基于改进YOLOv8的轻量化棉种外观品质识别方法OA

In the field of cotton seed appearance quality inspection,the widely adopted detection models are difficult to deploy on resource-constrained edge devices due to their large parameter count and high computational costs.To address this,this paper proposes a lightweight solution based on an improved YOLOv8n.By introducing the EMA Attention Mechanism into the backbone network,the model's ability to capture quality-related visual features is significantly enhanced.The Bottleneck layer in the C2f module is replaced with the lightweight StarBlock module,reducing the model's parameter count and improving computational efficiency.A Bidirectional Feature Pyramid Network(BiFPN)architecture is applied in the neck region,achieving more efficient and adaptive multi-scale feature fusion.Experimental results show that the improved YOLOv8n achieves an average precision of 92.7%,a 1%increase over the original version.The computational complexity is reduced by 0.9 GFLOPs,and the model size is reduced by 37.2%.Compared to other object detection models,this solution demonstrates superior computational efficiency and real-time performance while maintaining high accuracy.