Research On Seismic Damage Information Extraction Based On GF-2 Remote Sensing Image

As a natural disaster,earthquakes have caused huge loss of life and property due to their sudden and destructive characteristics.After the earthquake,due to the destruction of the affected area,there are many difficulties in getting into the disaster area quickly.Remote sensing is a technology that can acquire ground-related information without being restricted by ground conditions.It has the characteristics of fast,stable and multi-data sources.With the gradual maturity of remote sensing technology,it has become an increasingly important method in earthquake damage research.How to use modern remote sensing technology to quickly and accurately extract earthquake damage has become a key link in post-earthquake emergency work.The accuracy of traditional seismic damage extraction methods is often insufficient,and the extraction efficiency is not high.However,the pixel values of the self-image elements in the low-and medium-resolution remote sensing images are low,which also affects the accuracy of the classification results.With the advent of object-oriented feature extraction methods and the widespread use of high-resolution remote sensing images,the application of these is very important for the fast and accurate extraction of ground objects after earthquakes.significance.In this paper,the high-altitude remote sensing image is used to study the extraction of seismic features from August 8,2017 in Zhangzha Town,Jiuzhaigou County,Aba Tibetan Autonomous Prefecture,Sichuan Province.On the basis of summarizing the methods and precision evaluation of common features,using methods based on pixel,hierarchical extraction,object-based and a combination of different features,try to extract buildings,roads and landslides,and compare the results of different methods.Precision.The main conclusions of the paper are as follows:1.Three methods are used to extract roads and landslides after the earthquake.The results show that there are multiple misclassification points based on pixel extraction results,the overall accuracy is 80.45%;the spectral feature extraction results based on the object have higher leakage ratio,the overall accuracy is 79.68%;based on the texture feature extraction results of the object The leak rate is the highest,and the overall accuracy is 86.31%.In general,the object-based texture feature extraction accuracy is the best;in the case of damage,the landslide extraction of the pixel accounted for 14.5%,the road image is serious,the road damage ratio is 32.08% in the experimental area,the road accessibility is poor,affecting post-disaster relief.2.Using the combination of hierarchical grading and shape features to extract the building information of the three levels before the earthquake,and optimize the accuracy of each level of ground extraction by manually identifying the misclassification and the missing points.The results show that the accuracy of each grade is improved by about 10%,and the extraction accuracy of the building before the earthquake reaches 85.33%.3.Extract the post-earth building using object-oriented methods that combine different features.The results show that the spectral features combined with the texture features have the least points,the overall accuracy is 83.74%;the spectral features combined with the shape features have higher leakage ratio,the overall accuracy is 78.23%;the texture features combined with the shape feature errors The points are the most,and the overall accuracy is 79.54%.The spectral features combined with the texture features combined with the shape features have the highest leak rate and the worst overall accuracy is 75.14%.In general,the results of spectral features combined with texture feature extraction have lower leakage ratio,lowest error rate and highest precision.4.Through the comparative analysis of building extraction accuracy before and after the earthquake,the results show that the number of pixels extracted after the earthquake increased from 154712 pixels extracted before the earthquake to 175163 pixels,and the damage area ratio is about 1.04%,indicating that the earthquake has a small image of the buildings in Jiuzhaigou.In the change detection of buildings,three typical buildings were selected for change detection based on principal component analysis.The results show that the damage percentage of the three buildings is about 5%,and the buildings are basically intact after the earthquake.