Abstract
Radar tomographic imaging's foundation is the radar cross-section (RCS) of the pattern and material of the investigative shape. RCS varies when the target's permittivity and conductivity differ in material profile and shape from other objects in the environment. This paper investigates the use of Multi-Layer Perceptron (MLP) for pattern recognition of shapes in the frequency domain. Supervised and Unsupervised Locally Linear Embedding (LLE) algorithms were used as a robust means of addressing the difficulties of nonlinear dimensionality reduction of scattering fields when domains have frequency complexity. The proposed algorithm is capable of recognizing a variety of shape profiles based on the scattering field measured by using a geometrically diverse tomographic radar scanning transmitter/receiver set up. The algorithm is trained in recognition of the patterns of different scattering fields with a scattering field of radar tomographic handwritten digits across the frequency domain, which can be extended to the 3D model. The theoretical basis of this work is validated using the results of simulations.