Towards Precision Agriculture for Sustainable Chili Pepper Production: A Deep Learning Approach to Crop Disease Detection in Benin

Ensuring food security is a crucial priority for nations worldwide, but plant diseases significantly hinder this objective through their impacts on agricultural productivity. Chili pepper (Capsicum spp.) is a major crop in West Africa, including in Benin. However, its production is challenged by diseases such as anthracnose and Tomato Yellow Leaf Curl Virus (TYLCV), which severely impact yields and farmer livelihoods. While traditional methods for disease detection and management have been commonly used, they are no longer sufficient to combat rising pest infestations and declining agricultural productivity. To tackle this issue, we built a comprehensive dataset of 213 images of anthracnose-affected leaves, 202 images of TYLCV-infected leaves, and 119 images of healthy leaves, collected under diverse environmental conditions in Benin. The study compared the performance of thirteen (13) deep learning models, including YOLOv8, MobileNetV2, and DenseNet121, for the classification of chili diseases using transfer learning techniques. Performance was evaluated using metrics such as Accuracy, Precision, Recall, and F1-Score. Results show that YOLOv8 outperformed other models in real-time detection and localization of leaf diseases, achieving a mean Average Precision (mAP@0.5) of 0.995 and mAP@0.5-0.95 of 0.941, with precision and recall exceeding 99%. Among CNN models, MobileNetV2 and DenseNet121 achieved 96.25% accuracy. These findings demonstrate that deep learning models, particularly YOLOv8, hold immense potential for real-time, automated detection of chili pepper diseases. Future research should focus on expanding datasets, integrating climatic variations, and improving disease severity assessment for enhanced agricultural sustainability.