强降雨诱发滑坡群识别的光学遥感变化检测方法

Optical remote sensing change detection method for the identification of landslide clusters induced by heavy rainfall

针对强降雨诱发滑坡群的光学遥感干扰特征多、识别精度低的问题,该文提出一种基于灾前灾后遥感影像变化检测技术的滑坡识别精度提升方法:①通过波段比值预处理,消除道路、房屋等对滑坡识别的干扰特征;②通过相关参数优化,采用图像差值法和非监督分类法进行遥感变化检测滑坡识别,分析两种方法发生漏判误判的原因。结果表明,对遥感影像进行波段预处理,可以明显弱化道路、房屋等造成滑坡识别的干扰特征;图像差值法和非监督分类法识别滑坡的精确率、召回率和F1-score指标分别达到了0.815、0.647、0.721和0.680、0.788、0.730。基于滑坡特征影像预处理与参数优化的两种变化检测技术均明显提升了强降雨诱发滑坡群的自动识别效率,在光学遥感影像进行滑坡识别干扰特征消除和提高识别精度方面具有应用价值。

For researching the problem that the optical remote sensing of landslide clusters induced by heavy rainfall have many interference factors and the identification accuracy is low,this study proposes a method to improve the landslide identification accuracy based on the image change detection technology before and after the disaster:Eliminate the interference factors of roads and houses on landslide identification through band ratio preprocessing;Through the optimization of related parameters,image difference method and unsupervised classification method are used to identify landslide by remote sensing change detection.Analyze the reasons for misjudgment of the two methods.The results show that the interference factors of road and house which of the landslide identification can be significantly weakened by the band ratio preprocessing of remote sensing image.The Accuracy,Recall and F1-score of the image difference method and the unsupervised classification method are 0.815,0.647,0.721 and 0.680,0.788,0.730 respectively.The two change detection techniques based on landslide feature image preprocessing and parameter optimization have significantly improved the automatic identification efficiency of landslide clusters induced by heavy rainfall,and have more application value in eliminating the interference factors of landslide recognition and improving the recognition accuracy of optical remote sensing images.