| 基于复值深度神经网络的目标检测方法 |
| |
| 引用本文： | 王佳琛,吴亿锋.基于复值深度神经网络的目标检测方法[J].信号处理,2022,38(10):2021-2029. |
| |
| 作者姓名： | 王佳琛 吴亿锋 |
| |
| 作者单位： | 中山大学电子与通信工程学院，广东 深圳 518107 |
| |
| 基金项目： | 国家自然科学基金62101603中山大学中央高校基本科研业务费专项资金资助22qntd0401 |
| |
| 摘 要： | 针对雷达在检测概率要求严苛的多旁瓣干扰复杂场景下使用传统目标检测方法无法满足需求，性能有待进一步提升的问题，本文提出了一种基于多通道复值深度神经网络的雷达目标检测方法。传统脉冲体制阵列雷达的恒虚警率目标检测通常在和通道进行，在对回波信号进行空域相参预处理过程中获得了相参积累的同时丢失了阵元间的相位信息，而实际上目标回波在阵元间存在着一定的相位关系。本文利用神经网络强大的拟合能力和分类能力，将目标检测视为二元分类问题，设计复值深度神经网络深入挖掘目标与背景在不同阵元间的幅度及相位信息差异，从而在传统目标检测和通道-距离-多普勒空间的更前端更好地区分目标与背景的差异，提升了雷达目标检测性能。实验结果显示，所提方法在存在大量旁瓣干扰的场景下，相较传统方法具有更好的检测性能表现和抗干扰能力，且在杂波环境中也有良好的表现。
|
| 关 键 词： | ??雷达目标检测 相位信息 复值深度神经网络 多通道复值数据 |
| 收稿时间： | 2022-07-17 |
Target Detection Method Based on Complex-Valued Deep Neural Network |
| |
| Affiliation: | School of Electronics and Communication Engineering，Sun Yat-Sen University，Shenzhen，Guangdong 518107，China |
| |
| Abstract: | ? ?Traditional radar target detection method performs not good enough in the complex scene which consist of multi sidelobe jamming， non-homogeneous and non-stability clutter with strict probability of detection. The performance need to be further improved. Aiming at this problem， a radar target detection method based on multi-channel complex deep neural network is proposed in this paper. In narrow band target detection area， the constant false alarm rate target detection method of traditional pulse array radar is usually carried out with the channel. During the spatial coherent preprocessing of the echo signals， the coherent accumulation is obtained to rise the signal-noise ratio. In fact， the phase relationship between the target echo in spatial area is certain and it is different from the noise， the clutter， the side lobe jamming， etc. The coherent accumulation uses this relationship to rise the signal-noise ratio， but it may not the best way to use the phase information in spatial area， which means the traditional radar target detection method for narrow band target can be further improved. Deep learning and deep neural network have the powerful fitting ability and classification ability， and target detection can be regarded as a binary hypothesis testing， which can also regard as a binary classification. In this paper， by using the abilities of fitting and classification of neural networks， a complex-valued deep neural network is designed. This neural network can deeply mine the amplitude and phase information differences between target and background in different array elements. Thus， it can distinguish the differences between target and background better in the spatial-range-Doppler space， the more front end of traditional target detection， which means the performance of radar target detection is improved. The simulation results show that the complex-valued deep neural network proposed in this paper has better detection performance and anti-interference ability than the traditional method in the scene with a large number of side lobe interference， and also has good performance in the clutter environment. |
| |
| Keywords: | |
|
| 点击此处可从《信号处理》浏览原始摘要信息 |
|
点击此处可从《信号处理》下载全文 |
