2017年“8.8”九寨沟地震滑坡自动识别与空间分布特征

2017年8月8日发生的7.0级九寨沟地震诱发九寨沟熊猫海附近产生大量的滑坡体，造成道路阻塞，严重影响地震应急救援进度。为快速准确地识别滑坡分布范围，本文在深入分析滑坡遥感影像特征的基础上，引入面向对象分析方法，实现了基于无人机影像的震后滑坡体的自动识别。通过多尺度分割算法获取滑坡多层次影像对象，利用SEaTH算法自动构建每一层次特征规则集，实现基于不同层次分析的滑坡体自动识别。分析滑坡体在地形、活动断层等因子中的空间分布特征，为地震滑坡预测与危险性评价奠定基础。与人工目视解译结果相比较，基于面向对象的滑坡自动识别方法提取精度可达94.8%，Kappa系数为0.827，在电脑配置相同的情况下，自动识别方法的效率是人工目视解译效率的一倍。空间分布特征分析表明，地震滑坡的空间分布与斜坡坡度、地形起伏度呈正相关关系，与地表粗糙度存在负相关关系，研究区滑坡体分布存在明显的断层效应。

Recognition of earthquake-induced landslide and spatial distribution patterns triggered by the Jiuzhaigou earthquake in August 8, 2017

The Jiuzhaigou earthquake with the magnitude of 7.0 occurred in August 8, 2017 and resulted in a large number of landslides near the panda sea area in Jiuzhaigou. These landslides caused road congestion and seriously affected the progress of earthquake emergency rescue. The landslide caused by earthquake has wide distribution and large quantity. Given the urgency of the disaster and high resolution of unmanned aerial vehicle (UAV) images, the traditional artificial visual interpretation model cannot meet the needs of earthquake emergency response. Therefore, an automatic information identification method must be developed to identify the distribution range of landslide rapidly and accurately.
On the basis of comprehensive analysis of the features of remote sensing images of landslide, an automatic information identification model for object-oriented analysis was constructed. First, the remote sensing images were segmented at different scales to obtain different levels of image objects depending on different types and scales of land objects. Then, SEath algorithm was used to construct feature rule set automatically by the comprehensive utilization of the information of spectrum, texture, and shape of object at every level, and the distribution of earthquake-induced landslides was identified. Thereafter, the accuracy and efficiency of recognition were evaluated on the basis of artificial visual interpretation. Finally, the spatial distribution features of landslide body in topographic factor and fracture distribution layer were analyzed by statistical analysis.
Using the acquired aerial image data of UAV, the earthquake landslide near the panda sea area of Jiuzhaigou earthquake was identified. The overall accuracy was 94.8%, and the Kappa coefficient was 0.827. The present method was twice as efficient as the artificial visual interpretation method under the same configuration of computer.
The spatial distribution of landslide was positively related to slope and topographic relief but was negatively correlated with roughness. No evident relationship was found between the spatial distribution of landslide mass and the topographic factors such as slope and gradient of slope. Evident fault effects were observed in the distribution of landslide.
In this study, the object-oriented analysis method was developed to realize automatic identification of earthquake-induced landslide using UAV images. On the basis of the comprehensive utilization of spectrum, texture, and shape of image objects at each segmentation level, an automatic construction method of feature rule set based on SEaTH algorithm was established, Finally, an automatic, efficient extraction of earthquake-induced landslides was realized. Compared with the artificial visual interpretation method, the automatic method of object-oriented analysis could effectively improve the efficiency and timeliness of disaster information identification after earthquake, which could break the pattern of multiple interpretations and save time for earthquake emergency response.
The earthquake-induced landslide distribution features in elevation, slope, aspect, fault distance, and other factors were also analyzed. The correlation between landslide and topographic factors was found. Overall, the earthquake-induced landslide in the study area is mainly controlled by the Tazang fault. The spatial distribution rule can provide information support for landslide risk assessment, disaster investigation, prediction, and prevention.