在公共安全、**等领域高分辨率热红外图像能够提供更多的场景细节信息, 有着广泛的应用需求,但高昂的设备成本限制了高分辨率红外图像的获取。为此设计了一种多级跳线深层残差卷积神经网络(DR-CNN), 通过软件超分辨的方法重构出高分辨率的红外图像。采用多级跳线双通道注意力残差块增加卷积深度以解决卷积层间缺乏关联性的问题; 使用Concat模块实现局部特征信息的融合, 利用反卷积层进行特征图像的上采样, 使其直接从低分辨率图像学习到高分辨率图像以降低训练的复杂度, 加快运行速度。所提算法与SRCNN, FSRCNN和ADSR等算法进行对比测试, 使用峰值信噪比(PSNR)和结构相似度(SSIM)作为算法的评价指标。实验结果表明提出的RD-CNN算法优于其他对比算法, 生成的高分辨率图像细节丰富且清晰。

In public security and other fields, high-resolution thermal infrared image can provide more scene details, and has a wide range of application requirements.However, high equipment cost limits the acquisition of high-resolution infrared images.In this paper, a multistage skip Deep Residual Convolutional Neural Network (DR-CNN) is designed to reconstruct high resolution infrared images by software super-resolution method.Multistage skip dual-channel attention residual blocks are used to increase the convolution depth to solve the problem of lack of correlation between convolution layers.In order to reduce the complexity of training and speed up the operation, Concat module is used to realize the local feature information fusion and the deconvolution layer is used to upsample the feature images directly from low resolution images to high resolution images.Compared with SRCNN, FSRCNN and ADSR, the Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM) are used as evaluation indexes.Experimental results show that the DR-CNN algorithm proposed is superior to other algorithms, and the generated high-resolution images are rich in detail and clear.