The Research On Key Technologies Of Improving Infrared Image Quality Based On Deep Learning

With the increasing development and maturity of infrared imaging detection technology,based on its passive covert detection,strong anti-interference,all-time work and other characteristics,infrared imaging technology has been widely used in security,industrial detection,forest detection,power industry,aerospace,medical diagnosis and other fields.However,due to the low frequency of the infrared band and the small size of the focal plane array of the core device,the image obtained by infrared imaging technology has some problems,such as poor contrast,low resolution,obvious noise and so on.In order to avoid the defects caused by the above technical problems,improve the image quality of infrared imaging,and obtain infrared images with high contrast,clear details and high resolution,the key technologies of infrared image quality improvement based on depth learning are studied in this paper.The main contents are as follows.In order to solve the problem of infrared image non-uniformity fringe noise caused by the column signal readout mode of infrared detector,a non-uniformity correction method of infrared image based on convolution neural network is studied and realized.First of all,the convolution neural network is used to extract the features of the input image to learn the residual information between the noisy image and the noiseless image,and then the extracted feature image is reconstructed after nonlinear mapping.Finally,the noisy image is denoised based on the noisy image,and the final non-uniform correction result is obtained after filtering the noise.The performance of the method is verified by simulation experiments,and verified by several groups of data.Compared with the traditional correction methods,this method improves the image roughness and other performance indicators of the processing results.Aiming at the problems of high background,low contrast and blurred target outline of infrared image,an infrared image enhancement method based on generating countermeasure network is studied and realized.First of all,the original infrared image is input into the generator,and the enhanced results are output through feature extraction,nonlinear mapping and image reconstruction,and then the discriminator evaluates the output results of the generator.The quality of the output image of the generator is improved through multiple iterative learning,and the image results with high contrast and enhanced details are obtained.The joint loss function training network is used to better retain the semantic information of the image and improve the enhancement effect of the infrared image.Several infrared images of scenes are used to verify the performance of the algorithm.the results show that,compared with the traditional enhancement methods,the enhancement results of this method are richer in the expression of detail information,and the performance indexes such as image average gradient are significantly improved.Aiming at the problem of low spatial resolution of infrared image caused by the small size of infrared detector array,based on the advantage that residual structure can prevent gradient from disappearing when the network is deepened,an infrared image super-resolution reconstruction method based on generating countermeasure network is studied and implemented.First of all,in order to filter out the artifacts that may be introduced in the super-resolution process and improve the generalization ability of the network model,the batch normalization layer of the original residual module in the generator network is removed.then the number of input channels in the activation layer in the residual module is expanded to obtain the low-level feature information of the infrared image,which can not only reduce the computational complexity of the algorithm,but also improve the reconstruction quality and realize the end-to-end super-resolution reconstruction of the infrared image.The performance of the reconstruction method is verified by several groups of infrared image experiments.the results show that this method can effectively improve the spatial resolution and subjective visual effect of the infrared image,and the details of the reconstructed image are more abundant.