Research On Land Use Classification Based On Homogeneous Image Segments Analysis

The acquisition of land use information is of great significance to the land resource management department.Land use classification based on high-resolution remote sensing images is one of the current hotspots and main technical means.Among many classification methods,the image classification method based on homogeneous image segments can make full use of the spectrum,texture,shape and other features of the image,and the classification results have high accuracy and good application prospects.In the methods,image segmentation and feature selection are the key factors affecting classification accuracy.Therefore,for these two research-topics,taking the engineering practice of land use classification in Xingyang City,Henan Province as an example,combined with five common classification methods-Bayes,nearest neighbor,decision tree,random forest and fuzzy classification,the acquisition of optimal segmentation scale and feature selection mode were researched in this paper.And a set of remote-sensing-image-classification technology process,which was simple,practical and high accuracy,was designed.The main research contents and conclusions of this paper are as follows:(1)From two perspectives of single-level classification and multi-level classification,RMNE and RMAS indicators were constructed respectively to investigate the acquisition of global and local optimal segmentation scales for features.The experimental results showed that it could obtain good parameters with the multi-level local optimal segmentation scale based on RMAS,which could effectively remove the influence of under-segmentation and over-segmentation on the classification results,and improve the classification accuracy.(2)Two feature selection schemes were designed:(1)Relief F algorithm and J-M distance to construct feature subsets,(2)Wrapper algorithm to construct feature subsets according to specific classifiers.The experimental results showed that the first scheme was suitable to be used in combination with the fuzzy classification method to improve the classification efficiency and accuracy.(3)According to the experimental results of the cross-combination of the above two image segmentation strategies and feature selection schemes,the technical process of multi-level fuzzy classification was applied to the practice of land use classification.The results showed that the method could effectively determine the attribution of uncertain ground objects,and had high efficiency and high classification accuracy.So it had good application value for engineering practice.There are 25 figures,15 tables,and 90 references in this paper.