The users of RADARSAT images need ground control points (GCPs) to geometrically correct or to match the images. But the selection of GCPs has been done manually until now. Thus, the quality of the rectified images depends on the ability of the operator to select high quality GCPs. This selection is more difficult in SAR images due to their particular geometry and radiometry. We worked on the development of a method and software to automatically identify high quality GCPs in RADARSAT images using data from a topographic database. Different strategies have been elaborated to fit with different contexts. These strategies are gathered into two approaches used to match the two data sources: the vector and the raster approach. For both approach, a method has been developed to automatically identify GCPs from the category "stretches of water". A GCP software prototype has also been implemented according to these methods.

Users of Radarsat images need ground control points (GCPs) to georeference their images, to rectify them, and to fuse these images. Because the selection of GCPs is usually done manually, the efficiency of the process depends on the ability of the operator to select a GCP. The present paper introduces a new method to automatically identify GCPs in Radarsat images using a topographic database. First, different strategies are elaborated to fit different contexts (topography, cost, availability of data, etc.). These strategies are then grouped in two approaches, i.e., vector-based data matching and raster-based data matching. Both approaches rely on matching feature classes (in ISO/TC211 parlance), e.g., matching "water bodies" or matching "roads intersections" from the image with their equivalent on the topographic database. A user of this method can easily follow the various steps of the GCPs selection process with the support of software tools. This paper presents in detail the method developed to select GCPs with the feature class "water body" and the vector-based matching approach, and also presents some applications.

The users of Radarsat images need ground control points (GCP) to geometrically correct, to merge images or to facilitate the integration of extracted objects into a GIS. As the GCP selection is done manually, the quality of the rectified images depends on the ability of the operator to select good quality GCP. This selection is more difficult in SAR images due to their particular geometry and radiometry. We are working on the development of methods and software to automatically identify GCP in Radarsat images using a topographic database. Different strategies have been elaborated to deal with different contexts. The GCP identification is made according to the category to which the GCP belong. A method has been developed to automatically identify GCP belonging to the category "stretch of water" in a vector-based approach. It is based on recognition patterns which take into account the geometric deformations in SAR images.

Intelligent systems have to deal with expert's knowledge. Procedural representation is one of the most commonly used. The "if x then y" rules formalism is a part of this category. The event x is generally called observation. In a multi-sensor context, a single observation generally represents one point of view. As sensors can be concurrent or complementary, one should be able, at the end, to qualify the reliability of this observation. Certainty theory introduces the concept of confidence factors (CF) for rule's premise. CF can be used to represent the confidence in an observation. These CF have two major weaknesses. First, as the rules are generally issued from experts, uncertainty should be related to their CF attribution and to their ability to discriminate hypotheses. Second, CF are not suited for multi-sensor cases (i.e. no method is defined to combine efficiently single observation's CF taken from multiple sources).This study suggests a slight variation of the Dempster-Shafer theory using observation qualification in multi-sensor contexts. The uncertainty is placed on the rules instead of on sources. Thus, sensor's specialization is taken into account. By this approach, the masses are not directly attributed on the frame of discernment elements, but on the rules themselves that become the sources of knowledge, in the context of Dempster combining rule. It proposes then an approach for observation qualification in a multi-sensor context, as well as it suggests a new path for the delicate task of mass attribution.

This paper presents an overview of a satellite image fusion system for mapping applications. The map features to extract include roads, railroads, energy transmission lines and some types of rivers. Actually, the only data source used for their extraction is aerial black and white photographs. The objective here is to fuse multi-source and multi-type information in a semi-automatic intelligent system. The used information ranges from satellite images (visible and radar) to domain-based models and expert modeled knowledge, strategies and rules.

