Navegando por Autor "Silva, Pedro Henrique Lopes"
Agora exibindo 1 - 11 de 11
Resultados por página
Opções de Ordenação
Item CapsProm : a capsule network for promoter prediction.(2022) Moraes, Lauro Ângelo Gonçalves de; Silva, Pedro Henrique Lopes; Luz, Eduardo José da Silva; Moreira, Gladston Juliano PratesLocating the promoter region in DNA sequences is of paramount importance in bioinformatics. This problem has been widely studied in the literature, but it has not yet been fully resolved. Some researchers have shown remarkable results using convolutional networks that allowed the automatic extraction of features from a DNA chain. However, a single architecture schema that could learn the promoter prediction task competitively for several organisms has not yet been achieved. Thus, researchers must seek new architectures by hand-designing or by Neural Architecture Search for each new evaluated organism dataset. This work proposes a versatile archi- tecture based on a capsule network that can accurately identify promoter sequences in raw DNA data from five different organisms, eukaryotic and prokaryotic. Our architecture, the CapsProm, could help create models with minimum effort to learn the promoter identification task between different datasets. Furthermore, the CapsProm showed competitive results, overcoming the baseline method in five out of seven tested datasets (F1-score). The models and source code are made available at https://github.com/lauromoraes/CapsNet-promoter.Item Chimerical dataset creation protocol based on Doddington Zoo : a biometric application with face, eye, and ECG.(2019) Silva, Pedro Henrique Lopes; Luz, Eduardo José da Silva; Moreira, Gladston Juliano Prates; Moraes, Lauro Ângelo Gonçalves de; Gomes, David MenottiMultimodal systems are a workaround to enhance the robustness and effectiveness of biometric systems. A proper multimodal dataset is of the utmost importance to build such systems. The literature presents some multimodal datasets, although, to the best of our knowledge, there are no previous studies combining face, iris/eye, and vital signals such as the Electrocardiogram (ECG). Moreover, there is no methodology to guide the construction and evaluation of a chimeric dataset. Taking that fact into account, we propose to create a chimeric dataset from three modalities in this work: ECG, eye, and face. Based on the Doddington Zoo criteria, we also propose a generic and systematic protocol imposing constraints for the creation of homogeneous chimeric individuals, which allow us to perform a fair and reproducible benchmark. Moreover, we have proposed a multimodal approach for these modalities based on state-of-the-art deep representations built by convolutional neural networks. We conduct the experiments in the open-world verification mode and on two different scenarios (intra-session and inter-session), using three modalities from two datasets: CYBHi (ECG) and FRGC (eye and face). Our multimodal approach achieves impressive decidability of 7.20 ± 0.18, yielding an almost perfect verification system (i.e., Equal Error Rate (EER) of 0.20% ± 0.06) on the intra-session scenario with unknown data. On the inter-session scenario, we achieve a decidability of 7.78 ± 0.78 and an EER of 0.06% ± 0.06. In summary, these figures represent a gain of over 28% in decidability and a reduction over 11% of the EER on the intra-session scenario for unknown data compared to the best-known unimodal approach. Besides, we achieve an improvement greater than 22% in decidability and an EER reduction over 6% in the inter-session scenario.Item COVID-19 detection in CT images with deep learning : a voting-based scheme and cross-datasets analysis.(2020) Silva, Pedro Henrique Lopes; Luz, Eduardo José da Silva; Silva, Guilherme; Moreira, Gladston Juliano Prates; Silva, Rodrigo Pereira da; Lucio, Diego Rafael; Gomes, David MenottiEarly detection and diagnosis are critical factors to control the COVID-19 spreading. A number of deep learningbased methodologies have been recently proposed for COVID-19 screening in CT scans as a tool to automate and help with the diagnosis. These approaches, however, suffer from at least one of the following problems: (i) they treat each CT scan slice independently and (ii) the methods are trained and tested with sets of images from the same dataset. Treating the slices independently means that the same patient may appear in the training and test sets at the same time which may produce misleading results. It also raises the question of whether the scans from the same patient should be evaluated as a group or not. Moreover, using a single dataset raises concerns about the generalization of the methods. Different datasets tend to present images of varying quality which may come from different types of CT machines reflecting the conditions of the countries and cities from where they come from. In order to address these two problems, in this work, we propose an Efficient Deep Learning Technique for the screening of COVID-19 with a voting-based approach. In this approach, the images from a given patient are classified as group in a voting system. The approach is tested in the two biggest datasets of COVID-19 CT analysis with a patient-based split. A cross dataset study is also presented to assess the robustness of the models in a more realistic scenario in which data comes from different distributions. The cross-dataset analysis has shown that the generalization power of deep learning models is far from acceptable for the task since accuracy drops from 87.68% to 56.16% on the best evaluation scenario. These results highlighted that the methods that aim at COVID-19 detection in CT-images have to improve significantly to be considered as a clinical option and larger and more diverse datasets are needed to evaluate the methods in a realistic scenario.Item Exploiting a loss and a synthetic dataset protocol for biometrics system.