Background. Since historical handwritten documents have played important roles in promoting the development of human civilization, many of them have been preserved through digital versions for more scientific researches. However, various degradations always exist in these documents, which could interfere in normal reading. But, binarized versions can keep meaningful contents without degradations from original document images. Document image binarization always works as a pre-processing step before complex document analysis and recognition. It aims to extract texts from a document image. A desirable binarization performance can promote subsequent processing steps positively. For getting better performance for document image binarization, efficient binarization methods are needed. In recent years, machine learning centered on deep learning has gathered substantial attention in document image binarization, for example, Convolutional Neural Networks (CNNs) are widely applied in document image binarization because of the powerful ability of feature extraction and classification. Meanwhile, Support Vector Machine (SVM) is also used in image binarization. Its objective is to build an optimal hyperplane that could maximize the margin between negative samples and positive samples, which can separate the foreground pixels and the background pixels of the image distinctly. Objectives. This thesis aims to explore how the CNN based process of deep convolutional feature extraction and an SVM classifier can be integrated well to binarize handwritten document images, and how the results are, compared with some state-of-the-art document binarization methods. Methods. To investigate the effect of the proposed method on document image binarization, it is implemented and trained. In the architecture, CNN is used to extract features from input images, afterwards these features are fed into SVM for classification. The model is trained and tested with six different datasets. Then, there is a performance comparison between the proposed model and other binarization methods, including some state-of-the-art methods on other three different datasets. Results. The performance results indicate that the proposed model not only can work well but also perform better than some other novel handwritten document binarization method. Especially, evaluation of the results on DIBCO 2013 dataset indicates that our method fully outperforms other chosen binarization methods on all the four evaluation metrics. Besides, it also has the ability to deal with some degradations, which demonstrates its generalization and learning ability are excellent. When a new kind of degradation appears, the proposed method can address it properly even though it never appears in the training datasets. Conclusions. This thesis concludes that the CNN based component and SVM can be combined together for handwritten document binarization. Additionally, in certain datasets, it outperforms some other state-of-the-art binarization methods. Meanwhile, its generalization and learning ability is outstanding when dealing with some degradations.

This study investigates synthetic schematic generation using conditional generative adversarial networks, specifically the Pix2Pix algorithm was implemented for the experimental phase of the study. With the increase in deep neural network’s capabilities and availability, there is a demand for verbose datasets. This in combination with increased privacy concerns, has led to synthetic data generation utilization. Analysis of synthetic images was completed using a survey. Blueprint images were generated and were successful in passing as genuine images with an accuracy of 40%. This study confirms the ability of generative neural networks ability to produce synthetic blueprint images.

