Research On Change Detection Method Of High-Resolution Remote Sensing Image Combined With Land Covering Classification Vector Data

After the successful completion of the first geographic national census,the state began to carry out the normal national geographic state monitoring work.The national monitoring work is characterized by tight time,large scale of operations.So we need efficient,high-precision change detection methods to improve production speed.However,there are limitations in the traditional change detection methods on high-resolution remote sensing images which can't be applied directly to the national geographic state monitoring.This article is aimed at the actual production process for the national geographic state monitoring and focuses on the change detection method using high-resolution remote sensing image based on land covering classification vector data to assist the operator to quickly locate changes,improve operating speed,and reduce the workload.The main contents of this paper are as follows:(1)Select a best segmentation method for change detection.For the data characteristics of the national geographic state monitoring,this article chooses to segment the raster data with the vector data,and then do the multi-scale joint segmentation using the two periods of image.It is proved that the joint segmentation result is superior to the overlapping segmentation in terms of completeness through the experimental comparison of the two methods of the post-processing change detection method.(2)Fine-tuning based on deep learning and propose a hierarchical model to meet the classification system of geographic national census.Utilize Open Street Map vector data and Google Earth high resolution remote sensing image data to make simulation training data sets and fine-tuning based on the GoogLeNet to get a high-precision classification model.It is proved that the deep learning model has a much higher accuracy than the SVM classification model through comparing with the traditional SVM classifier.For the problem that there are some similarities in the characteristics of certain features,this paper provides a solution from two aspects.First,by training the UC Merced data set,it is proved that the hierarchical training method can significantly improve the classification accuracy and while the geographic national census classification system corresponds to a hierarchical model.Second,if the surface objects which are difficult to distinguish exist in the same model,we can strengthen the training through increasing the amount of data.(3)Propose an object-oriented change detection framework.According to the characteristics of geographic national census data,a high-resolution remote sensing image change detection framework is proposed by using the land covering classification vector data.For high-resolution image features,the framework uses multi-scale segmentation to obtain better homogeneity of the patch.In order to obtain the change category for vector update,this paper selects the post-processing change detection method.Finally,the classification model is used to post-processing change detection and gets a better change detection results.