This dissertation presents a novel features selection wrapper method based on neural networks, named the Binary Wrapper for Features Selection Technique. The major aim of this method is to reduce the computation time that is consumed during the implementation of the process of features selection and classifier optimization in the Heuristic for Features Selection (HVS) method. The HVS technique is a neural network based features selection technique that uses the weights of a well-trained neural network as relevance index for each input feature with respect to the target. The HVS technique consumes long computation time because it follows a sequential approach to discard irrelevant, low relevance, and redundant features. Hence, the HVS technique discards a single feature only at each training session of the classifier. In order to reduce the computation time of the HVS technique, a threshold was produced and used to implement the features selection process. In this dissertation, a new technique, named the replacement technique, was designed and implemented to produce an appropriate threshold that can be used in discarding a group of features instead of discarding a single feature only, which is currently the case with HVS technique. Since the distribution of the candidate features (i.e. relevant, low relevance, redundant and irrelevant features) with respect to the target in a dataset is unknown, the replacement technique produces low relevance features (i.e. probes) to generate a low relevance threshold that is compared to the candidate features and used to detect low relevance, irrelevant and redundant features. Moreover, the replacement technique is considered to be a novel technique that overcomes the limitation of another similar technique that is known as: random shuffling technique. The goal of the random shuffling technique is to produce low relevance features (i.e. probes) in comparison with the relevance of the candidate features with respect to the target. However, using the random shuffling technique, it is not guaranteed to produce such features, whereas this is guaranteed when using the replacement technique. The binary wrapper for features selection technique was evaluated by implementing it over a number of experiments. In those experiments, three different datasets were used, which are: Congressional Voting Records, Wave Forms, and Multiple Features. The numbers of features in the datasets are: 16, 40, and 649 respectively. The results of those experiments were compared to the results of the HVS method and other similar methods to evaluate the performance of the binary wrapper for features selection technique. The technique showed a critical improvement in the consumed time for features selection and classifier optimization, since the consumed computation time using this method was largely less than the time consumed by the HVS method and other methods. The binary wrapper technique was able to save 0.889, 0.931, and 0.993 of the time that is consumed by the HVS method to produce results identical to those produced by the binary wrapper technique over the former three datasets. This implies that the amount of the saved computation time by the binary wrapper technique in comparison with the HVS method increases as the number of features in a dataset increases as well. Regarding the classification accuracy, the results showed that the binary wrapper technique was able to enhance the classification accuracy after discarding features, which is considered as an advantage in comparison with the HVS which did not enhance the classification accuracy after discarding features.

Space weather has become an international issue due to the catastrophic impact it can have on modern societies. Solar flares are one of the major solar activities that drive space weather and yet their occurrence is not fully understood. Research is required to yield a better understanding of flare occurrence and enable the development of an accurate flare prediction system, which can warn industries most at risk to take preventative measures to mitigate or avoid the effects of space weather. This thesis introduces novel technologies developed by combining advances in statistical physics, image processing, machine learning, and feature selection algorithms, with advances in solar physics in order to extract valuable knowledge from historical solar data, related to active regions and flares. The aim of this thesis is to achieve the followings: i) The design of a new measurement, inspired by the physical Ising model, to estimate the magnetic complexity in active regions using solar images and an investigation of this measurement in relation to flare occurrence. The proposed name of the measurement is the Ising Magnetic Complexity (IMC). ii) Determination of the flare prediction capability of active region properties generated by the new active region detection system SMART (Solar Monitor Active Region Tracking) to enable the design of a new flare prediction system. iii) Determination of the active region properties that are most related to flare occurrence in order to enhance understanding of the underlying physics behind flare occurrence. The achieved results can be summarised as follows: i) The new active region measurement (IMC) appears to be related to flare occurrence and it has a potential use in predicting flare occurrence and location. ii) Combining machine learning with SMART¿s active region properties has the potential to provide more accurate flare predictions than the current flare prediction systems i.e. ASAP (Automated Solar Activity Prediction). iii) Reduced set of 6 active region properties seems to be the most significant properties related to flare occurrence and they can achieve similar degree of flare prediction accuracy as the full 21 SMART active region properties. The developed technologies and the findings achieved in this thesis will work as a corner stone to enhance the accuracy of flare prediction; develop efficient flare prediction systems; and enhance our understanding of flare occurrence. The algorithms, implementation, results, and future work are explained in this thesis.

