Application Of Deep Learning In Automatic Identification Of Earthquake-triggered Landsides

Moderate to strong earthquakes often result in severe co-seismic surface ruptures as well as secondary disasters such as landslides,collapses,debris flows,and soil liquefaction.In recent years,low-altitude unmanned aerial vehicle(UAV)remote sensing technology has greatly improved the ability to observe the ground,providing centimeter-level optical images or laser point cloud data for earthquake emergency response,disaster investigation,and post-earthquake scientific research.However,the rapid and automatic processing of high-precision and massive data greatly restricts the timeliness of data application,and the integration of deep learning in image recognition provides an effective solution to this constraint.This article proposes a machine learning framework for identifying earthquake-induced secondary disasters based on YOLO V5.Through the construction of a secondary disaster sample library(including collapsed buildings,landslides,etc.),machine learning and rule construction are carried out using a convolutional neural network model to achieve real-time feature extraction,target detection,and calibration analysis of earthquake-induced secondary disaster images and video streams.By optimizing the YOLO V5 framework,the feature extraction capability and target detection efficiency are greatly improved.To solve the accuracy limitation caused by the limited number of samples,this article uses spider technology to obtain a high-quality target dataset and construct a sample library,greatly improving the quality and quantity of samples.Finally,based on the QGIS open network map platform,an online visualization of the identification results is built to better display the identification results of secondary disasters,meeting the high timeliness and access requirements during earthquake emergency response.To test the optimized framework for automatic identification of secondary disasters,this article uses the 6.8-magnitude earthquake in Luding County,Ganzi Prefecture,Sichuan Province on September 5,2022 as an example to test UAV aerial photos and video streams with centimeter-level resolution in the earthquake zone.The experimental results show that the optimized YOLO V5 model has high accuracy and stability in identifying earthquake-induced secondary disaster types,and is superior to traditional methods in both recognition accuracy and efficiency,providing data support for early identification and prevention of earthquake-induced secondary disasters.