The Effectiveness of Deep Learning Methods on Groundnut Disease Detection

Abstract

Early detection of plant diseases in the agricultural sector is considered an important goal to increase productivity and minimize damage. This study deals with the use of deep learning methods to realize the automatic detection of leaf diseases in peanut plants and the explicability of the model with heatmap visualizations formed during the detection of diseases. In the study, a dataset containing 3058 images with 5 classes enriched with diseased and healthy samples of peanut leaves was used. The explainability property has also been studied to understand why the models detect a particular disease. The decision processes of deep learning models, which are usually described as the "magic box", were visualized with the heatmap method in this study. By highlighting the pixels that are effective in detecting diseased leaves with heatmap visualization, the decision-making process of the model has been tried to be made understandable. The results show that deep learning models have high performance in detecting peanut leaf diseases, and the explainability obtained by heatmap visualization is a reliable tool for agricultural specialists and producers. Thanks to the visual explanations provided by the model, the level of confidence in the detection of diseases has been increased and confidence in the decision processes of the model has been provided. This study constitutes an important step towards increasing efficiency in agricultural applications and providing a more efficient approach to disease management by investigating the impact and explicability of deep learning methods in the field of disease detection in peanut plants.