With the growth of computer power, Digital Image Processing plays a more and more important role in the modern world, including the field of industry, medical, communications, spaceflight technology etc. There is no clear definition how to divide the digital image processing, but normally, digital image processing includes three main steps: low-level, mid-level and highlevel processing. Low-level processing involves primitive operations, such as: image preprocessing to reduce the noise, contrast enhancement, and image sharpening. Mid-level processing on images involves tasks such as segmentation (partitioning an image into regions or objects), description of those objects to reduce them to a form suitable for computer processing, and classification (recognition) of individual objects. Finally, higher-level processing involves "making sense" of an ensemble of recognised objects, as in image analysis. Based on the theory just described in the last paragraph, this thesis is organised in three parts: Colour Edge and Face Detection; Hand motion detection; Hand Gesture Detection and Medical Image Processing. II In Colour Edge Detection, two new images G-image and R-image are built through colour space transform, after that, the two edges extracted from G-image and R-image respectively are combined to obtain the final new edge. In Face Detection, a skin model is built first, then the boundary condition of this skin model can be extracted to cover almost all of the skin pixels. After skin detection, the knowledge about size, size ratio, locations of ears and mouth is used to recognise the face in the skin regions. In Hand Motion Detection, frame differe is compared with an automatically chosen threshold in order to identify the moving object. For some special situations, with slow or smooth object motion, the background modelling and frame differencing are combined in order to improve the performance. In Hand Gesture Recognition, 3 features of every testing image are input to Gaussian Mixture Model (GMM), and then the Expectation Maximization algorithm (EM)is used to compare the GMM from testing images and GMM from training images in order to classify the results. In Medical Image Processing (mammograms), the Artificial Neural Network (ANN) and clustering rule are applied to choose the feature. Two classifier, ANN and Support Vector Machine (SVM), have been applied to classify the results, in this processing, the balance learning theory and optimized decision has been developed are applied to improve the performance.

Space weather has become an international issue due to the catastrophic impact it can have on modern societies. Solar flares are one of the major solar activities that drive space weather and yet their occurrence is not fully understood. Research is required to yield a better understanding of flare occurrence and enable the development of an accurate flare prediction system, which can warn industries most at risk to take preventative measures to mitigate or avoid the effects of space weather. This thesis introduces novel technologies developed by combining advances in statistical physics, image processing, machine learning, and feature selection algorithms, with advances in solar physics in order to extract valuable knowledge from historical solar data, related to active regions and flares. The aim of this thesis is to achieve the followings: i) The design of a new measurement, inspired by the physical Ising model, to estimate the magnetic complexity in active regions using solar images and an investigation of this measurement in relation to flare occurrence. The proposed name of the measurement is the Ising Magnetic Complexity (IMC). ii) Determination of the flare prediction capability of active region properties generated by the new active region detection system SMART (Solar Monitor Active Region Tracking) to enable the design of a new flare prediction system. iii) Determination of the active region properties that are most related to flare occurrence in order to enhance understanding of the underlying physics behind flare occurrence. The achieved results can be summarised as follows: i) The new active region measurement (IMC) appears to be related to flare occurrence and it has a potential use in predicting flare occurrence and location. ii) Combining machine learning with SMART's active region properties has the potential to provide more accurate flare predictions than the current flare prediction systems i.e. ASAP (Automated Solar Activity Prediction). iii) Reduced set of 6 active region properties seems to be the most significant properties related to flare occurrence and they can achieve similar degree of flare prediction accuracy as the full 21 SMART active region properties. The developed technologies and the findings achieved in this thesis will work as a corner stone to enhance the accuracy of flare prediction; develop efficient flare prediction systems; and enhance our understanding of flare occurrence. The algorithms, implementation, results, and future work are explained in this thesis.

