UAV Classification from Micro-Doppler Signature-based Time Frequency Images using SVM

Nowadays, UAVs are one of the most challenging flying objects to detect and classify. As their RCS is very low, traditional radar finds it very difficult to detect. To make the scenario even worse, birds are also exhibiting almost the same characteristics as UAV. Thus, this work tried to investigate the best options to identify UAV in all aspects. In such an endeavor, it delved into the area of sensor options, identifying options, distinguishing options, options to get better resolution, feature extracting options and image classifying options. This study identified that spectrogram-based micro-Doppler signature (m-DS) analysis retrieved from the radar is the most suitable option to do this task. To obtain good resolution in m-DS analysis, there is a grey area between STFT and covariance-based S- Transform. In addition to this, among several image classification options, CNN and SVM were found the most effective. Studies suggest that SVM has lower accuracy than CNN. However, as SVM has very little computation process, if the accuracy can be improved, SVM may be one of the best options in this regard. In pursuant to such aim, this paper identified the complete pathway of classifying UAV by X Band FMCW radar with the help of SVM with a few mathematical equations. This paper also checked all the viable options to use the SVM including the kernel.