基于改进YOLO11-Seg的轻量化端子接线分割模型

Lightweight Terminal Wiring Segmentation Model Based on Improved YOLO11-Seg

摘要: 为了解决复杂背景下端子接线分割模型准确率不足与边缘设备部署困难的问题，本文提出了一种改进的YOLOv11n网络分割模型。该模型引入洗牌网络版本2 (ShuffleNetv2)作为主干网络，结合点群卷积与通道混洗以降低参数量与运算量；同时，采用简化空间金字塔池化模块(Simplified Spatial Pyramid Pooling-Fast, SimSPPF)提取并融合多尺度上下文特征，从而增强模型对不同尺度目标区域的表征能力，提高分割结果的准确性。为了验证改进模型的准确性和快速性，本文构建了包含1290张图像的端子数据集，覆盖多同光照条件、拍摄角度和线缆颜色。实验结果表明，改进后的分割模型在参数量和计算量分别较YOLOv11n网络分割模型降低了18%和16.7%，在此基础上，其检测精度和mAP50-95分别达到98.4%和91.8%。同时，模型在嵌入式平台上的推理速度提升了29.9%，具备良好的实时性表现。在分割精度和轻量化性能上，该模型相较于现有轻量化算法均展现出显著优势，该研究可为线缆智能化检测提供技术支持。

Abstract: To address the issues of insufficient segmentation accuracy and deployment difficulty on edge devices in terminal block segmentation under complex backgrounds, this paper proposes an improved YOLOv11n-based segmentation network. ShuffleNetV2 is introduced as a lightweight backbone to reduce model parameters and computational cost through efficient pointwise convolutions and channel shuffling. Meanwhile, a Simplified Spatial Pyramid Pooling-Fast (SimSPPF) module is employed to enhance multi-scale contextual feature fusion, improving the representation of targets with varying scales. A terminal block dataset containing 1,290 images with diverse lighting conditions, viewpoints, and cable colors is constructed for evaluation. Experimental results demonstrate that the proposed segmentation model reduces the number of parameters and computational complexity by 18% and 16.7%, respectively, compared with the YOLOv11n segmentation model. On this basis, the model achieves a detection accuracy of 98.4% and an mAP@0.5-0.95 of 91.8%. Moreover, the inference speed on an embedded platform is improved by 29.9%, indicating favorable real-time performance. Overall, the proposed model exhibits clear advantages over existing lightweight methods in terms of segmentation accuracy and model efficiency, providing effective technical support for intelligent cable inspection.