| China is a country prone to geological disasters.According to statistics from the Ministry of Natural Resources,a total of 6181 geological disasters occurred in China in 2019,causing more than 200 deaths and direct economic losses of up to 2.77 billion yuan.Among them,Qinba Mountainous Area is one of the most frequent areas of geological disasters in China.Mountainous geological disasters are dominated by collapses,landslides,and debris flows.Among them,landslide disasters are the most common and seriously threaten the lives and property of local people.How to identify and prevent geological disasters is a common problem in the world.With the rise of artificial intelligence and big data in recent years,this paper attempts to combine artificial intelligence algorithms with geological disasters,and make full use of the advantages of multi-source and multi-temporal remote sensing data to explore new methods of regional landslide dynamic risk analysis.Due to the complex causes of landslide disasters and the changing geological environment,it is currently difficult to accurately predict.This article uses existing geological data,combined with field survey data,monitoring data,and remote sensing image data,to collect various impact factors related to landslide disasters in the area.Then,through high-precision multitemporal remote sensing images,the change information of inducing factors such as land use type,vegetation coverage,human engineering activities,etc.is extracted and incorporated into the evaluation index system to explore the direct and indirect laws of geological environment changes and geological disasters.Finally,using the inherent advantages of machine learning methods in data mining,a regional dynamic prediction model of landslide disasters is established to realize the dynamic risk analysis of landslide disasters in the region.Through the above methods,the main results and conclusions are as follows:(1)Successful use of multi-source multi-temporal remote sensing images to extract information on changes in landslide-induced factors such as land use,vegetation cover,and human engineering activities.This paper makes full use of the advantages of medium-and highresolution remote sensing images in the extraction of local information,incorporates various change factors into the evaluation index system of the landslide disaster spatial prediction model,and combines topography,geological conditions,rainfall and other influencing factors to achieve remote sensing Fusion and application of image and other data.(2)Establish a dynamic spatial prediction model of landslide disaster based on shallow machine learning and deep learning,and realize the dynamic risk analysis of landslide disaster.The relationship between landslide hazards and impact factors is difficult to express with algebraic formulas,and the intelligent algorithm of machine learning can automatically extract the hidden complex relationship between impact factors and landslides through data mining.The paper solves the problems of processing unit selection,hyperparameter optimization,and category imbalance in the process of constructing a spatial prediction model of landslide disasters by shallow machine learning methods(logistic regression,support vector machines)and deep learning methods(convolutional neural networks).Finally,three different models were established,and the prediction results of the different models were compared and analyzed.In this paper,the data of six time nodes is used as input.The model connects adjacent time nodes through small-scale fluctuations of hyperparameters,and a "dynamic model" is obtained,thereby achieving "dynamic risk analysis".(3)The characteristics of regional landslide disaster development and spatial distribution are obtained.In this paper,data from six time nodes of 2008,2010,2012,2014,2016,and 2018 are input to the three models and the results are obtained,and the development and distribution characteristics of landslide disasters in the area are obtained. |
PDF Full Text Request