Research On Keyesearch On Key Technologies Of Deep Learning In Flood Disaster Assessment

Chinese terrain is high in the west and low in the east,with basins and hills scattered across the country.Under the influence of the monsoon climate,rainfall is unevenly distributed and heavy rainfall cannot be discharged in time,leading to frequent floods,causing damage to crops and reducing production,damage to houses and other buildings,threatening people’s lives and property.It is of great significance to save people’s lives if accurate assessment of flood disaster can be carried out in time and accurate data support can be provided to government workers.Satellite remote sensing is a kind of non-contact remote sensing means,which can detect targets through sensors for a long distance.It has the characteristics of fast,large scale,high precision,etc.,and can meet the demand of large scope and real time in the extraction process of flood inundation range.At present,it has been widely used in the field of flood disaster.However,the current flood disaster information extraction methods based on remote sensing images still have problems of low accuracy and low efficiency.This paper studies the key technologies of deep learning in flood disaster assessment based on remote sensing satellite images.The main work and innovation of this paper are as follows:1.This paper investigates the current research status of flood disaster assessment at home and abroad,and uses remote sensing images to extract the two information of the affected area of the flood disaster and the type of ground features in the disaster area.The disaster situation is analyzed from the two aspects of the disaster area and the ground features in the disaster area.In order to solve the problem that optical remote sensing image is difficult to image due to thick cloud during the disaster,radar image is used to extract the scope of the disaster area during the disaster,and hyperspectral image is used to classify the ground features in the same area before the disaster,and the type of ground features in the disaster area is obtained through superposition analysis.Use multi-source data for information acquisition,focusing on the application of deep learning in the extraction of ground feature types in disaster-affected areas.2.This paper proposes an algorithm flow for extracting the range of the affected area of the radar image based on image processing.Aiming at the radar data before and after the disaster,there is a big difference between the water and the background in the image.Using the Otsu method as the segmentation method of the flooded area and the background can automatically obtain the optimal threshold of the image,saving the time of manually determining the threshold.Compared with the histogram threshold segmentation method and the watershed algorithm,it has better results and separates the water from the background with higher precision.At the same time,it combines the frame difference method and a variety of image denoising algorithms to remove the radar imaging and correction process.The error can accurately obtain the scope of the disaster-affected area.3.A lightweight neural network SDLN model based on Separable convolution and Dense connection is proposed.Using Indian Pines and other three public data sets and Zhuhai No.1 hyperspectral data,experiments are carried out in terms of algorithm accuracy and running time.Compared with other SOTA neural network algorithms and classic machine learning algorithms,the best results have been achieved.The overall accuracy is improved by about 2% on average,and the training time and testing time are shortened by more than 10%,proving the superiority and generalization ability of the deep learning algorithm model.At the same time,the algorithm input size experiment and ablation experiment are set to verify that the input data of 9×9 size is the best,and the effectiveness of separate convolution and dense connection is also verified.4.Based on the obtained disaster-affected area and feature category information of the disaster-affected area,count the submerged area of each type of feature based on the image resolution,obtain the pixel-level flood disaster loss results.Finally use the disaster intensity evaluation formula to make a preliminary estimate of the disaster.