Cette thèse s'inscrit dans le contexte de diagnostic assisté par ordinateur pour la localisation subcellulaire des protéines dans les images microscopiques. L'objectif est la conception et le développement d'un système de classification automatique permettant d'identifier le compartiment cellulaire dans lequel une protéine d'intérêt exerce son activité biologique. Afin de surmonter les difficultés rencontrées pour discerner les compartiments cellulaires présents dans les images microscopiques, les systèmes décrits dans la littérature proposent d'extraire plusieurs descripteurs associés à une combinaison de classifieurs. Dans cette thèse, nous proposons un schéma de classification différent répondant mieux aux besoins de généricité et de flexibilité pour traiter différentes bases d'images.Dans le but de fournir une caractérisation riche des images microscopiques, nous proposons un nouveau système de représentation permettant d'englober de multiples descripteurs visuels identifiés dans les différentes approches d'extraction de caractéristiques : locale, fréquentielle, globale et par région. Nous formulons ensuite le problème de fusion et de sélection des caractéristiques sous forme d'un problème de sélection de noyaux. Basé sur l'apprentissage de noyaux multiples (MKL), les tâches de sélection et de fusion de caractéristiques sont considérées simultanément. Les expériences effectuées montrent que la plateforme de classification proposée est à la fois plus simple, plus générique et souvent plus performante que les autres approches de la littérature. Dans le but d'approfondir notre étude sur l'apprentissage de noyaux multiples, nous définissons un nouveau formalisme d'apprentissage MKL réalisé en deux étapes. Cette contribution consiste à proposer trois termes régularisant liés à la résolution du problème d'apprentissage des poids associés à une combinaison linéaire de noyaux, problème reformulé en un problème de classification à vaste marge dans l'espace des couples. Le premier terme régularisant proposé assure une sélection parcimonieuse des noyaux. Les deux autres termes ont été conçus afin de tenir compte de la similarité entre les noyaux via une métrique basée sur la corrélation. Les différentes expérimentations réalisées montrent que le formalisme proposé permet d'obtenir des résultats de même ordre que les méthodes de référence, mais offrant l'avantage d'utiliser moins de fonctions noyaux. / This thesis arises in the context of computer aided analysis for subcellular protein localization in microscopic images. The aim is the establishment of an automatic classification system allowing to identify the cellular compartment in which a protein of interest exerts its biological activity. In order to overcome the difficulties in attempting to discern the cellular compartments in microscopic images, the existing state-of-art systems use several descriptors to train an ensemble of classifiers. In this thesis, we propose a different classification scheme wich better cope with the requirement of genericity and flexibility to treat various image datasets. Aiming to provide an efficient image characterization of microscopic images, a new feature system combining local, frequency-domain, global, and region-based features is proposed. Then, we formulate the problem of heterogeneous feature fusion as a kernel selection problem. Using multiple kernel learning, the problems of optimal feature sets selection and classifier training are simultaneously resolved. The proposed combination scheme leads to a simple and a generic framework capable of providing a high performance for microscopy image classification. Extensive experiments were carried out using widely-used and best known datasets. When compared with the state-of-the-art systems, our framework is more generic and outperforms other classification systems. To further expand our study on multiple kernel learning, we introduce a new formalism for learning with multiple kernels performed in two steps. This contribution consists in proposing three regularized terms with in the minimization of kernels weights problem, formulated as a classification problem using Separators with Vast Margin on the space of pairs of data. The first term ensures that kernels selection leads to a sparse representation. While the second and the third terms introduce the concept of kernels similarity by using a correlation measure. Experiments on various biomedical image datasets show a promising performance of our method compared to states of art methods.

