Navegando por Assunto "Redes neurais convolucionais"

Agora exibindo 1 - 7 de 7

Acesso aberto (Open Access)
5G MIMO and LIDAR data for machine learning: mmWave beam-selection using deep learning
(Universidade Federal do Pará, 2019-08-29) DIAS, Marcus Vinicius de Oliveira; KLAUTAU JÚNIOR, Aldebaro Barreto da Rocha; http://lattes.cnpq.br/1596629769697284
Modern communication systems can exploit the increasing number of sensor data currently used in advanced equipment and reduce the overhead associated with link configuration. Also, the increasing complexity of networks suggests that machine learning (ML), such as deep neural networks, can effectively improve 5G technologies. The lack of large datasets make harder to investigate the application of deep learning in wireless communication. This work presents a simulation methodology (RayMobTime) that combines a vehicle traffic simulation (SUMO) with a ray-tracing simulator (Remcom’s Wireless InSite), to generate channels that represents realistic 5G scenarios, as well as the creation of LIDAR sensor data (via Blensor). The created dataset is utilized to investigate beam-selection techniques on vehicle-to-infrastructure using millimeter waves on different architectures, such as distributed architecture (usage of the information of only a selected vehicle, and processing of data on the vehicle) and centralized architectures (usage of all present information provided by the sensors in a given moment, processing at the base station). The results indicate that deep convolutional neural networks can be utilized to select beams under a top-M classification framework. It also shows that a distributed LIDAR-based architecture provides robust performance irrespective of car penetration rate, outperforming other architectures, as well as can be used to detect line-of-sight (LOS) with reasonable accuracy.
Acesso aberto (Open Access)
Uma Análise do uso de informacões multiescala no mapeamento da PSNR para pontuacão perceptual
(Universidade Federal do Pará, 2019-11-18) GONÇALVES, Luan Assis; ZAMPOLO, Ronaldo de Freitas; http://lattes.cnpq.br/9088524620828017; BARROS, Fabrício José Brito; http://lattes.cnpq.br/9758585938727609
The prediction of visual quality is crucial in image and video systems. For this task, image quality metrics based on the mean squared error prevail in the field, due to their mathematical straightforwardness, even though they do not correlate well with the visual human perception. Latest achievements in the area support that the use of convolutional neural networks (CNN) to assess perceptual visual quality is a clear trend. Results in other applications, like blur detection and de-raining, indicate the combination of information from different scales improves the CNN performance. However, to the best of our knowledge, the best way to embody multi-scale information in visual quality characterization is still an open issue. Thus, in this work, we investigate the influence of using multi-scale information to predict the perceptual image quality. Specifically, we propose a single-stream dense network that estimates a spatially-varying parameter of a logistic function used to map values of a objective visual quality metric to subjective visual quality scores through the reference image. The proposed method achieved a reduction of 36.37% and 69.45% for the number of parameters and floating-point operations per second, respectively, and its performance is compared with a competing state-of-the-art approach by using a public image database.
Acesso aberto (Open Access)
Classificação de tumores cerebrais: um estudo comparativo entre rede neural convolucional e rede neural convolucional com mecanismo de atenção
(Universidade Federal do Pará, 2024-09-30) SILVA, Ulrich Kauê Mendes Alencar da; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
Brain tumors are neurological diseases with a high potential impact on the lives of affected individuals, requiring a rapid and accurate diagnosis through complementary imaging tests, such as magnetic resonance imaging, which is considered the gold standard in this process. Considering the need for faster diagnosis, classification systems based on Machine Learning have been developed and within this context, this dissertation aims to present a comparative study between a Convolutional Neural Network (CNN) and a CNN with an attention mechanism, developed for the classification of brain tumors from magnetic resonance images. The comparative study aims to identify the impact of the attention mechanism on the performance of the CNN for tumor classification. For the development and evaluation of the proposed models, a public database was used, collected from the Kaggle website and made available by Masoud Nickparvar, which is composed of 7023 brain magnetic resonance images, segmented into four classes: glioma, meningioma, no tumor and pituitary. As a result, from the performance metrics obtained, considering the image base used for testing in both CNNs, an improvement in the CNN performance was observed after the introduction of the attention mechanism, where the network with this mechanism presented an increase of 1.98% in the accuracy metric, 2.07% in the precision metric, 2.18% in the sensitivity metric and 1.72% in the F1-score metric in relation to the CNN without the attention mechanism. It is also possible to highlight the results obtained in particular for the meningioma tumor class, since the CNN without the attention mechanism presented difficulties in classifying this class and, after the integration of the attention mechanism, the model obtained an accuracy increase of 6.54% for this class.