在數據分析的領域中，尤其在大數據的領域之中，因常含有相當高維度的預測變數，故特徵選擇是一個很重要的主題。這個主題在半導體的應用上，已經獲得相當豐碩的成果，但在IC基板的應用上，成果就相對顯得貧乏。所以，此次的研究(以IC基板中鑽孔製程為例)將透過以下的試驗方法(含：GR-SNBC (Gain Ratio with Naive Bayes Classifier)、SU-SNBC (Symmetrical Uncer-tainty with Naive Bayes Classifier)與SU-CART (Symmetrical Uncer-tainty with Classification and Regression Tree Classifier))，來建立可應用於IC基板製程時間預測上的一組屬性。最後，此一研究的成果不僅在於，使用資料挖礦的方法，來找出一組具有顯著性，而且可以用來預測的IC基板製程時間的產品特徵屬性；而且，發現若為了縮短製程時間，來自產品結構本身的因子，會比來自產品在生產管理上的因子更具顯著的效果。 / Feature selection is significate subject in domain of data analysis, especially in big-data with a lot of high dimension predictive variables. In semi-conductor field, this subject has already gotten a plenty of achievement, but not in IC-substrate; so in this research for example of drilling operation, through experiments, it builds a group of se-lective features for this field to predict process time, and the methods used are GR-SNBC (Gain Ratio with Naive Bayes Classifier), SU-SNBC (Symmetrical Uncertainty with Naive Bayes Classifier) and SU-CART (Symmetrical Uncertainty with Classification and Regression Tree Classifier). The contributions of this research are not only a selective product characteristics subset suggested to predict process-time in IC-substrate fab via the data-mining methods here, but also an observation that in order to shorten the process time, the factors of product construction weighs more than production management.

La détection et l’évaluation des émotions sont des domaines qui suscitent un grand intérêt par de nombreuses communautés tant au niveau des sciences humaines que des sciences exactes. Dans cette thèse nous nous intéressons à la reconnaissance de l’anxiété sociale qui est une peur irrationnelle ressentie par une personne lors de toute forme de relation sociale. L’anxiété peut être révélée par un ensemble de traits physiques et physiologiques tels que l’intonation de la voix, les mimiques faciales, l’augmentation du rythme cardiaque, le rougissement… etc. L’avantage de l’utilisation des mesures physiologiques est que les individus ne peuvent pas les manipuler, c’est une source continue de données et chaque émotion est caractérisée par une variation physiologique particulière. Dans ce travail, nous proposons un système de mesure d’anxiété basé sur l’utilisation d’un seul signal physiologique « signal de pression sanguine volumique (Blood volume pulse BVP)». Le choix d’un seul capteur limite la gêne des sujets due au nombre de capteurs. De ce signal nous avons sélectionné des paramètres pertinents représentant au mieux les relations étroites du signal BVP avec le processus émotionnel de l’anxiété. Cet ensemble de paramètres est classé en utilisant les séparateurs à vastes marges SVM. Les travaux engagés dans le domaine de la reconnaissance des émotions utilisent fréquemment, pour support d’information, des données peu fiables ne correspondant pas toujours aux situations envisagées. Ce manque de fiabilité peut être dû à plusieurs paramètres parmi eux la subjectivité de la méthode d’évaluation utilisée (questionnaire, auto-évaluation des sujets, …etc.). Nous avons développé une approche d’évaluation objective des données basée sur les dynamiques des paramètres sélectionnés. La base de données utilisée a été enregistrée dans notre laboratoire dans des conditions réelles acquises sur des sujets présentant un niveau d’anxiété face aux situations sociales et qui ne sont pas sous traitement psychologique. L’inducteur utilisé est l’exposition à des environnements virtuels représentant quelques situations sociales redoutées. L’étape d’évaluation, nous a permis d’obtenir un modèle de données fiable pour la reconnaissance de deux niveaux d’anxiété. Ce modèle a été testé dans une clinique spécialisée dans les thérapies cognitives comportementales (TCC) sur des sujets phobiques. Les résultats obtenus mettent en lumière la fiabilité du modèle construit notamment pour la reconnaissance des niveaux d’anxiété sur des sujets sains ou sur des sujets phobiques ce qui constitue une solution au manque de données dont souffrent les différents domaines de reconnaissances / Detection and evaluation of emotions are areas of great interest in many communities both in terms of human and exact sciences. In this thesis we focus on social anxiety recognition, which is an irrational fear felt by a person during any form of social relationship. Anxiety can be revealed by a set of physical and physiological traits such as