Strawberry Maturity Recognition Algorithm Combining Dark Channel Enhancement and YOLOv5.

Area of Science:

• Agricultural Engineering
• Computer Vision
• Robotics

Background:

• Strawberry fruit picking accuracy is often limited by visual detection challenges.
• Existing methods struggle with varying ripeness stages, including 'bad fruits', and low-light conditions during nighttime harvesting.

Purpose of the Study:

• To develop a robust and accurate automated strawberry fruit picking system.
• To enhance the performance of the YOLOv5 object detection model for strawberry recognition.

Main Methods:

• A modified YOLOv5 model was developed, incorporating a 'bad fruit' criterion alongside ripeness, near-ripeness, and immaturity.
• Dark channel enhancement was applied to address low illumination issues in nighttime image collection.
• The proposed method was compared against five image enhancement algorithms and other object detection models (SSD, DSSD, EfficientDet).

Main Results:

• The YOLOv5 model with dark channel enhancement achieved training accuracy above 85% and testing accuracy above 90%.
• Dark channel enhancement demonstrated superior performance compared to other enhancement techniques under various conditions.
• The YOLOv5-based method outperformed SSD, DSSD, and EfficientDet in recognition accuracy, reaching over 90%.

Conclusions:

• The proposed YOLOv5 combined with dark channel enhancement significantly improves strawberry fruit picking accuracy and robustness.
• This approach is effective in complex environments with partial occlusion and multiple fruits, enabling all-day harvesting.
• The method offers a promising solution for automated harvesting in precision agriculture.