<p>Maintenance of transmission towers and substations requires professionals to wear safety equipment, so the use of safety equipment is critical to life safety. However, current safety equipment inspection technologies face limitations in detection accuracy and inspection content diversity. Motivated by this, this paper proposes an improved HAF-YOLO inspection model for aerial work safety equipment based on the YOLOv8 architecture to enhance the effectiveness and precision of inspections. First, to avoid increasing the model's computational cost while improving the accuracy of slender and small target detection, we replaced the Conv module in the model with DeepWise Convolution (DWConv). Additionally, GhostBottleneck was used to replace Bottleneck in C2f, and combined with the context anchor attention (CAA) to construct the CAA-GhostBottleneck-C2f (CGC) module. Based on the DWConv and CGC modules, we developed a new backbone network named CGC-CSPNet (Cross Stage Partial Network). Second, in the Neck network part, we added a detection layer for small targets (P2) and removed the layer for large targets (P5) along with the detection head in the downsampling layer. Finally, we designed the Channel Efficient Local Attention mechanism and deployed it at the fusion of shallow and deep features of small target to enhance the model’s detection accuracy for small targets. The observations from the experiment demonstrate that, in the context of the HAF-add dataset, the number of parameters of HAF-YOLO decreases by 42.67% compared to the original YOLOv8n, and improves by 8% on mAP (0.5). In addition, the generalization ability of the conceptualized model is evaluated using the VisDrone-2019 public dataset.</p>