基于改进YOLOv4算法的高压输电线路异物检测

Foreign object detection of high voltage transmission line based on improved YOLOv4 algorithm

针对传统输电线异物检测精度较低、误检率高等问题,提出了一种基于改进YOLOv4的高压输电线路异物检测算法,通过深度学习中的卷积神经网络让模型自主地关注和学习高压输电线上的异物特征。首先采集输电线路上异物图像,并采取多种数据增强方式构建高压输电线异物检测数据集;其次采用k-means聚类生成适配本数据集的锚点框,提升模型鲁棒性;再将网络的空间金字塔池化(spatial pyramid pooling,SPP)模块设计优化为快的空间金字塔池化(spatial pyramid pooling-fast,SPPF)模块,更高效地完成特征融合,加快网络运行速度;最后结合标签平滑、余弦退火衰减算法,在路径聚合网络和预测层部分采用Sigmoid加权线性单元(Sigmoid weighted linear unit,SiLU)激活函数替代普通卷积中的Leaky ReLU激活函数,更有利于模型收敛,使模型性能得到进一步提升。结果表明,模型在不增加参数量的前提下,精度提高到了97.57%,帧传输率达到42.4 f/s,满足实时检测的要求。

Aiming at the problems of low detection accuracy and high false detection rate of traditional transmission lines,a foreign object detection algorithm based on improved YOLOv4 is proposed in this paper,which makes the model autonomously pay attention to and learn the features of foreign objects on high-voltage transmission lines.It first collects images of foreign objects on the transmission line,and adopts a variety of data enhancement methods to construct a foreign object detection data set;then uses K-means clustering to generate an anchor frame adapted to this data set to improve the robustness of the model.Next,the algorithm optimizes the design of the spatial pyramid pooling(SPP)module of the network to a spatial pyramid pooling-fast(SPPF)module,which completes the feature fusion more efficiently and speeds up the network operation.Finally,the label smoothing and cosine annealing decay algorithms are combined,and the Sigmoid weighted linear unit(SiLU)activation function is used in the path aggregation network and prediction layer to replace the Leaky ReLU activation function in standard convolution,which is more conducive to model convergence,and further improves the performance.The result shows that the accuracy of the model is increased to 97.57%without increasing the amount of parameters,and the frame transmission rate reaches 42.4 f/s,which meets the requirement of real-time detection.