Hierarchical Deep Learning Approaches for Enhanced Tea Leaf Disease Analysis via Segmentation

Bangladesh is a leading global tea producer, yet declining production due to leaf diseases poses a significant challenge. Traditional disease detection methods, which rely on manual inspection, are both time-consuming and error-prone and researcher nowadays are trying to use artificial intelligence to ameliorate the issues. This study leverages deep learning techniques to improve the accuracy and efficiency of tea leaf disease detection and classification. Using advanced segmentation methods, the approach pinpoints diseased areas on tea leaves, surpassing the limitations of traditional bounding box-based detection. Three convolutional neural network models—YOLOv5s-Seg, YOLOv7-Seg, and YOLOv8x-Seg—were evaluated for instance segmentation, with their performance assessed through precision, recall, and F1 score metrics. The YOLOv8x-Seg model demonstrated superior performance, achieving a highest accuracy of 98.47%, making it the most promising among all the models. Additionally, this research introduces SylBD, a novel dataset that can serve as a benchmark for future studies and managing tea leaf diseases. By enabling more precise and efficient disease detection, the study aims to enhance crop health and productivity in the tea industry, demonstrating the practical benefits of deep learning for agricultural management.