Identificação de ninhos de cegonha em postes eléctricos

Dissertação para obtenção do Grau de Mestre em Engenharia Informática

Improvements on the KNN classifier

Dissertação para obtenção do Grau de Mestre em Engenharia Informática

Knowledge management framework based on brain models and human physiology

The life of humans and most living beings depend on sensation and perception for the best assessment of the surrounding world. Sensorial organs acquire a variety of stimuli that are interpreted and integrated in our brain for immediate use or stored in memory for later recall. Among the reasoning aspects, a person has to decide what to do with available information. Emotions are classifiers of collected information, assigning a personal meaning to objects, events and individuals, making part of our own identity. Emotions play a decisive role in cognitive processes as reasoning, decision and memory by assigning relevance to collected information. The access to pervasive computing devices, empowered by the ability to sense and perceive the world, provides new forms of acquiring and integrating information. But prior to data assessment on its usefulness, systems must capture and ensure that data is properly managed for diverse possible goals. Portable and wearable devices are now able to gather and store information, from the environment and from our body, using cloud based services and Internet connections. Systems limitations in handling sensorial data, compared with our sensorial capabilities constitute an identified problem. Another problem is the lack of interoperability between humans and devices, as they do not properly understand human’s emotional states and human needs. Addressing those problems is a motivation for the present research work. The mission hereby assumed is to include sensorial and physiological data into a Framework that will be able to manage collected data towards human cognitive functions, supported by a new data model. By learning from selected human functional and behavioural models and reasoning over collected data, the Framework aims at providing evaluation on a person’s emotional state, for empowering human centric applications, along with the capability of storing episodic information on a person’s life with physiologic indicators on emotional states to be used by new generation applications.

Hand gesture recognition for human computer interaction : a comparative study of different image features

Hand gesture recognition for human computer interaction, being a natural way of human computer interaction, is an area of active research in computer vision and machine learning. This is an area with many different possible applications, giving users a simpler and more natural way to communicate with robots/systems interfaces, without the need for extra devices. So, the primary goal of gesture recognition research is to create systems, which can identify specific human gestures and use them to convey information or for device control. For that, vision-based hand gesture interfaces require fast and extremely robust hand detection, and gesture recognition in real time. In this study we try to identify hand features that, isolated, respond better in various situations in human-computer interaction. The extracted features are used to train a set of classifiers with the help of RapidMiner in order to find the best learner. A dataset with our own gesture vocabulary consisted of 10 gestures, recorded from 20 users was created for later processing. Experimental results show that the radial signature and the centroid distance are the features that when used separately obtain better results, with an accuracy of 91% and 90,1% respectively obtained with a Neural Network classifier. These to methods have also the advantage of being simple in terms of computational complexity, which make them good candidates for real-time hand gesture recognition.

Segmentação do tecido tumoral cerebral em imagens de ressonância magnética

Dissertação de mestrado integrado em Engenharia Biomédica (área de especialização em Eletrónica Médica)

Sistema de apoio à decisão clínica: quality of life system

Dissertação de mestrado integrado em Engenharia e Gestão de Sistemas de Informação

Investigação das dificuldades na memória de trabalho em alunos do 2º ano do Ensino Fundamental em Manaus: um estudo com a Working Memory Rating Scale

Dissertação de mestrado em Educação Especial (área de especialização em Dificuldades de Aprendizagem Específicas)

Comparative Analysis of Signals Restoration by Different Kinds of Approximation

The comparative analysis of continuous signals restoration by different kinds of approximation is performed. The software product, allowing to define optimal method of different original signals restoration by Lagrange polynomial, Kotelnikov interpolation series, linear and cubic splines, Haar wavelet and Kotelnikov-Shannon wavelet based on criterion of minimum value of mean-square deviation is proposed. Practical recommendations on the selection of approximation function for different class of signals are obtained.

A novel gene expression signature in peripheral blood mononuclear cells for early detection of colorectal cancer.

BACKGROUND: Early detection and treatment of colorectal adenomatous polyps (AP) and colorectal cancer (CRC) is associated with decreased mortality for CRC. However, accurate, non-invasive and compliant tests to screen for AP and early stages of CRC are not yet available. A blood-based screening test is highly attractive due to limited invasiveness and high acceptance rate among patients. AIM: To demonstrate whether gene expression signatures in the peripheral blood mononuclear cells (PBMC) were able to detect the presence of AP and early stages CRC. METHODS: A total of 85 PBMC samples derived from colonoscopy-verified subjects without lesion (controls) (n = 41), with AP (n = 21) or with CRC (n = 23) were used as training sets. A 42-gene panel for CRC and AP discrimination, including genes identified by Digital Gene Expression-tag profiling of PBMC, and genes previously characterised and reported in the literature, was validated on the training set by qPCR. Logistic regression analysis followed by bootstrap validation determined CRC- and AP-specific classifiers, which discriminate patients with CRC and AP from controls. RESULTS: The CRC and AP classifiers were able to detect CRC with a sensitivity of 78% and AP with a sensitivity of 46% respectively. Both classifiers had a specificity of 92% with very low false-positive detection when applied on subjects with inflammatory bowel disease (n = 23) or tumours other than CRC (n = 14). CONCLUSION: This pilot study demonstrates the potential of developing a minimally invasive, accurate test to screen patients at average risk for colorectal cancer, based on gene expression analysis of peripheral blood mononuclear cells obtained from a simple blood sample.

