Research On High Resolution Radar Imaging Based On Deep Learning

Due to the capability of long-distance observation with the all-day and all-weather,radar is widely applied in civil and military fileds.For present the target accurately,improving the radar imaging quality is significant.Radar transmitted the linear frequency modulation(LFM)signals to obtain the high range resolution,which is proportional to the bandwidth of transmitted signal.High azimuth resolution can be achieve by coherently integrating the observations of radar signal.The quality of radar imaging not only be restricted by radar imaging system,but also be influented by the trandictional radar imaging method.In the paper,we apply the deep learning to improve the radar imaging quality.The mainly research include the ISAR super-resolution imaging,through-the-wall radar image enhancement,through-the-wall super-resolution imaging and SAR image enhancement.In first part,we research the inverse synthetic aperture radar(ISAR)super-resolution imaging based on deep learning.A general ISAR imaging based on deep learning is built.Some importance factors affecting the deep learning super-resolution imaging are analyzed.For the ISAR imaging of maneuvering target,the training data is generated by simulation method.Then,a suitable convolutional neural network(CNN)model is designed to fit the mapping function between the low-resolution time-frequency distribution and high-resolution distribution.Finally,the trained CNN model is intergeted into the process of ISAR maneuvering target imaging,which effectively improves the resolution of the ISAR maneuvering target imaging.In second part,we research through-the-wall radar imaging based on deep learning.An image enhancement model and super-resolution model based on deep learning are established.For through-the-wall radar image with gating-lobes and side-lobes,the radar image with gating-lobes and side-lobes is generated as the input image.Then,a suitable convolutional neural network(CNN)model is built to fit the mapping function between the high-quailty radar image and low-quailty radar image.Finally,the trained CNN model is intergeted into the process of through-the-wall radar imaging,which effectively improves the quality of through-the-wall radar imaging.For the problem of low resolution in radar images,this paper also establishes a mapping relationship between low-resolution radar images and high-resolution radar images through a neural network modelIn third part,we research the synthetic aperture radar(SAR)imaging based on deep learning.An image enhancement model based on deep learning is established.Some importance factors affecting the deep learning super-resolution imaging are analyzed.For the defocus problem of SAR imaging,the training data is generated by simulation method.Then,a suitable convolutional neural network(CNN)model is designed to fit the mapping function between the original SAR images and defocus SAR images.Finally,the trained CNN model is intergeted into the process of SAR imaging,which effectively improves the quality of the SAR imaging.