针对传统的基于统计特性的目标检测方法统计建模困难和机器学习目标检测方法特征提取复杂的问题,提出了深度迁移学习的雷达目标检测方法。采集雷达的IQ 数据,经过脉冲压缩处理后形成雷达原始图像,通过对雷达图像进行放大、裁剪等步骤构建飞机目标数据集;设计R-CNN、SPP-Net、Fast R-CNN 和Faster R-CNN 等两阶段深度学习目标检测模型,对雷达图像中的飞机目标进行自动检测;模型训练时,引入迁移学习思想,使用预训练过的卷积神经网络自动提取图像中的深层特征,以达到减少雷达图像训练样本量的目的。某型航管雷达实测数据的实验结果表明:与传统的恒虚警率检测方法相比,该方法提高了雷达目标检测率,降低了虚警率,解决了检测率与虚警率的矛盾。

Aiming at the difficulty of statistical modeling of traditional target detection methods based on statistical characteristics and the complex feature extraction of machine learning target detection methods, a radar target detection method based on deep transfer learning was proposed. The IQ data of the radar was collected, and the original radar image was formed after pulse compression processing. The aircraft target data set was constructed by amplifying and clipping the radar image; R-CNN, SPP-Net, Fast R-CNN and Faster R-CNN two-stage deep learning target detection models were designed to automatically detect aircraft targets in radar images; during model training, the idea of transfer learning is introduced, and the pre-trained convolutional neural network is used to automatically extract the deep features in the images, so as to reduce the training sample size of radar images. The experimental results of the measured data of a certain type of air traffic control radar show that compared with the traditional constant false alarm rate detection method, this method improves the radar target detection rate and reduces the false alarm rate, and solves the contradiction between the detection rate and the false alarm rate.

TN957

军事类研究生重点资助项目(JY2020B150)