Flood disaster monitoring and disaster situation analysis method based on visual Transformer

Abstract

The invention discloses flood disaster monitoring and disaster situation analysis based on a visual Transformer. The flood disaster monitoring and disaster situation analysis comprises the following steps: (1) constructing a double-time-phase image change detection model based on the visual Transformer; (2) selecting a dual-temporal remote sensing image, and making a flood disaster label; and (3) carrying out flood monitoring and disaster analysis according to the double-time-phase image change detection model constructed in the step (1). According to the method, a dual-time-phase image change detection model based on advanced vision Transformer in deep learning and radar data which is not influenced by time and weather and is high in penetrability are combined, data when flood occurs can be obtained, and the recognition precision can be improved.

Description

technical field [0001] The invention relates to flood disaster monitoring and disaster analysis methods, in particular to a visual Transformer-based flood disaster monitoring and disaster analysis method. Background technique [0002] Natural disasters occur frequently in my country, among which floods occur most frequently. Flood disasters are extremely destructive and pose a huge threat to the ecological environment, human life and property. Floods have the characteristics of wide range, frequent occurrence, strong suddenness and large losses. The submergence of large areas caused by floods is dynamic, and real-time monitoring of flooded areas is very important. Based on 3S technology and its mutual cooperation, it is the main means of monitoring flood disasters. Flooded area is the basis of disaster assessment, which depends on timely and effective data. With the rapid development of remote sensing technology, a large amount of satellite data is available for flood mo...

AI Technical Summary

Problems solved by technology

However, there are still the following shortcomings in the above-mentioned technologies, which make it challenging to monitor floods: 1. The accuracy of traditional remote sensing flood monitoring methods is not high, it is difficult to exceed 90%, and is limited by factors such as available data; 2. Semantics based on CNN Segmentation networks are often used in tasks such as change detection and target recognition in remote sensing images, but they are not satisfactory for long-range spatio-temporal information.

[0003] In view of the fact that the current flood monitoring using remote sensing technology mainly relies on algorithms such as index method, threshold segmentation and machine learning, these methods are simple and are easily limited by factors such as image quality, terrain and human intervention, and it is difficult to achieve accurate and efficient flood monitoring. ; In addition, because floods are often accompanied by clouds and rain, optical satellites capture a lot of useful water body information through clouds and fog, and it is inevitable that there is a lack of available data for flood monitoring

Images