Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Micro-doppler classification of humans and animals using FMCW radar
This research attempts to address the issue of animal poaching by exploring human and animal target classification by making use of micro-Doppler data generated by a Frequency-Modulated Continuous Wave (FMCW) radar. Raw Analogue to Digital Converter (ADC) data was collected of human and animal speci...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English English |
| Published: |
Department of Electrical Engineering
2025
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| Summary: | This research attempts to address the issue of animal poaching by exploring human and animal target classification by making use of micro-Doppler data generated by a Frequency-Modulated Continuous Wave (FMCW) radar. Raw Analogue to Digital Converter (ADC) data was collected of human and animal species including dogs, horses and cows. Signal processing techniques such as creating range-Doppler and Constant False Alarm Rate (CFAR) maps for detection and using Short-Time Fourier Transform (STFT) for spectrogram generation were applied. Principal Component Analysis (PCA) was employed as the method for data reduction. The dataset was subsequently classified and evaluated across various target class configurations, comparing both the full dataset and its PCA-reduced versions, using Convolutional Neural Networks (CNN), k-Nearest Neighbors (kNN), Random Forest (RF) and Support Vector Machine (SVM) models. Following this, a two-stage classification process was implemented to further refine the analysis. In the first stage, the 4 above-mentioned classifiers were used to distinguish between humans and animals. In the second stage, these classifiers differentiated among the specific animal species. The study experimented with all 16 permutations of classifier combinations, such as SVM-CNN and kNN-RF. Notably, the SVM-SVM combination achieved the highest accuracy at 97.66%, closely matching the 97.5% accuracy outcome obtained in the multi-class classification of humans versus dogs, horses and cows. The PCA-reduced set yielded results closely comparable to those from the full dataset evaluation, confirming its effectiveness. This study highlights the challenges of data collection in natural settings and the need for publicly accessible micro-Doppler data of animal targets to further this research area. Recommendations for future work include developing tracking algorithms tailored to various animal movement patterns. The findings indicate that low-cost radar surveillance systems with micro-Doppler classification technology hold promising application in animal conservation-oriented efforts. |
|---|