The increasing energy consumption is one of the greatest environmental challenges of our time. Residential buildings account for a considerable part of the total electricity consumption and is further a sector that is shown to have large savings potential. Non Intrusive Load Monitoring (NILM), i.e. the deduction of the electricity consumption of individual home appliances from the total electricity consumption of a household, is a compelling approach to deliver appliance specific consumption feedback to consumers. This enables informed choices and can promote sustainable and cost saving actions. To achieve this, accurate and reliable appliance load disaggregation algorithms must be developed. This Master's thesis proposes a novel approach to tackle the disaggregation problem inspired by state of the art algorithms in the field of speech recognition. Previous approaches, for sampling frequencies <img src="http://www.diva-portal.org/cgi-bin/mimetex.cgi?%5Cleq" />1 Hz, have primarily focused on different types of hidden Markov models (HMMs) and occasionally the use of artificial neural networks (ANNs). HMMs are a natural representation of electric appliances, however with a purely generative approach to disaggregation, basically all appliances have to be modelled simultaneously. Due to the large number of possible appliances and variations between households, this is a major challenge. It imposes strong restrictions on the complexity, and thus the expressiveness, of the respective appliance model to make inference algorithms feasible. In this thesis, disaggregation is treated as a factorisation problem where the respective appliance signal has to be extracted from its background. A hybrid model is proposed, where a convolutional neural network (CNN) extracts features that correlate with the state of a single appliance and the features are used as observations for a hidden semi Markov model (HSMM) of the appliance. Since this allows for modelling of a single appliance, it becomes computationally feasible to use a more expressive Markov model. As proof of concept, the hybrid model is evaluated on 238 days of 1 Hz power data, collected from six households, to predict the power usage of the households' washing machine. The hybrid model is shown to perform considerably better than a CNN alone and it is further demonstrated how a significant increase in performance is achieved by including transitional features in the HSMM. / Den ökande energikonsumtionen är en stor utmaning för en hållbar utveckling. Bostäder står för en stor del av vår totala elförbrukning och är en sektor där det påvisats stor potential för besparingar. Non Intrusive Load Monitoring (NILM), dvs. härledning av hushållsapparaters individuella elförbrukning utifrån ett hushålls totala elförbrukning, är en tilltalande metod för att fortlöpande ge detaljerad information om elförbrukningen till hushåll. Detta utgör ett underlag för medvetna beslut och kan bidraga med incitament för hushåll att minska sin miljöpåverakan och sina elkostnader. För att åstadkomma detta måste precisa och tillförlitliga algoritmer för el-disaggregering utvecklas. Denna masteruppsats föreslår ett nytt angreppssätt till el-disaggregeringsproblemet, inspirerat av ledande metoder inom taligenkänning. Tidigare angreppsätt inom NILM (i frekvensområdet <img src="http://www.diva-portal.org/cgi-bin/mimetex.cgi?%5Cleq" />1 Hz) har huvudsakligen fokuserat på olika typer av Markovmodeller (HMM) och enstaka förekomster av artificiella neurala nätverk. En HMM är en naturlig representation av en elapparat, men med uteslutande generativ modellering måste alla apparater modelleras samtidigt. Det stora antalet möjliga apparater och den stora variationen i sammansättningen av dessa mellan olika hushåll utgör en stor utmaning för sådana metoder. Det medför en stark begränsning av komplexiteten och detaljnivån i modellen av respektive apparat, för att de algoritmer som används vid prediktion ska vara beräkningsmässigt möjliga. I denna uppsats behandlas el-disaggregering som ett faktoriseringsproblem, där respektive apparat ska separeras från bakgrunden av andra apparater. För att göra detta föreslås en hybridmodell där ett neuralt nätverk extraherar information som korrelerar med sannolikheten för att den avsedda apparaten är i olika tillstånd. Denna information används som obervationssekvens för en semi-Markovmodell (HSMM). Då detta utförs för en enskild apparat blir det beräkningsmässigt möjligt att använda en mer detaljerad modell av apparaten. Den föreslagna Hybridmodellen utvärderas för uppgiften att avgöra när tvättmaskinen används för totalt 238 dagar av elförbrukningsmätningar från sex olika hushåll. Hybridmodellen presterar betydligt bättre än enbart ett neuralt nätverk, vidare påvisas att prestandan förbättras ytterligare genom att introducera tillstånds-övergång-observationer i HSMM:en.

This thesis investigates the use of Convolutional Neural Networks (CNNs) toperform semantic segmentation of feet during endovascular surgery in patientswith Critical Limb Threatening Ischemia (CLTI). It is currently being investigatedwhether objective assessment of perfusion can aid surgeons during endovascularsurgery. By segmenting feet, it is possible to perform automatic analysis of perfusion data which could give information about the impact of the surgery in specificRegions of Interest (ROIs). The CNN was developed in Python with a U-net architecture which has shownto be state of the art when it comes to medical image segmentation. An imageset containing approximately 78 000 images of feet and their ground truth segmentation was manually created from 11 videos taken during surgery, and onevideo taken on three healthy test subjects. All videos were captured with a MultiExposure Laser Speckle Contrast Imaging (MELSCI) camera developed by Hultman et al. [1]. The best performing CNN was an ensemble model consisting of10 sub-models, each trained with different sets of training data. An ROI tracking algorithm was developed based on the Unet output, by takingadvantage of the simplicity of edge detection in binary images. The algorithmconverts images into point clouds and calculates a transformation between twopoint clouds with the use of the Iterative Closest Point (ICP) algorithm. The resultis a system that perform automatic tracking of manually selected ROIs whichenables continuous measurement of perfusion in the ROIs during endovascularsurgery.

