Establishment Of Target Feature Library Based On Multi-resource Data And The Research Of Extraction Method Of Natural Resource Asset Information

Since the Third Plenary Session of the 18th CPC Central Committee,the departure audit of natural resources assets is becoming an important management measure,accurate and timely extraction of natural resource assets information is an important guarantee for the successful implementation of the departure audit of natural resource assets.Remote sensing has the advantages of wide coverage,low cost,and strong current situation,taking remote sensing images as data sources to extract auditing essential information,which can provide effective technical support for the departure audit of natural resources assets.The spatial resolution and spectral resolution of remote sensing images are mutually restricted,it's difficult to take into account the spatial information and spectral attributes of the objects with a single data source.This paper attempts to combine ZY3,Landsat8 and Digital Elevation Model?DEM?to form a more complete description of natural resource targets,which can assist the effective extraction of natural resource asset information.The object-oriented classification method is used to extract the natural resource asset information with the interpretation features of the image object such as spectrum,shape,texture,etc.,which can fully fuse the advantages of multi-source data,avoids the problem of easy to produce"pepper and salt noise"and single interpretation features with pixel-oriented classification,it has become one of the major methods in the remote sensing information extraction.It is the basis of object-oriented classification to obtain the image object that is close to the real object through image segmentation,the multi-scale image object layer constructed by the Fractal Net Evaluation Approach?FNEA?can better reflect the actual composition of the target,however,FNEA uses a single segmentation scale which is difficult to solve the problem of different optimal segmentation scales for different targets;secondly,auditors often lack expertise in the extraction of remote sensing information,the traditional information extraction methods require to select training samples through visual discrimination,the sampling process is time-consuming and labor-intensive,it also has strong dependence on expert interpretation knowledge,resulting in the extraction of natural resource asset information lack efficiency and automation,finally,multi-source data face the challenge of high dimensionality of interpretation features,dimension reduction is a feasible solution,but feature selection also requires certain prior knowledge.Therefore,this paper focus on how to reduce the level of human participation in the extraction of natural resource assets,realize the accurate and rapid extraction of natural resource asset information,and solve the problem that non-professional auditors lack interpretation knowledge.Regarding these issues above,this paper studies the selection of optimal segmentation scheme based on multi-source data and multi-segmentation methods,the establishment of natural resource target feature library based on multi-source data,the construction of classification model based on high-dimensional features,and an experiment to extract natural resource assets is conducted.The main research work of this paper is as follows:?1?Integrating with ZY3,Landsat8 and DEM,used FNEA to segment data in different combinations,and Spectral Difference Segmentation?SDS?was used to solve the problem of different optimal segmentation scales with different targets,and improved the segmentation effect;based on the modified Euclidean Distance 3(ED3_modified)to quantitative evaluate the above multiple segmentation schemes,and obtained optimal segmentation scheme based on multi-source data and multiple segmentation methods;?2?Comprehensive consideration of natural resource audit requirements and land cover information that can be reflected by remote sensing images,designed a natural resource classification system,used PostGIS database management system to store and manage object-level audit objectives and interpretation features,realized the construction of target feature library based on multi-source data;?3?Based on the target feature library,boosting technology was used to integrate decision tree C5.0,and a robust classification model based on high-dimensional features was constru-cted to solve the problem that feature selection depends on prior knowledge;?4?Used multi-temporal and multi-source data to carry out natural resource asset information extraction experiments,which verified the feasibility of the target feature library assisted extraction of natural resource asset information.It is shown that:multi-source data collaborates and multiple segmentation methods collaborates both can improve image segmentation quality;the rules are robust which using decision tree C5.0 with boosting technology based on the target feature library;accuracy of extracting natural resources asset information in multiple phases is relatively high,the extraction of natural resource asset information assisted by the target feature library can reduce the human participation to a certain extent,and increase the efficiency and semi-automation of natural resource asset information extraction.The research results of this paper solve the problem that non-remote sensing auditors lack expert knowledge,it has important practical value for the accurate and semi-automatic extraction of natural resource assets information,and it also has a good application prospect in the departure audit of natural resource assets.