Method Research Of Extraction Of Natural Resource Assets Information Based On High Resolution Remote Sensing Imagery And LiDAR Data

Natural resource assets departure "off-site audit" need to extract the high aging and precision information of typical natural resource assets,object-oriented information extraction technology based on high-resolution images is an efficient solution.The first target of this solution is to obtain the resource object which is close to the real earth object through the image segmentation technique.And existing optimal segmentation scale selection technique based on the fractal network evolution algorithm focused on the single scale parameter.Most optimal processes ignored the shape parameter and the compactness parameter,which resulted a certain number of mixed-objects.Although the segmentation which is under the best parameter combination can reduce the number of mixed-objects,mixed-objects are still there because of different spectral characteristics with the same objects in the images and the edge of the bridge in the natural.Then,the solution need to extract the robust features of different types of resource objects for classification.The images from the limited platform due to partly cloudy area are usually lack of the infrared wave band.This situation makes accurate extraction of vegetation more difficult.At the last,the solution need to classify the different types of resource objects based on features.A large of different resources brings a huge of features.Simple accumulation of features will not only increase the amount of calculation,and features of redundancy,but also reduce the extraction precision of the natural resource assets.So,this solution need to solve those problems by studying the optimization strategy of segmentation parameters based on the fractal network evolution algorithm,the method of optimizing segmentation by adding the features derived from the LiDAR data,the building of feature which can extract vegetation effectively and the optimal feature subset which can improve the accuracy of resource information extraction.In response to the problems in the solution,this article has done the following:1)Summarizes the existing methods of evaluation of image segmentation.Aiming at that the optimal segmentation parameter selection based on fractal network evolution algorithm mostly focused on a single scale parameter,study the method of determining the optimal segmentation parameters based on the Taguchi method to get better result from the segmentation.2)Aiming at segmentation result with the problem of mixed-objects based on a single image,study adding normalized height feature derived from LiDAR data and the slope feature from DSM to image for the segmentation.This method aims at reducing the number of mixed-objects.3)Aiming at the ineffective extraction of vegetation due to the lack of near infrared band data,build the TBCVI which is based on the visible bands.Study TBCVI and the other six indexes to extract the vegetation information from images which token in the urban,the wasteland,the water and the urban whose images are with a large area of shadow.Compare the extraction results.4)Aiming at simple accumulation of a large number of features resulting the classification accuracy reduction,study the ReliefF and mRMR algorithm to get feature subset which is good for the extraction of resource information.For the sudden reduction of classification accuracy based on each single algorithm,study the RFmRMR algorithm which gets weights using the ReliefF first and eliminate redundant features with mRMR second.Compare the strength and weakness of feature subsets that get from these algorithms with the classification accuracy.5)Study the strategy of extraction of natural resource assets based on the object-oriented extraction process with RBF-SVM classifier.Gain the classification accuracy and area of natural resource assets by getting the fine classification first,merging the objects then.Verify the feasibility of this solution.The result of the study shows that the optimal segmentation parameters get from Taguchi method can get better edges in some regions,and adding features get from the LiDAR point cloud can reduce the number of mixed-object in the segmentation.These outcomes can assist researchers get better segmentation result.This study also shows that extract vegetation with TBCVI is effectively in different environments,the RFmRMR algorithm can obtain a more favorable feature subset for classification faster,and the solution which accurately extracts the natural resource assets based on high resolution image and LiDAR point cloud data is feasible.Natural resource assets departure "off-site audit" can realize based on the method in this thesis.It also shows some important reference values for natural resource assets departure "off-site audit"with multi-source and multi-scale remote sensing data in the future.