Scholarlink

<p>Infrared composite material imaging is essential in the inspection of industries, detection of defects, and thermal monitoring, but the lack of spatial resolution and noise can usually impair the image quality and reduce the accuracy of the analysis. In this case, this paper suggests an Attention-based Infrared Composite Image Improved Super-Resolution Generative Adversarial Network (Attention-ICI-ISRGAN) to improve the quality of infrared image reconstruction. The proposed model combines Residual-in-Residual Dense Blocks (RRDB), perceptual loss, adversarial loss, and attention mechanism so as to enhance feature extraction, structural details, and attention to important infrared features. The model was trained and tested on the large-scale Inf-590&#xa0;K Infrared Image Dataset, which consists of about 590,000 infrared images in different thermal modes, composite material structures, and environmental conditions. To provide sound performance testing, the dataset was divided into 70 per cent training, 15per cent validation, and 15 per cent testing subsets. Experimental findings indicate that the proposed Attention-ICI-ISRGAN is much better than traditional and deep learning-based super-resolution, such as Bicubic, SRCNN, SRGAN, EDSR, and Lightweight CNN. The model proposed attained the highest signal-to-noise ratio (PSNR) of 38.92&#xa0;dB, structural similarity index (SSIM) of 0.962, and root mean square error (RMSE) of 0.0115, and this means high reconstruction accuracy and structural preservation. Besides, the model scored a total reconstruction error of 97.28% and the inference time of 0.058&#xa0;s per image, which indicates high performance and reliability. The combination of the attention mechanism enhanced the feature localization and reconstruction quality as important infrared areas were strengthened selectively. The findings prove that the suggested Attention-ICI-ISRGAN is a strong and efficient method of infrared image super-resolution, and it is much more convenient in practice in the context of composite materials examination, industrial monitoring, and other intelligent thermal imaging systems.</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

An evaluation of super resolution reconstruction quality for infrared composite material images using deep neural networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月3日

查看全文
  • Jing Xiao,
  • Yubo Zhang,
  • Chong Liu,
  • Haimei Liu,
  • Yalei Dong

摘要

Infrared composite material imaging is essential in the inspection of industries, detection of defects, and thermal monitoring, but the lack of spatial resolution and noise can usually impair the image quality and reduce the accuracy of the analysis. In this case, this paper suggests an Attention-based Infrared Composite Image Improved Super-Resolution Generative Adversarial Network (Attention-ICI-ISRGAN) to improve the quality of infrared image reconstruction. The proposed model combines Residual-in-Residual Dense Blocks (RRDB), perceptual loss, adversarial loss, and attention mechanism so as to enhance feature extraction, structural details, and attention to important infrared features. The model was trained and tested on the large-scale Inf-590 K Infrared Image Dataset, which consists of about 590,000 infrared images in different thermal modes, composite material structures, and environmental conditions. To provide sound performance testing, the dataset was divided into 70 per cent training, 15per cent validation, and 15 per cent testing subsets. Experimental findings indicate that the proposed Attention-ICI-ISRGAN is much better than traditional and deep learning-based super-resolution, such as Bicubic, SRCNN, SRGAN, EDSR, and Lightweight CNN. The model proposed attained the highest signal-to-noise ratio (PSNR) of 38.92 dB, structural similarity index (SSIM) of 0.962, and root mean square error (RMSE) of 0.0115, and this means high reconstruction accuracy and structural preservation. Besides, the model scored a total reconstruction error of 97.28% and the inference time of 0.058 s per image, which indicates high performance and reliability. The combination of the attention mechanism enhanced the feature localization and reconstruction quality as important infrared areas were strengthened selectively. The findings prove that the suggested Attention-ICI-ISRGAN is a strong and efficient method of infrared image super-resolution, and it is much more convenient in practice in the context of composite materials examination, industrial monitoring, and other intelligent thermal imaging systems.