Automated Identification of Begomovirus in Tomato and Chili Plants using Deep Learning

Abstract

The purpose of this case was to identify and classify diseases of tomato and chilli plants caused by Begomoviruses using AI-based learning techniques. Along with other viruses, it is one of the most destructive in the plant kingdom in Pakistan. Begomovirus is the most common viral infection in both tomato and chilli plants and lowers their productivity. It is not possible to detect Begomovirus through morphological symptoms observation because it shares many symptoms with other viral infections. Leaves may also exhibit other symptoms, such as yellowing and curling, and the entire plant may become stunted. Pathogen spread needs to be controlled as early as possible to contain the virus. Machine learning and artificial intelligence now have several tools to measure the detection of objects with great precision. This research focused on developing an advanced deep learning algorithm for the early identification and intervention of diseases caused by Begomovirus. Screening of the Visual Geometry Group 16 (VGG-16), Residual Network-50 (ResNet-50), and the third version of the Google Inception CNN (Inception-v3) techniques were applied to diagnose Begomovirus, and possible solutions were suggested. Using VGG-16 yields the highest detection accuracy of 98% for plants infected with Begomovirus, while Inception-v3 and ResNet-50 achieve 95% and 80%, respectively. This technique involves capturing data images from an algorithm for those features that correlate and merge the images to teach the algorithm more efficiently without explicitly instructing it on how to do so. This research aims to train a deep learning model to automatically diagnose Begomovirus based on disease severity measurements.