The availability of multi-sensed data, especially in remote sensing, leads to new possibilities in the area of target recognition. In fact, the information contained in an individual sensor represents only one facet of the reality. The use of several sensors aims at covering different facets of real world objects. In this study, the targets to recognize are the planimetric features (i.e. roads, energy transmission lines, railroads and rivers). The sensors used are visible type satellite sensors (SPOT Panchromatic and Landsat TM) as well as radar satellites (Radarsat fine mode and ERS-1). Sensor resolutions range from 8 to 30 meters/pixel. In this study, the modeling is not limited, as it is generally the case, to the problem feature's reality, but to each sensor that will be used. Moreover, the decision space (here a 3D symbolic map) has to be modeled in the same way as the reality and sensors to lead to a coherent and uniform system. Each model is developed using an object-oriented approach. Each reality-object is defined through its radiometric, geometric and topologic feature. The sensor model objects are defined in accordance to image acquisition and definition, including the stereo image cases (for SPOT and Radarsat). Finally, the decision space objects define the resulting 30 symbolic map where, for instance, a pixel attributes contain classification information as well as position, accuracy, reality object's membership values, etc.

This paper describes the use of the wavelet transform for multiscale texture analysis. One of the basic problems is that texture measures have to adapt to the peculiarity of radar images that contain multiplicative speckle noise. In this paper, the focus is on the effect of speckle on the wavelet transform. The effect is first assessed analytically. It is shown that the wavelet coefficients are modulated by the multiplicative character of the speckle in a manner that is proportional to the target mean backscattering coefficient. The effect of speckle correlation is also demonstrated. Wavelet decomposition is then applied to a simulated radar image generated by a Monte Carlo approach and based on a statistical model. Modeling shows that the correlation properties of speckle have an effect up to a scale that corresponds to its granular size. The results also show that the main contribution to the wavelet transform for an homogeneous area is the first-order statistical distribution of speckle, which remains important even at large scales. The results are then compared to a ERS-1 synthetic aperture radar (SAR) image of a primary tropical forest region.

Modelling in agriculture has been widely used for the prediction of various soil and crop growth variables. However, in order to obtain precise estimations, models often require large data sets for the initialisation and for the adjustment of the soil-crop parameters during the growing season. Remote sensing, especially radar sensors can be useful for temporal and spatial monitoring of the soil and plant parameters. A set of seven multitemporal RADARSAT images was used to determine the optimal conditions to extract parameters such as LAI, crop height, percent soil cover. In particular, these images were used to evaluate the possibility of integrating the RADARSAT soil-crop parameters in the SWAP model in order to obtain more accurate yield estimates for four crops (corn and soybean, carrots and onions).

Nonparametric alternatives based on the Wilcoxon-Mann-Whitney statistics are evaluated for edge detection in synthetic aperture radar (SAR) images. First, the power of the nonparametric test is compared to that of the ratio-of-averages detector. Then, a nonparametric change-point test is evaluated to identify the true edge position inside a fixed window.

Abstract In this letter, we establish a link that exists between several low level methods for the evaluation of segmentation results. It is demonstrated that, although they probe different characteristics of segmentation outputs, the methods examined in this study exploit the same mathematical entity: the overlapping area matrix. The impact of this observation on segmentation evaluation is discussed in the letter.

As part of our ongoing multisource data research program a groundwater vulnerability model was implemented within Arc/InfoTM, for the Lachute Area, near Montreal in Quebec, Canada. The model used was the DRASTIC model, initially developed in the United States and now recognized by Canadian and American government agencies (Aller et al., 1987). The objective of the present project was to develop an error propagation model for the evaluation of the reliability of the DRASTIC model. DRASTIC uses seven different parameters which influence groundwater vulnerability. These parameters are: depth to water, recharge, aquifer media, soil type, impact of the vadose zone and hydraulic conductivity. Each element is weighted according to its own influence on vulnerability. The DRASTIC index is obtained by the weighted summation of the rating attributed to the different parameters according to the relations incorporated in the DRASTIC model. For a homogeneous area: Index equals DwDr plus RwRr plus AwAr plus SwSr plus TwTr plus IwIr plus CwCr where Xw is the weight attributed to each parameter and Xr is the rating attributed to each parameter according to DRASTIC charts. The necessary data used for evaluating each model parameter originates from various sources including point data from drill holes, digitized maps, a digital elevation model and a Landsat TM image classification. As a final product, a vulnerability surface of the area in raster format is obtained. Two different approaches could have been considered for the characterization of error: a stochastic simulation of the model or analytical error propagation. In this paper, the second approach is presented because the DRASTIC type of model contains too many parameters for straightforward simulation.

