Font Size: a A A

Study On Remote Sensing Image Segmentation Based On Clonal Selection And Clustering Algorithm

Posted on:2015-12-04Degree:DoctorType:Dissertation
Country:ChinaCandidate:H J GuanFull Text:PDF
GTID:1228330452453723Subject:Cartography and Geographic Information Engineering
Abstract/Summary:PDF Full Text Request
With the development of remote sensing technology, high-resolution images fromnew remote sensing platform not only have abundant spectral information, but alsocontain more shape, texture details of ground surface. The huge challenges followed byare how to effectively deal with those data and how to prepare for the specificapplication of the subsequent. Segmenting remote sensing image is the key step beforeanalysis and understand the image data after processing. Good or bad imagesegmentation results can directly influence the following process that describing thefeature extraction, recognition and classification of the targets.The paper focus on the key problem that how to improve the segmentationaccuracy of high-resolution image and study three theory such as clonal selectionalgorithm, fuzzy C-means clustering (FCM) and spectral embedded clusteringalgorithm (SEC) which often used on image segmentation. The paper proposes threeimproved image segmentation algorithms for high-resolution remote sensing imagebased on the previous research on image segmentation algorithms and gives severalexperiments on QuickBird high-resolution remote sensing image.In this paper, research results are as follows:Firstly, through in-depth analysis the clonal selection theory of artificial immunesystems, aiming at the deficiency of basic clonal selection algorithm, adding crossoverand regulating the concentration of antibodies to control population size, an improvedclonal selection algorithm is proposed. The algorithm can enhance antibody diversityand improve the global search ability. The image segmentation experiments thatimproved clonal selection algorithm combining respectively with2-D maximumentropy image segmentation method (ICS2DME) and multiple spatial constructionimage segmentation method (ICSMSC) show that improved algorithm better than thebasic one.Secondly, aiming at the deficiency of spectral clustering (SC) and spectralembedded clustering algorithm (SEC), a kernel function-based spectral embeddedclustering algorithm (KSEC) is proposed by adding kernel function into spectralembedded clustering algorithm. The paper compares three usually used kernel functionand uses spectral embedded clustering and kernel function-based spectral embeddedclustering algorithm firstly in the segmentation of high-resolution remote sensingimages. Experiments of segmentation on high-resolution remote sensing images show that KSEC can effectively improve the accuracy of segmentation than K-means, SCand SEC.Thirdly, based on studying fuzzy C-means clustering (FCM) and some improvedfuzzy C-means clustering algorithms, the paper proposes a novel fuzzy localneighborhood-attraction-based information C-means clustering algorithm (FLNAICM)which incorporated the local spatial and gray information based on neighborhoodattractions between the center pixel and its neighborhood pixels. In FLNAICM,neighborhood attractions are used as a measure to balance the influences on the centerpixel from its neighborhood pixels, aiming to incorporate appropriate local spatial andgray information for improving the robustness and noise insensitiveness of theconventional FCM. Experiments show that FLNAICM can effectively improve theaccuracy of segmentation on high-resolution remote sensing images than the above twomethods.At last, the paper compares the image segmentation accuracy on high-resolutionremote sensing images by the three improved algorithms which proved thatsegmentation method based on fuzzy theory is more effective than the other two.
Keywords/Search Tags:high-resolution remote ssensing, image segmentation, clonal selection, fuzzy C-means clustering, spectral embedded clustering, kernel function
PDF Full Text Request
Related items