| The mountains and hills in our country are widely distributed,and geological disasters represented by landslides frequently occur,which will cause great harm to the people's lives and property safety and social and economic losses.With the rapid development of my country's economy and society at the current stage,slope instability and other disasters have become more and more obvious constraints on the sustainable development of my country's economy and environment,and the harm to society has become more and more serious.Due to the widespread existence of mountain slopes,the complexity of stability discrimination has always been a difficult problem in geological disaster prevention and control.In recent years,the research of artificial intelligence has been a hot spot in various scientific research fields,and the judgment of slope stability has also become a research hot spot.With the continuous maturity of artificial intelligence,machine learning and deep learning algorithms have become regional slope stability classification evaluations and an important means of landslide identification.The artificial intelligence method does not need to fully acquire the parameters of the physical mechanics method like the traditional method,and the stability of the slope can be evaluated early on the premise of ensuring the amount of data and the optimized model.It provides a faster and more accurate method for early identification of landslide hidden danger points and automatic identification of mass landslides.The thesis mainly has two aspects of work.On the one hand,from the remote sensing satellite image,the convolutional neural network is used to identify landslides from the satellite images of Xinyuan County,Xinjiang,which provides a basis for the automatic identification of mass landslides after earthquakes and heavy rains.On the other hand,from the aspect of slope data,it verifies the substitutability of remote sensing data and engineering survey data,uses engineering survey data to replace remote sensing data for modeling training,and preliminary evaluation of the stability of the earth-rock slopes of Chongqing Wanzhou section using machine learning algorithms,and preliminary screening of hidden dangers of the slopes.The specific research content and research results are:(1)The study obtained a total of 199 soil-rock slope data in the Wanzhou section of the Three Gorges Reservoir.A total of 254 satellite images were intercepted in Xinyuan County,Xinjiang,including 102 landslide images and 152 other topographical images.(2)Sort out and analyze the collected data,select slope height and slope angle,which are important and easily accessible factors that affect the soil-rock slopes,as the early evaluation factors of slope stability,and perform statistics and visualization of the evaluation index data characteristics.(3)The three machine methods of k-nearest neighbor,random forest and support vector machine were used to model the soil-rock slopes of the Three Gorges Reservoir section in Wanzhou District and select the optimal parameters.The obtained optimal model was performed on the 40 slopes of the test set.The evaluation is compared with the actual stability state of the slope.The results show that the evaluation accuracy of k-nearest neighbor is 67.5%,the evaluation accuracy of random forest evaluation is 70%,and the evaluation accuracy of support vector machine model is 65%.Combining the evaluation results of the three machine learning algorithms is of great significance for the " audition" of hidden danger points and landslide warning for largescale slopes.(4)For the collected remote sensing satellite images of Xinyuan County,Xinjiang,the convolutional neural network algorithm based on GoogLeNet is used for image recognition,and the prediction accuracy rate is 81.25%.It can realize the rapid identification of large-scale mass landslides,and provide the basis for the emergency rescue and disposal of landslide disasters in the later period. |
PDF Full Text Request