Adaptive image segmentation using genetic and hybrid search methods

This paper describes an adaptive approach for the important image processing problem of image segmentation that relies on learning from experience to adapt and improve the segmentation performance. The adaptive image segmentation system incorporates a feedback loop consisting of a machine learning subsystem, an image segmentation algorithm, and an evaluation component which determines segmentation quality. The machine learning component is based on genetic adaptation and uses (separately) a pure genetic algorithm (GA) and a hybrid of GA and hill climbing (HC). When the learning subsystem is based on pure genetics, the corresponding evaluation component is based on a vector of evaluation criteria. For the hybrid case, the system employs a scalar evaluation measure which is a weighted combination of the different criteria. Experimental results for pure genetic and hybrid search methods are presented using a representative database of outdoor TV imagery. The multiobjective optimization demonstrates the ability of the adaptive image segmentation system to provide high quality segmentation results in a minimal number of generations     

Neural controllers for systems with unknown dynamics

This study presents a methodology for specifying a neural controller for a system about which no a priori model information is available. The neural design presumes that a finite duration input/output (I/O) histogram on the system is available. The design procedure extracts from the histogram sufficient information to specify the neural feedback controller. The resultant controller will drive the system along a general output reference profile (unknown during the design). The resultant controller also exhibits the capability of disturbance rejection and the capacity to stabilize unstable plants     

Presupernova evolution of the most massive stars

Taking as example a 60M _⊙ star of solar metallicity, the state of the art of model calculations for very massive, from the main sequence to the supernova stage, is reviewed. It is argued that — due to the simple internal structure of Wolf-Rayet stars — the post main sequence evolutionary phases are currently those which are better understood. A brief discussion of the supernova outcome from very massive stars is given. Then, the more uncertain main sequence evolution is discussed. A first attempt to incorporate results about pulsational instabilities of very massive stars in stellar evolutionary calculations is performed. On its basis, a new type of evolutionary sequence for very massive stars is obtained, namely O-star → Of-star → H-rich WNL → LBV → H-poor WNL → WNE → WC → SN. This scenario is shown to correspond better to many observed properties of very massive stars than the standard one. It includes a model for the prototype LBV P Cygni.     

Wind asymmetries in massive stars

We are in the process of surveying the linear polarization in luminous, early-type stars. We here report on new observations of the B [e] stars S 18 and R 50, and of the Luminous Blue Variables HR Car, R 143, and HD 160529. Together with previously published data, these observations provide clear evidence for the presence of intrinsic polarization in 1 B[e] star (HD 34664) and in 5 LBVs ( Car, P Cyg, R 127, AG Car, and HR Car). The data indicate that anisotropic stellar winds are a common occurrence among massive stars in these particular evolutionary stages. For such stars, mass-loss rates estimated using the assumption of a spherical, homogeneous and stationary outflow may be in error.     

The impact of semi-convection and overshooting on the surface abundances in massive stars

We analyse the effects of semi-convection and overshooting on the predicted surface abundances after the first and second dredge-ups in 15 and 20 M Pop. I stars. Overshooting is applied either to the core boundary or to the boundaries of all convective zones. It is shown that the surface abundances are sensitive to the mixing scheme adopted in the interior. The models including semi-convection lead to lower¹²C/¹³C ratios than the other mixing schemes, while models with overshooting predict higher enhancements of sodium at the surface.     

A modified Gaussian sum approach to estimation of non-Gaussiansignals

A Gaussian sum estimation algorithm has previously been developed to deal with noise processes that are non-Gaussian. Inherent in this algorithm is a serious growing memory problem that causes the number of terms in the Gaussian sum to increase exponentially at each iteration. A modified Gaussian sum estimation algorithm using an adaptive filter is developed that avoids the growing memory problem of the previous algorithm while providing effective state estimation. The adaptive filter is comprised of a fixed set of estimators operating in parallel with each individual estimate possessing its own corresponding weighting term. A simulation example illustrates the new non-Gaussian estimation technique     

Interacting multiple model tracking with target amplitude feature

A recursive tracking algorithm is presented which uses the strength of target returns to improve track formation performance and track maintenance through target maneuvers in a cluttered environment. This technique combines the interacting multiple model (IMM) approach with a generalized probabilistic data association (PDA), which uses the measured return amplitude in conjunction with probabilistic models for the target and clutter returns. Key tracking decisions can be made automatically by assessing the probabilities of target models to provide rapid and accurate decisions for both true track acceptance and false track dismissal in track formation. It also provides the ability to accurately continue tracking through coordinated turn target maneuvers     

On phased-array radar tracking and parameter control

Based on a simple model of a ground-based phased-array radar used for a multiple-target surveillance task, beam scheduling, positioning, and radar parameters like signal-to-noise ratio, track sharpness, and detection threshold have been optimized with respect to the radar/computer load induced by tracking. From this the minimum energy necessary for track maintenance during surveillance can be derived     

Forward error correction for radiobeacon broadcast of differentialGPS data

Marine radiobeacon networks are being used to broadcast differential Global Positioning System (DGPS) corrections to marine users. The correction data digitally modulate signals from some of the existing marine radiobeacons, which operate in the 285 to 325 kHz band, creating DGPS/radiobeacons. The corrections improve the accuracy of the GPS fix from 100 m to 5-10 m, and provide position fixing service for many marine applications which are too demanding for the normal GPS service. Forward error correction can be used to improve the reliability or range of the DGPS/radiobeacon signal. The improvements made possible by channel coding are analyzed, and a code for DGPS/radiobeacons is recommended     

Optimal multiple level decision fusion with distributed sensors

A data fusion model consisting of several levels of parallel decision fusions is considered. Global optimization of such a model is discussed to obtain the fusion rules for overall optimal performance. The reliability analysis of the proposed model is carried out to establish its superiority over the existing parallel and serial fusion models