Catálogo de publicaciones - libros

ANN-based supervised classification systems are very popular when dealing with high dimensional datasets, like multi or hyperspectral images. Typical approaches require a highly time-consuming preprocessing stage where the dimensionality is reduced through the deletion or averaging of redundant information and the establishment of a processing “window” that is displaced over the dataset. Only after this stage, the ANN-based system can perform the classification process the success of which, as a consequence, depends on the quality of the preprocessed data. In this paper, we propose a classification system that automatically obtains the optimal window size and dimensional transformation parameters for a given set of categorization requirements while it is performing the training of the ANN. In addition, the parameters of the ANN in terms of number of inputs are also adapted on line. To test the system, it was applied to a hyperspectral image classification process of real materials where the pixel resolution implies that a material is characterized by spectral patterns of combinations of pixels.

Soaring is a recreational activity and competitive sport where individuals fly un-powered aircrafts known as gliders. Soaring place selection process depends on a number of factors, resulting in a complex decision-making task. In this paper, we propose the use of the dmFSQL language for fuzzy queries as one of the techniques of Data Mining, which can be used to solve the problem of offering the better place for soaring given the environment conditions and customer characteristics. After doing a process of clustering and characterization of a Customers Database in a Data Warehouse we are able of classify next customer in a cluster and offer an answer according it.

Steganalysis consists in classifying documents as steganographied or genuine. This paper presents a methodology for steganalysis based on a set of 193 features with two main goals: determine a sufficient number of images for effective training of a classifier in the obtained high-dimensional space, and use feature selection to select most relevant features for the desired classification. Dimensionality reduction is performed using a forward selection and reduces the original 193 features set by a factor of 13, with overall same performance.

It is well known that Genetic Algorithms (GA) is an optimization method which can be used in problems where the traditional optimization techniques are difficult to be applied. Sonic Crystals (SC) are periodic structures that present ranges of sound frequencies related with the periodicity of the structure, where the sound propagation is forbidden. This means that in the acoustic spectrum there are ranges of frequencies with high acoustic attenuation. This attenuation can be improved producing vacancies in the structure. In this paper we use a parallel implementation of a GA to optimize those structures, by means of the creation of vacancies in a starting SC, in order to obtain the best acoustic attenuation in a predetermined range of frequencies. The cost function used in GA is based on the Multiple Scattering Theory (MST), which is a self consistent method for calculating acoustic pressure in SCs. As a final result we achieve a quasi ordered structures that presents a high acoustic attenuation in a predetermined range of frequencies, independent of the periodicity of the SC.

A new method for spectral analysis, based on ADALINE artificial neural networks (ANNs), is proposed. The network is able to calculate accurately the fundamental frequency and the harmonic content of the input signal. This method is especially useful in high precision digital measurement systems in which periodical signals are involved, i.e. digital watt meters. Most of these system use spectrum analysis algorithms for the computation of the magnitudes of interest. The traditional spectrum analysis methods require synchronous sampling, which introduce limitations to the sampling circuitry. Sine-fitting multiharmonics algorithms resolve the hardware limitations concerning the synchronous sampling but have some limitations with regard to the phase of the array of samples. The new implementation of sine-fitting multiharmonics algorithms based in ANN, eliminates these limitations.

MultiLayer Perceptrons (MLPs) trained in a supervised way to minimize the Mean Square Error are able to approximate the Neyman-Pearson detector. The known target detection in a Weibull-distributed clutter and white Gaussian noise is considered. Because the difficulty to obtain analytical expressions for the optimum detector under this environment, a suboptimum detector like the Target Sequence Known A Priori (TSKAP) detector is taken as reference. The results show a MLP-based detector dependency with the training algorithm for low MLP sizes, being the Levenberg-Marquardt algorithm better than the Back-Propagation one. On the other hand, this dependency does not exist for high MLP sizes. Also, this detector is sensitive to the MLP size, but for sizes greater than 20 hidden neurons, very low improvement is achieved. So, the MLP-based detector is better than the TSKAP one, even for very low complexity (6 inputs, 5 hidden neurons and 1 output) MLPs.

This paper presents an application of neural networks in pattern recognition of defects in sonic signals from non-destructive evaluation by multichannel impact-echo. The problem approached consists in allocating parallelepiped-shape materials in four levels of classifications defining material condition (homogeneous or defective), kind of defects (holes and cracks), defect orientation, and defect dimension. Various signal features as centroid frequency, attenuation and amplitude of the principal frequency are estimated per channel and processed by PCA and feature selection methods to reduce dimensionality. Results for simulations and experiments applying Radial Basis Function, Multilayer Perceptron and Linear Vector Quantization neural networks are presented. Neural networks obtain good performance in classifying several 3D finite element models and specimens of aluminum alloy.

This paper presents the validation of the expressive content of an acted corpus produced to be used in speech synthesis. The use of acted speech can be rather lacking in authenticity and therefore its expressiveness validation is required. The goal is to obtain an automatic classifier able to prune the bad utterances –with wrong expressiveness–. Firstly, a subjective test has been conducted with almost ten percent of the corpus utterances. Secondly, objective techniques have been carried out by means of automatic identification of emotions using different algorithms applied to statistical features computed over the speech prosody. The relationship between both evaluations is achieved by an attribute selection process guided by a metric that measures the matching between the misclassified utterances by the users and the automatic process. The experiments show that this approach can be useful to provide a subset of utterances with poor or wrong expressive content.

Unit-selection based Text-to-Speech synthesis systems aim to obtain high quality synthetic speech by optimally selecting previously recorded units. To that effect these units are selected by a dynamic programming algorithm guided through a weighted cost function. Thus, in this context, weights should be tuned perceptually so as to be in agreement with perception from listening users. In previous works we have proposed to subjectively tune these weights through an interactive evolutionary process, also known as Active Interactive Genetic Algorithm (aiGA). The problem comes out when different users, although being consistent, evolve to different weight configurations. In this proof-of-principle work, Generative Topographic Mapping (GTM) is introduced as a method to extract knowledge from user specific preferences. The experiments show that GTM is able to capture user preferences, thus, avoiding selecting the best evolved weight configuration by means of a second preference test.

This paper presents a model of the retina with its properties with respect to sampling, spatiotemporal filtering, color-coding and non-linearity, and their consequences on the processing of visual information. It’s formalism points out the architectural and algorithmic principles of neuromorphic circuits which are known to improve compactness, consumption, robustness and efficiency, leading to direct applications in engineering science. It’s biological aspect, strongly based neural and cellular descriptions makes it suitable as an investigation tool for neurobiologists, allowing the simulation of experiences difficult to set up and answering fundamental theoretical questions.