CD-YOLO: An Enhanced YOLOv8n-Based Algorithm for Accurate Crop Disease Detection

Timely detection of crop diseases is crucial for precise fertilization. To address challenges such as difficulty in distinguishing disease spots and low recognition efficiency, this paper proposes an improved disease detection algorithm based on the YOLOv8n network, termed CD-YOLO. The enhancements focus on three main aspects. First, the Convolutional Block Attention Module (CBAM) is added at the end of the backbone network to improve the model's ability to extract features related to crop diseases. Second, the Bidirectional Feature Pyramid Network (BiFPN) is incorporated into the Neck layer to enable efficient bidirectional feature fusion and adaptive weight allocation, enhancing detection performance across crops of various scales. Finally, the Efficient Channel Attention (ECA) mechanism is introduced in the output layer, allowing the model to dynamically focus on key channel features with low computational cost. This improves the distinction of disease categories and details, particularly in complex backgrounds or when disease spots share similar features. Experimental results indicate that the improved CD-YOLO model achieves a 2.8% increase in mAP@50 compared to the traditional YOLOv8, reaching 92.0%. In tomato leaf disease detection, the model demonstrates higher accuracy and stability, significantly reducing missed and false detections while providing more precise localization and recognition of disease targets. This study offers valuable insights for improving crop productivity.