基于RCNN的双极化气象雷达天气信号检测

Weather signal detection for dual polarization weather radar based on RCNN

为检测混杂在地杂波、生物杂波中的天气信号,提高定量降水精度,提出了基于残差卷积神经网络(residual convolutional neural network, RCNN)的天气信号检测算法。首先,将采集的极化参数水平反射率因子、差分反射率、相关系数、差分相移率堆叠为三维数组后进行预处理,将其分为天气信号与杂波信号。然后,开发并优化RCNN,给出详细的网络结构。最后,通过多次实际的降水过程对所提算法的检测效果进行评价。结果表明,相比支持向量机以及卷积神经网络(convolutional neural network, CNN),所提算法对天气信号的检测效果更好,并且在不同仰角以及全年的实测数据上均表现出良好的检测性能。

To detect weather signals submerged in ground clutter and biological clutter and improve the accuracy of quantitative precipitation, a weather signal detection algorithm based on residual convolutional neural network(RCNN) is proposed. Firstly, the collected polarization parameters: horizontal reflectivity, differential reflectivity, correlation coefficient, and differential propagation phase constant stack into a three dimensional array which are then divided into weather signals and clutter signals for preprocessing. Then, RCNN is developed and optimized, and the detailed structure of network is provided. Finally, the detection effect of the proposed algorithm is evaluated through several actual precipitation processes. As demonstrated by testing results, compared to support vector machines and convolutional neural network(CNN), the proposed algorithm has better detection effects on different elevation angles and measured data throughout the year.