Show simple item record

dc.contributor.authorLAU, KING SHING ALBERT
dc.contributor.otherFaculty of Artsen_US
dc.date.accessioned2012-08-10T08:23:31Z
dc.date.available2012-08-10T08:23:31Z
dc.date.issued1992
dc.identifierNot availableen_US
dc.identifier.urihttp://hdl.handle.net/10026.1/1131
dc.description.abstract

Cloud motion wind (CMW) determination requires tracking of individual cloud targets. This is achieved by first clustering and then tracking each cloud cluster. Ideally, different cloud clusters correspond to diiferent pressure levels. Two new clustering techniques have been developed for the identification of cloud types in multi-spectral satellite imagery. The first technique is the Global-Local clustering algorithm. It is a cascade of a histogram clustering algorithm and a dynamic clustering algorithm. The histogram clustering algorithm divides the multi-spectral histogram into'non-overlapped regions, and these regions are used to initialise the dynamic clustering algorithm. The dynamic clustering algorithm assumes clusters have a Gaussian distributed probability density function with diiferent population size and variance. The second technique uses graph theory to exploit the spatial information which is often ignored in per-pixel clustering. The algorithm is in two stages: spatial clustering and spectral clustering. The first stage extracts homogeneous objects in the image using a family of algorithms based on stepwise optimization. This family of algorithms can be further divided into two approaches: Top-down and Bottom-up. The second stage groups similar segments into clusters using a statistical hypothesis test on their similarities. The clusters generated are less noisy along class boundaries and are in hierarchical order. A criterion based on mutual information is derived to monitor the spatial clustering process and to suggest an optimal number of segments. An automated cloud motion tracking program has been developed. Three images (each separated by 30 minutes) are used to track cloud motion and the middle image is clustered using Global-Local clustering prior to tracking. Compared with traditional methods based on raw images, it is found that separation of cloud types before cloud tracking can reduce the ambiguity due to multi-layers of cloud moving at different speeds and direction. Three matching techniques are used and their reliability compared. Target sizes ranging from 4 x 4 to 32 x 32 are tested and their errors compared. The optimum target size for first generation METEOSAT images has also been found.

en_US
dc.description.sponsorshipMeteorological Office, Bracknellen_US
dc.language.isoenen_US
dc.publisherUniversity of Plymouthen_US
dc.titleAPPLICATION OF IMAGE ANALYSIS TECHNIQUES TO SATELLITE CLOUD MOTION TRACKINGen_US
dc.typeDoctorateen_US
plymouth.versionFull versionen_US


Files in this item

Thumbnail
Thumbnail

This item appears in the following Collection(s)

Show simple item record


All items in PEARL are protected by copyright law.
Author manuscripts deposited to comply with open access mandates are made available in accordance with publisher policies. Please cite only the published version using the details provided on the item record or document. In the absence of an open licence (e.g. Creative Commons), permissions for further reuse of content should be sought from the publisher or author.
Theme by 
@mire NV