Development and Validation of Novel data Fusion Techniques for Environmental Remote Sensing

COFIN-2005-Bruzzone DIT-PRJ-06-002

Status NOT active project
DISI role Partner
Project type Research Project
Dimension National
Acquisition date 2005-12-23
Start date 2006-01-30
End date 2007-12-31

Project details

Project astract The main objective of this research program is the development of a set of advanced techniques for multisensor and multiresolution fusion of remote-sensing optical and radar data. Such a topic is of great interest for the international scientific community. In fact data-fusion is becoming more and more important in the context of scientific and applicative fields, thanks to the development of sensors and of methodologies for data analysis and processing, and to the availability of computers with better and better cost-performance ratios. In addition, remote-sensing data can have an increased importance for environmental applications (e.g., monitoring), if endowed with effective data-fusion methodologies.<br/>The proposed research program aims at overcoming the problems related to the heterogeneous characteristics of remote sensing data acquired either by different sensors or by the same sensor in different bands and/or different operational modalities. Such heterogeneity can consist in differences in the spatial resolution, the soil characteristics highlighted by the images, or the signal properties. These differences do not allow an easy definition of a correspondence between the available images and a joint model for analysis and processing.<br/>In order to state clearly the requirements and the operational modalities of the methodologies to be developed, data acquired by some of the most interesting sensors of recent and future missions will be considered, including optical (panchromatic with high spatial resolution, multispectral, and hyperspectral) and Synthetic Aperture Radar (SAR). In particular, the application context is focused on environmental monitoring. The theoretical basis of the proposed data-fusion methodologies take origin from random process theory, pattern recognition, and remote sensing systems (all relevant scientific areas within the Telecommunication field).<br/>Specifically, the following aspects will be considered:<br/>(a) multisensor fusion, with a particular focus on the fusion of optical and SAR data and of optical data acquired in spectral regions characterized by different phenomena (i.e., emission and reflection). Innovative techniques will be developed for: image registration of optical and SAR data; SAR despeckling; feature extraction; multispectral-SAR fusion; optical-SAR joint modelling and classification; pixel-level and decision-level fusion for detection purposes.<br/>(b) multiresolution fusion: the following techniques will be considered and will be applied to optical data of different spatial resolutions: "pan-sharpening"; hierarchical image segmentation; multiscale/multilevel feature extraction; high spatial resolution data classification.
Keywords Remote Sensing, Data Fusion, Environmental Monitoring, Multiresolution Data Fusion, Multisensor Data Fusion, Image Segmentation, Image Modelling and Classification, Decision Fusion, Multiscale Analysis
Fundings 160000 €
Partners
  • DIT - UniTN
  • University of Genoa
  • University of Siena
  • University of Pisa
  • University of Naples "Federico II"
  • University of Naples Federico II

DISI Sub-project details

Project astract The activity proposed by DIT-UniTN is mainly focused on the development of classification techniques based on multiscale/multilevel fusion analysis of high-spatial resolution remote-sensing images (from 4 meters to less than 1 meter). The choice of addressing this research activity depends on the key role played by high-spatial resolution sensors in the last generation of Earth Observation satellite missions (e.g., Ikonos and Quickbird). The importance of this research topic is confirmed by the great attention devoted from the remote-sensing community to the development of advanced methodologies for the analysis of high-spatial resolution images. This is explained by two main reasons: 1) methodologies developed for the classification of traditional low-spatial resolution images are not suitable to be used with high-spatial resolution images (high resolution images have a heterogeneous information content and intrinsic multiscale nature); 2) high-spatial resolution images allow to significantly increase the range of applications of remote sensing (e.g., applications related to a detailed analysis of urban scenes). In this context, it becomes particularly important to study and integrate in a proper way the information content conveyed by the geometric component that characterizes such images in the automatic classification process.<br/>Accordingly, DIT-UniTN intends to develop advanced analysis and classification techniques for high-spatial resolution images, which aim at improving the performances (in terms of classification accuracy and robustness) of the techniques proposed in the literature. A particular attention will be devoted to the design of effective feature-extraction methods capable to characterize the information available at different levels of details in the considered scene. In this context, it is planned to develop models that permit to provide a simple, general and robust representation of the spatial information that characterize each object in the scene. This will be done by defining novel multiscale and hierarchical techniques that exploit spectral, textural and geometrical features extracted from the considered images. The project activity will be organized according to two main objectives. The first one consists in the development of feature-extraction methodologies based on hierarchical, adaptive and multiscale representations of the images. The second objective addresses the problem of the fusion of the feature extracted at various resolution/scale levels by using classification strategies capable to overcome the problems of the statistical heterogeneity and the intrinsic high dimensionality of the obtained parameters.
Keywords Remote Sensing, Data Fusion, Automatic Classification, Pattern Recognition, Image Processing, Very High Gemetrical Resolution Images
Fundings 40000 €
Manager Lorenzo Bruzzone