针对铁塔摄像头视频中挖掘机等工程机械目标极为稀疏、小目标居多、背景复杂、目标遮挡等问题，提出一种改进的YOLOv5s算法，首先将数据增强方式改为mosaic-9，有效增加了小目标样本量；其次，优化了特征提取与特征融合过程，增加了一个检测尺度，提高了小目标的检测精度；最后，引入了SKNet注意力模块，使模型更加关注感兴趣区域，抑制无用信息和复杂背景的干扰，提高算法的检测准确率。实验结果表明，改进后算法AP值达到了79.63%，相对YOLOv5s的AP值提升了6.8%，同时小目标检测精度更高，漏检情况更少。

The construction equipment targets such as excavators in the tower-based video are extremely sparse, mostly small and with complex backgrounds, and part of them is occluded. So, we proposed an improved detection algorithm based on YOLOv5s. Firstly, we changed the data augmentation mode to Mosain-9, to increase the number of small target. Then, we optimized the process of feature extraction and feature fusion,and added a detection scale to improve the detection ability of small targets and feature fusion. Finally, we introduced an attention mechanism to pay more attention to the region of interest, suppress useless information and background noise, and improve the accuracy of target detection.The experimental results show that AP value of the improved algorithm reaches 79.63%, which is 6.8% higher than that of YOLOv5s. At the same time, the detection accuracy of small targets is higher, and the missing target is less.