基于区域生长算法巴谢河流域典型滑坡自动识别

Automatic identification of typical landslides in Baxie River Basin based on region growing

黄土高原在夏季集中降雨条件下易发生大量滑坡,对当地居民生命财产造成巨大损失,快速提取滑坡信息对灾害应急有至关重要的作用。传统的人工解译可以有效识别滑坡,但是其效率极低。以黄土高原巴谢河流域为研究区,利用Google Earth软件提取研究区图片,通过研究区已解译滑坡体后缘灰度值范围作为种子点自动选取条件,利用区域生长算法和形态学操作实现对研究区未解译典型滑坡的识别和滑坡后缘边界提取。发现本方法能够自动识别研究区典型滑坡,当灰度生长阈值为10时,识别滑坡后缘范围精度达到78.94%。结果表明:本文方法能够应用于黄土高原巴谢河流域典型滑坡识别,为滑坡灾害评估、救援和重建提供方向。

[Background]The Loess Plateau is prone to a large number of landslides under the condition of concentrated rainfall in summer,which causes huge damage to local people's lives and property,rapid extraction of landslide information plays an important role in disaster response.At present,there are a large number of landslides in the Baxie River Basin of the Loess Plateau,and most of them are typical landslide.Traditional manual interpretation can effectively identify landslides,but its efficiency is extremely low.In addition,due to the restrictions of terrain and weather,the manual interpretation of typical landslides is more difficult.[Methods]On study area Baxie River Basin in the Loess Plateau,using Google Earth software to extract the images of study area,and Gaussian filter was adopted to reduce the influence of image noise on recognition results.The gray value range of the trailing edge of the interpreted landslide in the study area was used as the condition for automatic selection of seed points,then using region growing algorithm and morphological operations to realize recognition and trailing edge boundary extraction of landslides in typical landslides that has not been interpreted in the study area.Optimal region growth conditions were obtained by adjusting the growth threshold.[Results]This method automatically identified the typical landslide in the study area.The combination of morphologic open operation and close operation removed the obvious sawtooth and small burrs in the results and link the external broken lines to smooth the edges.Sobel operator combined with watershed algorithm extracted effectively the landslide's trailing edge boundary.Correct identification,wrong identification and missing identification existed in the extraction process of landslide trailing edge boundary.With the increase of the growth threshold,the divergence factor increased and the error factor decreased,indicating that the over-recognition increased and the missing recognition decreased.Three images of the Baxie River Basin in the Loess Plateau were extracted by Google Earth.The three images were identified and the landslide trailing edge was extracted by this method.Meanwhile,the three images were interpreted manually and the landslide trailing edge was drawn to verify the identification results.When the gray growth threshold is 10,the accuracy of identifying the trailing edge of landslide reached 78.94%.For new landslides,the grayscale characteristics at the back edge of the landslide body were obvious,and the identification results were good.For landslides with a long occurrence time,the surface weathering and new vegetation growth will reduce the identification accuracy of this method.[Conculsions]These suggest that the proposed method is effective to identify the typical landslides in Baxie River Basin of Loess Plateau,which provides a direction for landslide disaster assessment,rescue and reconstruction.