Abstract
A general framework for testing the quality of the segmentation of a multi-spectral satellite image is proposed. The method is based on the production of synthetic images with the spectral characteristics of the image pixels extracted from a signature multi-spectral image. The knowledge of the location of objects in the synthetic image provides a reference segmentation, which allows for a quantitative evaluation of the quality provided by a segmentation algorithm. The Hammoude metric and three external similarity indices (Rand, Corrected Rand, and Jaccard) were chosen to perform this evaluation, but other metrics can also be used. The proposed methodology can be used for any type of satellite image (or multi-spectral image), set of land cover types, and segmentation algorithms. A practical application was carried out to illustrate the value of the proposed method. A SPOT satellite image was used to extract the spectral signature of 8 land cover types. Three test images were produced using the 8 land cover classes and two different 5 class sub-sets. The segmentation results provided by a standard algorithm were compared with the reference or expected segmentation. The results clearly indicate that the quality of a segmentation obtained from a multi-spectral image not only depends on the geometric properties of the objects present in the image, but also on their spectral characteristics. The results suggest that a specific evaluation should be carried out for each particular experiment, as the segmentation results are very dependent on the choice of land cover types.

Abstract
Image data from Earth Observation Satellites (EDS) were used to analyse mountain landscape changes in Northeast Portugal. Three Landsat images, from April 30th 1979, March 14th 1989 and May 29th 2002 were used. A supervised classification was performed for each image based on the radiometric information and the Normalised Difference Vegetation Index (NDVI). Eleven classes were selected considering the main land cover types in the region. The classification results showed high overall accuracy (above 92.5%) and kappa coefficient (above 0.91). Broadly, the range of dates of the Landsat images used allowed for the differentiation between classes. Nevertheless, some problems occurred in differentiating between classes of forest and shrub vegetation due to similar characteristics and vegetation conditions in some periods of the year, and also due to the effects of topographic shadows associated to mountain areas. Meadows and annual crops were the classes having greater changes from 1979 to 2002: meadows area increased 60% while annual crops decreased 43.5%. The increase in meadows area was likely due to policies supporting agroenvironmental conservation and autochthon bovine livestock production. Differently, the decrease in annual crops was likely due to the loss of economical competitiveness of main annual crops and to the rural population decrease and ageing, which favoured the replacement of arable lands by permanent meadows. These results may help developing policies and measures for sustainable management of traditional mountain rural landscapes.

Abstract
The revision of the 1995 land cover dataset for the Vale do Sousa region, in the northwest of Portugal, was carried out by supervised classification of a multi-spectral image from the Advanced Spaceborne Thermal Emission and Reflectance Radiometer ( ASTER) sensor. The nine reflective bands of ASTER were used, covering the spectral range from 0.52-2.43 μ m. The image was initially ortho-rectified and segmented into 51 186 objects, with an average object size of 135 pixels ( about 3 ha). A total of 582 of these objects were identified for training nine land cover classes. The image was classified using an algorithm based on a fuzzy classifier, Support Vector Machines (SVM), K Nearest Neighbours (K-NN) and a Logistic Discrimination (LD) classifier. The results from the classification were evaluated using a set of 277 validation sites, independently gathered. The overall accuracy was 44.6% for the fuzzy classifier, 70.5% for the SVM, 60.9% for the K-NN and 72.2% for the LD classifier. The difficulty in discriminating between some of the forest land cover classes was examined by separability analysis and unsupervised classification with hierarchical clustering. The forest classes were found to overlap in the multi-spectral space defined by the nine ASTER bands used.

Abstract
Polar orbiting satellites with low spatial resolution sensors, such as the AVHRR, provide repeated global coverage of the Earth. The data is directly transmitted to ground stations, and in some cases distributed immediately after the data acquisition. Near real time applications can be implemented if the adequate processing tools are available. This paper presents a near real time processing system, developed for NOAA/AVHRR data acquired from the Dundee satellite station. The system performs image calibration, geometric corrections and atmospheric corrections with minimum operator intervention. The geometric corrections consist of an orbital-based correction refined by the automatic identification of Ground Control Points (GCPs) by image matching. The atmospheric correction is based on simulations performed on the 6S radiative transfer code using a set of typical and expected values for the most significant parameters. An attempt to evaluate the error associated with the simplified atmospheric correction method was carried out. As an illustration, 3 AVHRR images from NOAA 16 were processed. The ranges of values encountered for the most relevant parameters were analysed. The range and average values for the reflectance channels 1 and 2 with and without the atmospheric correction are compared. These were used to produce standard Normalized Difference Vegetation Index (NDVI) images and atmospheric corrected NDVI images.

Abstract
A Bayesian segmentation approach for hyperspectral images is introduced in this paper. The method improves the classification performance of discriminative classifiers by adding contextual information in the form of spatial dependencies. The technique herein presented builds the class densities based on Fast Sparse Multinomial Logistic Regression and enforces spacial continuity by adopting a Multi-Level Logistic Markov-Gibs prior. State-of-art performance of the proposed approach is illustrated in a set of experimental comparisons with recently introduced hyperspectral classification/segmentation methods.

Abstract
The availability of in effective internal similarity index to determine the "natural" number of classes in a multi spectral satellite image would benefit the process of unsupervised image classification. Two similarity indices (DB and Xu) were tested in sections of multi-spectral satellite images from Landsat TM, SPOT HRVIR, ASTER and IKONOS. The images were initially clustered into a manageable number of classes using, an unsupervised classification algorithm. These results were then structured hierarchically, and the internal similarity indices computed for each level. The inspection of the DB and Xu index plots were used to select the "natural" number of classes for each test image.