发电技术 ›› 2024, Vol. 45 ›› Issue (3): 558-565.DOI: 10.12096/j.2096-4528.pgt.24003

• 智能电网 • 上一篇    下一篇

异构图像融合算法及其在电力设施检测中的应用研究

贾梦涵1, 刘刚1, 徐世杰2, 吴双应2   

  1. 1.上海电力大学自动化工程学院, 上海市 杨浦区 200090
    2.重庆大学能源与动力工程学院, 重庆市 沙坪坝区 400044
  • 收稿日期:2024-03-11 修回日期:2024-05-01 出版日期:2024-06-30 发布日期:2024-07-01
  • 通讯作者: 刘刚
  • 作者简介:贾梦涵(1997),女,硕士研究生,主要研究方向为图像处理与应用,jmh9910@163.com
    刘刚(1977),男,博士,教授,主要研究方向为图像融合、缺陷检测和深度学习等,本文通信作者,liugang@shiep.edu.cn
    徐世杰(1996),男,博士研究生,主要研究方向为太阳能综合利用,xsj9629@163.com
    吴双应(1968),男,博士,教授,主要研究方向为热工理论及其工程应用、太阳能热电转换与利用等,shuangyingwu@126.com
  • 基金资助:
    国家自然科学基金项目(62203224);上海市地方院校能力建设专项计划项目(22010501300)

Heterogeneous Image Fusion Algorithm and Its Application in Power Facility Detection Research

Menghan JIA1, Gang LIU1, Shijie XU2, Shuangying WU2   

  1. 1.Department of Automation, Shanghai University of Electric Power, Yangpu District, Shanghai 200090, China
    2.School of Energy and Power Engineering, Chongqing University, Shapingba District, Chongqing 400044, China
  • Received:2024-03-11 Revised:2024-05-01 Published:2024-06-30 Online:2024-07-01
  • Contact: Gang LIU
  • Supported by:
    National Natural Science Foundation of China(62203224);Shanghai Special Plan for Local Colleges and Universities Capacity Building(22010501300)

摘要:

目的 电力设施的及时、准确检测对保障能源供应的可靠性至关重要,而单一传感器在电力设施检测中存在一定的局限性,为此,提出了一种基于显著性检测的多尺度特征异构图像融合算法。 方法 采用边缘制导网络从红外图像中提取显著目标,生成显著目标掩模;在每个区域建立特定的损失函数,结合显著目标掩模引导网络进行特征提取;基于特征层次的定向异构融合方法,将不同尺度的深度特征进行定向结合,最大限度地减少信息丢失。 结果 在TNO数据集上进行的主观与客观实验表明,该算法在大多数评估指标上优于其他方法,验证了其在电力设施检测领域应用的有效性。 结论 该算法有效解决了检测率较低和信息丢失的问题,使电力设施的检测更全面准确,对提高电力设备故障检测的准确度和诊断效率具有重要意义。

关键词: 电力设施检测, 异构图像融合, 目标检测, 深度学习

Abstract:

Objectives Timely and accurate detection of power facilities is very important to ensure the reliability of energy supply. However, a single sensor has certain limitations in the detection of power facilities. Therefore, a multi-scale feature heterogeneous image fusion algorithm based on saliency detection was proposed. Methods Firstly, the edge guidance network was used to extract the salient target from the infrared image to generate the salient target mask. Secondly, a specific loss function was established in each region, and the salient target mask was used to guide the network for feature extraction. Finally, a directional heterogeneous fusion method based on feature hierarchy was proposed, which combined the depth features of different scales to minimize information loss. Results Subjective and objective experiments on the TNO dataset show that the algorithm is superior to other methods in most evaluation indicators, which verifies the effectiveness of its application in the field of power facilities detection. Conclusions The algorithm effectively solves the problems of low detection rate and information loss, and makes the detection of power facilities more comprehensive and accurate. It is of great significance to improve the accuracy and diagnostic efficiency of power equipment fault detection.

Key words: power facility detection, heterogeneous image fusion, target detection, deep learning

中图分类号: