Classification of Radar Targets Based on Micro-Doppler Features Using High Frequency High Resolution Radar Signatures

Target classification for radar detections play a vital role in identifying unauthorized objects which pose security threats. Modern radars have a potential to extract Micro Doppler features that helps in identification and classification of various types of radar targets like drones, human, birds etc. In this aspect, a framework for pre-processing followed by deep convolutional neural network (DCNN) based architecture is proposed. The proposed framework is trained and tested for FMCW radar dataset which contains measurements from birds, human and six different drone types. The signature is received from the target range bin and adjacent range bins over coherent pulse integration (CPI) period. A Joint time frequency transform such as Short Time Fourier Transform (STFT) matrix is formed for the signature received and is used as training input to the target classifier. The classifier is able to report the targets type with an accuracy of 97.4% on test data. The ability to classify a radar target is subject to radar parameters such as frequency of operation, dwell time, transmitted power, pulse duration, revisit rates etc. The proposed method is designed keeping in mind the need for classification of radar targets with less dwell time in scan-based (mechanically/Electronically stirred) radars.