Skip to main content

Ana Carolina

UFPA - Federal University of Pará, Institute of Technology, Department Member

Followers

37

Following

13

Public Views

Ms. Siravenha received the master’s degree and doctor’s degree in electrical engineering from the Federal University of Para, Brazil, in 2010 and 2017, respectively. From 2013 and 2017 worked as a Fadesp (Fundação de Amparo e Desenvolvimento da Pesquisa) Fellow at the Vale Institute of Technology. Currently is Researcher assistant at SENAI Innovation Institute for Mineral Technologies. Acting mainly in signal processing projects, since 2008 dedicates to different branches of image processing. During her post-graduate studies worked in the processes of noise reduction, segmentation, and classification of images from remote sensors. During the Fadesp fellowship performed activities concerning plant species recognition from plants leaves, from the database building to species identification by machine learning algorithms. Also, directly contributed to the development of the app that runs the developed algorithms.
Since the last period as a Fadesp fellowship until now dedicates to projects of electroencephalography (EEG) signals processing and neuro-cognitive training, and objects inspection from computational vision tools.
Her interests are related to machine learning algorithms and its applications, mental states monitoring using EEG and performance enhancing by neuro-cognitive training.
Phone: 91988886021
Address: http://lattes.cnpq.br/4383482501456728

less

Uploads

Papers by Ana Carolina

A Novel Procedure for Classification of Early Human Actions from EEG Signals

We introduce a novel procedure that extends the time feasibility for classification of early huma... more We introduce a novel procedure that extends the time feasibility for classification of early human actions. Its major characteristic is to use epoch training data from a wider time duration before action onset (i.e., within the intention period) instead of data from localized sliding windows. This is the case of time-specific and selected fixed classifiers. Our approach models human actions from EEG signals and leverages on amplitudes and power frequencies to construct fifteen groups of action vectors, which were subjected to a set of classifiers. Regarding early classification our approach did it earlier than both time-specific and selected fixed classifiers. Moreover, our results reported an increase in classification performance.

Utilização de Redes Neurais do tipo Extreme Learning Machine na classificação da cobertura de solo do Município de Novo Progresso-PA

The municipality of Novo Progresso, in Brazil, has been a frequent topic in discussions about env... more The municipality of Novo Progresso, in Brazil, has been a frequent topic in discussions about environmental issues because having one of the highest deforestation rates in the Amazon region. These high rates of deforestation has a direct relationship with the development of policies implemented in the region over the years. In this context, the aim of this work is to develop a supervised architecture for image classification by Artificial Neural Networks (RNA), using the extreme learning machine algorithm (ELM). This learning machine has a set of parameters able to identify the characteristics of deforestation over the years and, consequently, the different features of use and land cover in remote sensing images. The acquired scenes, from the satellite Landsat-5 TM, represent part of the municipality of Novo Progresso, through the period 1984 to 2011. The results were evaluated through the confusion matrices, where the coefficients quantify the correctness of the classes, the Kappa index and the index of overall accuracy, beyond the Mean Square Error. Although the methodology effectiveness, there are some features to be considered in future works, mainly considering the influence of the mid infrared during the feature vector formation.

The use of Discrete Cosine Transform for satellite images segmentation and comparison to statistical metrics

In the last years a large amount of methods have been proposed to segment images, and it is show ... more In the last years a large amount of methods have been proposed to segment images, and it is show that this task is still an open problem. Remote sensing is one field on Digital Image Processing (DIP) that is becoming increasingly widespread. In this paper we describe a novel approach based on Discrete Cosine Transformation (DCT) and compare it with some approaches based on statistical metrics. Both, the DCT and statistical approaches, make a texture analysis based on components of a gray scale satellite image from QuickBird QB2 and Word View WV02 sensors. The texture analysis are based on the similarity property within a group of pixels that allows simple sub-blocks manipulation. The DCT is applied for the image and the AC energy component is calculated individually for square blocks. The segmentation rules of the proposed method are defined based on those values of energy. The statistical metrics carries out a local analysis of pixel intensities where the statistical information of pixels intensities are calculated in image split into N × N blocks. The statistical metrics adopted for comparison are the mean, standard deviation, median and variance. The segmentation of statistical metrics is done according a maximum likelihood classifier and the approaches evaluation are done by an entropy analysis. The results demonstrate the effectiveness of the proposed DCT-based method when compared to the other statistical metrics.