(2022) Silva, Pedro Henrique Lopes; Moreira, Gladston Juliano Prates; Luz, Eduardo José da Silva; Moreira, Gladston Juliano Prates; Luz, Eduardo José da Silva; Queiroz, Rafael Alves Bonfim de; Silva, Rodrigo César Pedrosa; Oliveira, Luciano Rebouças de; Santos, Thiago Oliveira dosOs sistemas biométricos são um assunto comum no cotidiano do ser humano. Os esforços para aumentar a segurança desses sistemas estão aumentando a cada ano devido à sua necessidade por robustez. Os sistemas baseados em uma modalidade biométrica não tem um desempenho próximo da perfeição em ambientes não cooperativos, o que exige abordagens mais complexas. Devido a isso, novos estudos são desenvolvidos para melhorar o desempenho de sistemas baseados em biometria, criando novas formas de ensinar um algoritmo de machine learning a criar novas representações. Atualmente, vários pesquisadores estão direcionando seus esforços para desenvolver novas abordagens de metric learning para arquiteturas de deep learning para uma ampla gama de problemas, incluindo biometria. Neste trabalho, propõe-se uma função de perda baseada em dados biométricos para criar representações profundas a serem utilizadas em sistemas biométricos, chamada de D-loss. Os resultados mostram a eficácia da função de perda proposta com a menor taxa de equal-error rate (EER) de 5,38%, 13,01% e 7,96% para MNIST-Fashion, CIFAR-10 e CASIA-V4. Uma estratégia diferente para aumentar a robustez de um sistema é a fusão de duas ou mais modalidades biométricas. No entanto, é impossível encontrar um conjunto de dados com todas as combinações de modalidades biométricas possíveis. Uma solução simples é criar um conjunto de dados sintéticos, embora a metodologia para criar um ainda seja um problema em aberto na literatura. Neste trabalho, propõe-se a criação de um critério para mesclar duas ou mais modalidades de tal forma a criar conjuntos de dados sintéticos semelhantes: o critério de Doddington Zoo. Várias estratégias de mesclagem são avaliadas: fusões ao nível de score (mínimo, multiplicação e soma) e nível de características (concatenação simples e metric learning). Um EER próximo a zero também é observado usando os critérios de fusão propostos com a fusão de soma de pontuação e as modalidades de Eletrocardiograma (CYBHi), olho e face (FRGC). Dois conjuntos de dados com mais de 1.000 indivíduos (UFPR-Periocular e UofTDB) são usados para avaliar os critérios de mesclagem junto com a D-loss e outras funções de metric learning. Os resultados mostram o aspecto do critério do Doddington Zoo de criar conjuntos de dados semelhantes (pequeno desvio padrão em relação ao critério randômico) e a robustez do D-loss (2,50% EER contra 2,17% da função de perda triplets e 5,74, da função de perda multi-similarity).Item Multi-objective approach for multiple clusters detection in data points events.(2019) Bodevan, Emerson Cotta; Duczmal, Luiz Henrique; Duarte, Anderson Ribeiro; Silva, Pedro Henrique Lopes; Moreira, Gladston Juliano PratesThe spatial scan statistic is a widely used technique for detecting spatial clusters. Several extensions of this technique have been developed over the years. The objectives of these techniques are the detection accuracy improvement and a flexibilization on the search clusters space. Based on Voronoi-Based Scan (VBScan), we propose a biobjective approach using a recursively VBScan method called multiobjective multiple clusters VBScan (MOMC-VBScan), alongside a new measure called matching. This approach aims to identify and delineate all multiple significant anomalies in a search space. We conduct several experiments on different simulated maps and two real datasets, showing promising results. The proposed approach proved to be fast and with good precision in determining the partitions.Item Seleção de features em representações profundas para a íris e a região periocular como modalidades biométricas.