Abstract
In this paper, we experimented with k-nearest neighbors (KNN), support vector machines (SVM), and random forests (RF) machine learning (ML) algorithms to classify multiple-shape object detection using mm-wave radar. For the demonstration purpose, we generated a linear frequency modulated (LFM) radar signal at 77GHz to 81GHz with bandwidth ranges of 1GHz–4GHz and detected flat, cylindrical, and spherical objects of various shapes and dimensions. The beat signals are recorded by shifting the radar transmitter in a step of 1cm. The 1D range-FFT of thus obtained beat signal power levels are recorded for various radar shifts and the required data set is prepared. Acquired power level data sets are divided into 80% training and 20% test data sets through random permutation. The trained model for each ML algorithm is evaluated in the test stage. Through the analysis, the KNN and SVM provided about 70.59% accuracy and RF provided about 64.71% accuracy in recognizing different types of objects. The presented approach with a lower bandwidth radar signal, can be performed with a photonically generated high bandwidth radar signal which provides finer resolution in object detection which is under investigation. It is anticipated that with photonically generated high-bandwidth radar signal (>8GHz), an accuracy percentage as close to 90% can be obtained.