Deep learning-based classification of visual symptoms of bacterial wilt disease caused by Ralstonia solanacearum in tomato plants

Abstract

Classification of plant diseases based on computer vision is a multidisciplinary challenge that involves technical and data-related complexities. Artificial Intelligence (AI) has increasingly found its application in plant pathology, disease, and anomaly visual characterization. Specifically, Machine Learning (ML) and Deep Learning (DL) algorithms have proven to be highly effective for tasks such as plant disease classification, detection, diagnosis, and management. In this work, we present a comparative analysis of multiple DL models based on Convolutional Neural Networks (CNNs) for visual symptoms classification of the phytopathogen Ralstonia solanacearum in tomato plants. We demonstrate that by implementing DL classification algorithms based on CNNs, it is possible to identify Ralstonia solanacearum potentially affected plants. This was possible due to the main virulence factor of Ralstonia solanacearum, the exopolysaccharide (EPS), which obstructs the plant’s xylem limiting water absorption and consequently inducing visual wilting symptoms. For this, we implemented, trained, and evaluated fourteen different CNN-based models. We evaluated the models by using different metrics such as precision, recall, accuracy, specificity, and F1-score. The models that obtained the best accuracy results were MobileNet-v2 and Xception, with an accuracy of 97.7% for both models. The presented findings significantly contribute to the visual symptoms classification of Ralstonia solanacearum in tomato plants, which may contribute to the control of this disease and its spread to healthy crops or other susceptible hosts in the future.