Infrared image super-resolution reconstruction method based on EDSR network

Abstract

The invention discloses an infrared image super-resolution reconstruction method based on an EDSR network, and the method employs a dual-loss function training network for protecting the edge of a reconstructed image. The method includes the following steps: preprocessing an input image, and performing depth expansion on the input image in combination with an image channel number; constructing a deep learning network structure by adopting a local residual block cascading and global residual mode so as to extract residual features of a low-resolution image and a high-resolution image; taking L2 loss and an edge image difference value of an input image and an output image as a loss function to train a network, so that a high PSNR (peak signal-noise ratio) and a high-resolution infrared image with a clear edge are ensured; and restoring the trained features to the number of input image channels to generate a reconstructed super-resolution infrared image. According to the method, the double-loss function training network is adopted, the super-resolution reconstruction of the infrared image can be quickly completed with high quality, and the method has high application value.

Description

technical field [0001] The invention relates to an infrared image super-resolution reconstruction method based on an EDSR network, belonging to the field of computer vision. [0002] technical background [0003] In recent years, infrared and visible light sensors, as the two most commonly used image sources, are widely used in medicine, remote sensing, power systems, human body detection and tracking and other fields. Visible light imaging systems rely on light reflection imaging, which can provide background information. The obtained images have rich color, geometry, and texture details, but are easily affected by weather conditions such as light and smoke. The infrared camera relies on the temperature difference imaging between the target and the background, and can detect the target in the case of poor light. However, the infrared image is limited by the inherent factors such as the size of the detection unit and the sampling density of the detector array, and there are r...

AI Technical Summary

Problems solved by technology

[0005] The technical problem to be solved by the present invention is: to propose a super-resolution reconstruction method of infrared images based on EDSR network, according to the problem of low spatial resolution of infrared images at present, to achieve 3 times super-resolution of low-resolution infrared images Reconstruction, and realize the edge protection of the reconstructed image through the double loss function, and improve the image quality and visual effect of the infrared image

Images