Analyzing the micro-doppler effect of UAVs and birds

• Main Author: Chua, Xi Hong
• Other Authors: Jiang Xudong
• Format: Final Year Project
• Language: English
• Published: 2016
• Subjects: DRNTU::Engineering
• Online Access: http://hdl.handle.net/10356/67591
• Institution: Nanyang Technological University

The micro-Doppler phenomenon is gaining much attention and interest due to its ability to retain information of the target in the return radar signal. The micro-Doppler signatures which are generated upon contact with a target in motion and it can be used to differentiate different objects. In this project, analysis of the micro-Doppler effect of UAVs and birds was done, with the aim of classifying them with high accuracy. Firstly, a spectrogram was used as a method of feature extraction. Then, DTW was utilized to obtain a distance matrix, built from the global distance between any two signals. Subsequently, classification was carried out by the nearest-neighbour classifier. An improved method of classification was proposed, which was to treat the distances of one sample to all gallery samples as the feature vector, and to classify it using the SVM, to achieve better performance. Evaluation was done on both approaches on the dataset that consisted of 69 samples of birds and 854 samples of UAVs. Evaluation was also done on the impacts of the system parameters on the final classification performance. Pitting 1-NN classifier against SVM classifier, SVM classifier turned out to perform better. The possible reasons were discussed. To end off, several possible future research directions were pointed out, which could further improve the classification performance.