ISAR Imaging in the Presence of Quasi-Random Multiplicative Noise Using Convolutional Deep Learning

Inverse Synthetic Aperture Radar (ISAR) imaging methods are well established. These methods utilize a variety of tools to estimate the spatial distribution of target energy from measurements in the k-space domain. These methods include inverse Fourier techniques, subspace methods such as MUSIC, and sparse optimization such as Compressive Sensing (CS). All these methods assume a linear signal model with a tolerable amount of additive Gaussian noise. However, in many real-world ISAR measurement scenarios, a significant amount of multiplicative noise or clutter may be present. Current linear imaging methods are not generally well suited for multiplicative noise, as they rely significantly on phase information that can be heavily distorted or randomized under random multiplicative processes. This paper will present a method for imaging a target in the presence of a specific type of time-varying multiplicative noise using a convolutional classification neural network [1] – [4] .