Classification of micro-Doppler signatures with the use of orthogonal moment-based features

Radar micro-Doppler signatures are crucial for classification and recognition of many targets in a wide range of applications. In this paper, a novel processing pipeline is proposed to comprehensively compute 21 different orthogonal moments and evaluate them as features for classification of micro-Doppler signatures. Using a publicly available dataset for human activity recognition, it is shown that orthogonal moments can achieve very good classification performance above 90%, while remaining more explainable and more robust to noise added to the testing set compared to convolutional neural networks.