【摘要】 针对建筑工地场所范围广、环境复杂、人工巡检效率低的问题，提出一种以无人机为移动巡检单元，实现施工区域安全生产设施全覆盖的动态巡查方法，以突破传统固定监控的局限，为后续智能监测提供高质量、高时效性的数据输入。针对无人机视角下存在的检测目标尺度变化大、目标遮挡严重的问题，提出一种以YOLOv7模型为基础的改进目标检测方法。首先，在主干网络末端与各尺度预测前嵌入置换注意力（SA）模块，构建小目标特征增强结构，提升无人机采集图像中小尺度目标的检测精度；然后，融合CoTNet与高效层聚合网络（ELAN）网络，加强特征提取能力，同时，充分利用上下文信息与全局依赖关系，改善遮挡目标的识别性能；最后，通过消融实验及对比试验对该方法的性能进行验证。结果表明，该方法能够在复杂施工环境中有效完成安全防护措施检测。更多还原

【Abstract】 To address the issues of large area coverage, complex environments, and low efficiency of manual inspections at construction sites, a dynamic inspection method utilizing unmanned aerial vehicles as mobile inspection units is proposed to achieve full coverage of safety production facilities in construction areas. This approach breaks through the limitations of traditional fixed monitoring systems and provides high-quality, timely data input for subsequent intelligent monitoring. In response to challenges of significant scale variations and severe target occlusion in perspectives from unmanned aerial vehicles, an improved object detection method based on the YOLOv7 model is introduced. Firstly, a Shuffle Attention(SA) module is embedded at the end of the backbone network and before each scale prediction to construct a small-target feature enhancement structure, thereby improving the detection accuracy of small-scale targets in images captured by unmanned aerial vehicles. Then, the CoTNet and Efficient Layer Aggregation Network(ELAN) are fused to strengthen feature extraction capabilities while fully leveraging contextual information and global dependencies to enhance the recognition performance of occluded targets. Finally, the performance of this method is validated through ablation experiments and comparative tests. Results demonstrate that this method can effectively accomplish the detection of safety protection measures in complex construction environments.更多还原