Font Size: a A A

Research On Remotely Sensed Image Classification Techniques

Posted on:2002-03-10Degree:DoctorType:Dissertation
Country:ChinaCandidate:N HuangFull Text:PDF
GTID:1118360032451970Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
Research On Remotely Sensed Image Classification TechniquesA dissertation subm itted for the degree of Doctor of Philosophy Ning HUANG (Signal and Information Processing)Directed by Professor ZHU Min-HuiProfessor ZHANG Shou-RongWith the development of DSP technology, computer technology and communication technology, remote sensing takes more important effects in the field of military and civil applications. Pattern classification is a key technique in remotely sensed image processing. Although the research history of pattern classification techniques is quite long, users require for more accurate classification result and more smaller computing load now. So there is an urgent need for modem pattern classification methods to solve the modem remote sensing applications.The major work of the thesis aims to the study of remotely sensed image classification. It describes the architecture of pattern classification, theories, ideas and methods, and gives a brief review about the current remotely sensed image classification techniques.There are two pattern classification techniques studied in the thesis. Considering the complexity and fuzziness of remotely sensed image data, Fuzzy clustering methods are right choices for the remote sensing applications. But Fuzzy clustering algorithms make decisions on a pixel-by-pixel basis and does not benefit from the spatial information, regardless of the pixels?correlation. In this paper we introduce a novel Fuzzy C-means algorithm which is based on image抯 neighborhood system. During classification procedure, the novel algorithm refers neighboring pixels?fuzzy membership information. The experiment shows that the algorithm gives a more ~smooth classification result and cuts the computation time.We use the Markov Random Field Model in the remote sensing applications. The image model we use in the thesis is the Hidden Markov random field (HMRF) model with respect to the observable image pixel intensity and unobservable image class label. In our classification algorithm framework, the finite Gaussian mixture(FGM) model is used to describe the density function of the image pixel intensity, the expectation-maximization (EM) algorithm used for the HMRF model parameters, and maximum a posteriori (MAP) estimation used for estimate of the image class label. The experiments show that our novel HMRF-EM image classification method gives a more accurate and robust image classification comparing to the classical classification methods...
Keywords/Search Tags:Pattern Classification, Clustering, Fuzzy Clustering, Markov Random Field, EM Algorithm
PDF Full Text Request
Related items