<div> <p>People spend most of their time indoors. Because people’s health and productivity are highly dependent on the quality of the indoor thermal environment, it is important to provide occupants with healthy, comfortable and productive indoor thermal environment. However, inappropriate thermostat temperature setpoint settings not only wasted large amount of energy but also make occupants less comfortable. This study intended to develop a new control strategy for HVAC systems to adjust the thermostat setpoint automatically and accordingly to provide a more comfortable and satisfactory thermal environment.</p> <p>This study first trained an image classification model based on CNN to classify occupants’ amount of clothing insulation (clothing level). Because clothing level was related to human thermal comfort, having this information was helpful when determining the temperature setpoint. By using this method, this study performed experimental study to collect comfortable air temperature for different clothing levels. This study collected 450 data points from college student. By using the data points, this study developed an empirical curve which could be used to calculate comfortable air temperature for specific clothing level. The results obtained by using this curve could provide environments that had small average dissatisfaction and average thermal sensation closed to neutral.</p> <p>To adjust the setpoint temperature according to occupants’ thermal comfort, this study used mean facial skin temperature as an indicator to determine the thermal comfort. Because when human feel hot, their body temperature would rise and vice versa. To determine the correlation, we used a long wave infrared (LWIR) camera to non-invasively obtain occupant’s facial thermal map. By processing the thermal map with Haar-cascade face detection program, occupant’s mean facial skin temperature was calculated. By using this method, this study performed experimental study to collect occupant’s mean facial skin temperature under different thermal environment. This study collected 225 data points from college students. By using the data points, this study discovered different intervals of mean facial skin temperature under different thermal environment. </p> <p>Lastly, this study used the data collected from previous two investigations and developed a control platform as well as the control logic for a single occupant office to achieve the objective. The measured clothing level using image classification was used to determine the temperature setpoint. According to the measured mean facial skin temperature, the setpoint could be further adjusted automatically to make occupant more comfortable. This study performed 22 test sessions to validate the new control strategy. The results showed 91% of the tested subjects felt neutral in the office</p> </div> <br>

<div>Maize (Zea mays L.) is one of the most important crops worldwide for its critical importance in agriculture, economic stability, and food security. Many agricultural research and commercial breeding programs target the efficiency of this crop, seeking to increase productivity with fewer inputs and becoming more environmentally sustainable and resistant to impacts of climate and other external factors. For the purpose of analyzing the performance of the new varieties and management strategies, accurate and constant monitoring is crucial and yet, still performed mostly manually, becoming labor-intensive, time-consuming, and costly.<br></div><div>Flowering is one of the most important stages for maize, and many other grain crops, requiring close attention during this period. Any physical or biological negative impact in the tassel, as a reproductive organ, can have significant consequences to the overall grain development, resulting in production losses. Remote sensing observation technologies are currently seeking to close the gap in phenotyping in monitoring the development of the plants’ geometric structure and chemistry-related responses over the growth and reproductive cycle.</div><div>For this thesis, remotely sensed hyperspectral imagery were collected, processed and, explored to detect tassels in maize crops. The data were acquired in both a controlled facility using an imaging conveyor, and from the fields using a PhenoRover (wheel-based platform) and a low altitude UAV. Two pixel-based classification experiments were performed on the original hyperspectral imagery (HSI) using Spectral Angle Mapper (SAM) and Support Vector Machine (SVM) supervised classifiers. Feature reduction methods, including Principal Component Analysis (PCA), Locally Linear Embedding (LLE), and Isometric Feature Mapping (Isomap) were also investigated, both to identify features for annotating the reference data and in conjunction with classification.</div><div>Collecting the data from different systems allowed the identification of strengths and weaknesses for each system and the associated tradeoffs. The controlled facility allowed stable lighting and very high spatial and spectral resolution, although it lacks on supplying information about the plants’ interactions in field conditions. Contrarily, the in-field data from the PhenoRover </div><div>and the UAV exposed the complications related to the plant’s density within the plots and the variability in the lighting conditions due to long times of data collection required. The experiments implemented in this study successfully classified pixels as tassels for all images, performing better with higher spatial resolution and in the controlled environment. For the SAM experiment, nonlinear feature extraction via Isomap was necessary to achieve good results, although at a significant computational expense. Dimension reduction did not improve results for the SVM classifier.</div>

Le Transfert de Connaissance (Knowledge Transfer or Transfer Learning) est une solution prometteuse au difficile problème de l’apprentissage des réseaux profonds au moyen de bases d’apprentissage de petite taille, en présence d’une grande variabilité visuelle intra-classe. Dans ce travail, nous reprenons ce paradigme, dans le but d’étendre les capacités des CNN les plus récents au problème de la classification. Dans un premier temps, nous proposons plusieurs techniques permettant, lors de l’apprentissage et de la prédiction, une réduction des ressources nécessaires – une limitation connue des CNN. (i) En utilisant une méthode hybride combinant des techniques classiques comme des Bag-Of-Words (BoW) avec des CNN. (iv) En introduisant une nouvelle méthode d’agrégation intégrée à une structure de type CNN ainsi qu’un modèle non-linéaire s’appuyant sur des parties de l’image. La contribution clé est, finalement, une technique capable d’isoler les régions des images utiles pour une représentation locale. De plus, nous proposons une méthode nouvelle pour apprendre une représentation structurée des coefficients des réseaux de neurones. Nous présentons des résultats sur des jeux de données difficiles, ainsi que des comparaisons avec des méthodes concurrentes récentes. Nous prouvons que les méthodes proposées s’étendent à d’autres tâches de reconnaissance visuelles comme la classification d’objets, de scènes ou d’actions. / Knowledge transfer is a promising solution for the difficult problem of training deep convolutional neural nets (CNNs) using only small size training datasets with a high intra-class visual variability. In this thesis work, we explore this paradigm to extend the ability of state-of-the-art CNNs for image classification.First, we propose several effective techniques to reduce the training and test-time computational burden associated to CNNs:(i) Using a hybrid method to combine conventional, unsupervised aggregators such as Bag-of-Words (BoW) with CNNs;(ii) Introducing a novel pooling methods within a CNN framework along with non-linear part-based models. The key contribution lies in a technique able to discover useful regions per image involved in the pooling of local representations;In addition, we also propose a novel method to learn the structure of weights in deep neural networks. Experiments are run on challenging datasets with comparisons against state-of-the-art methods. The methods proposed are shown to generalize to different visual recognition tasks, such as object, scene or action classification.

The early discovery of gastrointestinal (GI) disorders can significantly decrease the fatality rate of severe afflictions. Video capsule endoscopy (VCE) is a technique that produces an eight hour long recording of the GI tract that needs to be manually reviewed. This has led to the demand for AI-based solutions, but unfortunately, the lack of labeled data has been a major obstacle. In 2020 the Kvasir-Capsule dataset was produced which is the largest labeled dataset of GI abnormalities to date, but challenges still exist.The dataset suffers from unbalanced and very similar data created from labeled video frames. To avoid specialization to the specific data the creators of the set constructed an official split which is encouraged to use for testing. This study evaluates the use of transfer learning, Data augmentation and binary classification to detect GI abnormalities. The performance of machine learning (ML) classification is explored, with and without official split-based testing. For the performance evaluation, a specific focus will be on achieving a low rate of false negatives. The proposition behind this is that the most important aspect of an automated detection system for GI abnormalities is a low miss rate of possible lethal abnormalities. The results from the controlled experiments conducted in this study clearly show the importance of using official split-based testing. The difference in performance between a model trained and tested on the same set and a model that uses official split-based testing is significant. This enforces that without the use of official split-based testing the model will not produce reliable and generalizable results. When using official split-based testing the performance is improved compared to the initial baseline that is presented with the Kvasir-Capsule set. Some experiments in the study produced results with as low as a 1.56% rate of false negatives but with the cost of lowered performance for the normal class.

