Prototype-Based Classification for Image Analysis and Its Application to Crop Disease Diagnosis

In this paper, provide an application of Learning Vector Quantization (LVQ)-based techniques for solving a real-world problem. We apply LVQ for automated diagnosis of crop disease in cassava plants using features extracted from images of plants’ leaves. The problem reduces to a five class problem in which we attempt to distinguish between a leaf from a health plant and leaves representing four different viral and bacterial diseases in cassava. We discuss the problem under additional constraints that the solution must easily be deployable on a mobile device with limited processing power. In this study we explore the right configuration of type of algorithm and type of features extracted from the leaves that optimally solves the problem. We apply different variations of LVQ and compare them with standard classification techniques (Naive Bayes, SVM and KNN). Results point to a preference of color feature representations and LVQ-based algorithms.