tone of voice, facial expressions, increased heart rate, flushing ... etc. The interest to the physiological measures is motivated by them robustness to avoid the artifacts created by human social masking, they are a continuous source of data and each emotion is characterized by a particular physiological variation. In this work, we propose a measurement system based on the use of a single physiological signal "Blood volume pulse BVP". The use of a single sensor limits the subjects’ discomfort. From the BVP signal we selected three relevant features which best represents the close relationship between this signal and anxiety status. This features set is classified using support vector machine SVM. The work undertaken in the field of emotion recognition frequently use, for information support, unreliable data do not always correspond to the situations envisaged. This lack of reliability may be due to several parameters among them the subjectivity of the evaluation method used (self-evaluation questionnaire, subjects…etc.). We have developed an approach to objective assessment of data based on the dynamics of selected features. The used database was recorded in our laboratory under real conditions acquired in subjects with a level of anxiety during social situations and who are not under psychological treatment. The used stimulus is the exposition to virtual environments representing some feared social situations. After the evaluation stage, we obtained a reliable model for the recognition of two levels of anxiety. The latter was tested in a clinic specializing in cognitive behavioral therapy (CBT) on phobic subjects. The results highlight the reliability of the built model specifically for the recognition of anxiety levels in healthy subjects or of phobic subjects, what constitutes a solution to the lack of data affecting different areas of recognition

Στα πλαίσια της διατριβής, πραγματοποιήθηκε μελέτη και ανάλυση σημάτων πηγών ακουστικής εκπομπής, προτάθηκαν νέες ολοκληρωμένες μεθοδολογίες βασισμένες σε συμβατικές αλλά και προχωρημένες τεχνικές επεξεργασίας και ανάλυσης δεδομένων για την εξαγωγή εκείνων των χαρακτηριστικών που διαχωρίζουν τα σήματα ακουστικής εκπομπής από τον περιβάλλοντα θόρυβο. Εξετάσθηκαν ποια χαρακτηριστικά γνωρίσματα (παράμετροι) περιέχουν σημαντικό τμήμα της “πληροφορίας” έτσι ώστε στη συνέχεια χρησιμοποιώντας προχωρημένες μεθόδους αναγνώρισης προτύπων να επιτευχθεί ανίχνευση και χαρακτηρισμός ρωγμοειδών αστοχιών σε θορυβώδεις συνθήκες αλλά και σε σύνθετες κατασκευές. Συνοπτικά στην παρούσα διατριβή προτάθηκε και αξιολογήθηκε μια νέα μέθοδος για την εκτίμηση της βέλτιστης τοποθέτησης αισθητήρων. Προτάθηκαν δύο μέθοδοι για τον εντοπισμό θέσης πηγής ακουστικής εκπομπής. Πραγματοποιήθηκε για πρώτη φορά εξαγωγή ενενήντα παραμέτρων, εκ’ των οποίων οι εξήντα επτά προσδιορίστηκαν μετά από επεξεργασία του σήματος στο πεδίο του χρόνου ενώ οι υπόλοιπες είκοσι τρεις με επεξεργασία του σήματος στο πεδίο της συχνότητας. H μείωση του αριθμού των παραμέτρων, χωρίς όμως να μειώνεται ταυτόχρονα και η αξιοπιστία του ταξινομητή, αποτελεί ένα μεγάλος μέρος έρευνας που πραγματοποιήθηκε στα πλαίσια εκπόνησης της παρούσας διατριβής. Προτάθηκαν και αξιολογήθηκαν τέσσερις μέθοδοι επιλογής παραμέτρων. Για πρώτη φορά κατασκευάστηκαν και αξιολογήθηκαν ολοκληρωμένα συστήματα ανίχνευσης αστοχιών τα οποία έχουν την δυνατότητα να ανιχνεύουν τη δημιουργία ρωγμών λόγω καταπόνησης σε καιρικές συνθήκες βροχής. Στο τελευταίο μέρος της διατριβής κατασκευάστηκε και αξιολογήθηκε ένα καινοτόμο σύστημα χαρακτηρισμού ρωγμοειδών γεγονότων για τις ενισχύσεις πλοίων, υπό προσομοιωμένες συνθήκες λειτουργίας του πλοίου. / The present PhD thesis dealt with the following subjects: best sensors position, source location, features extraction and features selection, crack detection on raining conditions, crack characterization in ship structures. A new method, for the estimation of the best sensors position that used for accurate acoustic emission source location on empty spherical surfaces, is presented. Two acoustic emission source location methods are presented and evaluated. In this thesis, an extensive set of ninety features (forty-one novel features) are extracted from acoustic emission signals, sixty-seven in the time domain and twenty-three by processing the signal in the frequency domain. The features are estimated for two time-frames the first has 1msec duration (typically the signal does not contain all the