Plant Disease Detection Using Clustering Based Segmentation and Neural Networks

Abstract

Farmer suicides in India had ranged between 1.4 and 1.8 per hundred thousand people, accounting to 11.2 % of all suicides in India due to reasons like debt, low produce prices, crops failure and alcohol addiction. Among these, crop failure is attributed to various factors including unpredictable weather conditions, poor farming practices, pests and diseases along withill use of fertilizers and late disease diagnosis. Various systems have been proposed and implemented for immediate identification of the disease, using mobile devices for disease identification and consequent action, but the majority of proposed approaches involve segmentation techniques coupled with classical machine learning algorithms, which focused on the entire plant or fruit image, not primarily on the diseased part, thus embedding pixels which introduce possible bias in each data point leading to an imprecise training dataset and consequently faulty training. In this paper we propose a method of leveraging a combination of clustering based segmentation for identification of the diseased part exclusively and consequent feature extraction over it along with using neural networks over classical algorithms, thereby increasing feature complexity and thus better training, increasing training accuracy and leaving scope for further integration of huge amount of data which can added later on.