Advanced techniques for remote sensing image processing and recognition

ITPAR - ATRS DIT-PRJ-03-012

Status NOT active project
DISI role Coordinator
Project type Education and Mobility Project
Dimension International
Acquisition date 2003-07-01
Start date 2003-07-01
End date 2005-07-01
SAP code 30300068

Project details

Project astract In the last decade, remote sensors have shown a significant quantitative as well as qualitative development. The quantitative aspect is expressed by the numerous launches of remote sensing satellites such as the Landsat 7, Ikonos, IRS-1D and ERS 2 and the future missions/satellites like the Italian Cosmo-SkyMed and the Indian IRS-P5; while the qualitative aspect is reflected by a constant improvement of the spatial, spectral, radiometric and temporal resolutions. In particular, high spatial resolution imagery provides a great informative potential for a better identification, quantification and monitoring of man-made and natural resources. However, from the methodological viewpoint, the complexity of such kind of data makes them somewhat challenging with respect to the old generation data. In this context, in order to exploit fully high spatial resolution data, different methodological aspects require to be reviewed for adaptation such as the feature extraction, the classification and change detection aspects.<br/><br/>The present project represents a good opportunity to exploit the complementary competences of the University of Trento and the different Indian Institutions that work in this field, useful to face the considered complex and very important problem. Indeed, on the one hand, the University of Trento has acquired a long and valuable experience in the development of statistical and neural approaches for the processing and automatic analysis of remote sensing images. This is in part explained by the fact that the region of Trento dedicates a particular attention to the involvement of remote sensing technologies in many applications of potential interest for the local development. On the other hand, the Indian partners have demonstrated a deep know-how in the knowledge-based (soft computing) and pattern recognition fields. Furthermore, India is internationally well-known for the valuable technological achievements and growing interest it reached during the past years in the field of remote sensing. The future Indian remote sensing satellite system IRS-P5 represents a significant indicator of the interest shown by India in the applicative potential of high spatial resolution remote sensing imagery. In this project, it is intended to develop advanced software technologies that are able to deal with the complexity of the analysis and understanding of the high spatial resolution images, due to the wide range of patterns that can appear on the scene. A promising way to do that consists in combining knowledge-based approaches with well-theoretically founded mathematical tools like statistical tools. In such a way, it is expected to reach reliable, accurate and robust processing and analysis systems for such a data. From the application point of view, the idea of a web-based implementation of the developed processing and analysis tools aims at giving a particular emphasis to the support of local and national institutions in Italy and in India at low cost but also at an international level by, for instance, identifying and encouraging end-users of developing and developed countries in the use of such an advanced and useful technology.<br/>
Keywords Remote Sensing, Pattern Recognition, Image Processing
Fundings 159000 €
Partners
  • DIT - UniTN
  • IIT- Bombay
  • Jadavpur University - Kolkata
  • ISI - Kolkata

DISI Sub-project details

Project astract Areas for the University of Trento (UniTN) component of the Research<br/><br/>a) Development of multiresolution fusion techniques: Generally the panchromatic channel is characterized by a high spatial resolution with respect to the multispectral channels, while these latter have the advantage to provide a better spectral resolution. The fusion of both kinds of channel aims at providing a resulting image of higher quality integrating both desired characteristics (i.e. high spatial and spectral resolution).<br/><br/>b) Development of classification techniques: The classification of high spatial resolution imagery allows to produce maps with thematic information of high level (e.g., discrimination between houses and buildings, between streets and highways�). Obviously, the analysis and interpretation complexity increases considerably making the use of semantic networks (knowledge-based approaches) almost inevitable in such a context. However, the performances of such a heuristic approach can be considerably enhanced if combined with well-theoretically founded mathematical tools like statistical tools. The development of combined classification techniques represents a valuable way to follow in order to be able to analyze in an automatic and efficient way high spatial resolution imagery.<br/><br/>c) Development of change-detection techniques: The application fields and advantages of high resolution remote sensing data are numerous. In particular, the monitoring of urban and extra-urban areas represents an important input in decision making for local and national, public and private institutions. The supervised or unsupervised analysis of a sequence of temporal images of such typology raises many methodological issues like the data temporal models, the registration noise problem that must be faced before thinking in the adaptation of change-detection techniques developed for data of lower spatial resolution. Depending on whether only the localization or also the recognition of changed areas are desired, one may think to techniques that merge or not statistical tools with the knowledge-based approach.<br/>
Keywords Image Processing, Pattern Recognition, Remote Sensing
Fundings 159000 €
Manager Lorenzo Bruzzone