| 引用本文: | 赵洪山,王惠东,刘婧萱,等.考虑局部纹理特征和全局温度分布的电力设备红外图像超分辨率重建方法[J].电力系统保护与控制,2025,53(02):89-99.[点击复制] |
| ZHAO Hongshan,WANG Huidong,LIU Jingxuan,et al.Super-resolution reconstruction method for infrared images of power equipment considering local texture features and global temperature distribution[J].Power System Protection and Control,2025,53(02):89-99[点击复制] |
|
| 摘要: |
| 针对传统电力设备红外图像超分辨率重建方法缺乏对设备局部纹理特征和全局温度分布的考虑导致重建后图像分辨率较低的问题，提出一种基于Transformer-GAN聚合网络的电力设备超分辨率重建方法。首先，基于移位卷积设计电力设备局部特征提取模块，在不增加参数情况下扩展卷积的感受野，提取电力设备局部纹理及其周围不同空间维度特征的信息。然后，引入全局特征提取模块，通过深度卷积和空间注意力机制捕捉图像不同区域间温度分布的关联性。最后，采用UNet编解码器网络融合各层局部特征和全局表示，生成清晰自然的电力设备红外图像。算例结果表明，所提方法的峰值信噪比(peak signal-to-noise ratio, PSNR)、结构相似性(structural similarity, SSIM)、和视觉信息保真度(visual information fidelity, VIF)三项评价指标均优于其他方法。同时它具有良好的主观视觉效果，泛化能力较强。 |
| 关键词: 电力设备 红外图像 超分辨率重建 局部纹理特征 全局温度分布 Transformer-GAN |
| DOI：10.19783/j.cnki.pspc.240741 |
| 投稿时间：2024-06-15修订日期：2024-07-24 |
| 基金项目:国家自然科学基金项目资助(52077078)；国家电网公司科技项目资助(52018K22001P) |
|
| Super-resolution reconstruction method for infrared images of power equipment considering local texture features and global temperature distribution |
| ZHAO Hongshan1,WANG Huidong1,LIU Jingxuan1,YANG Weixin2,LI Zhonghang1,LIN Shiyu1,YU Yang1,LÜ Tingyan1 |
| (1. School of Electrical and Electronic Engineering, North China Electric Power University, Baoding 071003, China;
2. State Grid Jibei Electric Power Research Institute, Beijing 100045, China) |
| Abstract: |
| To address the problem of low resolution in reconstructed image due to the lack of consideration for local texture features and global temperature distribution in traditional super-resolution reconstruction methods for infrared images of power equipment, a super-resolution reconstruction method based on Transformer-GAN aggregation network is proposed. Firstly, a local feature extraction module for power equipment is designed based on shift convolution, which expands the receptive field of the convolution without increasing parameters, extracting local texture and surrounding spatial dimension features of the power equipment. Then, a global feature extraction module is introduced to capture the correlation of temperature distributions between different regions of the image through deep convolution and spatial attention mechanisms. Finally, the UNet encoder-decoder network is used to fuse the local features and global representations at each layer, generating clear and natural infrared images of power equipment. The case study results show that the proposed method outperforms other methods in terms of peak signal to noise ratio (PSNR), structural similarity (SSIM), and visual information fidelity (VIF). It also has good subjective visual effects and strong generalization ability. |
| Key words: power equipment infrared image super-resolution reconstruction local texture features global temperature distribution Transformer-GAN |
