Manuel Rosa-Zurera

Transient modeling by matching pursuits with a wavelet dictionary for parametric audio coding

In this letter, we propose a novel matching pursuit-based method for transient modeling with application to parametric audio coding. The overcomplete dictionary for the matching pursuit is composed of wavelet functions that implement a wavelet-packet filter bank. The proposed transient modeling method is suitable to be integrated into a parametric audio coder based on the three-part model of sines, transients, and noise (STN model). Comparative analysis between wavelet and exponentially damped sinusoidal functions are shown in experimental results. The mean-squared-error performance of the proposed approach is better than that obtained with damped sinusoids.

Automatic classification of background EEG activity in healthy and sick neonates.

The overall aim of our research is to develop methods for a monitoring system to be used at neonatal intensive care units. When monitoring a baby, a range of different types of background activity needs to be considered. In this work, we have developed a scheme for automatic classification of background EEG activity in newborn babies. EEG from six full-term babies who were displaying a burst suppression pattern while suffering from the after-effects of asphyxia during birth was included along with EEG from 20 full-term healthy newborn babies. The signals from the healthy babies were divided into four behavioural states: active awake, quiet awake, active sleep and quiet sleep. By using a number of features extracted from the EEG together with Fishers linear discriminant classifier we have managed to achieve 100% correct classification when separating burst suppression EEG from all four healthy EEG types and 93% true positive classification when separating quiet sleep from the other types. The other three sleep stages could not be classified. When the pathological burst suppression pattern was detected, the analysis was taken one step further and the signal was segmented into burst and suppression, allowing clinically relevant parameters such as suppression length and burst suppression ratio to be calculated. The segmentation of the burst suppression EEG works well, with a probability of error around 4%.

Application of Fisher Linear Discriminant Analysis to Speech/Music Classification

This paper proposes the application of Fisher linear discriminants to the problem of speech/music classification. Fisher linear discriminants can classify between two different classes, and are based on the calculation of some kind of centroid for the training data corresponding with each one of these classes. Based on that information a linear boundary is established, which will be used for the classification process. Some results will be given demonstrating the superior behavior of this classification algorithm compared with the well-known K-nearest neighbor algorithm. It will also be demonstrated that it is possible to obtain very good results in terms of probability of error using only one feature extracted from the audio signal, being thus possible to reduce the complexity of this kind of systems in order to implement them in real-time

Sea clutter reduction and target enhancement by neural networks in a marine radar system.

The presence of sea clutter in marine radar signals is sometimes not desired. So, efficient radar signal processing techniques are needed to reduce it. In this way, nonlinear signal processing techniques based on neural networks (NNs) are used in the proposed clutter reduction system. The developed experiments show promising results characterized by different subjective (visual analysis of the processed radar images) and objective (clutter reduction, target enhancement and signal-to-clutter ratio improvement) criteria. Moreover, a deep study of the NN structure is done, where the low computational cost and the high processing speed of the proposed NN structure are emphasized.

Spatial-Range Mean-Shift Filtering and Segmentation Applied to SAR Images

The mean-shift (MS) algorithm is applied for reducing speckle noise and segmenting synthetic aperture radar (SAR) images. Two coastal images acquired by Envisats advanced SAR (ASAR) [European Space Agency (ESA)] are used. Studies of the MS parameters are carried out according to the desired product: a speckle filtered image where textures and edges are preserved, or a segmented image, where land and sea are distinguished, as a previous stage for obtaining a land mask and detecting the coastal line. In all cases, Gaussian kernels are used. Speckle filtering results are compared with those obtained using uniform kernels, proving that the former provides better results than the latter. A segmentation approach based on the positions and frequencies at which the MS converges is applied. The use of a combined spatial-range processing and the corresponding bandwidths makes the MS suitable for the two proposed problems. The solid theoretical basis of this procedure allows designing a guided search of the best parameters according to the desired solution, avoiding a tedious trial-and-error process. Although the used images have different characteristics, results prove that similar sets of parameters can be used, showing some degree of robustness with respect to the image, for a given sensor and image acquisition mode.

Multilayer perceptrons applied to traffic sign recognition tasks

The work presented in this paper suggests a Traffic Sign Recognition (TSR) system whose core is based on a Multilayer Perceptron (MLP). A pre-processing of the traffic sign image (blob) is applied before the core. This operation is made to reduce the redundancy contained in the blob, to reduce the computational cost of the core and to improve its performance. For comparison purposes, the performance of the a statistical method like the k-Nearest Neighbour (k-NN) is included. The number of hidden neurons of the MLP is studied to obtain the value that minimizes the total classification error rate. Once obtained the best network size, the results of the experiments with this parameter show that the MLP achieves a total error probability of 3.85%, which is almost the half of the best obtained with the k-NN.

