An Unprecedented Approach for Deep Learning Assisted Web Application to Diagnose Plant Disease
Shahina Anwarul, M Trivedi Mohan, Radhika Agarwal
Abstract
Crop diseases are becoming one of the major threats to food security at an alarming rate, and timely detection is challenging due to the shortage of infrastructure in many areas of the world. Plant diseases affect crop yield on a large scale, as different pathogens from bacteria to viruses, prove to be major, and possibly irreparable, causes of food loss. The situation has worsened by the fact that diseases are now more easily transmitted globally than ever before. Therefore, an accurate system is required which would provide this required assistance to the professionals that could ensure the error-free diagnosis of the plant disease and an accessible tool that would help farmers and even simple gardeners to have access to agronomic advice. The proposed system uses Convolutional Neural Networks (CNN) on simple digital images of diseased as well as healthy plants, to perform plant disease diagnoses. The authors intend to develop a highly accurate web application implemented using a deep learning model, to provide the user with a platform to identify and mitigate this issue. The proposed model is deployed with 7 convolutional layers, 2 densely connected layers, and 4 pooling layers. All the experiments are conducted on the PlantVillage dataset available on Kaggle and achieved 94% recognition accuracy in comparison to the other state-of-the-art approaches. The intended model is also evaluated on pre-trained models and surpasses in terms of accuracy and storage requirements.