| 摘要: |
| 针对海杂波背景下海面小目标检测精度低的问题，本文提出一种基于SAE?GA?XGBoost算法的海面小目标检测方法。使用堆栈自编码器对待检测信号自适应提取深层次特征，将时域、频域分别提取的特征组合为高维特征，增强特征的差异性。引入遗传算法对XGBoost中超参数进行寻优并更新，利用超参数更新后的模型对高维特征进行评估分类。利用IPIX数据集进行实验验证，结果表明：与多组分类检测方法相比，所提检测方法对于高海况数据具有更好的检测效果，HH极化下#17与#280数据的检测率分别达到了80.02%与82.73%。 |
| 关键词: 目标检测 堆栈自编码器 高维特征 极端梯度提升 |
| DOI：DOI:10.3969/j.issn.1672-2337.2023.01.011 |
| 分类号:TN911.7 |
| 基金项目:国家自然科学基金项目资助（No.62171228）；国家重点研发计划项目资助（No.2021YFE0105500） |
|
| Sea⁃Surface Small Target Detection Based on SAE⁃GA⁃XGBoost Algorithm |
|
ZHAO Di， XING Hongyan， WANG Haifeng， YAN Yan
|
|
1. Collaborative Innovation Center on Forecart and Evaluation of Meteorological Disasters, Nanjing University of Information Science and Technology, Nanjing 210044, China;2. Jiangsu Key Laboratory of Meteorological Detection and Information Processing, Nanjing University of Information Science and Technology, Nanjing 210044, China
|
| Abstract: |
| To solve the problem of low accuracy of small target detection in the background of sea clutter, this paper proposes a sea?surface small target detection method based on SAE?GA?XGBoost algorithm. Stacked autoencoder is used to adaptively extract deep features from detected signals, and the features extracted from time domain and frequency domain are combined into high?dimensional features to enhance the difference of features. Genetic algorithm is introduced to optimize and update the hyperparameters in XGBoost, and the model after updating hyperparameters is used to evaluate and classify the high?dimensional features. Experimental verification using IPIX data set shows that the detection rate of #17 and #280 data under HH polarization reaches 80.02% and 82.73% respectively compared with several sets of classification detection methods. |
| Key words: target detection stacked autoencoder high⁃dimensional features XGBoost |
