Object Localization Using Machine Learning

This study explores the method of using machine learning techniques to localize objects in a three-dimensional space based on signatures of their scattered field. The antenna systems that stimulate and collect the scattered field are kept small to avoid a high cost and deployment difficulties. The cases of localizing a single object and simultaneously localizing two objects are reported, and two types of artificial neural networks are considered. It is found that a multilayer perceptron outperforms a convolutional neural network for the system under test. With a single object to locate, a trained network delivers very good performance. When there are two objects, the achieved performance is much lower, for which the constraints on the antenna systems and the amount of available training data are believed to be major limiting factors.