Enhancing Instance Segmentation in Agriculture: An Optimized YOLOv8 Solution.

Area of Science:

• Computer Vision
• Agricultural Technology
• Machine Learning

Background:

• Traditional segmentation algorithms struggle with complex agricultural scenes.
• Need for improved small object detection in precision agriculture.

Purpose of the Study:

• To enhance the YOLOv8n-seg model for improved agricultural scene segmentation.
• To boost the detection accuracy of small objects and overall feature extraction.

Main Methods:

• Introduced a dedicated small object detection layer.
• Replaced the C2f module with a C2f_CPCA module featuring channel prior attention mechanism (CPCA).
• Integrated a C3RFEM module utilizing dilated convolutions and weighted layers.

Main Results:

• Achieved 1.4% and 4.0% increases in precision and recall on private datasets.
• Improved mAP@0.5 by 3.0% and mAP@0.5:0.95 by 3.5%.
• Demonstrated superior performance compared to YOLOv5, YOLOv7, YOLOv8n, YOLOv9t, YOLOv10n, YOLOv10s, Mask R-CNN, and Mask2Former.

Conclusions:

• The improved YOLOv8n-seg model offers an optimal balance between computational efficiency and detection performance.
• The model shows significant potential for research and development in small intelligent precision operation technology and equipment.