输电塔作为电力输送网络的关键基础设施，其结构安全性直接关系到电网的稳定运行。在长期服役过程中，输电塔螺栓受风荷载、温差效应及材料老化等多因素耦合作用逐渐松动。本文提出了一种基于改进YOLOv5s的输电塔螺栓松动智能检测模型（CCSGS-YOLO）。采用坐标卷积（Coordinate Convolution）替代主干网络中的标准卷积层，增强模型对目标位置信息的获取能力；引入卷积注意力模块（CBAM），通过通道与空间双重注意力机制强化模型在复杂背景下的特征鉴别能力；构建Slim-Neck特征融合结构，通过跨阶段部分连接与深度可分离卷积的优化组合，维持检测精度的同时减少计算复杂度；采用GIoU损失函数与Soft-NMS的联合优化策略，通过考虑预测框与真实框的重叠几何特性，提升目标检测的定位精度。结果表明，CCSGS-YOLO精确率达91.7%，召回率为89.4%，平均精度均值达到95.3%，F1分数提升至90.0%，较基准模型YOLOv5s分别提高1.6%、3.0%、1.4%和1%。在计算效率方面，模型检测速度达到74.8 f/s，推理时延降低至13.4ms，较原模型提升11.6%。通过现场实验对CCSGS-YOLO在不同场景下的检测鲁棒性进行了验证，为输电塔螺栓松动的智能巡检提供了一种新的思路。

As critical infrastructure of the power transmission network, the structural safety of transmission towers directly influences the stable operation of the power grid. During long-term service, transmission tower bolts loosen gradually due to multi-factors coupling such as wind load, temperature difference effect, and material aging. A transmission tower bolt looseness intelligent detection model (CCSGS-YOLO) based on improved YOLOv5s is proposed in the paper. The coordinate convolution is used to replace the standard convolution layer in the backbone network to enhance the model's ability to obtain target location information. The convolutional block attention mechanism (CBAM) is introduced to strengthen the model's feature discrimination ability in complex backgrounds through a dual attention mechanism of channels and space. A Slim-Neck feature fusion structure is designed, which optimally combines cross-stage partial connections and depth-separable convolutions to maintain detection accuracy while reducing computational complexity. A combination of the Generalized Intersection over Union (GIoU) loss function and the Soft Non-Maximum Suppression (Soft-NMS) algorithm is used to optimize the positioning accuracy of object detection by considering the geometric characteristics of the overlap between the predicted box and the true box. The results show that CCSGS-YOLO achieves a precision of 91.7%, a recall of 89.4%, a mean average precision of 95.3%, and an F1-score of 90.0%, representing improvements of 1.6%, 3.0%, 1.4%, and 1% over the baseline model YOLOv5s, respectively. In terms of computational efficiency, the model achieves a detection speed of 74.8 FPS and reduces inference latency to 13.4 ms, which is an 11.6% improvement over the baseline model. The detection robustness of CCSGS-YOLO in different scenarios is verified through field experiments, which provides a new idea for intelligent inspection of transmission tower bolt looseness.