Landslide Disaster Evaluation Based On Deep Learning And Remote Sensing And Socialized Emergency Resource Sharing

When large range natural disasters happen,an accurate and enough guarantee of emergency resources is positive to minimize the development and deterioration of the disaster and reduce casualties and property damage in the affected area.This largely rely on the comprehensive and adequate emergency resource guarantees.However,the variety and quantity of emergency resources in national emergency material reserve system are limited,they often cannot fully meet the needs of emergency rescue.Therefore,in the “National Emergency Plan for Natural Disaster Relief” and “National Emergency Plan for Earthquake” issued by the state,it is clearly proposed to encourage and call on social forces to participate in emergency rescue,and to concentrate the whole society through socialized emergency resources sharing to reduce the loss of lives and property in the disaster area as far as possible.This dissertation aims to discuss the issues such as disaster awareness,basic geospatial information acquisition,emergency resource demands analysis and effective sharing in the socialized emergency resource sharing using deep learning,remote sensing.As a result,a feasible solution can be proposed.The research missions are shown as follows:(1)In terms of disaster awareness and basic geospatial data acquisition in the affected area,the remote sensing technology and Deep Learning was combined to propose a recognition method of landslides and the hazard-affected bodies based on an SG-Bi TLSTM model.The model consists of a U-Net and two interacted LSTMs.The U-Net receives remote sensing images and output semantic segmentation maps and multi-channel feature map.On the other hand,the bi-temporal LSTM is used to generate the image caption.In order to recognize and locate the hazard-affected bodies of landslides more accurately,a semantic gate was designed in this dissertation,which can generate the image caption dynamically and adaptively based on the image,context information and the output of prediction LSTM.As a result,the disaster information of a wide-range area can be acquired accurately,the emergency commanders can therefore make a scientific response plan.The effect of recognition of this model was verified through a remote sensing image of Wenchuan taken in 2008.The experimental results show that the model is helpful for the recognition of landslides and the hazard-affected bodies.(2)In terms of the demand analysis of resources,we evaluate the disaster first based on the recognition of the hazard-affected bodies,then analyze the demands of emergency resources quantitatively.Then we take the demand of road rescue as an example,demonstrate that both the professional equipment and human resource support are needed in the rescue process,these needs cannot be meet by the national reserve.As a result,the necessity of socialized emergency resource sharing can be come out.(3)In terms of the emergency resource sharing,this dissertation established a flat resource sharing platform where the demander and supplier of resource can communicate with each other to clear the actual resource demand including the types,numbers and standard that must be obeyed.After the communication,the two parts can sign the sharing agreement.When the execution of the agreement is finished,the demander is responsible to check and evaluate the behavior of the supplier,which can ensure the successful implementation of the sharing.The multi-objective dynamic sharing model of resources based on integrity and charity is established according to the demand information obtained by remote sensing analysis of landslides and their hazard-affected bodies,which can optimize the traditional distribution plan of emergency resources.This dissertation chooses socialized emergency resource sharing,which is a part of the emergency rescue process,to execute the research,takes quantitative analysis of emergency resource based on hazard-affected bodies as a link,integrate the rapid recognition of landslides and their hazard-affected bodies using deep learning and remote sensing and the socialized emergency resource sharing into a complete technical process,which realized the whole process of exploration from disaster awareness and quantitative analysis of emergency resources to the final sharing,and provide a feasible and effective solution for the socialized emergency resource sharing of major natural disasters.