Object-oriented Research On Typical Landslide Identification Based On High-resolution Remote Sensing Images

Landslide have become one of the major geological hazards that seriously threaten people’s life and property safety and the healthy development of regional economy and society in China.The high-precision identification of landslides is an important basis for the study of landslide hazards,and can provide an important reference basis for landslide disaster prevention and mitigation.In recent years,with the rapid development of high-resolution remote sensing satellites,the high-resolution images obtained by them provide an important data source for landslide hazard identification,and the geographic object-oriented image analysis technology can synthesize various information in high-resolution remote sensing images to identify landslides more accurately and efficiently,but the construction of the current classification rule set has become one of the main problems restricting the development of the automation of object-oriented landslide identification.Based on this,this paper develops two new algorithms for building classification rule sets based on SEa TH,a semi-automated information extraction method commonly used in current geo-object-oriented image analysis technology,and addresses the problems of this algorithm in practice,one is an optimized SEa TH feature selection method,and the other is a classification rule set building method based on SEa TH and SVM.Based on the remote sensing images of Gaofen 2 and Gaofen 6,the two new algorithms were used to construct the classification rule sets and further identify the landslides in the study area by the object-oriented image analysis technology.The research results,compared with the SEa TH method,not only automate and efficiently obtain accurate feature subsets,but also further reduce the reliance on the subjectivity of classifiers in the landslide extraction process,which is of good reference value significance for landslide hazard investigation and early warning in complex scenarios.The main innovative work and research results of this paper include:(1)An optimized SEa TH feature selection method(OPSEa TH)is developed to solve the problems that the SEa TH algorithm only uses J-M distance to judge a single feature without taking into account the influence of dispersion of feature values on feature merit and the possible strong correlation between features,and that the SEa TH algorithm does not have a strategy for determining the classification order.Firstly,the redundant features with high correlation are removed by correlation analysis,and then the features are judged in terms of inter-class distance and intra-class dispersion,and a scheme to find the best combination of features is developed and the classification order can be determined automatically.(2)In order to solve the shortcomings of SEa TH and OPSEa TH algorithms in practical applications that require the original feature set itself to have good classification ability for existing features and the final obtained threshold results cannot be applied to soft classifiers,a method of constructing a classification rule set for high-resolution remote sensing images based on optimized SEa TH combined with SVM is proposed.The new algorithm,compared with the OPSEa TH algorithm,further obtains new features and soft thresholds based on the preferred features using the linear indivisible support vector machine principle,in addition to feature de-correlation and construction of new feature evaluation indexes.(3)Object-oriented landslide identification research was carried out in this region using OPSEa TH algorithm based on the remote sensing image of Ba Dong County with high resolution 6 on April 9,2020.The results show that the overall accuracy of classification based on the classification rule set constructed by the OPSEa TH algorithm is 91.3% and the Kappa coefficient is 0.886,which are 9.3% and 0.124 higher than those of the traditional SEa TH algorithm,respectively;the landslide extraction results of Badong County discriminated by the optical image based on the OPSEa TH algorithm match the unstable area revealed by the In SAR monitoring results The unstable area near the loess slope landslide cluster reflected by the In SAR monitoring results in the same period is thus indicated to be mainly caused by new landslide events.(4)An object-oriented landslide identification study was carried out in this region based on the Hefangtai area’s high-fraction 2 remote sensing images on November 9,2020 and June16,2021,using a new method of constructing a high-fraction remote sensing image classification rule set based on optimized SEa TH combined with SVM.The results show that the overall accuracy of the new algorithm is 93.3% and the kappa coefficient is 0.912,which is 9.3% and 0.121 higher than that of the traditional SEa TH algorithm,respectively.algorithm.In addition,the changes of the landslide area of Hefangtai discriminated by optical images based on the new algorithm are in good agreement with the significant deformation areas revealed by In SAR monitoring in the same period.