Atmospheric components: analyzing Inpainting methods to removal these occurrences in remote sensing images

This paper presents the evaluation of two approaches, widely used in the inpainting literature, a... more This paper presents the evaluation of two approaches, widely used in the inpainting literature, applied in the context of the atmospheric noise removal, such as fog, clouds-dense and sparse-and shadows, which often occur in remote sensing images. The presence of such elements affect, in many ways, the image processing in an environmental or urban monitoring, and also in the steps of the digital image processing, suchlike segmentation and classification. Whilst one approach uses a technique of interpolation for the dissemination of information by a multidimensional Discrete Cosine Transform (DCT) smoothing method, the other one is based on second-order partial differential equations methods (PDE). This PDE-based development uses the heat diffusion and thin-plate spline methods to achieve their solutions with the aid of the finite-difference method. To proceed the methods evaluation, this work uses the kappa coefficient and the Peak Signal-to-Noise Ratio index (PSNR). The metrics indicate the effectiveness of the DCT strategy, which produces higher quality images, specially when comparing the results obtained by the use of differential equations modeled by thin-plate spline. The visual aspect of images is clearly an important factor for measuring the effectiveness of any image processing, so, in addition to numerical metrics, are presented the nine images used to evaluate both methods.

A Graphical Open Source Tool for Preprocessing Satellites20190422 80636 11ep3tj

The digital image processing is widely studied in the scientific literature. Techniques for image... more The digital image processing is widely studied in the scientific literature. Techniques for image treatment derived from satellites are continuously developed and improved. Although most of these techniques are readily available to the researchers, it is known that a minimum knowledge in programming should be purchased by those who wish to use these techniques. In this context, this paper proposes an open source graphical application which refers to preprocessing of images coming from satellites. The main focus is to become simple the use of preprocessing techniques without requiring the user to have knowledge about programming. To achieve this goal, technologies such as Qt framework (used for the development of graphical interface) and TerraLib (used for the development of preprocessing techniques) were used together and Faç ade design pattern was chosen to perform the communication between the technologies. This design pattern consists of developing a single interface to access resources developed and available in TerraLib library. The communication between the technologies is explained in detail and this work concludes with expectations about the proposal graphical application. Index Terms-Faç ade design pattern, Qt framework, TerraLib library.

The Development of a Hybrid Solution to Replacement of Clouds and Shadows in Remote Sensing Images

Nowadays, many works are dedicated to improve the research results, previously achieved manually,... more Nowadays, many works are dedicated to improve the research results, previously achieved manually, by computational solutions. On light of this, the presented work aims to overcome a common problem in many satellite images, which is the presence of undesirable atmospheric components such as clouds and shadows at the time of scene capture. The presence of such elements hinders the identification of meaningful information for applications like urban and environmental monitoring, exploration of natural resources, etc. Thus, it is presented a new way to perform a hybrid approach toward removal and replacing of these elements. The authors propose a method of regions decomposition using a nonlinear median filter, in order to map regions of structure and texture. These types of regions will explain which method will be applied to region redefinition. At structure region, will be applied the method of inpainting by a smoothing based on DCT, and at texture one, will be applied the exemplar-based texture synthesis. To measure the effectiveness of this proposed technique, a qualitative assessment was presented, at the same time that a discussion about quantitative analysis was made.

