宽带高分辨雷达较窄带雷达可获得更多的目标和环境信息,且可更精确地测量目标的位置和运动参数,同时也具有更好的低截获性能。但目标在宽带波形下通常表现为扩展目标,基于点目标假设的传统的信噪比阈值检测方法难以充分利用扩展目标的回波特性。因此,文中提出了基于支持向量机和卷积神经网络两种机器学习方法的海面扩展目标检测器。通过建立仿真平台生成样本数据,对两种机器学习模型进行了训练和测试。阈值检测、支持向量机和卷积神经网络三种方法的检测性能曲线的对比展现了两种基于机器学习方法在海面扩展目标检测上的优势。实测场景的测试进一步表现出卷积神经网络能有效提升点迹质量,从而有利于提升航迹质量,也表明了采用仿真生成的样本数据集对应用于海面扩展目标检测的机器学习模型进行训练和验证的有效性。

Compared to narrow-band radars, high-resolution wide-band radars obtain more information of targets and environment, measure the location and movement parameters more accurately, and achieve lower probability of interception. However, targets are typically presented as distributed targets when their dimensions are greater than radars′ resolution. This invalidates the presumption of point targets in the traditional threshold detection method, and makes it difficult to make full use of the characteristics of distributed targets. Therefore, this work investigates the distributed sea-surface target detection based on two machine learning (ML) approaches, i. e. supporting vector machine and convolutional neural network (CNN). The machine learning models are trained and tested with the synthetic data generated by the established simulation platform. The comparison of receiver operating characteristics curves illustrates the advantages of two ML methods over threshold detection. The evaluation with measured data further highlights that CNN can improve the detection performance, helping to gain better track result than threshold detection. Furthermore, this work also demonstrates that machine learning models can be trained and validated by simulated data for the distributed sea-surface target detection.

TN29