Catálogo de publicaciones - libros

Vehicle classification is an important issue in the domain of ITS (Intelligent Transportation Systems). In this paper we presents an improved one based on Gabor features, which contains three consecutive stages: vehicle segmentation, Gabor features extraction and template matching. A novel non-even sampling of Gabor features is proposed. The experimental data show that this method can heavily reduce the computation and memory requirements, and illustrate good performance both in discrimination ability and robustness.

One of the central problems in machine learning is how to effectively combine unlabelled and labelled data to infer the labels of unlabelled ones. In recent years, there has a growing interest on the transduction method. In this article, the transductive learning machines are described based on a so-called affinity rule which comes from the intuitive fact that if two objects are close in input space then their outputs should also be close, to obtain the solution of semi-supervised learning problem. By using the analytic solution for this problem, an incremental learning algorithm adapting to on-line data processing is derived.

Dynamic Programming (DP) has been widely used as an approach solving the Markov Decision Process problem. This paper takes a well-known gambler’s problem as an example to compare different DP solutions to the problem, and uses a variety of parameters to explain the results in detail. Ten C++ programs were written to implement the algorithms. The numerical results from gamble’s problem and graphical output from the tracking car problem support the conceptual definitions of RL methods.

In this paper, the design methodology of neural network hardware has been discussed, and two weighted neural network implemented by this method been applied for object recognition. It was pointed out that the main problem of the two weighted neural network hardware implementation lies in three aspects. At final, two weighted neural network implemented by this method is applied for object recognition, and the algorithm were presented. We did experiments on recognition of omnidirectionally oriented rigid objects on the same level, using the two weighted neural networks. Many animal and vehicle models (even with rather similar shapes) were recognized omnidirectionally thousands of times. For total 8800 tests, the correct recognition rate is 98.75%, the error rate and the rejection rate are 0.5 and 1.25% respectively.

This paper presents an approach that discovers clusters of Web pages based on Web log data and Web page contents as well. Most existing Web log mining techniques are access-based approaches that statistically analyze the log data without paying much attention on the contents of the pages. The log data contains various kinds of noise which can significantly influence the performance of pure access-based web log mining. The method proposed in this paper not only considers the frequence of page co-occurrence in user access logs, but also takes into account the web page contents to cluster Web pages. We also present a method of using information entropy to prune away irrelevant papges which improves the performance of the web page clustering.

How to provide a means or organize the information used in making exploration decisions in petroleum exploration is an important task. In this paper, a machine learning method is put forward to collect experiences and estimate or prediction the absent data. The well logging experiments show that the method is efficiently and accurately.