Objectif
In project FLIERS we propose to develop a new approach to mapping land cover from remotely sensed data and for incorporating such land cover information in Geographical Information Systems.
The project is based on the reality that natural landscapes present a continuum of variety at many different spatial scales and that a high proportion of the discretely-sampled pixels within an image contain mixed spectral signatures, and are not easily placed into fixed thematic classes. Since the approach advocated here contains an implicit recognition of the reality of the continuum we believe that it will be considerable improvement over the traditional concept of "thematic classification" of satellite imagery.
The approach involves three complementary activities:
1. neuro-fuzzy classifiers - use of classifiers which admit recognition of uncertainty in class allocation, and store that uncertainty for later manipulation. Research will examine fuzzy neural network classifiers, texture-based neural computing, and other alternative classifiers, and all outcomes will be compared with standard classifiers;
2. scientific visualisation - user interaction in the decision making process, via visualisation, at all stages. This will involve projection pursuit to assist data compression and selection for classification, interactive visualisation by hypermap linked views of multiple images and source data to assist combination of the outcomes of classification, and interactive manipulation to produce tailor-made thematic map products for particular purposes, and
3. ground verification - comparison of the outcome of classification with ground information is crucial, and this activity will include detailed vegetation, land cover and land use mapping from aerial photographs and in the field. The approach to field mapping will require development of novel approaches to mapping the continua of vegetation so that it can be used to verify multiple resolution remotely sensed data.
The research partnership established to address these issues includes five organizations which bring to bear a unique combination of skills, including those in advanced computational image processing involving neuro computing and texture analysis (Southampton, VTT and JRC), scientific visualisation of spatial information (JRC and Leicester), and innovative vegetation and land use mapping in remote sensing (Leicester and Thessaloniki).
Above all the approach advocated here is a total environment for the recognition that the allocation of pixels to classes is uncertain and that acceptance of that uncertainty and its manipulation could strengthen the use of remote sensing as an input to subsequent activity.
Keywords - Fuzzy sets, fuzzy mapping, fuzzy neuro classifiers, texture classifiers, hyperrnap, interactive decision making, projection pursuits, ground verification, multiple sensor resolutions.
Champ scientifique
- engineering and technologymechanical engineeringvehicle engineeringaerospace engineeringsatellite technology
- engineering and technologyenvironmental engineeringremote sensing
- natural sciencesearth and related environmental sciencessoil sciencesland-based treatment
- engineering and technologyelectrical engineering, electronic engineering, information engineeringelectronic engineeringsensors
- natural sciencescomputer and information sciencesartificial intelligencecomputational intelligence