Maria do Carmo Nicoletti

A genetic programming based system for the automatic construction of image filters

The manual selection of linear and nonlinear operators for producing image filters is not a trivial task in practice, so new proposals that can automatically improve and speed up the process can be of great help. This paper presents a new proposal for constructing image filters using an evolutionary programming approach, which has been implemented as the IFbyGP software. IFbyGP employs a variation of the Genetic Programming algorithm GP and can be applied to binary and gray level image processing. A solution to an image processing problem is represented by IFbyGP as a set of morphological, convolution and logical operators. The method has a wide range of applications, encompassing pattern recognition, emulation filters, edge detection, and image segmentation. The algorithm works with a training set consisting of input images, goal images, and a basic set of instructions supplied by the user, which would be suitable for a given application. By making the choice of operators and operands involved in the process more flexible, IFbyGP searches for the most efficient operator sequence for a given image processing application. Results obtained so far are encouraging and they stress the feasibility of the proposal implemented by IFbyGP. Also, the basic language used by IFbyGP makes its solutions suitable to be directly used for hardware control, in a context of evolutionary hardware. Although the proposal implemented by IFbyGP is general enough for dealing with binary, gray level and color images, only applications using the first two are considered in this paper; as it will become clear in the text, IFbyGP aims at the direct use of induced sequences of operations by hardware devices. Several application examples discussing and comparing IFbyGP results with those obtained by other methods available in the literature are presented and discussed.

Using Bayesian networks with rule extraction to infer the risk of weed infestation in a corn-crop

This paper describes the modeling of a weed infestation risk inference system that implements a collaborative inference scheme based on rules extracted from two Bayesian network classifiers. The first Bayesian classifier infers a categorical variable value for the weed-crop competitiveness using as input categorical variables for the total density of weeds and corresponding proportions of narrow and broad-leaved weeds. The inferred categorical variable values for the weed-crop competitiveness along with three other categorical variables extracted from estimated maps for the weed seed production and weed coverage are then used as input for a second Bayesian network classifier to infer categorical variables values for the risk of infestation. Weed biomass and yield loss data samples are used to learn the probability relationship among the nodes of the first and second Bayesian classifiers in a supervised fashion, respectively. For comparison purposes, two types of Bayesian network structures are considered, namely an expert-based Bayesian classifier and a naive Bayes classifier. The inference system focused on the knowledge interpretation by translating a Bayesian classifier into a set of classification rules. The results obtained for the risk inference in a corn-crop field are presented and discussed.

Constructive Neural Network Algorithms for Feedforward Architectures Suitable for Classification Tasks

This chapter presents and discusses several well-known constructive neural network algorithms suitable for constructing feedforward architectures aiming at classification tasks involving two classes. The algorithms are divided into two different groups: the ones directed by the minimization of classification errors and those based on a sequential model. In spite of the focus being on two-class classification algorithms, the chapter also briefly comments on the multiclass versions of several two-class algorithms, highlights some of the most popular constructive algorithms for regression problems and refers to several other alternative algorithms.

An empirical evaluation of constructive neural network algorithms in classification tasks

Unlike conventional Neural Network (NN) algorithms that require the definition of the NN architecture before learning starts, Constructive Neural Network (CoNN) algorithms enable the network architecture to be constructed along with the learning process. This paper presents and discusses the results of an empirical evaluation of seven two-class CoNN algorithms, namely Tower, Pyramid, Tiling, Upstart, Shift, Perceptron Cascade (PC) and Partial Target Inversion (PTI) in 12 knowledge domains. The way each particular algorithm approaches the growing of the network determines their differences. This paper also presents and analyses empirical results of five multiclass CoNN algorithms in five knowledge domains, namely MTower, MPyramid, MTiling, MUpstart and MPerceptron Cascade, which can be considered extensions of their two-class counterparts. Results obtained with the Pocket with the Ratchet Modification (PRM) algorithm, with its multiclass version, the PRMWTA algorithm and with the back propagation algorithm, are presented for comparison.

Using a modified genetic algorithm to minimize the production costs for slabs of precast prestressed concrete joists

Genetic algorithms (GAs) are search methods that have been successfully applied to a variety of tasks. This paper describes the use of a modified GA as an optimization method in structural engineering for minimizing the production costs of slabs using precast prestressed concrete joists. The work initially identifies and describes the multiple costs involved in the production of these slabs and then combines them into a function subjected to 28 equality and inequality constraints. The experiments conducted address the minimization of this function using GA, where constraints are treated using a penalty technique. In addition, results obtained with a conventional optimization method are presented, for comparison.

