Title:Single Infrared Image Super-resolution And Enhancement Based On Fusion ESRGAN And Gradient Network

Infrared imaging systems can capture objects' infrared images which show good imaging effects penetrability in the conditions of insufficient light such as night,rain,fog,and other complex weather conditions.And infrared imaging systems now widely used in military,medical,public safety and other fields.In the vehicle-mounted safety assistance system,the lnfrared imaging system can assist drivere to understand the road conditions in front of the vehicle clearly,and avoid the dangers timely in light changes and adverse weather effects.The internal space of the vehicle is limited,so the infrared imaging system should not be too large,and image size is limited.Research on the reconstruction and enhancement of in-vehicle infrared images can improve the quality and resolution of infrared images,which will help reduce the probability of car accidents and improve driving safcty.In this paper,the high-resolution infrared image data set at FLIR-ADAS collected on the driving recorder is used for image degradation processing of center-point motion blur,The degradation process includes motion blur and image downsampling.In the paper,the degraded image is used as the low-resolution image of the algorithm,and the source image is the ideal reference image for the reconstructed image.This paper proposes an infrared image rcconstruction and enhancenent algorithum based on a fusion gradient network.First,ESRAN network reconstruction is pcrformed on the low-resolution image,and the optimization parameters are adjusted to generate the initial high-resolution image.The gradient conversion module uses the input low-resolution image using the sobel operator,The gradient layer is extracted,and the low-resolution gradient image is input into the gradient reconstruction network to generate the gradient high-resolution image.Infrared image super-resolution reconstruction based on fusion gradient network has designed a fusion network for the fusion module.Fusion network input the infrared image and gradient image,then generates a high-resolution image through the adjustment layer and the reconstruction layer.It can realize high-resolution reconstruction of infrared image of crystalline silicon photovoltaic cells,reduce the requirements on the resolution of infrared imager,cut down cost,reduce the excitation intensity and action time of electromagnetic induction,and improve the detection rate of defects in photovoltaic cells,and verifies the effectiveness of inputting gradient image to network can enhance the infrared image high-frequency details.In order to enhance the details of the reconstructed image,the fusion of the rolling guide filter is proposed,in which the original detail layer and the original foundation of the fusion image are designed by the fusion module.The layer extracts the image saliency map,and uses the guide filter to construct the weight map of the fused image,which is fused into the final high-resolution image.The fused image is expanded to 4 times that of the low-resolution image,and the edge contour of the image is clear,image enhanced,and contrast enhanced.