基于SVM模型的星载GNSS-R内陆水体识别方法研究

Research on Inland Water Body Recognition Method Based on SVM Model for Spaceborne GNSS-R

内陆水体探测是遥感领域中的重要研究方向之一,涵盖河流形态变化监测、实时洪水监测和地表水变化分析等领域。支持向量机(Support Vector Machine,SVM)在非线性分类问题上表现优秀并且能够得到较低的错误率。提出了一种基于SVM的全球导航卫星系统反射(GNSS-R)信号内陆水体识别方法,使用星载GNSS-R信噪比(Signal to Noise Ratio,SNR)作为像素构建遥感图像,并应用合成少数类过采样技术(Synthetic Minority Oversampling Technique,SMOTE)进行数据处理。在此基础上,通过SVM模型的方式实现水体信息提取,并在刚果盆地区域的旋风全球导航卫星系统(Cyclone Global Navigation Satellite System,CYGNSS)数据进行了验证。结果表明,未进行SMOTE算法处理的分类准确率是65.1%,其中水体分类准确率5.39%,而经过SVM模型和SMOTE算法处理后,数据分类准确率提高至96.49%,水体分类准确率提高至96.32%,准确率和水体查准率都得到了提高,显示了利用基于SVM的星载GNSS-R内陆水体识别方法的有效性。

Inland water body detection is one of the important research directions in the field of remote sensing,covering river morphology change monitoring,real-time flood monitoring,surface water change analysis and other fields.Support Vector Machine(SVM)performs well for resolving nonlinear classification problems and can get a low error rate.A method of inland water body recognition based on Global Navigation Satellite System Reflectometry(GNSS-R)signals is proposed,which uses satellite-borne GNSS-R Signal to Noise Ratio(SNR)as pixels to construct remote sensing images,and applies the Synthetic Minority Oversampling Technique(SMOTE)to perform data processing.On this basis,the water body information is extracted by the SVM model,which is verified by the Cyclone Global Navigation Satellite System(CYGNSS)data in the Congo Basin.The results show that the classification accuracy without SMOTE algorithm processing is 65.1%and the water body classification accuracy is 5.39%,while after SVM model and SMOTE algorithm processing,the data classification accuracy is improved to 96.49%and the water body classification accuracy is improved to 96.32%.The accuracy and water body precision rates are improved,which shows the effectiveness of the spaceborne GNSS-R inland water body identification method based on SVM.