Computational Modeling of Objects Presented in Images

Nowadays, many works are dedicated to improve the research results, previously achieved manually,... more Nowadays, many works are dedicated to improve the research results, previously achieved manually, by computational solutions. On light of this, the presented work aims to overcome a common problem in many satellite images, which is the presence of undesirable atmospheric components such as clouds and shadows at the time of scene capture. The presence of such elements hinders the identification of meaningful information for applications like urban and environmental monitoring, exploration of natural resources, etc. Thus, it is presented a new way to perform a hybrid approach toward removal and replacing of these elements. The authors propose a method of regions decomposition using a nonlinear median filter, in order to map regions of structure and texture. These types of regions will explain which method will be applied to region redefinition. At structure region, will be applied the method of inpainting by a smoothing based on DCT, and at texture one, will be applied the exemplar-based texture synthesis. To measure the effectiveness of this proposed technique, a qualitative assessment was presented, at the same time that a discussion about quantitative analysis was made.

The use of high-pass filters and the Inpainting method to clouds removal and their impact on satellite images classification

This paper proposes a new technique to smooth undesirable elements of the atmosphere, such as fog... more This paper proposes a new technique to smooth undesirable elements of the atmosphere, such as fogs, clouds and shadows, which damage and lead to loss of image data. In our approach, an efficient way to detect clouds and shadows is presented. The method applies constants related to such undesirable elements, as well as a High boost Filter in the homomorphic filtering for scattered clouds removal. We highlight the use of the Inpainting method, which replaces contaminated pixels using a nearest neighbor interpolation. Beside this, the proposed algorithm adopts a morphologic opening of the image that aims to suppress remaining occurrences. The results are evaluated by Kappa coefficient and PSNR index, proving the good performance of the method.

Evaluating Inpainting methods to the satellite images clouds and shadows removing

This paper presents the evaluation of two approaches widely used in the inpainting literature, ap... more This paper presents the evaluation of two approaches widely used in the inpainting literature, applied in the context of atmospheric noise removal, such as fog, clouds dense and sparse and shadows, which often occurs in remote sensing images. One approach uses the technique of nearest neighbor interpolation for the information dissemination by a DCT-based smoothing method, and the other is based on second-order partial differential equations methods that uses the heat diffusion and thin-plate spline methods, achieving their solutions by using the finitedifference method. Finally, the evaluation uses the Kappa coefficient and the PSNR index. The metrics indicate the effectiveness of the nearest neighbor interpolation strategy, which produces higher quality images, specially when comparing the results obtained by the use of differential equations modeled by thin-plate spline.

Uso de Regras Fonol´ogicas com Determinação de Vogal Tônica para Conversão Grafema-Fone em Português Brasileiro

A conversão de uma sequência de caracteres em sequências de fones é um importante pré-requisito p... more A conversão de uma sequência de caracteres em
sequências de fones é um importante pré-requisito para serviços
que envolvem reconhecimento e/ou s´ıntese de voz. Contudo, a
tarefa não é trivial e diversas técnicas de conversão vêm sendo
adotadas ao longo da última década. Existe um número bem
menor de estudos na área dedicados ao Português Brasileiro (PB)
quando comparado ao Inglês, por exemplo. Este trabalho discute
esforços para reduzir esta deficiência, enfatizando-se a conversão
grafema-fone, e apresenta um sistema para reconhecimento de
voz em PB usando o novo corpus West Point. Os seguintes recursos
encontram-se dispon´ıveis: HTK scripts, dicionário fonético,
modelos ac´ ustico e de linguagem.

Using Concepts of CSCW to Validate Requirements Elicited for a Brazilian Geographic Information System

This work is based on the connection between two important and growing fields in computing: Compu... more This work is based on the connection between two important and growing fields in computing: Computer-Supported Cooperative Work (CSCW) and Geographic Information Systems (GIS). In this kind of system is important the concept of teamwork, due to the complexity and size of the projects underway around the world, where there may be situations where users need to access, modify and share spatial information in real time from different geographical locations, and also perform classifications, georeferencing and other activities synchronously. In the study summarized in this paper, was conducted a process of requirements engineering about the collaborative work in the TerraAmazon system, a multiuser workspace, developed with the open source geoprocessing library called Terralib, used in major important projects of urban and environmental research institutions, like the National Institute for Space Research (INPE). With regards to the requirements elicitation technique, it was made a kind of half-open interview with introspection and discussion of scenarios with photo interpreters (potential users) from INPE (The Amazon Regional Center) and Brazilian Enterprise for Agricultural Research (EMBRAPA). The requirements are validated and solutions are proposed using important concepts in CSCW, like awareness, coordinators and communicators. Finally, the conclusions are presented.

