基于CNN的机载气象雷达气象目标检测方法

机载气象雷达系统进行气象探测时易受到强地杂波的干扰，从而导致目标信息丢失。为准确检测地杂波中的气象目标，获取完整的目标信息，本文提出了一种基于卷积神经网络(Convolution Neural Networks, CNN)的机载气象雷达目标检测方法。该方法联合时域、多普勒域和俯仰维空域信息，将杂波相位对准指标、多普勒速度和干涉相位作为CNN的输入，并给出详细的网络结构。本文通过模拟雷达回波仿真产生训练集和测试集，并对所提网络进行训练和测试。仿真结果表明，与目前的气象目标检测方法相比，该方法具有较高的检测概率，而且在谱矩信息变化的情况下仍可维持较好的检测性能，具有很好的鲁棒性。此外，仿真结果表明CNN比传统的贝叶斯分类器和支持向量机等分类网络具有更好的分类性能。

CNN-Based Meteorological Target Detection Algorithm for Airborne Weather Radar

Airborne weather radar system is easy to be interfered by strong ground clutter during weather detection, which leads to loss of target information. In order to detect the meteorological target in ground clutter accurately and obtain complete information of weather target, a target detection algorithm for airborne weather radar based on convolution neural networks (CNN) is proposed in this paper. This algorithm combines time domain, Doppler domain and elevation dimensional spatial information, which takes the clutter phase alignment index, Doppler velocity and interferometric phase as the input of CNN. The detailed network structure is given in this paper and the training dataset and test dataset are generated by simulation radar echoes, by which the proposed network is trained and tested. Compared with current weather target detection methods, the simulation results show that the proposed algorithm has a higher detection probability and can maintain good detection performance and good robustness under the condition that spectral moment information changes. In addition, simulation results show that CNN has better classification performance than traditional Bayesian classifier and support vector machine.