Deep learning method for cucumber disease detection in complex environments for new agricultural productivity

Area of Science:

• Agricultural technology
• Computer vision
• Plant pathology

Background:

• Cucumber disease detection is challenging due to variations in scale, background clutter, and hardware constraints.
• Accurate and efficient disease identification is crucial for crop management and yield optimization.

Purpose of the Study:

• To develop an improved, lightweight algorithm for cucumber disease detection.
• To enhance the accuracy, speed, and compactness of disease detection models for agricultural applications.

Main Methods:

• The study proposes YOLO-Cucumber, an enhanced lightweight detection algorithm based on YOLOv11n.
• Key innovations include Deformable Convolutional Networks (DCN), a P2 prediction layer, Target-aware Loss (TAL), and Channel Pruning via Batch Normalization (CPBN).

Main Results:

• YOLO-Cucumber achieved a 6.5% improvement in mAP@50 (93.8%) on a cucumber disease dataset.
• The model size was reduced by 3.87 MB, and inference speed increased to 218 FPS.
• The algorithm demonstrated effectiveness in handling symptom variability and complex detection scenarios.

Conclusions:

• YOLO-Cucumber offers superior accuracy, speed, and compactness compared to mainstream detection algorithms.
• The developed model is well-suited for embedded agricultural applications requiring efficient disease detection.