Plant Classification from Leaf Textures

This work describes a methodology for plant classification based on the analysis of leaf textures... more This work describes a methodology for plant classification based on the analysis of leaf textures by combining a multi-resolution technique, such as the two-dimensional (2D) Discrete Wavelet Transform (2D-DWT), statistical models and Gray-Level Co-occurrence Matrices (GLCM) in which some invariance (e.g. rotation and scale) are achieved. As a second step, an Artificial Neural Network (ANN) model is trained for automatic classifying plant species. The proposed approach was tested on the Flavia database. An overall classification accuracy of 91.85% was achieved which demonstrates that plants can be reliably classified using texture samples extracted from leaf tissues.

A hybrid approach to clouds and shadows removal in satellite images

This work aims to overcome a common problem in many satellite images, which is the presence of un... more This work aims to overcome a common problem in many satellite images, which is the presence of undesirable atmospheric components such as clouds and shadows at the time of scene capture. The presence of such elements can hinder the identification of image objects, urban environmental monitoring, and subsequent steps of the digital image processing like segmentation and classification. Thus, this work presents a new way to perform a hybrid approach toward removal and replacing of these elements in satellite images. The authors propose a method of regions decomposition using nonlinear median filter in order to map the regions of structure and texture. In this areas will be applied the methods of inpainting, by smoothing based on DCT, and exemplar-based texture synthesis, respectively. Finally, it was found the effectiveness of this technique through a qualitative evaluation at the same time that a discussion about quantitative analysis is made.

A Deep Learning Approach for Classification of Reaching Targets from EEG Images

In this paper, we propose a new approach for the classification of reaching targets before moveme... more In this paper, we propose a new approach for the classification of reaching targets before movement onset, during visually-guided reaching in 3D space. Our approach combines the discriminant power of two-dimensional Electroencephalography (EEG) signals (i.e., EEG images) built from short epochs, with the feature extraction and classification capabilities of deep learning (DL) techniques, such as the Convolutional Neural Networks (CNN). In this work, reaching motions are performed into four directions: left, right, up and down. To allow more natural reaching movements, we explore the use of Virtual Reality (VR) to build an experimental setup that allows the subject to perform self-paced reaching in 3D space while standing. Our results reported an increase both in classification performance and early detection in the majority of our experiments. To our knowledge this is the first time that EEG images and CNN are combined for the classification of reaching targets before movement onset.

A histogram-based approach to color satellite image segmentation

Hierarchical approaches for image segmentation are largely used in literature. Images from satell... more Hierarchical approaches for image segmentation are largely used in literature. Images from satellite, in general, provides a lager amount of information because the resolution allowed by image sensors. Considering this fact, the algorithm directed to process these images must be efficiently able to work with larger quantity of information, reading and writing matrices with million of pixels. The proposed approach uses histogram analysis for segmentation in color domain, and it is a preparatory task for a classifier algorithm that will be applied for satellite image components identification.

Exploring the Use of Leaf Shape Frequencies for Plant Classification

Fig. 1. Leaf shape modeling in the frequency domain. This figure shows the digital image process ... more Fig. 1. Leaf shape modeling in the frequency domain. This figure shows the digital image process in combination with the Fourier descriptor for extracting a leaf shape signature. From left to right: original image; binary representation; extracted leaf contour; edge-centroid distance measure; normalised Fourier coefficients. Abstract-Plant identification and classification play an important role in ecology, but the manual process is cumbersome even for experimented taxonomists. Technological advances allows the development of strategies to make these tasks easily and faster. In this context, this paper describes a methodology for plant identification and classification based on leaf shapes, that explores the discriminative power of the contour-centroid distance in the Fourier frequency domain in which some invariance (e.g. rotation and scale) are guaranteed. In addition, it is also investigated the influence of feature selection techniques regarding classification accuracy. Our results show that by combining a set of features vectors-in the principal components space-and a feedforward neural network, an accuracy of 97.45% was achieved.