Super-resolution Reconstruction Of Infrared Images Based On Convolutional Neural Networks

Image Super-resolution technology refers to the reconstruction of one or several lowresolution images into a high-resolution image.This method reduces the requirements for image acquisition equipment and can obtain higher-resolution images by image processing methods.Infrared images are important sources of information in military,medical,security and other fields.However,infrared images usually contain low resolution,with which human eyes cannot obtain a good visual experience.Traditional infrared image processing methods can improve the quality of infrared images to a certain extent.But,the quality can still be improved by modern technologies.With the development of Deep Learning,the current image superresolution technology based on convolutional neural networks(CNN)has achieved a lot of results in the field of visible light images.But in the field of infrared images,work has not been done properly yet.Therefore,this article proposes to apply image super-resolution technology based on CNN to infrared images.The main contents include:Three image super-resolution reconstruction algorithms based on CNN are introduced in this paper and an infrared image test data set is proposed.The three reconstruction algorithms are applied on infrared images to observe the differences of reconstruction ability.Experiments show that the image super-resolution reconstruction algorithm can also obtain better detail recovery capabilities than traditional algorithms on infrared images.However,for some lowcontrast infrared images,it is still difficult for human eyes to see the detailed information directly after super-resolution reconstruction.Therefore,the low-contrast infrared images should be pre-processed before image super-resolution.Image enhancement preprocessing is applied to certain low-contrast images.Experiments show that the pre-processed low-contrast infrared images show better results on image super-resolution,which obtain better visual effects,contain more high-frequency information and can effectively improve image quality.An improvement of image super-resolution reconstruction algorithm is proposed,combining the channel attention mechanism and the residual network.The channel attention mechanism is added to the residual structure to increase the level of the convolutional neural network so that the network can learn the weights and bias of different channels by itself.Different weights improve the network's ability to extract image features.The deeper the network,the better the generalization ability is.The residual block is used to learn the residual information between high and low frequency,which reduces the computational cost during the training of the deep learning model.It also helps reduce the probability of gradient disappearance and gradient explosion during training.Experiments show that the proposed algorithm can obtain better image reconstruction results in the super-resolution reconstruction of infrared images.Using this algorithm to perform super-resolution reconstruction of infrared images and videos,human eyes can intuitively see that the details(such as edges)in the images are improved,and it has better image reconstruction capabilities than the SR algorithms introduced in this article.