Disease detection in herbal agricultural products based on image processing and deep learning

Climate change and drought conditions at both global and local levels have increased the importance of sustainable agricultural practices. This study examines the applicability of artificial intelligence and deep learning technologies as effective tools for detecting plant diseases in crops such as corn and wheat, and for preventing yield losses that may occur in these crops. During the growth stages of corn and wheat, a significant factor determining the health of plants is the detection of diseased leaves. In this study, artificial intelligence-based classification models were developed using 4,188 images for corn and 977 images for wheat. The "Residual Network (ResNet)" model was employed for corn, while the "You Only Look Once (YOLO)" models were used for wheat to conduct classification processes. For corn, accuracies of 85.30% with ResNet50, 94.56% with ResNet101, and 94.14% with ResNet152 were achieved. For wheat, precision and recall rates of 68.9% and 70.4% with YOLOv5s, 95.4% precision and 88% recall with YOLOv8, and a 93.4% mAP50 (mean average precision) were attained. Additionally, a comparative analysis of ResNet-50/101/152 and YOLO-v5s/v8 models was presented, revealing that the ResNet101 model for corn and the YOLOv8 model for wheat achieved the highest success rates when compared with similar datasets.