1 / 21

CSC400W Honors Project Proposal

CSC400W Honors Project Proposal. Understanding ocean surface features from satellite images Jared Tilanus Nemanja Spasic. Project Background. Project Supervisor: Dr. Anet Potgieter

liluye
Télécharger la présentation

CSC400W Honors Project Proposal

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. CSC400W Honors Project Proposal Understanding ocean surface features from satellite images Jared Tilanus Nemanja Spasic

  2. Project Background • Project Supervisor: Dr. Anet Potgieter • Proposed by Mr. Laurent Drapeau, member of the French company De l’Institut de Recherche pour le Développement , and Prof. J. Field of the UCT Oceanography department • Mr. Drapeau’s company is comparing the ocean features of South Africa and South America • Prof. J. Field has a lot of oceanographic data that he needs visual representations for

  3. Understanding ocean surface features from satellite images • Develop a system to automatically detect features from thermal images • Fronts (where cold and warm water meet) • Eddies • Upwelling • Gather information about these features • Important to the study of the ocean as these features determine lots about ocean life

  4. Understanding ocean surface features from satellite images • Our system will give quantitative information on current conditions • System also aims to detect patterns in how these features occur • Seasonal averages • Seasonally persistent features • Predict how features evolve

  5. Understanding ocean surface features from satellite images • Jared will do develop image processing software to detect (and possibly identify) features from the original images • Nemanja will develop a Bayesian network to identify features and recognise patterns

  6. Image Processing • Working with the satellite images • Make the computer recognize fronts • Position • Temperatures • Size • Detect features • Eddies etc.

  7. Image Processing • Essentially edge detection, segmentation and feature recognition • Many algorithms exist • My project is to select ones that will work on the noisy data we have and implement them • Algorithms need to be tuned to work optimally

  8. Image Processing • Data is noisy by nature and incomplete • Features are messy and hard to distinguish exactly • Areas are often covered by cloud • Will probably use an algorithm that tracks features across multiple images • Eliminates some noise • Temporal changes are clearer

  9. Image Processing • This section alone will be useful to Oceanographic researchers • Accurate information about these features current status will be valuable for other research

  10. Image Processing • Success of this section will be best evaluated by eye • By overlaying detected features on the original images one will be able to see how effective the software is

  11. Output Format • Will be a challenge representing data that is output • Initially will probably be stored in some XML format • Perhaps topic maps • Would be useful to represent it as an image • Easy to see lots at once

  12. Output Format • Difficult to represent temporal information in an image • Will do user requirements gathering to see what information is important • Will evaluate intuitiveness and informativeness on users • Expert and non-expert

  13. Bayesian networks summary • A directed acyclic graph (DAG) • Consists of a set of nodes: variables or uncertain quantities • Nodes are linked by directional arcs , where the parent node is the cause and the child node is the effect • Links represent informational or casual dependencies among nodes, which are given in terms of conditional probabilities • Each variable has a finite set of mutually exclusive propositions - states

  14. Bayesian networks summary 2 • Bayesian networks can be singly-connected (without loops) or multiply-connected (loops) • A Dynamic Bayesian network handles varying values for each variable over a time period and is probably best suited to the project

  15. Bayesian network software • Open source software will be used initially to learn how to use a Bayesian network • Potential software would be : BayesiaLab and Bayesian network tool in java – BNJ • Available open source packages are very slow to train and do not handle temporal data patterns

  16. Temporal Bayesian Inference • The data we will have access to is temporal and thus software will have to be designed to allow the Bayesian network to handle temporal data • Dr. A. Potgieter has algorithms that can be used to develop software for temporal data inference by a Bayesian network • Research will have to be extensively done to design the required software.

  17. Bayesian network data input interface • A user friendly interface will be designed to enable quick, efficient and easy entry of data into the Bayesian network • User Centered Design will be used to accomplish the use friendly interface goal. • Probable software for implementing the user interface would be visual c++, visual j++ builder of Flash MX

  18. Output visualization • The output of the Bayesian network will probably be stored in xml or topic map format • The stored output data will probably be converted to a bmp format to allow most graphical software packages to open them and • bmp format is a binary rasta (pixel based) format so it is easy to work with

  19. Project Benefits • Beneficial to research being done by Mr. Drapeau • Beneficial to the UCT oceanographic department as they will have visual representations of their data • Allow researchers to easily access information contained in thermal images of the ocean surface • Beneficial to local fishermen as they will be able to detect which ocean surface patterns attract the most fish • May be used by a person studying migration of fish to determine which ocean feature makes fish migrate

  20. Project Successfulness • Comparing the output data of the Bayesian network and the input satellite images will give a clear indication of the success of the prediction and inference of the Bayesian network • Comparison to an existing Oceanographic model will also be used as a success rating • A non-experts opinion of the final output visual representation will give a good idea of the projects visual representation success

More Related