Automatic learning of image filters using Cartesian genetic programming

This paper proposes a computational modeling for image filtering processes based on the Cartesian Genetic Programming CGP methodology, suitable for hardware devices. A computational system named ALIF-CGP Automatic Learning of Image Filters Using Cartesian Genetic Programming was designed as a simulator for automatically constructing a sequence of operators, mainly morphological and logical, which can filter a particular shape of image. ALIF-CGP is a convenient option for executing the non-trivial task, usually manually done by human experts, of selecting the sequence of nonlinear operators to be used in morphological filters. ALIF-CGP has already a built-in pool of morphological and logical operators, which can be used by default. The user, however, has the flexibility of choosing only those operators which are of interest or then, conveniently introduce new ones. The system expects as input a pair of images input-target. The flexibility given by the CGP-based computational modeling used by ALIF-CGP as well as its efficiency and satisfactory results, obtained in various image processing case studies, recommend its use when developing a hardware implementation for the purposes of image filtering. A few case studies using ALIF-CGP are presented and comparatively analyzed in relation to previous results available in the literature.

Enhancing Constructive Neural Network Performance Using Functionally Expanded Input Data

Abstract Constructive learning algorithms are an efficient way to train feedforward neural networks. Some of their features, such as the automatic definition of the neural network (NN) architecture and its fast training, promote their high adaptive capacity, as well as allow for skipping the usual pre-training phase, known as model selection. However, such advantages usually come with the price of lower accuracy rates, when compared to those obtained with conventional NN learning approaches. This is, perhaps, the reason for conventional NN training algorithms being preferred over constructive NN (CoNN) algorithms. Aiming at enhancing CoNN accuracy performance and, as a result, making them a competitive choice for machine learning based applications, this paper proposes the use of functionally expanded input data. The investigation described in this paper considered six two-class CoNN algorithms, ten data domains and seven polynomial expansions. Results from experiments, followed by a comparative analysis, show that performance rates can be improved when CoNN algorithms learn from functionally expanded input data.

Feature-weighted k-Nearest Neighbor Classifier

This paper proposes a feature weighting method based on X2 statistical test, to be used in conjunction with a k-NN classifier. Results of empirical experiments conducted using data from several knowledge domains are presented and discussed. Forty four out of forty five conducted experiments favoured the feature weighted approach and are empirical evidence that the proposed weighting process based on X2 is a good weighting strategy

Quantitative correlation between ability to compute and student performance in a primary school

Many studies have provided qualitative arguments for the premise that the ability to compute is fundamental and therefore should be treated at all levels of education. This work presents a quantitative indication of the validity of this premise for primary school students. In order to precisely define what ability a student should have to use a model of computation to compute, this work reports an experimental study that shows a significant correlation between the students ability to compute and his/her academic performance in a primary school.

Solar flare detection system based on tolerance near sets in a GPU-CUDA framework

This article presents a unique application of tolerance near sets (TNS) for detecting solar flare events in solar images acquired using radio astronomy techniques. In radio astronomy (RA) applications, the interferometric array processing of data streams presents algorithmic and response time challenges as well as a high volume of data. The radio interferometer is an RA instrument composed of an array of antennas. Radio signals emitted by a celestial object are captured by the antennas and are subsequently processed in such a way that each pair of antennas produces correlated data. The overall correlated data is then accumulated and, after an integration period, the spectral image of the observed object is obtained. The process of deconvolution of the spectral image produces the desired spatial image of the celestial object. The proposed solar flare detection system is embedded in a computational platform framework suitable for dealing with huge volumes of data, based on a cluster of CPU-GPU pairs. The experimental results presented in the paper include comparison of the TNS-based algorithm (implemented as the SOL-FLARE system) with the K-means algorithm using significant samples of test images to validate the detection system. The performances of both systems are comparatively analyzed using Receiver Operating Characteristic (ROC) curves. The images used in the experiments were selected from a data repository produced by the Nobeyama Radioheliograph, in Japan, during the years 2004 up to 2013. The main contribution of the article is a novel approach to solar flare detection in a GPU-CUDA framework.