Hydrometeor Classification for Dual Polarization Radar Based on Multi-Sample Fusion SVM

Abstract

In order to enhance the accuracy of dual polarization radar in hydrometeor classification, a hydrometeor classification algorithm based on multi-sample fusion Support Vector Machine (SVM) is proposed in this paper after considering that traditional fuzzy logic algorithm has the defect of over relying on expert experience to set parameters. The data of four polarization parameters (horizontal reflectivity factor, differential reflectivity, correlation coefficient and differential propagation phase constant) detected by the KOHX radar were taken as the feature information of hydrometeors. The dataset was collected, and the model was trained. According to the classification results of SVM model and combined with the distribution characteristics of target particles in the rainfall area, a classification system that can effectively identify four types of particles (dry snow, moderate rain, big drops and hail possibly with rain) was established This model greatly reduced the misidentification of dry snow (DS) and moderate rain (RA)) in the precipitation area, and significantly improved the overall classification effect of hydrometeors in the area. The 0.5-degree elevation scanning data of the radar at a certain time were tested, and the classification accuracy of system model was up to 97.21%. The average accuracy of other elevation scanning data was approximately 97%, which showed strong robustness.