基于深度学习的电力线路航拍图像检索分析方法

Aerial Image Retrieval and Analysis Method of Power Lines Based on Deep Learning

摘要:

电力行业的快速发展使电力线路巡检变得愈加重要。传统巡检方法面临效率低和精度不足等问题，对基于航拍图像的巡检进行研究，提出了一种基于深度学习的图像检索方法。通过构建深度残差注意力网络哈希模型，采用通道和空间注意力模块细化特征，自动识别并聚焦图像中的关键区域；利用基准样本的三元组损失函数进行端到端学习，优化图像间的距离度量，精准区分相似图像。与已有方法的对比实验表明，该系统能够有效地对电力线路航拍图像进行检索和分析，为巡检工作提供支持。

Abstract:

The rapid development of the power industry has made power line inspection increasingly important. Traditional inspection methods face issues such as low efficiency and accuracy. A deep learning-based image retrieval method is proposed for aerial image-based inspection. By constructing a deep residual attention network hashing model, channel and spatial attention modules are used to refine features, automatically identifying and focusing on key areas in the images. A triplet loss function with reference samples is employed for end-to-end learning to optimize the distance metric between images, enabling precise differentiation of similar images. Experimental tests demonstrate the effective retrieval and analysis of aerial images of power lines, providing support for inspection tasks.