Research On Method Of Landslide Recognition Based On Object-Oriented Image Analysis Technology

As a highly destructive and frequent geological disaster,landslides have caused huge economic losses and survival disasters to the production,life and economic construction of human beings in our country and even the world.The research in this paper is based on remote sensing technology,machine learning and neural network,using Sentinel-2 optical remote sensing image and digital elevation model as data sources,using object-oriented image analysis technology,through remote sensing data preprocessing,segmentation,fusion,and classification Waiting for multiple steps to realize the identification of landslides and build a model of landslide identification.Experiments show that the method in this paper can effectively identify landslides and achieve a better identification effect.The main research content and work of this paper are as follows:(1)Based on optimized support vector machine for landslide recognition,preprocess the Sentinel-2 optical remote sensing image and combine the elevation information obtained from the digital elevation model to form the characteristics of landslide recognition.The method of object-oriented image analysis technology is adopted through K-means algorithm performs remote sensing image segmentation,using the segmented object instead of pixels as the basic identification unit.After segmentation,there are a large number of non-landslide sections.The region growth algorithm is used to merge the non-landslide sections.The region growth algorithm seed point selection and growth criteria are based on the landslide with unique information.After segmentation,merging,and calculation of different segment features,the pixel-based landslide recognition problem is transformed into a classification problem based on landslide features.On this basis,a variety of machine learning algorithms are explored for classification.Finally,a support vector machine algorithm combining principal component analysis and Bayesian optimization was selected for classification,and a good classification accuracy was achieved.(2)Landslide recognition based on a fully connected network,after exploring a variety of machine learning algorithms,the same preprocessing,segmentation,and merging are performed on the elevation information obtained from the Sentinel-2 optical remote sensing image and the digital elevation model.After the steps,the neural network method was explored during the classification,and the fully connected neural network with weight sharing and widely used was used as the algorithm for training the classification model.The experimental results showed that the landslide recognition method based on the fully connected neural network has achieved good results.The accuracy index of the landslide recognition model can reach more than 80%,which can realize the effective identification and detection of landslides,and perform the training of the fully connected neural network Factor weight analysis to analyze the contribution of different characteristics to landslide identification.