| Title |
Radar Target Detection in Cluttered Environments using Faster R-CNN |
| Authors |
김일석(Ilseok Kim) ; 이상언(Sang-Eon Lee) ; 이희재(Hee-Jae Lee) ; 이창기(Changki Lee) ; 원종민(Jongmin Won) ; 장준혁(Joon-Hyuk Chang) |
| DOI |
https://doi.org/10.5573/ieie.2025.62.8.29 |
| Keywords |
Radar target detection; Range-velocity domain; Deep learning model; Knowledge distillation |
| Abstract |
In radar target detection, accurate position estimation and robust detection performance in cluttered environments are crucial. This paper compares the performance of conventional CFAR detection techniques with deep learning-based Faster R-CNN detection methods in the R-V domain. We experimentally demonstrate that Faster R-CNN achieves superior detection performance compared to CFAR, particularly in cluttered environments. While CFAR shows vulnerability in target detection under clutter due to its fixed threshold approach, Faster R-CNN maintains reliable detection performance even in complex environments with significant clutter. The proposed Faster R-CNN model employs a lightweight encoder structure for computational efficiency and utilizes knowledge distillation from VGG16 to minimize performance degradation from model compression. Experimental results showed that Faster R-CNN demonstrated superior detection performance compared to CFAR across various clutter environments, while maintaining stable detection accuracy even in high-clutter conditions. |