基于深度学习与视觉显著性的输电线路鸟巢识别  

Deep Learning and Visual Saliency-based Bird Nest Identification for Transmission Lines

作　　者：张辉 ZHANG Hui(PowerChina Guizhou Electric Power Engineering Co.,Ltd.,Guiyang 550081,China)

机构地区：[1]中国电建集团贵州电力设计研究院有限公司,贵州贵阳550081

出　　处：《电工技术》2024年第21期130-135,共6页Electric Engineering

基　　金：中国电建集团贵州电力设计研究院有限公司科技项目“新一代输电线智慧运维关键技术的研究与推广”(编号GZEDKJ-2024-17)。

摘　　要：探索从无人机巡线影像中自动识别鸟巢对于输电线路的安全运行具有重要意义,因此提出了一种融合视觉显著性与深度学习的鸟巢识别方法。首先利用视觉显著性算法提取显著性图,将其与可见光图像进行融合,使得融合图像既具备可见光图像特征信息丰富的优点,又具备显著性图鸟巢目标显著的优点;然后将融合图像输入深度学习模型进行训练,模型构建了特征金字塔以满足多尺度鸟巢目标识别的需要;最后利用分类和回归子模型进行输出,得到鸟巢识别结果。实验结果表明,该方法能准确识别不同背景、塔型、拍摄角度和拍摄距离的图像,稳健性和泛化性较好,Precision、Recall及IoU精度指标值分别为0.9765、0.9651及0.9579,精度要优于几种流行的深度学习模型。Exploring automatic identification method for bird nest from drone inspection images is of great significance to safe operation of transmission lines.Therefore a bird nest identification method combining visual saliency and deep learning is proposed.First the saliency map is extracted by the visual saliency algorithm and fused with the visible image,so that the fused image has the advantages of both rich feature information and significant nest target.Then the fused images are input into the deep learning model for training.The model constructs a feature pyramid to meet the needs of multi-scale bird nest target identification.Finally the classification and regression sub-model are used to output the identification results.The experimental results show that the proposed method can accurately recognize images with different backgrounds,tower types,shooting angles and shooting distances,and has good robustness and generalization.The Precision,Recall and IoU accuracy index values are 0.9765,0.9651 and 0.9579 respectively,superior to several popular deep learning models.

关 键 词：视觉显著性 深度学习 鸟巢识别 图像融合 特征金字塔 

分 类 号：TM755[电气工程—电力系统及自动化]