基于极化神经网络的雷达舰船检测识别方法

Radar Ship Target Detection and Recognition Based on Polarimetric Neural Networks

相参雷达捕获的全极化海面目标距离-多普勒(RD)回波数据中,目标区域占比小、信噪比低,且海况环境与干扰种类多变,使得经典的深度神经网络在此种条件下检测识别精度较低。为此,本文提出了一种基于极化深度神经网络的全极化相参雷达海面目标检测识别算法。首先,引入极化特征提取模块挖掘目标与干扰的差异化特征;其次,通过特征金字塔网络解决小目标检测识别的问题;最后,使用级联结构进一步提升算法性能。在全极化相参雷达回波数据集上的测试结果表明:基于特征值与特征矢量的极化特征对于数据集中两类舰船目标的平均精度分别达到0.907 9与1.0,相比不采用极化特征有着显著提高。

In the fully polarimetric range-Doppler(RD) radar echo data of sea objects captured by coherent radar, the target area is small, the signal-to-noise ratio is low, and the sea conditions and disturbances are complicated. This makes classical deep neural networks have low detection and recognition accuracy under such conditions. Therefore, this paper proposes a fully polarimetric coherent radar sea target detection and recognition algorithm based on the polarimetric deep neural networks. Firstly, a polarimetric feature extraction module is introduced to mine the differential features between the targets and disturbances. Secondly, the feature pyramid networks(FPN) are used to better complete small target detection and recognition tasks. Finally, a cascade structure is used to improve the algorithm performance further. The test results on the fully polarimetric coherent radar data set show that the average precision of the polarimetric features based on the eigenvalues and eigenvectors for the two classes of targets in the data set is 0.907 9 and 1.0, respectively, which is a significant improvement compared with that obtained under the situation without polarimetric features.