Micronic Mydata AB utvecklar och tillverkar maskiner för att automatisk montera elektroniska komponenter på kretskort, s.k. ”Pick and place” (PnP) maskiner. Komponenterna blir lokaliserade och inspekterade optiskt innan de monteras på kretskorten, för att säkerhetsställa att de monteras korrekt och inte är skadade. En komponent kan t.ex. plockas på sidan, vertikalt eller missas helt. Det nuvarande systemet räknar ut uppmätta parametrar så som: längd, bredd och kontrast.Projektet syftar till att undersöka olika maskininlärningsmetoder för att klassificera felaktiga plock som kan uppstå i maskinen. Vidare skall metoderna minska antalet defekta komponenter som monteras samt minska antalet komponenter som felaktigt avvisas. Till förfogande finns en databas innehållande manuellt klassificerade komponenter och tillhörande uppmätta parametrar och bilder. Detta kan användas som träningsdata för de maskininlärningsmetoder som undersöks och testas. Projektet skall även undersöka hur dessa maskininlärningsmetoder lämpar sig allmänt i mekatroniska produkter, med hänsyn till problem så som realtidsbegräsningar.Fyra olika maskininlärningsmetoder har blivit utvärderade och testade. Metoderna har blivit utvärderade för ett test set där den nuvarande metoden presterar mycket bra. Dels har de nuvarande parametrarna använts, samt en alternativ metod som extraherar parametrar (s.k. SIFT descriptor) från bilderna. De nuvarande parametrarna kan användas tillsammans med en SVM eller ett ANN och uppnå resultat som reducerar defekta och monterade komponenter med upp till 64 %. Detta innebär att dessa fel kan reduceras utan att uppgradera de nuvarande bildbehandlingsalgoritmerna. Genom att använda SIFT descriptor tillsammans med ett ANN eller en SVM kan de vanligare felen som uppstår klassificeras med en noggrannhet upp till ca 97 %. Detta överstiger kraftigt de resultat som uppnåtts när de nuvarande parametrarna har använts. / Micronic Mydata AB develops and builds machines for mounting electronic component onto PCBs, i.e. Pick and Place (PnP) machines. Before being mounted the components are localized and inspected optically, to ensure that the components are intact and picked correctly. Some of the errors which may occur are; the component is picked sideways, vertically or not picked at all. The current vision system computes parameter such as: length, width and contrast.The project strives to investigate and test machine learning approaches which enable automatic error classification. Additionally the approaches should reduce the number of defect components which are mounted, as well as reducing the number of components which are falsely rejected. At disposal is a large database containing the calculated parameters and images of manually classified components. This can be used as training data for the machine learning approaches. The project also strives to investigate how machine learning approaches can be implemented in mechatronic systems, and how limitations such as real-time constraints could affect the feasibility.Four machine learning approaches have been evaluated and verified against a test set where the current implementation performs very well. The currently calculated parameters have been used as inputs, as well as a new approach which extracts (so called SIFT descriptor) parameters from the raw images. The current parameters can be used with an ANN or a SVM and achieve results which reduce the number of poorly mounted components by up to 64 %. Hence, these defects can be decreased without updating the current vision algorithms. By using SIFT descriptors and an ANN or a SVM the more common classes can be classified with accuracies up to approximately 97 %. This greatly exceeds results achieved when using the currently computed parameters.

