| Infrared imaging technology plays an important role in the field of public security.However,limited by infrared imaging equipment,infrared images have the features of low resolution and low signal-to-noise ratio in the process of acquisition,compression and transmission,which leads to the blurred visual effect of infrared images,and then affects the quality of infrared images.At this stage,there are still challenges to improve the quality of infrared images.To solve the above problems,this thesis mainly focuses on two algorithms based on deep learning: the first is the infrared image denoising algorithm;The second is the infrared and visible image fusion algorithm.The details are as follows:(1)To remove the noise of infrared image,based on U-Net network and attention mechanism,this thesis proposes an infrared image denoising algorithm based on improved UNet network.Among them,U-Net network is mainly used to solve the pixel problem.This algorithm uses U-Net network to extract the detail noise of infrared image.At the same time,the algorithm combines the attention mechanism to establish the long-distance dependence between features and improve the performance of the network to extract infrared image noise in complex background.The infrared image denoising algorithm in this thesis mainly includes two modules: noise extraction module and residual module.Firstly,the noise of the infrared image is extracted by the noise extraction module,and then the residual module is used to calculate the residual between the extracted noise and the original image to obtain the infrared image with high signal-to-noise ratio.(2)To solve the problem of blurred visual effect of infrared image,this thesis uses the complementary characteristics of infrared image and visible image to fuse them.To further improve the quality of fused images,a fusion algorithm of infrared and visible images based on dense cross network is proposed in this thesis.Firstly,the algorithm uses dense cross network to extract image features;Then an adaptive fusion strategy is designed to fuse the features of the image;Finally,the fused image is obtained by using the reconstruction module.Among them,the dense cross network in this thesis can not only extract the shallow features of the image,but also extract the deep features of the image.In addition,the network combines residual learning to strengthen the transfer between features.At the same time,it can effectively avoid the phenomenon of network degradation when the number of layers of convolutional neural network increases,so as to enhance the performance of the network.The proposed infrared image denoising algorithm based on improved U-Net network and the fusion algorithm of infrared and visible images based on dense cross network are verified respectively,the experimental results show that the infrared image denoising algorithm in this thesis has good image denoising performance in terms of subjective and objective evaluation.At the same time,it is proved that the noise extraction module in this thesis can effectively extract the noise of infrared image.In addition,this thesis compares the denoised infrared image with the non-denoised infrared image for fusion.The subjective and objective evaluation results show that the denoised infrared image is better for fusion.Finally,by comparing the fusion algorithm in this thesis with the advanced fusion algorithm,it is verified that the fusion algorithm in this thesis can better improve the quality of the fused image.At the same time,it is also verified that the dense cross network in this thesis can effectively improve the ability of network feature expression and the fusion effect of infrared and visible images,and overcome the problems of loss of detailed features and blurred visual effect of infrared images. |
