基于KNN分类器的地表冻融判别研究

RESEARCH ON DISCRIMINATION OF SURFACE FREEZING AND THAWING BASED ON KNN CLASSIFIER

以青藏高原的5个地区为研究对象,分别是阿里观测网、黑河观测网、玛曲观测网、那曲观测网和帕里观测网,结合SMAP被动微波辐射计L波段的亮温数据和AMSR2被动微波辐射计Ka波段的亮温数据构建了二维时间序列,然后用KNN分类器对5个观测网的土壤冻融状态进行分类。最后根据实测的土壤5 cm深处的温度数据分别验证5个观测网的分类结果,并计算准确率。实验结果表明在湿润的那曲观测网中分类效果最好,分类的准确率为97.14%;在比较干旱的阿里地区分类精度最低,分类的准确率为88.98%;五个观测网的平均准确率为94.23%。

Taking the five areas of Qinghai-Tibet High as the research objects,they are the Ali Observation Network,Heihe Observation Network,Maqu Observation Network,Naqu Observation Network and Pali Observation Network,combined with the brightness temperature data of the L-band of the SMAP passive microwave radiometer and the AMSR2 passive The Ka-band brightness temperature data of the microwave radiometer constructs a two-dimensional time series,and then uses the KNN classifier to classify the soil freeze-thaw state of the five observation networks.Finally,the classification results of the five observation networks were verified according to the measured soil temperature data at a depth of 5 cm,and the accuracy rate was calculated.The experimental results show that the classification effect is the best in the humid Nagqu observation network,with a classification accuracy of 97.14%;the classification accuracy is the lowest in the relatively arid Ali area,with a classification accuracy of 88.98%;the average accuracy of the five observation networks is 94.23%.