The Study Of SAR Image Super-resolution Restoration Base On Deep Learning

The image field has a huge demand for high-quality and high-resolution images,A large number of researchers have conducted extensive research on how to obtain high-resolution images.Although the super resolution method has achieved great success in the field of natural images,there are still many difficulties in remote image super resolution reconstruction.Especially SAR image,it has the advantages of all-day,all-weather work,due to the limitations and influence of the imaging mechanism,we often get SAR image with low resolution.Based on the National special support program for high-level talents of China(The interpretation and object identification of SAR image).This paper study the SAR image identification method for super-resolution problems based on deep learning and generative adversarial network.The main content of this paper is as follow:A super-resolution reconstruction method of SAR image base on convolution is realized.The network uses multi-scale convolution kernel operation to ensure the diversity of features.For the speckle noise interference,the perception loss is added in the network and the semantic-level image reconstruction is achieved.The experimental results show that the reconstructed SAR image has better reconstruction on edge informationA SAR image super-resolution method based on the Generative Adversarial network is realized.This method introduces classification labels as auxiliary information into the training process of the generator so that the super-resolution results are more useful for later image classification.The noise reduction encode GAN network can be used to learn the probability distribution of noisy data.This method can be applied to image super-resolution problems such as SAR images,which have large noise e.The method uses an artificially synthesized noise image for testing.The experimental results show that the network can achieve super resolution reconstruction of images under noisy conditions.A super-resolution image reconstruction base on deep ensemble model is realized.This method constructs a deep ensemble integration network including three channels of coded network,residual network,and convolutional network.By adding an ensemble network layer,an image super-resolution framework that can comprehensively utilize multiple network models is realized.Super-resolution results for real images show that this method can obtain better image super-resolution results than a single network.