SmartAgriDoctor: Plant and Crop Disease Detection and Diagnosis Using Deep Learning

Abstract

 Agriculture is still a staple of food security but the global productivity is threatened by the disease on plants. Traditional methods of manual inspection are usually subjective, time consuming and unproductive and are not conducive to being monitored industrial scale. This paper introduces SmartAgriDoctor which is a Deep learningbased system for automated Crop Disease Detection and Diagnosis using Convolutional Neural Networks (CNNs). The proposed framework passes through the phases of preprocessing, feature extraction and classification of the images from the New Plant Disease Dataset in order to identify between healthy and diseased leaves of plants. The dataset was augmented using random rotation, flip and contrast normalization to augment the generalization skills. The optimized CNN model that was trained and validated with 80,000 labelled images was able to achieve the Knowledge Level of 97.8% overall classification with the precision, recall and F1 score of 97.4%, 97.6% and 97.5% respectively. Experimental results demonstrate the robustness of the model under different class of diseases and environment variations; Due to low weight design of the system, it can be sent to the mobile devices and edge devices to conduct real-time diagnose to farmers and agronomists. Overall, SmartAgriDoctor is a very interesting example of the real-life application of Deep learning in precision agriculture leading to an early detection of the diseases, lesser losses of crop and sustainable agriculture.