Real-Time Plant Leaf Disease Detection and Identification using a Convolutional Neural Networks on an Embedded Platform

Abstract

Damage in yield due to crop diseases is a great concern to the farmers and India being an agriculture-based country, a large portion of the country’s GDP also depends on it. Loss of yield due to undetected or untimely dispersion of pesticides affects the overall crop production. In this project, a method to identify the type of disease present in a crop based on leaf images using machine learning is proposed. First, the leaves are individually detected in real-time from the field using Single Shot Detector (SSD). To classify the type of disease present in the crop, a convolutional neural networks architecture is proposed which is trained on the PlantVillage dataset and the proposed hybrid network is deployed on the embedded platforms namely NVIDIA Jetson TX1 and NVIDIA Jetson Nano, for real-time detection and identification. The disease classification accuracy achieved is around 96.88%. Moreover, the proposed Classification model’s accuracy is compared with the AlexNet model’s accuracy. As a result, the proposed classification model accuracy is higher than AlexNet CNNs model accuracy, which is 95.53%. Furthermore, the proposed plant disease detection system also tested at a tomato farm. This in-field real-time testing shows that the proposed model is robust and gives efficient results in real-time testing.

This project aims to aid the farmers by embedding the machine learning model for detection and classification in a handheld device that would inform the farmers about the disease type on time, thereby providing them sufficient help to take precautions and countermeasures for its removal.