| 摘要: |
| 为了提升雷达对海面小目标的探测能力，本文提出一种基于双重虚警控制的极限梯度提升（eXtreme Gradient Boosting, XGBoost）的目标检测方法，解决高维特征空间中分类器设计难的问题。首先，从时域、频域、时频域中挖掘了海杂波和含目标回波的精细化差异，并将这些差异凝聚为7个特性，进而构建高维特征空间。然后，发展了一种双重虚警控制的两分类器。在第一重中，重新定义XGBoost的损失函数，通过迭代调整两类错误率的惩罚因子，实现结构层面上的粗虚警控制。在第二重中，将分类概率作为统计量，实现参数层面上的精虚警控制。最后，经实测数据验证，所提出的检测器能精准控制虚警且具备稳健的探测能力。 |
| 关键词: 海杂波 目标检测 特征提取 虚警控制 极限梯度提升树 |
| DOI：DOI:10.3969/j.issn.1672-2337.2023.03.010 |
| 分类号:TN911.23 |
| 基金项目:国家自然科学基金（No.61901224）； 江苏省高等学校基础科学（自然科学）研究项目(No.21KJB510036) |
|
| Sea⁃Surface Small Target Detection Based on XGBoost with Dual False Alarm Control |
|
SHI Sainan， JIANG Li， LI Dongchen， WU Xuzi
|
|
1. School of Electronic and Information Engineering, Nanjing University of Information Science & Technology, Nanjing 210044, China;2. System Engineering Research Institute, China State Shipbuilding Corporation, Beijing 100094, China;3. School of Information Engineering, Jiangsu Open University, Nanjing 210036, China
|
| Abstract: |
| Due to the difficulty to design high?dimensional classifier, a detection method using extreme gradient boosting （XGBoost） with dual false alarm control is proposed to improve the detection performance of marine radars for small targets on the sea surface. Firstly, the refined differences between the sea clutter and the returns containing target are extracted from time domain, frequency domain and time?frequency domain. These differences are condensed into seven features and then the high?dimensional feature space is constructed. Secondly, a binary classifier with dual false alarm control is developed. In the first part, the loss function of XGBoost is redefined to realize the rough false alarm control at the structural level by adjusting iteratively the penalty factor of the two types of error rate. In the second part, the classification probability is used as test statistics to obtain the precise false alarm control at the parameter level. Finally, it is verified by measured data that the proposed detector can accurately control false alarm and has robust detection capability. |
| Key words: sea clutter target detection feature extraction false alarm control extreme gradient boosting |