Acesso aberto (Open Access)
Detecção de danos em superfícies geotécnicas com redes neurais convolucionais de baixa complexidade
(Universidade Federal do Pará, 2024-05-29) ARAÚJO, Thabatta Moreira Alves de; FRANCÊS, Carlos Renato Lisboa; ttp://lattes.cnpq.br/7458287841862567
Most natural disasters result from geodynamic events, such as landslides and collapse of geotechnical structures. These failures are catastrophic that directly impact the environment and cause financial and human losses. Visual inspection is the main method for detecting surface flaws in geotechnical structures. However, visits to the site can be risky due to the possibility of soil’s instability. Furthermore, the terrain design, hostile environment and remote installation conditions make access to these structures impractical. When a quick and safe assessment is necessary, computer vision analysis becomes a potential alternative. However, studies on computer vision techniques still need to be explored in this field due to the particularities of geotechnical engineering, such as limited, redundant and scarce public data sets. In this context, this thesis presents a redes neurais convolucionais, do inglês Convolutional Neural Network (CNN) approach for identifying defects on the surface of geotechnical structures to reduce dependence on human-led on-site inspections. To this end, images of surface failure indicators were collected on slopes on the banks of a Brazilian highway, with the help of UAVs and mobile devices. Next, low-complexity CNN architectures were explored to build a binary classifier capable of detecting flaws apparent to the naked human eye in images. The architecture composed of three convolutional layers, each with 32 filters, followed by two fully connected layers, each composed of 128 neurons and output with one neuron, showed an accuracy of 94.26%. The performance evaluation of the model with the test set obtained AUC metrics of 0.99, confusion matrix, and a AUPRC curve that indicates robust performance of the classifier in detecting damage, while maintaining a low computational complexity, making it suitable for applications field practices. The contributions of the thesis include the provision of an image database, the obtaining of a classification model suitable for scarce data and limited computational resources, and the exploration of strategies for remote inspection and detection of signs of failure in geotechnical structures.
Acesso aberto (Open Access)
Metodologia para a classificação automática de doenças em plantas utilizando redes neurais convolucionais.
(Universidade Federal do Pará, 2019-11-07) REZENDE, Vanessa Castro; OLIVEIRA, Roberto Célio Limão de; http://lattes.cnpq.br/4497607460894318; SANTOS, Adam Dreyton Ferreira dos; http://lattes.cnpq.br/2616572481839756
Convolutional neural networks (CNNs) are one of the deep learning techniques that, due to the computational advancement of recent years, have enhanced the field of computer vision by enabling substantial gains in various classification problems, especially those involving digital images. Given the advantages of using these networks, a variety of applications for automatic plant diseases identification have been developed for specialized assistance or automated screening tools, contributing to more sustainable farming practices and improved food production security. In this context, this work aims to propose a methodology for the classification of multiple pathologies from distinct plant species, having as input a database composed of digital images of plant diseases. Initially, this methodology involved image preprocessing activities on the plant disease database to provide the appropriate input for selected CNN models (VGG16, RestNet101v1, ResNet101v2, ResNetXt50 and DenseNet169), as well as to generate ten new bases, ranging from 50 to 66 classes with greater representativeness, to submit the models to different situations. After model training, a comparative study was conducted based on widely used classification metrics such as test accuracy, f1-score, and area under the curve. To attest the significance of obtained results, the Friedman nonparametric statistical test and two post-hoc procedures were performed, which showed that ResNetXt50 and DenseNet169 obtained superior results when compared with VGG16 and ResNets.
Acesso aberto (Open Access)
Previsão de geração de energia fotovoltaica utilizando transformação de séries temporais em imagens e redes neurais convolucionais bidimensionais
(Universidade Federal do Pará, 2023-10-26) MONTEIRO, Diego Ramiro Melo; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
This research presents a novel approach based on a Bidimensional Convolutional Neural Network (CNN) and techniques for transforming time series data into images, such as Gramian Angular Field (GAF) and Recurrence Plot (RP), for short-term forecast of electricity generation from a photovoltaic microgrid connected to the electrical grid, located at the Center of Excellence in Energy Efficiency of the Amazon (Centro de Excelência em Eficiência Energética da Amazônia –CEAMAZON) at the Federal University of Pará (Universidade Federal do Pará –UFPA). The GAF and RP techniques were employed to transform the time series data into images, which were used as input for the CNN. More accurate electricity generation forecasts enable users to better estimate the potential costs for grid implementation and the payback periods, as well as assess the available load capacity that can be connected to the system with higher precision. The prediction results using GAF and RP with a 2D CNN were compared with results obtained using other established neural network architectures in the field, such as Multilayer Perceptron and 1D CNNs, yielding satisfactory Root Mean Square Error (RMSE) values. This demonstrates the applicability of using images generated from the transformation of photovoltaic time series data in a 2D CNN for this problem.
Acesso aberto (Open Access)
Rede neural convolucional aplicada à identificação de equipamentos residenciais para sistemas de monitoramento não-intrusivo de carga
(Universidade Federal do Pará, 2018-04-03) PENHA, Deyvison de Paiva; CASTRO, Adriana Rosa Garcez; http://lattes.cnpq.br/5273686389382860
This research presents the proposal of a new methodology for the identification of residential equipment in non-intrusive load monitoring systems. The system is based on a Convolutional Neural Network to classify residential equipment, which uses directly as inputs to the system, the transient power signal data of 7 equipment obtained at the moment they are connected in a residence. The methodology was developed using data from a public database (REED) that presents data collected at a low frequency (1 Hz). The results obtained in the test database show an accuracy of more than 90%, indicating that the proposed system is capable of performing the task of identification. In addition, the results presented are considered satisfactory when compared with the results already presented in the literature for the problem in question.