为了提升高空螺母的识别准确率，减少螺栓螺母的误检、漏检率，提出了一种基于改进YOLOv5的高空螺母识别模型。首先，在骨干网络端添加了高效多尺度注意力机制（efficient multi-scale attention, EMA），以此融合更多的信息。其次，将颈部网络的PANet更换为BiFPN（bidirectional feature pyramid network），以加强网络的特征提取能力。最后，将原损失函数CIoU（complete intersection-over-union）更换为SIoU（structured intersection over union），以加快模型的收敛速度并提高模型的分类准确率。结果表明，相比于YOLOv5原模型，改进后的模型拥有更好的性能，其中准确率提升了0.92%，召回率提升了0.16%，平均精度1（mAP_0.5）提升了0.53%，平均精度2（mAP_0.5：0.95）提升了2.26%。再用改进前后的模型进行实际识别对比实验，结果表明，改进后的模型识别效果更好，漏检、误检率下降，实际的识别率更高。改进后的模型能够很好地满足高空螺母的识别和图像数据采集，也为后续的螺母维护提供了数据基础。

In order to improve the recognition accuracy of high-altitude nuts and reduce the false detection rate of bolts and nuts, a high-altitude nut recognition model based on improved YOLOv5 was proposed. Firstly, a new attention mechanism efficient multi-scale attention(EMA) was added to the backbone network to integrate more information. Secondly, in order to enhance the network's feature extraction capability, bidirectional feature pyramid network (BiFPN) was used to replace the PANet of the neck network. Finally, structured intersection over union(SIoU) was used to replace the original loss function complete intersection-over-union(CIoU) to accelerate the convergence of the model and improve its classification accuracy. The results show that the improved model has better performance than the original YOLOv5 model. The accuracy of the improved model increases by 0.92%. The recall increases by 0.16%. The average precision 1 (mAP_0.5:0.5) increases by 0.53%. And the average precision 2 (mAP:0.95) increases by 2.26%. An actual recognition comparison experiment between the improved model and the original YOLOv5 model was carried out. The experimental results show that the improved model has better recognition performance, which reduces the missed detection rate and the false detection rate, and improves the actual recognition rate. The improved model can well meet the recognition and image data acquisition of high-altitude nuts. And it also provide a data foundation for subsequent nut maintenance.