(2018) Silva, Pedro Henrique Lopes; Moreira, Gladston Juliano Prates; Luz, Eduardo José da Silva; Moreira, Gladston Juliano Prates; Gomes, David Menotti; Santos, Jeferson Alex dosA biometria já é um tema bem consolidado na literatura. Há diversos trabalhos que são baseados em uma única modalidade biométrica, contudo, sistemas que fazem uso de somente uma modalidade são suscetíveis a ataques de diversas naturezas e ruídos de todos os tipos, especialmente em ambientes não cooperativos. Como os ambientes não-cooperativos estão se tornando cada vez mais comuns, técnicas para contornar esse problema estão ganhando mais atenção, dentre elas, técnicas multimodais. A forma de como fusionar os dados de diferentes modalidades ainda é um problema em aberto. Neste trabalho, propõe-se um modelo unimodal para a íris treinado com uma CNN e o fusionamento bimodal da íris e da região periocular. Testou-se dois baselines para o fusionamento: fusão a nível de scores com três regras (soma, multiplicação e mínimo) e fusão a nível de features com concatenação simples. Propõe-se também uma seleção de características utilizando PSO sobre o fusionamento a nível de features. Os resultados são reportados usando as imagens da competição NICE.II no cenário de galeria aberta. Para a íris reportamos 2,21 (com desvio padrão de 0.019) de decidibilidade e EER de 14,59% (com desvio padrão de 0.22%), enquanto para o fusionamento da íris e da região periocular, reportamos decidibilidade de 3,43 (com desvio padrão de 0.015) e EER de 5,72% (com desvio padrão de 0.12%), atingindo resultados estatisticamente superiores aos estados-da-arte encontrados na literatura.Item Self-supervised learning for arrhythmia classification.(2023) Silva, Guilherme Augusto Lopes; Luz, Eduardo José da Silva; Silva, Pedro Henrique Lopes; Luz, Eduardo José da Silva; Silva, Pedro Henrique Lopes; Freitas, Vander Luis de Souza; Meneghini, Ivan ReinaldoArrhythmias, heart diseases that are commonly diagnosed through electrocar- diograms (ECG), require computational methods for detection and classification to improve the physician’s diagnosis. Although there is abundant literature on the subject, the high intra-patient variability and noise of ECG signals pose challenges in developing practical machine-learning models. To address this, we propose a cus- tomized adjustment of machine learning models through self-supervised learning with human-in-the-loop. Our approach introduces a pretext task called ECGWavePuzzle, which improves classification performance through better generalization. Evaluation metrics on the MIT-BIH database demonstrate the effectiveness of our approach, which improved the ECGnet global accuracy by over 10% and the Mousavi’s CNN by over 13%. Additionally, the experimental results demonstrated that the proposed approach improved the sensitivity and positive predictive value of the arrhythmic classes for certain patients.Item Towards an effective and efficient deep learning model for COVID-19 patterns detection in X-ray images.(2021) Luz, Eduardo José da Silva; Silva, Pedro Henrique Lopes; Silva, Rodrigo Pereira da; Silva, Ludmila; Ananias, João Víctor Gomes Guimarães; Miozzo, Gustavo; Moreira, Gladston Juliano Prates; Gomes, David MenottiPurpose Confronting the pandemic of COVID-19 is nowadays one of the most prominent challenges of the human species. A key factor in slowing down the virus propagation is the rapid diagnosis and isolation of infected patients. The standard method for COVID-19 identification, the Reverse transcription polymerase chain reaction method, is time-consuming and in short supply due to the pandemic. Thus, researchers have been looking for alternative screening methods, and deep learning applied to chest X-rays of patients has been showing promising results. Despite their success, the computational cost of these methods remains high, which imposes difficulties to their accessibility and availability. Thus, the main goal of this work is to propose an accurate yet efficient method in terms of memory and processing time for the problem of COVID-19 screening in chest X-rays. Methods To achieve the defined objective, we propose a new family of models based on the EfficientNet family of deep artificial neural networks which are known for their high accuracy and low footprints. We also exploit the underlying taxonomy of the problem with a hierarchical classifier. A dataset of 13,569 X-ray images divided into healthy, non-COVID-19 pneumonia, and COVID-19 patients is used to train the proposed approaches and other 5 competing architectures. We also propose a cross-dataset evaluation with a second dataset to evaluate the method generalization power. Results The results show that the proposed approach was able to produce a high-quality model, with an overall accuracy of 93.9%, COVID-19 sensitivity of 96.8%, and positive prediction of 100% while having from 5 to 30 times fewer parameters than the other tested architectures. Larger and more heterogeneous databases are still needed for validation before claiming that deep learning can assist physicians in the task of detecting COVID-19 in X-ray images, since the cross-dataset evaluation shows that even state-of-the-art models suffer from a lack of generalization power. Conclusions We believe the reported figures represent state-of-the-art results, both in terms of efficiency and effectiveness, for the COVIDx database, a database of 13,800 X-ray images, 183 of which are from patients affected by COVID-19. The current proposal is a promising candidate for embedding in medical equipment or even physicians’ mobile phones.Item Towards better heartbeat segmentation with deep learning classification.