<div>Every year the most effective Deep learning models, CNN architectures are showcased based on their compatibility and performance on the embedded edge hardware, especially for applications like image classification. These deep learning models necessitate a significant amount of computation and memory, so they can only be used on high-performance computing systems like CPUs or GPUs. However, they often struggle to fulfill portable specifications due to resource, energy, and real-time constraints. Hardware accelerators have recently been designed to provide the computational resources that AI and machine learning tools need. These edge accelerators have high-performance hardware which helps maintain the precision needed to accomplish this mission. Furthermore, this classification dilemma that investigates channel interdependencies using either depth-wise or group-wise convolutional features, has benefited from the inclusion of Bottleneck modules. Because of its increasing use in portable applications, the classic inverted residual block, a well-known architecture technique, has gotten more recognition. This work takes it a step forward by introducing a design method for porting CNNs to low-resource embedded systems, essentially bridging the difference between deep learning models and embedded edge systems. To achieve these goals, we use closer computing strategies to reduce the computer's computational load and memory usage while retaining excellent deployment efficiency. This thesis work introduces HBONext, a mutated version of Harmonious Bottlenecks (DHbneck) combined with a Flipped version of Inverted Residual (FIR), which outperforms the current HBONet architecture in terms of accuracy and model size miniaturization. Unlike the current definition of inverted residual, this FIR block performs identity mapping and spatial transformation at its higher dimensions. The HBO solution, on the other hand, focuses on two orthogonal dimensions: spatial (H/W) contraction-expansion and later channel (C) expansion-contraction, which are both organized in a bilaterally symmetric manner. HBONext is one of those versions that was designed specifically for embedded and mobile applications. In this research work, we also show how to use NXP Bluebox 2.0 to build a real-time HBONext image classifier. The integration of the model into this hardware has been a big hit owing to the limited model size of 3 MB. The model was trained and validated using CIFAR10 dataset, which performed exceptionally well due to its smaller size and higher accuracy. The validation accuracy of the baseline HBONet architecture is 80.97%, and the model is 22 MB in size. The proposed architecture HBONext variants, on the other hand, gave a higher validation accuracy of 89.70% and a model size of 3.00 MB measured using the number of parameters. The performance metrics of HBONext architecture and its various variants are compared in the following chapters.</div>

Today's smartphones are capable of accomplishing far more advanced tasks than reading emails. With the modern framework TensorFlow, visual object recognition becomes possible using smartphone resources. This thesis shows that the main challenge does not lie in developing an artifact which performs visual object recognition. Instead, the main challenge lies in developing an ecosystem which allows for continuous improvement of the system’s ability to accomplish the given task without laborious and inefficient data collection. This thesis presents four design principles which contribute to an efficient ecosystem with quick initiation of new object classes and efficient data collection which is used to continuously improve the system’s ability to recognize smart meters in varying environments in an automated fashion.

This doctoral thesis deals with the issue of solving massive tasks in GIS. These tasks process large volumes of geographic data with different formats. The thesis describes a theoretical analysis of the complexity of tasks and the possibilities to optimize sub-processes which lead to an acceptable solution. It considers the possibility of using parallelism in GIS, which leads to an acceleration in the processing of large volumes of geographic data. It also proposes a method for the optimization of processes through an algorithm which determines the number of means necessary for the successful solution of a task at a specified time and assigns processes to these means. Additionally, there is a proposed algorithm for the optimization of the preparation of data for extensive GIS projects. The algorithms have been validated by the results of a research project, the aim of which was to analyse the terrain surface above a gas line in the Czech Republic. The primary method of analysis was the classification of an orthophoto image, which was further refined through filtration using the ZABAGED layers. Therefore, the thesis deals with the possibility of improving the results of image classification using GIS instruments as well as dealing with the determination of the error rate in analysis results. The results of the analysis are now used for the strategic planning of maintenance and the development of gas facilities in the Czech Republic. The results of the work have general importance regarding the performance of other operations of the same class in GIS.

<p>    </p> <p>Mutual learning algorithm is a machine learning algorithm where multiple machine learning algorithms learns from different sources and then share their knowledge among themselves so that all the agents can improve their classification and prediction accuracies simultaneously. Mutual learning algorithm can be an efficient mechanism for improving the machine learning and neural network efficiency in a multi-agent system. Usually, in knowledge distillation algorithms, a big network plays the role of a static teacher and passes the data to smaller networks, known as student networks, to improve the efficiency of the latter. In this thesis, it is showed that two small networks can dynamically and interchangeably play the changing roles of teacher and student to share their knowledge and hence, the efficiency of both the networks improve simultaneously. This type of dynamic learning mechanism can be very useful in mobile environment where there is resource constraint for training with big dataset. Data exchange in multi agent, teacher-student network system can lead to efficient learning.  </p>