Regiones salientes complejas para aplicaciones de seguimiento facial

Este trabajo presenta una metodología para detectar y realizar el seguimiento de características faciales. En el primer paso del procedimiento se detectan caras mediante Adaboost con cascadas de clasificadores débiles. El segundo paso busca las características internas de la cara mediante el CSR, detectando zonas de interés. Una vez que estas características se capturan, un proceso de tracking basado en el descriptor SIFT, que hemos llamado pseudo-SIFT, es capaz de guardar información sobre la evolución de movimiento en las regiones detectadas. Además, un conjunto de datos públicos ha sido desarrollado con el propósito de compartirlo con otras investigaciones sobre detección, clasificación y tracking. Experimentos reales muestran la robustez de este trabajo y su adaptabilidad para trabajos futuros.

Effect of training-sample size and classification difficulty on the accuracy of genomic predictors.

Introduction: As part of the MicroArray Quality Control (MAQC)-II project, this analysis examines how the choice of univariate feature-selection methods and classification algorithms may influence the performance of genomic predictors under varying degrees of prediction difficulty represented by three clinically relevant endpoints. Methods: We used gene-expression data from 230 breast cancers (grouped into training and independent validation sets), and we examined 40 predictors (five univariate feature-selection methods combined with eight different classifiers) for each of the three endpoints. Their classification performance was estimated on the training set by using two different resampling methods and compared with the accuracy observed in the independent validation set. Results: A ranking of the three classification problems was obtained, and the performance of 120 models was estimated and assessed on an independent validation set. The bootstrapping estimates were closer to the validation performance than were the cross-validation estimates. The required sample size for each endpoint was estimated, and both gene-level and pathway-level analyses were performed on the obtained models. Conclusions: We showed that genomic predictor accuracy is determined largely by an interplay between sample size and classification difficulty. Variations on univariate feature-selection methods and choice of classification algorithm have only a modest impact on predictor performance, and several statistically equally good predictors can be developed for any given classification problem.

Structured output SVM for remote sensing image classification

In the recent years, kernel methods have revealed very powerful tools in many application domains in general and in remote sensing image classification in particular. The special characteristics of remote sensing images (high dimension, few labeled samples and different noise sources) are efficiently dealt with kernel machines. In this paper, we propose the use of structured output learning to improve remote sensing image classification based on kernels. Structured output learning is concerned with the design of machine learning algorithms that not only implement input-output mapping, but also take into account the relations between output labels, thus generalizing unstructured kernel methods. We analyze the framework and introduce it to the remote sensing community. Output similarity is here encoded into SVM classifiers by modifying the model loss function and the kernel function either independently or jointly. Experiments on a very high resolution (VHR) image classification problem shows promising results and opens a wide field of research with structured output kernel methods.

Hierarchical conditional random fields for parts based models matching

A parts based model is a parametrization of an object class using a collection of landmarks following the object structure. The matching of parts based models is one of the problems where pairwise Conditional Random Fields have been successfully applied. The main reason of their effectiveness is tractable inference and learning due to the simplicity of involved graphs, usually trees. However, these models do not consider possible patterns of statistics among sets of landmarks, and thus they sufffer from using too myopic information. To overcome this limitation, we propoese a novel structure based on a hierarchical Conditional Random Fields, which we explain in the first part of this memory. We build a hierarchy of combinations of landmarks, where matching is performed taking into account the whole hierarchy. To preserve tractable inference we effectively sample the label set. We test our method on facial feature selection and human pose estimation on two challenging datasets: Buffy and MultiPIE. In the second part of this memory, we present a novel approach to multiple kernel combination that relies on stacked classification. This method can be used to evaluate the landmarks of the parts-based model approach. Our method is based on combining responses of a set of independent classifiers for each individual kernel. Unlike earlier approaches that linearly combine kernel responses, our approach uses them as inputs to another set of classifiers. We will show that we outperform state-of-the-art methods on most of the standard benchmark datasets.

Active Learning of Very-High Resolution Optical Imagery with SVM: Entropy vs Margin Sampling

An active learning method is proposed for the semi-automatic selection of training sets in remote sensing image classification. The method adds iteratively to the current training set the unlabeled pixels for which the prediction of an ensemble of classifiers based on bagged training sets show maximum entropy. This way, the algorithm selects the pixels that are the most uncertain and that will improve the model if added in the training set. The user is asked to label such pixels at each iteration. Experiments using support vector machines (SVM) on an 8 classes QuickBird image show the excellent performances of the methods, that equals accuracies of both a model trained with ten times more pixels and a model whose training set has been built using a state-of-the-art SVM specific active learning method

Meta-analysis based SVM classification enables accurate detection of Alzheimer's disease across different clinical centers using FDG-PET and MRI.

The application of support vector machine classification (SVM) to combined information from magnetic resonance imaging (MRI) and [F18]fluorodeoxyglucose positron emission tomography (FDG-PET) has been shown to improve detection and differentiation of Alzheimer's disease dementia (AD) and frontotemporal lobar degeneration. To validate this approach for the most frequent dementia syndrome AD, and to test its applicability to multicenter data, we randomly extracted FDG-PET and MRI data of 28 AD patients and 28 healthy control subjects from the database provided by the Alzheimer's Disease Neuroimaging Initiative (ADNI) and compared them to data of 21 patients with AD and 13 control subjects from our own Leipzig cohort. SVM classification using combined volume-of-interest information from FDG-PET and MRI based on comprehensive quantitative meta-analyses investigating dementia syndromes revealed a higher discrimination accuracy in comparison to single modality classification. For the ADNI dataset accuracy rates of up to 88% and for the Leipzig cohort of up to 100% were obtained. Classifiers trained on the ADNI data discriminated the Leipzig cohorts with an accuracy of 91%. In conclusion, our results suggest SVM classification based on quantitative meta-analyses of multicenter data as a valid method for individual AD diagnosis. Furthermore, combining imaging information from MRI and FDG-PET might substantially improve the accuracy of AD diagnosis.