High performance deep learning architecture for early detection and classification of plant leaf disease

Abstract

Agriculture is primary source of income for farmers worldwide and they face a significant annual production loss owing to plant leaf disease. In an agroecosystem, staying aware of plant leaf diseases is a very important concern. The losses that farmers experience as a result of numerous plant leaf diseases can be reduced with earlier detection of these diseases. Manual recognition becomes very tedious in case of huge dataset and also recognition accuracy is greatly affected based on human expertise. To overcome this, automated intelligent strategies using deep learning (DL) convolution neural network (CNN) approach has gained popularity in recent times owing to the accurate diagnosis with reduced time and resource complexity. Combining with image processing, the patterns of leaf images at specific times are used to identify plant leaf diseases. For disease identification, classification and diagnosis, tomato plant is considered for current research work. Dataset for our research is captured from the real time environment from the agricultural fields of Jalgaon city. The proposed method is able to classify diseases with high precision by automatically extracting features thereby eliminating the feature engineering and threshold segmentation process. We adopt and extend our network utilizing spatial images captured in adverse environmental conditions. Automated disease diagnosis has been made possible by recent developments in computer vision through DL. Overall, a clear pathway for crop disease diagnosis on a gigantic worldwide scale is presented by the method of training deep learning models on progressively larger and publicly available and real time image datasets.