Search

This paper is an attempt to compare two interferometric processings. The first one is applied to traditional space-borne radar (SAR) and the second on recent interferometric sonar data. Few comparisons between those tech-niques have already been made, despite the fact that they share many similar principles, only a. Thus, the key idea of this article is to present both techniques with assets, drawbacks and specific "tricks" used in data processing. The first part introduces briefly both sensors and compares signal and processing techniques used for both of them. The second part deals with interferometry, and more precisely with underwater and satellite interferometry. Then a noisepollution analysis is performed on both techniques followed by bias removal methods for getting interferometric information. The conclusion summarizes the similarities between sonar & radar processing, point-ing at the techniques that can applied to both. The abstract is to be in fully-justified italicized text, at the top of the left-hand column as it is here, below the author information. Use the word "Abstract" as the title, in 12-point Times, boldface type, centered relative to the column, initially capitalized. The abstract is to be in 10-point, single-spaced type, and may be up to 7.5 cm long. Leave two blank lines after the abstract, then begin the main text.

We previously (1995) proposed a new error function tailored to train one hidden-layer perceptron for classification problems, using a two-stage mechanism (splitting up the input space into regions which are then combined according to the most representative class of the encompassed patterns). This crisp error function relies on the number of patterns of different classes in a same region. In this paper, we enhance this function by incorporating a fuzzy component which allows all the region boundaries to tend to an optimal Bayesian separation. Genetic algorithms were previously involved in computing the weights of the network. A genetic enhancement is here introduced with a phenotypic based operator, through fresh blood strategies. This method takes advantage of coarse a priori knowledge about the problem to drive more efficiently the population evolution. Numerous tests using our new approach show a dramatic additional improvement compared to the already efficient former one. Interesting perspectives emerge from these performance results associated with the relevant knowledge offered by the region boundaries.

The radial basis function (RBF) network is an efficient function approximator. Theoretical researches focus on the capabilities of the network to reach an optimal solution. Unfortunately, few results concerning the design, and training of the network are available. When dealing with a specific application, the performances of the network dramatically depend on the number of neurons and on the distribution of the hidden neurons in the input space. Generally, the network resulting from learning applied to a predetermined architecture, is either insufficient or over-complicated. In this study, we focus on genetic learning for the RBF network applied to prediction of chaotic time series. The centers and widths of the hidden layer neurons basis function-defined as the barycenter and distance between two input patterns-are coded into a chromosome. It is shown that the basis functions which are also coded as a parameter of the neurons provide an additional degree of freedom resulting in a smaller optimal network. A direct matrix inversion provides the weights between the hidden layer and the output layer and avoids the risk of getting stuck into a local minimum. The performances of a network with Gaussian basis functions is compared with those of a network with genetic determination of the basis functions on the Mackey-Glass delay differential equation.

International audience; Accurately simulating sonar images and swath bathymetry profiles enables to render the reality of the underwater environment. This approach is strongly motivated by the operational constraints for data acquisition. Such a tool is especially attractive for tuning algorithms (i.e. detection and classification ones) according to sensors settings and sea-bottom nature and relief. Therefore, we developed such a simulator relying on acoustical ray tracing techniques tailored for considering specific underwater propagation properties (multi-path sound propagation, sound velocity profiles, adaptive ray oversampling to modelize acoustic lobe configuration). It aims at rendering data similar to what real sensors provide in a natural contest. In the proposed approach, the virtual environment gathers both static and dynamic behaviours, namely simulator inputs may be a priori known or updated all along time. Sea-ground truth is provided by either a real elevation map or fractal simulation allowing a fine macro-relief description (flat, rocky patch, rippled sea-bottom). In addition, micro-relief is modelled through sea-bed textures resulting from adequate spatial layouts of ground resolution cells. Sets of natural or manufactured bottom-laying items may be introduced using mesh descriptions (ship wreck, ...) or structured geometric primitives (anchors, pipelines...). Furthermore, the material of the previous inputs tunes their acoustic response. Taking into account carrier trajectory and attitude along time warrants multi-sensor realistic outputs. A great interest of our parallel and modular implementation is an easier incorporation in a more global rendering system. However, its interface proposes real time visualisation of the acquired environment through windows showing current sonar images, swath bathymetry, vehicle attitude and trajectory. Given this flexibility, a promising 3D sea-bottom and objects reconstruction method already benefit from this simulator virtual reality.

Depending on sonar characteristics (resolution, ...) and tow fish configuration (location, heading, altitude, ...), sonar images present different properties. An algorithm which registers directly all these images would be very complex and hardly evolutive. In order to overcome these problems, the authors propose a two-step registrating system: its first stage is acquisition-dependent and extracts information through processed tailored for each sonar type. Then, this symbolic knowledge feeds a generic system whose goal is to perform the matching task between current information and previously recorded data. Examples of such data range from coarse detected objects (low resolution sensor) to finely delimited homogeneous sea-bed regions (texture analysis relying on higher resolution sonar outputs). Thus, multiple symbolic information can be extracted from the same sea-bed region when various tracks overlap. Nevertheless, precision on tow-fish navigational data recorded with sonar images, do not allow one to directly match common information. Thus, the authors implement such a matching mechanism using genetic algorithms with variable-length chromosomes coding a field of transformations the size of which is not a priori known. This new flexibility allows the registration of different tracks covering a same sea-bed region with multiple distortions.