Super-resolution reconstruction of infrared images based on convolutional neural network

Due to the technical limitations of devices and processes, the resolution of infrared images is relatively low compared with that of visible light images, and there are shortcomings such as blurred detailed texture features. In this paper, a super-resolution method for infrared images is proposed based on convolutional neural network. The method consists of a two-branch infrared image information extraction module by combining dense connections and the multi-scale attention residual block proposed in this paper, and the module is used to construct an infrared image super-resolution network. Experimental results show that the infrared images reconstructed by this paper's method outperform the SRCNN, EDSR, RDN, and MFDRN super-resolution reconstruction models based on deep learning in terms of peak signal-to-noise ratio. The average PSNR of the reconstructed images is improved by 2.27 dB and 2.14 dB over the traditional Bicubic interpolation method at scale factors of x2 and x4, respectively.