面向实验教学场景导线端点轻量化检测算法

Lightweight detection algorithm for wire endpoints for experimental teaching scenarios: MHS-YOLO

摘要:

导线端点检测是实现电气实验接线检查的核心前提。针对现有电气实验场景中导线端点尺寸小、背景复杂且遮挡频繁导致的漏检、误检等问题，提出一种改进YOLOv11n轻量化检测算法MHS-YOLO:使用多尺度卷积块MSCB优化C3k2模块，增强对导线端点局部区域的细节感知与上下文信息捕捉能力；针对原始HSFPN结构中空间感知能力不足的问题，融合CAA注意力机制,在通道注意力基础上使用锚点引导的空间注意力分布，实现导线端点区域的二次加权与全局语义依赖建模。在检测头之前引入SEAM注意力模块，通过融合多尺度路径的局部细节与全局上下文信息，提升遮挡或模糊区域导线端点的可检测性。实验结果表明，改进模型在自建导线端点数据集上，平均精度达到98.6%，对比基准模型提高了1.8%，参数量、浮点计算量和模型体积分别降低了26.1%、11.1%和23.1%，兼顾精度与轻量化需求，为实验接线检查提供了技术支撑。

Abstract:

Wire-endpoint detection is essential for wiring inspection in electrical experimental teaching, where small target size, complex backgrounds, and frequent occlusions often result in missed or false detections. This paper proposes MHS-YOLO, an improved YOLOv11n-based detection method. The C3k2 module is enhanced using a Multi-Scale Convolutional Block (MSCB) to improve local detail perception and contextual feature extraction. To address spatial awareness limitations in the original HSFPN, a Context Anchor Attention (CAA) mechanism is integrated, introducing anchor-guided spatial attention to reinforce regional weighting and global semantic modeling. Additionally, a Self-Ensemble Attention Mechanism (SEAM) is employed before the detection head to improve endpoint detectability under occlusion and blur by fusing multi-scale local and global features. Experiments on a custom wire-endpoint dataset demonstrate a mean average precision of 98.6%, outperforming the baseline by 1.8%, while reducing parameter count, GFLOPs, and model size by 26.1%, 11.1%, and 23.1%, respectively. These results validate the method"s effectiveness in balancing accuracy and lightweight deployment for experimental wiring inspection.