Study of Two Error Functions to Approximate the Neyman–Pearson Detector Using Supervised Learning Machines

A study of the possibility of approximating the Neyman-Pearson detector using supervised learning machines is presented. Two error functions are considered for training: the sum-of-squares error and the Minkowski error with R = 1. The study is based on the calculation of the function the learning machine approximates to during training, and the application of a sufficient condition previously formulated. Some experiments about signal detection using neural networks are also presented to test the validity of the study. Theoretical and experimental results demonstrate, on one hand, that only the sum-of-squares error is suitable to approximate the Neyman-Pearson detector and, on the other hand, that the Minkowski error with R = 1 is suitable to approximate the minimum probability of error classifier.

Combining MLPs and RBFNNs to Detect Signals With Unknown Parameters

The detection of Gaussian signals with an unknown correlation coefficient rhos is considered. Solutions based on neural networks (NNs) are studied, and a strategy for designing committee machines in a composite hypothesis test is proposed. A single multilayer perceptron (MLP) has been trained with rhos uniformly varying in [0, 1]. Considering the decision boundaries for rhos = 0 and rhos = 1 and how an MLP approximates them, a detection scheme composed of two MLPs has been proposed. One of them MLP1 has been trained with rhos uniformly varying in [0, 0.5], and the other one MLP2 has been trained with rhos uniformly varying in [0.5, 1]. For making a decision, the higher output is compared to a threshold for each false-alarm probability (P FA). This strategy simplifies the task of finding a compromise solution between the computational cost and the approximation error and outperforms the single-MLP detector. When MLP1 is substituted with a radial basis function NN (RBFNN), a new combination strategy of the outputs is required. We propose separately thresholding the outputs and applying them to an or logic function. The performance of this detector is slightly better than the two-MLP one, and the computational cost is significantly reduced.

Artificial neural network-based clutter reduction systems for ship size estimation in maritime radars

The existence of clutter in maritime radars deteriorates the estimation of some physical parameters of the objects detected over the sea surface. For that reason, maritime radars should incorporate efficient clutter reduction techniques. Due to the intrinsic nonlinear dynamic of sea clutter, nonlinear signal processing is needed, what can be achieved by artificial neural networks (ANNs). In this paper, an estimation of the ship size using an ANN-based clutter reduction system followed by a fixed threshold is proposed. High clutter reduction rates are achieved using 1-dimensional (horizontal or vertical) integration modes, although inaccurate ship width estimations are achieved. These estimations are improved using a 2-dimensional (rhombus) integration mode. The proposed system is compared with a CA-CFAR system, denoting a great performance improvement and a great robustness against changes in sea clutter conditions and ship parameters, independently of the direction of movement of the ocean waves and ships.

Modified LMS-Based Feedback-Reduction Subsystems in Digital Hearing Aids Based on WOLA Filter Bank

Digital hearing aids usually suffer from acoustic feedback. This feedback corrupts the speech signal, causes instability, and damages the speech intelligibility. To solve these problems, an acoustic feedback reduction (AFR) subsystem using adaptive algorithms such as the least mean square (LMS) algorithm is needed. Although this algorithm has a reduced computational cost, it is very unstable. To avoid this situation, other AFR subsystems based on modifications of the LMS algorithm are used. Such algorithms are given as follows: 1) normalized LMS (NLMS); 2) filtered-X LMS (FXLMS); and 3) normalized FXLMS (NFXLMS). These algorithms are tested in three digital hearing aid categories: 1) in the ear (ITE); 2) in the canal (ITC); and behind the ear (BTE). The first and second categories under study suffer from great feedback effects due to the short distance between the loudspeaker and the microphone, whereas the third category suffers from these effects due to the high signal level at the hearing aid output; thus, robust AFR subsystems are needed. The added stable gains (ASGs) over the limit gain when AFR subsystems are working in the digital hearing aids are studied for all the categories. The ASG is determined as a tradeoff between two measurements: 1) segmented signal-to-noise ratio (objective measurement) and 2) speech quality (subjective measurement). The results show how the digital hearing aids that work with AFR subsystems adapted with the NLMS or the NFXLMS algorithms can achieve up to 18 dB of increase over the limit gain. After analyzing the results, it is observed that the subjective measurement always limits the achieved ASG, but when the NLMS algorithm is used, it is appreciated that the objective measurement is a good approximation for estimating the maximum achieved ASG. Finally, taking into consideration the hearing aid performances and the computational cost of each AFR subsystem implementation, an AFR subsystem based on the NLMS algorithm to adapt feedback-reduction filters that are 128 coefficients long is proposed.