基于YOLOv5s剪枝模型的输电线路全景监测研究

随着人工智能技术的快速发展，基于深度学习的目标检测算法在诸多行业得到了应用。针对当前输电线路影像中典型障碍物目标识别对人工要求较高的问题，提出了基于YOLOv5模型剪枝的算法来对输电线路附近典型目标进行检测。首先，算法进行基础训练后得到一个检测精度和推理速度两种性能比较均衡的网络模型，再进行稀疏训练以获得参数较为稀疏的神经网络模型，最后，采取不同剪枝策略对网络进行修剪，达到压缩模型大小提高推理速度的目的。在自制数据集上使用多种算法进行对比试验，实验结果表明：相较于YOLOv4、CenterNet和SSD算法，所提算法在保持相对较高检测精度条件下提高了检测速度，能够满足实际需要。

Panoramic Monitoring of Transmission Line Based on YOLOv5s Pruning Model

With the rapid development of artificial intelligence technology,target detection algorithm based on deep learning has been applied in many industries.Aiming at the high manual requirement of typical obstacle targets recognition in current transmission line images,a pruning algorithm based on the YOLOv5 model is proposed to detect typical targets near the transmission line.Firstly,a network model with balanced performance of detection accuracy and inference speed is obtained after basic training.Then,sparse training is carried out to obtain a neural network model with relatively sparse parameters.Finally,different pruning strategies are adopted to prune the network to achieve the purpose of reducing model size and improving inference speed.The experimental results show that compared with YOLOv4,CenterNet and SSD algorithm on self-made datasets,the proposed algorithm can improve the detection speed while maintaining relatively high detection accuracy,and the actual needs can be met.