reflections from the material edges) and the second has 32msec of the normalized signal, which is not separated by its reflections, in small structures. To achieve robust performance both in accuracy and computational complexity of any classification method, it is necessary to pick up the most relevant features. Four features selection methods are proposed and evaluated. In outside constructions (e.g bridges, tanks, ships etc) real-life noises reduce significantly the capability of location and characterization acoustic emission sources. Among the most important types of noise is the rain, producing signal similar to crack. A completed system of detection crack on condition of rain is estimated. An efficient system for automatic and real-time characterization of crack events using a robust set of features to monitor crack events in ship structures is presented. In normal operation of ship, real-life noises (e.g engines, sea waves, weather conditions etc) reduce significantly the capability of location and characterization of crack events.

Submitted by JÚLIO HEBER SILVA (julioheber@yahoo.com.br) on 2017-05-31T17:56:45Z No. of bitstreams: 2 Dissertação - Rhelcris Salvino de Sousa -2017.pdf: 12280322 bytes, checksum: 2985f69ec9d4b79ed4266baba761bd15 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) / Approved for entry into archive by Luciana Ferreira (lucgeral@gmail.com) on 2017-06-01T11:00:44Z (GMT) No. of bitstreams: 2 Dissertação - Rhelcris Salvino de Sousa -2017.pdf: 12280322 bytes, checksum: 2985f69ec9d4b79ed4266baba761bd15 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) / Made available in DSpace on 2017-06-01T11:00:44Z (GMT). No. of bitstreams: 2 Dissertação - Rhelcris Salvino de Sousa -2017.pdf: 12280322 bytes, checksum: 2985f69ec9d4b79ed4266baba761bd15 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) Previous issue date: 2017-04-20 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / Machine learning problems usually involve a large number of features or variables. In this context, feature selection algorithms have the challenge of determining a reduced subset from the original set. The main difficulty in this task is the high number of solutions available in the search space. In this context, genetic algorithm is one of the most used techniques in this type of problem due to its implicit parallelism in the exploration of the search space of the problem considered. However, a binary type representation is usually used to encode the solutions. This work proposes an implementation solution that makes use of integer representation called intEA-MLR instead of binary. The integer representation optimizes the understanding of the data, as the features to be selected are represented by integer values, reducing the size of the chromosome used in the search process. The intEA-MLR in this context is presented as an alternative way of solving high dimensional problems in regression problems. As a case study, three different sets of data are used concerning problems involving determination of properties of interest in samples of 1) Grain Wheat, 2) Medicine tablets and 3) petroleum. Such sets were used in competitions held at the International Diffuse Reflectance Conference (IDRC) (http://cnirs.clubexpress.com/content.aspx?page_id=22&club_ id=409746&module_id=190211), in the years 2008, 2012 and 2014, respectively. The results showed that the proposed solution was able to improve the obtained solutions when compared to the classical implementation that makes use of binary coding, with both more accurate prediction models and with reduced number of features. IntEA-MLR also outperformed the competition winners, reaching 91.17% better than the competition winner for the petroleum data set. In addition, the results also indicated that the computation time required by the intEA-MLR is relatively smaller as more features are available. / Problemas de aprendizado de máquina geralmente envolvem um grande número de características ou variáveis. Nesse contexto, algoritmos de seleção de características tem como desafio determinar um subconjunto reduzido a partir do conjunto original. A principal dificuldade nesta tarefa é o elevado número de soluções disponíveis no espaço de busca. Nesse contexto, algoritmo genético é uma das técnicas mais utilizadas nesse tipo de problema em razão de seu paralelismo implícito na exploração do espaço de busca do problema considerado. Entretanto, geralmente