| 摘要: |
| 雷达目标检测识别是雷达信号处理的重要方向。雷达距离-多普勒图是雷达目标运动状态和速度的量化描述,是进行雷达目标检测识别的重要数据。本文提出了一种基于距离-多普勒图的雷达目标智能检测识别算法。算法基于双步式检测识别范式,针对距离-多普勒图的特点,采用随机变换和标签混合的样本增强策略扩充训练样本,提升网络的鲁棒性。其次,结合重合度和目标尺寸对位置回归损失进行改进,提升算法的位置计算精度。消融实验表明,本文提出的改进能有效地提升算法的检测识别性能。在权威公开数据集上的测试表明,相对于已有的Faster RCNN和DAROD算法,本文提出的算法在性能上有显著的提升。 |
| 关键词: 距离-多普勒图、雷达目标检测识别、样本增强、自适应位置回归损失 |
| DOI： |
| 分类号:TN957.51 |
| 基金项目: |
|
| An Intelligent Radar Target Detection and Recognition Algorithm Based on Range-Doppler Map |
|
|
|
|
| Abstract: |
| Radar target detection and recognition represents a significant area of focus within the broader field of radar signal processing. Range doppler map is a quantized representation of the radar target"s motion state and velocity, which constitutes crucial data for the detection and recognition of radar targets. This paper proposes an intelligent detection and recognition algorithm for radar targets based on range doppler map. The algorithm is based on a two-stage detection and recognition paradigm. In order to take advantage of the characteristics of range doppler map, a sample enhancement strategy of random transformation and label mixing is employed to expand the training samples and improve the robustness of the network. Secondly, the position regression loss is enhanced by combining IOU and target size, thereby improving the accuracy of positioning. Ablation experiments demonstrate that the improvements proposed in this paper effectively enhance the algorithm"s detection and recognition performance. Tests on authoritative public datasets indicate that the algorithm proposed in this paper exhibits a notable improvement in performance relative to existing Faster RCNN and DAROD algorithms. |
| Key words: range-doppler map, radar target detection and recognition, sample enhancement, adaptive position regression loss |