The ratio of the arithmetic to the geometric mean is introduced as a general index to test SAR image homogeneity. Application to a multilook airborne SAR image shows that this first-order statistical measure compares advantageously and in some typical cases outperforms the coefficient of variation for the detection of random and structural heterogeneity.

An edge detector based on a second-order texture measure is presented. It is demonstrated that the operator is well adapted for SAR image analysis: it possesses good sensitivity to edges and its response does not depend on the image brightness. In the second part of the paper, an adaptive speckle filter based on the presented edge detector is introduced. The potential of the filter is illustrated using an airborne SAR image. It is shown that the filter preserves edges better than filters based on the coefficient of variation and still provides good sensitivity to textural patterns other than speckle. Considering the recognized importance of texture in the interpretation of radar imagery, these filters may be useful for pre-interpretation image enhancement.

The Gamma-Gamma Maximum a posteriori filter provides an estimate of the speckle-free radar cross-section when both the radar reflectivity and the speckle distributions are gamma-distributed. In this Letter, it is shown that a factor greater than two in precision can be typically gained in the radar cross-section estimate in the case of natural land Synthetic Aperture Radar (SAR) clutters. This is accomplished by using the normalized logarithmic estimator which is a more accurate index than the coefficient of variation to deduce the local heterogeneity parameter.

Most image analysis techniques, originally designed for optical images, rely on measures based on differences between pixel intensities. When applied to SAR images, these methods suffer from noise artifacts because the distribution of the difference between pixel intensities depends on the underlying image intensity. Some solutions are proposed to deal with the multiplicative nature of the speckle noise. The cases of edge detection, texture measurements and clustering techniques are discussed

Presents the results obtained from the MIMICS (Michigan Microwave Canopy Scattering) forest backscattering model which was modified to accommodate agricultural parameters. In the MIMICS model, the forest canopy is divided into three regions: the crown layer, the trunk region, and the underlying rough ground. The crop cover situation is simulated by a rough ground and a crown layer composed of scatterers with different forms, distributions, and dielectric constants. The authors omit trunks from the final agricultural representation because these components are of the same order as the wavelength, contrary to the model's implicit assumptions. The simulation results are compared to ground based scatterometer data of wheat and canola. The paper describes the simulation results for the two crops at L and C bands and the two like polarizations. An analysis of the different backscattering mechanisms is also given for each crop. Good simulation results were obtained at L and C bands for HH polarization for both these crops throughout the growing season. An error analysis indicates that the soil moisture can be predicted with a precision better than 0.03 g/cm/sup 3/ for both crops, if all other model parameters are known. In addition, if the moisture is known, the height of the stems and the diameter of the leaves of the canola crop can be estimated with a precision better than /spl plusmn/5 cm and /spl plusmn/0.5 cm, respectively.< >

The multi-date data set discussed in this paper was acquired over an agricultural site near Melfort, Saskatchewan, in 1983. The Synthetic Aperture Radar (SAR) data are C-band (5.26GHz) with vertical polarization and two incidence angles, 53° and 30°. A comparison between the data at the two incidence angles shows that the pixel statistics of the image data are different for the two angles. Examination of the statistical distribution shows that the 53° data are more influenced by vegetation than the 30° angle data. The classification accuracy of the 53° data was higher than that obtained with the 30° data. In addition, the classification accuracies obtained using multi-date combinations with the 53° data were superior to accuracies obtained using multi-angle combinations (53°+ 30°). An overall improvement in classification accuracy was obtained using a post segmentation field classifier.