utiliza-se uma representação do tipo biná- ria para codificar as soluções. Neste trabalho é proposto uma solução de implementação que faz uso de representação inteira denominada intEA-MLR em detrimento da binária. A representação inteira otimiza o entendimento dos dados, na medida em que as características a serem selecionadas são determinadas por valores inteiros reduzindo o tamanho do cromossomo utilizado no processo de busca. O intEA-MLR nesse contexto, se apresenta como uma forma alternativa de resolução de problemas de alta dimensionalidade em problemas de regressão. Como estudo de caso, utiliza-se três diferentes conjuntos de dados referente a problemas envolvendo determinação de propriedades de interesse em amostra de 1) Grãos de Trigo, 2) Comprimidos de remédio e 3) Petróleo. Tais conjuntos foram utilizados nas competições realizadas no International Diffuse Reflectance Conference (IDRC) (http://cnirs.clubexpress.com/content.aspx?page_id=22&club_ id=409746&module_id=190211), nos anos de 2008, 2012 e 2014, respectivamente. Os resultados mostraram que a solução proposta foi capaz de aprimorar as soluções obtidas quando comparadas com a implementação clássica que faz uso da codificação binária, tanto com modelos de predição mais acurados quanto com número reduzido de características. intEA-MLR também obteve resultados superiores aos dos vencedores das competições, chegando a obter soluções 91,17% melhores do que o vencedor da competição para o conjunto de dados de petróleo. Adicionalmente, os resultados também indicaram que o tempo de computação requerido pelo intEA-MLR é relativamente menor a medida em que um número maior de características estão disponíveis.

The aim of this semestral work is to summarize the existing knowledge from the area of comparison of musical recordings and to implement an evaluation system for determining the period of creation using the music signal parameterization. In the first part of this work are describe representations which can music take. Next, there is a cross-section of parameters that can be extracted from music recordings provides information on the dynamics, tempo, color, or time development of the music’s recording. In the second part is described evaluation system and its individual sub-blocks. The input data for this evaluation system is a database of 56 sound recordings of the first movement of Beethoven’s 5th Symphony. The last chapter is dedicated to a summary of the achieved results.

Afin de poursuivre et d'améliorer la tâche de détection qui est en cours à l'INSA, nous nous sommes concentrés sur la fusion des informations visibles et infrarouges du point de vue de reconnaissance des obstacles, ainsi distinguer entre les véhicules, les piétons, les cyclistes et les obstacles de fond. Les systèmes bimodaux ont été proposées pour fusionner l'information à différents niveaux: des caractéristiques, des noyaux SVM, ou de scores SVM. Ils ont été pondérés selon l'importance relative des capteurs modalité pour assurer l'adaptation (fixe ou dynamique) du système aux conditions environnementales. Pour évaluer la pertinence des caractéristiques, différentes méthodes de sélection ont été testés par un PPV, qui fut plus tard remplacée par un SVM. Une opération de recherche de modèle, réalisée par 10 fois validation croisée, fournit le noyau optimisé pour SVM. Les résultats ont prouvé que tous les systèmes bimodaux VIS-IR sont meilleurs que leurs correspondants monomodaux. / To continue and improve the detection task which is in progress at INSA laboratory, we focused on the fusion of the information provided by visible and infrared cameras from the view point of an Obstacle Recognition module, this discriminating between vehicles, pedestrians, cyclists and background obstacles. Bimodal systems have been proposed to fuse the information at different levels:of features, SVM's kernels, or SVM’s matching-scores. These were weighted according to the relative importance of the modality sensors to ensure the adaptation (fixed or dynamic) of the system to the environmental conditions. To evaluate the pertinence of the features, different features selection methods were tested by a KNN classifier, which was later replaced by a SVM. An operation of modelsearch, performed by 10 folds cross-validation, provides the optimized kernel for the SVM. The results have proven that all bimodal VIS-IR systems are better than their corresponding monomodal ones.