Research On Remote Sensing Image Change Detection Method Based On Multi-feature Synthesis Under Multi-scale Segmentation

With the steady development of social economy in our country,problems such as unbalanced development of urban and rural areas,unreasonable industrial structure,contradiction between resources and environment have become increasingly prominent.Geographical survey and monitoring work plays an increasingly important role in serving social and economic development and ecological civilization construction.By carrying out the geographic survey and monitoring work,we can grasp the relevant trends of China's geographic conditions comprehensively,real-time and efficiently,and profoundly reveal the internal relationship between social and economic development,resources and ecological environment and their development trends.In the major engineering applications of geographic condition monitoring,change region recognition is an important part of its technical process,and change region information recognition mainly relies on the support of remote sensing image change detection technology.Therefore,based on the object-oriented image analysis technology,the change detection method is deeply discussed and studied in this paper through highresolution remote sensing images.This paper systematically summarizes the research status,existing problems and basic theory of change detection technology of remote sensing image,and summarizes its main technical process.Starting from the basic unit of remote sensing image change detection and processing,it is divided into two categories of change detection methods based on pixel and object-oriented image analysis.The principles,contents and corresponding conventional methods of these two kinds of change detection methods are discussed in depth.With the rapid development of high spatial resolution remote sensing images,the feature information provided by a single pixel becomes more and more limited,Traditional pixel-based change detection only uses spectral features of image pixels.But for high spatial resolution images,there will be a large number of phenomena of "foreign body isospectral" and "foreign body isospectral",resulting in certain errors in detection accuracy;data redundancy,slow processing speed,and detection results are subject to noise images.The object-oriented change detection method divides the image into objects.By extracting and analyzing the object features,the change information can be obtained quickly and accurately.This method can make up for the deficiency of the pixel-based change detection method.For object-oriented change detection method,image segmentation,feature extraction and automatic classification are the key and difficult points of this technology.Starting from these problems,the main research contents are as follows:(1)The main technological processes of object-oriented high-resolution remote sensing image change detection include image preprocessing,image segmentation,image object feature extraction,automatic classification and change detection.(2)The principle and content of image segmentation technology are deeply studied.Two kinds of methods based on change detection of image segmentation are summarized.The multi-scale segmentation algorithm based on eCognition is deeply studied.Three main parameters of the multi-scale segmentation algorithm are discussed experimentally,especially the determination of the optimal segmentation scale,The GLCM homogeneity(Homogeneity)and Entropy(Entropy)based on texture features are proposed to construct the evaluation model function,and to construct the change curves of the segmentation scale and the evaluation indexes.The comparison and analysis with other evaluation indexes verify the feasibility of the HOEN evaluation index method proposed in this paper.Finally,the optimal segmentation scale of each object is judged.Using the determined parameters to guide the image segmentation under multiple optimal segmentation scales,the image object layer of multi-level network is constructed,which makes the expression of the image object more appropriate to the actual objects after segmentation,and effectively avoids the situation of over-segmentation or lack of segmentation.(3)Based on the establishment of image object layer under multi-scale segmentation,the spectral,texture and spatial structure features of image objects and the corresponding features of real objects are comprehensively analyzed.Aiming at the image problems of classification results and change detection accuracy in objectoriented image analysis technology,the quality and dimension of feature selection are discussed,and the segregation threshold method SEaTH and FSO feature space optimization algorithm are deeply studied.Through the FSO algorithm,the redundant features are preliminarily screened to reduce the feature dimension and synthesize the optimal feature combination.Then,SEaTH algorithm is used to automatically select the optimized feature set and classify it after determining the threshold value.(4)Finally,combined with the example of high-resolution remote sensing image,the key technologies in(2),(3)are applied to optimize the scale segmentation and feature space optimization of T1 and T2 temporal images.After completing the classification of temporal images,change detection is carried out to distinguish the change information and non-change information,and automatic acquisition of change information is realized.