A Modified Feature Optimization Approach with Convolutional Neural Network for Apple Leaf Disease Detection

Abstract

AbstractMonitoring plant health and detecting various plant diseases is an important discipline of plant pathology. Some diseases can even be detected by observing pathological characteristics. However, it requires high expertise and is very time consuming when inspecting large farms. Therefore, automated disease detection methods are required for rapid analysis. In this paper, an automated method for detecting apple leaf disease is proposed, and image processing techniques are used to analyze leaf samples. The paper uses the approach of Contrast Limited Intensity Adjustment (CLIA) for image enhancement followed by segmentation of the diseased region using k-means. Features of this region are extracted using Speeded Up Robust Features (SURF) that is further optimized using Particle Swarm Optimization (PSO) to select the best feature set among the extracted features. At training and classification stage, CPU based Convolutional Neural Network (CNN) is implemented whose performance is evaluated on a dataset comprising of 3,171 leaf images. Simulation analysis demonstrates that the proposed work outperforms existing works, with an average precision of 0.97 and accuracy of 99.26%.KeywordsApple leaf diseasek-meansSURFPSOConvolutional neural network (CNN)