pixel value adjustments and contrast enhancement of an image were performed to improve the image quality. Histogram with 100 bins was constructed for recognizing the diseased tissue from the healthier part of a leaf image. At last, segmentation of diseased leaf was found based on the histogram bins. Such bins were found manually which is not easy for all cases. Moreover, detection accuracy was reduced the quality by the influence of reflection light and distortion regions in an acquired image. Hence reflection light and distortion from image were removed using Quality Assessment Method Scheme (QAMS) algorithm. For automatic separation of diseased part from the healthier regions in a leaf image an optimization algorithm is required. To automatically define the histogram bins and separate diseased part from the healthier regions in a leaf image the Convolutional Neural Networks (CNN) algorithm used. After segmenting the diseased leaf image, the classification is done by Support Vector Machine (SVM) to detect the leaf diseases. The method provides better detection accuracy and computational time is reduced. Keywords-CNN, Diseases detection, Distortion removal, Reflection light, SVM.