Improved plant leaf disease detection architecture using unmanned aerial vehicle images and hybrid YOLOv11

Plant diseases affect crop quality and quantity, which significantly reduces crop production. Grapes are a popular fruit, and pumpkin is a widely consumed vegetable in the world. Early detection of grape and pumpkin leaf disease is crucial to avoiding large losses in crop quality and yield. Even though the use of unmanned aerial vehicle (UAV) remote sensing can assist in reaching a broader scale for plant leaf disease identification, the work of detection is significantly hampered by the overlapping leaves, and blurring of UAV images. This paper suggests combining object detection techniques with Real-World Enhanced Super-Resolution Generative Adversarial Network (Real-ESRGAN) for enhancing image quality to detect grape and pumpkin leaf diseases in UAV-captured images. The Atrous Spatial Pyramid Pooling (ASPP) technique is incorporated into YOLOv11 architecture to enhance the network’s feature representation capabilities by efficiently incorporating depth-wise separable convolutions and dilated convolutions with different rates. The proposed approach has shown remarkable effectiveness. In the validation, the ASPP-YOLOv11 outperforms the baseline YOLOv11 architecture in precision achieving 84% with 5.2% increase, 94.7% with 3.3% increase in mAP50, and 74% with 0.9% increase in mAP50-95. Additionally, in the testing set the ASPP architecture significantly enhances mAP50 to 91.6%, attaining 1.6% gain and obtaining 1.6% increase in mAP50-95.