Imaging of complex targets by machine learning based inverse synthetic aperture radar technique

When imaging complex targets with radar systems, these systems are highly affected by environmental noise and cannot provide high resolution images. In order to provide a solution to this situation, a combination of Inverse Synthetic Aperture Radar (ISAR), which is a good signal processing technique, and Convolutional Neural Networks (CNN), a deep learning based structure, are used. In this way, it is aimed to overcome the problems of traditional TAR imaging methods such as lack of time and data. A total of 400 backscattered electric field data is collected from 7 different complex targets illuminated by electromagnetic waves at different angles and polarizations covering the X band (8-12 GHz) using ANSYS HFSS software. These data are processed with MATLAB to create ISAR images with the scattering center and used as input image data for the CNN structure with the help of range-Doppler. The CNN structures are trained to overcome the noise problem and to improve the resolution. Two different CNN structures consisting of 21 layers and 11 lavers are built and compared to solve the noise problem. Then, the reconstructed images with the scattering center and the unconstructed images are compared using these structures. Finally, these images are randomly sampled and distorted to solve the resolution problem and the training results are compared. As a result of these three different tests, CNN structures yielded successful image results and demonstrated that effective results can be obtained with less data in future studies.