A thermal infrared image super-resolution reconstruction method

Abstract

The embodiment of the application provides a kind of thermal infrared image super-resolution reconstruction method.Application in computer vision technical field, this method is by obtaining image dataset, image dataset includes high-resolution visible light image and high-resolution visible light image corresponding high-resolution thermal infrared image;Bilinear downsampling operation is executed to high-resolution thermal infrared image, and low-resolution thermal infrared image is obtained;High-resolution visible light image is input into fine-tuned diffusion model and is analyzed and handled, and pseudo high-resolution thermal infrared image is obtained;Low-resolution thermal infrared image, high-resolution visible light image and pseudo high-resolution thermal infrared image are input into thermal infrared image reconstruction network and are analyzed and handled, and the high-resolution thermal infrared image of reconstruction is obtained, and the thermal infrared image reconstruction network includes: shallow feature extraction module, Mamba coding module, Cross-Mamba cross-modal fusion module, feature fusion module and decoder module.The method improves the accuracy of reconstructed thermal infrared image.

Images