【摘要】 为了满足光学航拍图像中电力线检测的实时性和高精度，提出了一种轻量级FastUnet网络电力线检测方法。它以Unet语义分割网络为基础，添加金字塔池化结构增强特征上下文信息的融合。设计深度可分离残差卷积运算，增加了网络深度且进一步减少了网络参数量。使用多损失函数训练Fast-Unet网络，缓解图像中前景与背景类别分布极度不平衡的问题。实验结果表明，相较于Unet算法，模型参数量大幅减少，运算速度明显提升。Fast-Unet满足了实际应用需求，且模型参数体积得到了有效压缩，更容易部署于各种嵌入式系统，对于提高直升机与无人机的低空飞行安全有一定的现实意义。更多还原

【Abstract】 A lightweight Fast-Unet network power line detection method is proposed to meet the real-time and high-precision power line detection in optical aerial images. It is based on the Unet semantic segmentation network, adding a pyramid pooling structure to enhance the fusion of feature context information. The design depth can separate the residual convolution operation, which increases the network depth and further reduces the amount of network parameters. The Fast-Unet network is trained with multiple loss functions to alleviate the extremely unbalance distribution of foreground and background categories in the image. The experimental results show that compared with the Unet algorithm, the amount of model parameters is greatly reduced, and the calculation speed is significantly improved. Fast-Unet meets the actual application requirements, and the model parameter volume is effectively compressed, making it easier to deploy in various embed systems, which has certain practical significance for improving the safety of helicopters and UAVs in low-altitude flight.更多还原