Research On Flood Disaster Assessment Technology Based On Multi-source Data

Flood disaster is one of the most destructive natural disasters,and the occurrence of flood events will cause serious damage to the regional ecological environment and social development.With the strengthening of global climate change and human activities,the impact of flood disaster events has become more severe,and the damage to social and economic development has become more profound.In order to realize the comprehensive management of flood disaster events,it is necessary to make comprehensive,accurate and timely response to flood events.Flood disaster land assessment is the main basis for disaster prevention and mitigation.For the time before,during and after the disaster,it is worthy of our in-depth discussion to realize the task of regional flood disaster assessment.How to make full use of multi-source data,comprehensively evaluate regional flood events,quantify regional flood risk,achieve scientific statistics on disasters,and improve the objectivity of flood event evaluation are the key points of this study.This paper takes Hubei Province as the research area,Due to the topography,climate,temperature and rainfall factors in this area,frequent floods occur,which cause great harm to the society and seriously restrict the economic development of the region and the safety of citizens’ lives and properties.In the summer of 2020,a flood event occurred in southern China,and Hubei Province suffered from it.Based on multi-source data,this paper evaluates floods in the study area from various time points of flood events,and applies machine learning and deep learning methods to complete the evaluation tasks.The main work and innovations are as follows:1.In view of the establishment of disaster risk assessment model,this paper proposes to combine GIS technology and web crawler technology,based on meteorological raster data,geographic information data,socio-economic data and medical index data and other data,establish an index system through index calculation and data preprocessing,and then use principal component analysis（PCA）to confirm the weight of each index,and finally establish the flood disaster risk assessment model of Hubei Province,and realize the flood disaster risk assessment of Hubei Province at the county level.2.When acquiring flood area,this paper proposes an algorithm flow for extraction of waterlogging range based on sentinel-1 radar image data.In order to enhance the characteristics of water bodies,the radar images are processed by the Sentinel-1 water body index method（SWI）.In order to obtain the range of water body,the OTSU threshold method（OTSU）is used to automatically select the threshold value of the image data after the index operation,and the water body information data before and after disaster is obtained.The SWI-OTSU water body extraction algorithm can automatically select the threshold value with the most obvious difference between water body and background.The automatic threshold selection saves time cost,and reduces the error interference of manual threshold selection,making the water body extraction more accurate.Finally,the change area of the water body is obtained by frame difference method,and after the image denoising process is performed on the change result,the flooded area data of the flood event in Wuhan is obtained.3.In the task of ground object classification,this paper proposes a deep learning algorithm for ground objects classification based on multispectral remote sensing image data.Through residual mechanism,dense connection mechanism and attention mechanism,a dual-branch lightweight neural network model RDDA is designed to realize the recognition and classification of typical ground objects in Wuhan.Using L8 SPARCS multispectral open data set and Wuhan Sentinel-2 multispectral data set,experiments are carried out from the recognition accuracy and training time of algorithm operation,and the results are compared with other classical network algorithms.Finally,it is proved that the RDDA model proposed in this paper has better performance in the multispectral remote sensing data classification task.4.Aiming at specific flood disaster events,this paper proposes an algorithm flow to realize the loss statistics of flood events through the deep learning method.After obtaining the inundation range of the flood event in Wuhan and the classification results of the ground objects in Wuhan.Unify the spatial resolution of the result data and superimpose the data,so as to count the types of disaster-affected objects within the flooded area,and realize the loss assessment of the disaster situation.