Scholarlink

<p>To better adapt to the non-uniformity variations in infrared images and capture their key features, a super-resolution non-uniformity correction method for substation infrared images based on a convolutional neural network (CNN) with an adaptive learning rate is studied. The contrast of the infrared image is enhanced through an image contrast enhancement module. The convolutional layers are used to extract the non-uniformity features from the enhanced substation infrared image. By incorporating non-local similarity modules via skip connections, a non-local residual convolutional structure is constructed specifically to enhance, extract, and reconstruct non-uniform features in substation infrared images. The accurate capture of non-uniform variations is critical in infrared imaging, as it directly affects the accuracy of temperature field analysis. Unlike conventional CNNs that capture only local features, this method establishes long-range dependencies through non-local operations and can simultaneously model spatial non-uniformity and radiation characteristics. The innovative integration of residual learning and non-local attention mechanisms effectively addresses the limitation of existing methods that ignore global correlations during feature extraction, which often leads to blurred reconstruction in non-uniform regions, and significantly enhances the representation of non-uniformity in infrared images under complex scenes. The original substation infrared image is used to offset the reconstructed non-uniform infrared image, thereby improving image resolution and producing a super-resolved infrared image, which completes the super-resolution non-uniformity correction. The curvature change of the CNN loss function is computed via gradient differences, and an adaptive learning rate is designed. Combined with the gradient descent algorithm, the CNN parameters are optimized, improving the super-resolution non-uniformity correction performance for infrared images. Experiments show that the proposed method can effectively extract non-uniformity features from substation infrared images with improved feature extraction performance. It effectively corrects infrared image non-uniformity, enhances image resolution, and accomplishes super-resolution non-uniformity correction.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Super-Resolution Correction Method for Substation Infrared Image Non-uniformity Based on Adaptive Learning Rate CNN

  • 期刊论文

  • 发表日期：

    2025年12月8日

查看全文
  • Kuiping Ding,
  • Bo Ma,
  • Hailong Zhang,
  • Bo Li,
  • Tian Zuo,
  • Fanlin Ma

摘要

To better adapt to the non-uniformity variations in infrared images and capture their key features, a super-resolution non-uniformity correction method for substation infrared images based on a convolutional neural network (CNN) with an adaptive learning rate is studied. The contrast of the infrared image is enhanced through an image contrast enhancement module. The convolutional layers are used to extract the non-uniformity features from the enhanced substation infrared image. By incorporating non-local similarity modules via skip connections, a non-local residual convolutional structure is constructed specifically to enhance, extract, and reconstruct non-uniform features in substation infrared images. The accurate capture of non-uniform variations is critical in infrared imaging, as it directly affects the accuracy of temperature field analysis. Unlike conventional CNNs that capture only local features, this method establishes long-range dependencies through non-local operations and can simultaneously model spatial non-uniformity and radiation characteristics. The innovative integration of residual learning and non-local attention mechanisms effectively addresses the limitation of existing methods that ignore global correlations during feature extraction, which often leads to blurred reconstruction in non-uniform regions, and significantly enhances the representation of non-uniformity in infrared images under complex scenes. The original substation infrared image is used to offset the reconstructed non-uniform infrared image, thereby improving image resolution and producing a super-resolved infrared image, which completes the super-resolution non-uniformity correction. The curvature change of the CNN loss function is computed via gradient differences, and an adaptive learning rate is designed. Combined with the gradient descent algorithm, the CNN parameters are optimized, improving the super-resolution non-uniformity correction performance for infrared images. Experiments show that the proposed method can effectively extract non-uniformity features from substation infrared images with improved feature extraction performance. It effectively corrects infrared image non-uniformity, enhances image resolution, and accomplishes super-resolution non-uniformity correction.