(2020) Luz, Eduardo José da Silva; Silva, Pedro Henrique Lopes; Silva, Rodrigo César Pedrosa; Silva, Ludmila; Guimarães, João; Miozzo, Gustavo; Moreira, Gladston Juliano Prates; Gomes, David MenottiPurpose Confronting the pandemic of COVID-19 is nowadays one of the most prominent challenges of the human species. A key factor in slowing down the virus propagation is the rapid diagnosis and isolation of infected patients. The standard method for COVID-19 identification, the Reverse transcription polymerase chain reaction method, is time-consuming and in short supply due to the pandemic. Thus, researchers have been looking for alternative screening methods, and deep learning applied to chest X-rays of patients has been showing promising results. Despite their success, the computational cost of these methods remains high, which imposes difficulties to their accessibility and availability. Thus, the main goal of this work is to propose an accurate yet efficient method in terms of memory and processing time for the problem of COVID-19 screening in chest X-rays. Methods To achieve the defined objective, we propose a new family of models based on the EfficientNet family of deep artificial neural networks which are known for their high accuracy and low footprints. We also exploit the underlying taxonomy of the problem with a hierarchical classifier. A dataset of 13,569 X-ray images divided into healthy, non-COVID-19 pneumonia, and COVID-19 patients is used to train the proposed approaches and other 5 competing architectures. We also propose a cross-dataset evaluation with a second dataset to evaluate the method generalization power. Results The results show that the proposed approach was able to produce a high-quality model, with an overall accuracy of 93.9%, COVID-19 sensitivity of 96.8%, and positive prediction of 100% while having from 5 to 30 times fewer parameters than the other tested architectures. Larger and more heterogeneous databases are still needed for validation before claiming that deep learning can assist physicians in the task of detecting COVID-19 in X-ray images, since the cross-dataset evaluation shows that even state-of-the-art models suffer from a lack of generalization power. Conclusions We believe the reported figures represent state-of-the-art results, both in terms of efficiency and effectiveness, for the COVIDx database, a database of 13,800 X-ray images, 183 of which are from patients affected by COVID-19. The current proposal is a promising candidate for embedding in medical equipment or even physicians’ mobile phones.Item Transformações geométricas no plano : atividades para o 8.º ano do ensino fundamental.(2020) Silva, Pedro Henrique Lopes; Viana, Marger da Conceição VenturaEste artigo apresenta atividades de um minicurso apresentado aos participantes do XII Encontro Nacional de Educação Matemática, realizado na cidade de São Paulo, de 13 a 16 de julho de 2016. Tais atividades integram uma pesquisa de natureza qualitativa de que uma turma do 8.º ano do Ensino Fundamental permitiu a constituição da amostra. Resulta, então, de uma pesquisa qualitativa que objetivou contribuir para o desenvolvimento do processo de ensino/aprendizagem da congruência de figuras planas, por meio de transformações geométricas tendo como fundamentação teórica o Paradigma Histórico Cultural de Vigotski. A amostra foi baseada em critérios pragmáticos e teóricos. Na análise dos dados foi utilizado o enfoque indutivo procurando compreender os fenômenos segundo a perspectiva dos participantes da situação em estudo. Explica-se que tais atividades, que envolvem transformações geométricas, tiveram a intenção de contribuir para o desenvolvimento do processo de ensino e aprendizagem de conteúdos da Geometria Euclidiana Plana relacionados com a congruência de figuras planas.Item A VNS algorithm for PID controller : hardware-in-the-loop approach.(2021) Silva, Guilherme Augusto Lopes; Silva, Pedro Henrique Lopes; Santos, Valéria; Rêgo Segundo, Alan Kardek; Luz, Eduardo José da Silva; Moreira, Gladston Juliano PratesTuning the Proportional Integral Derivative, or PID, controller in cyber-physical systems is a major challenge as it requires advanced mathematical skills. Several authors in the literature have shown that optimization algorithms are efficient for auto-adjust PID controller constants, especially when there is no mathematical modeling. However, the literature lacks works that show the efficiency of the Variable Neighborhood Search (VNS) algorithm to auto-adjust the PID. In this work, we investigate the efficiency of the Variable Neighborhood Algorithm to fine-tune a PID controller of a real cyber physical-system: a birotor flying drone. The approach consists of applying a numerical neighborhood structure to optimize the three constants of the PID, according to a proposed fitness function. Experiments reveal the feasibility of fine-tuning the PID controller and the birotor balancing with the Variable Neighborhood Algorithm with reduced time. We compared the VNS-approach against one based on genetic algorithms, and on average, the VNS-approach achieves better results with lower computational and memory costs. Results suggest that the approach may be used in real or commercial systems, helping to fine-tune the controller to new environment changes or even last-minute project modifications.