Having enough data about the usage of tyre types on the road can lead to a better understanding of the consequences of studded tyres on the environment. This paper is focused on training and testing a machine learning model which can be further integrated into a larger system for automation of the data collection process. Different machine learning algorithms, namely CNN, SVM, and Random Forest, were compared in this experiment. The method used in this paper is an empirical method. First, sound data for studded and none-studded tyres was collected from three different locations in the city of Gävle/Sweden. A total of 760 Mel spectrograms from both classes was generated to train and test a well-known CNN model (AlexNet) on MATLAB. Sound features for both classes were extracted using JAudio to train and test models that use SVM and Random Forest classifi-ers on Weka. Unnecessary features were removed one by one from the list of features to improve the performance of the classifiers. The result shows that CNN achieved accuracy of 84%, SVM has the best performance both with and without removing some audio features (i.e 94% and 92%, respectively), while Random Forest has 89 % accuracy. The test data is comprised of 51% of the studded class and 49% of the none-studded class and the result of the SVM model has achieved more than 94 %. Therefore, it can be considered as an acceptable result that can be used in practice.

No description available.

Modern handheld devices rely on specialized hardware for evaluating machine learning algorithms. This thesis investigates the feasibility of using the digital signal processor, a part of the modem of the device, as an alternative to this specialized hardware. Memory management techniques and implementations for evaluating the machine learning primitives convolutional, max-pooling and fully connected layers are proposed. The implementations are evaluated based on to what degree they utilize available hardware units. New instructions for packing data and facilitating instruction pipelining are suggested and evaluated. The results show that convolutional and fully connected layers are well-suited to the processor used. The aptness of the convolutional layer is subject to the kernel being applied with a stride of 1 as larger strides cause the hardware usage to plummet. Max-pooling layers, while not ill-suited, are the most limited in terms of hardware usage. The proposed instructions are shown to have positive effects on the throughput of the implementations.

Indiana University-Purdue University Indianapolis (IUPUI) / The Convolutional Neural Network (CNN) is a mathematical model designed to distill input information into a more useful representation. This distillation process removes information over time through a series of dimensionality reductions, which ultimately, grant the model the ability to resist noise, and generalize effectively. However, CNNs often contain elements that are ineffective at contributing towards useful representations. This Thesis aims at providing a remedy for this problem by introducing Absolute Cosine Value Regularization (ACVR). This is a regularization technique hypothesized to increase the representational power of CNNs by using a Gradient Descent Orthogonalization algorithm to force the vectors that constitute their filters at any given convolutional layer to occupy unique positions in in their respective spaces. This method should in theory, lead to a more effective balance between information loss and representational power, ultimately, increasing network performance. The following Thesis proposes and examines the mathematics and intuition behind ACVR, and goes on to propose Dynamic-ACVR (D-ACVR). This Thesis also proposes and examines the effects of ACVR on the filters of a low-dimensional CNN, as well as the effects of ACVR and D-ACVR on traditional Convolutional filters in VGG-19. Finally, this Thesis proposes and examines regularization of the Pointwise filters in MobileNetv1.

Mapping of earth resources, environmental monitoring, and many other systems require high-resolution wide-area imaging. Since images often have to be captured at night or in inclement weather conditions, a capability is provided by Synthetic Aperture Radar (SAR). SAR systems exploit radar signal's long-range propagation and utilize digital electronics to process complex information, all of which enables high-resolution imagery. This gives SAR systems advantages over optical imaging systems, since, unlike optical imaging, SAR is effective at any time of day and in any weather conditions. Moreover, advanced technology called Interferometric Synthetic Aperture Radar (InSAR), has the potential to apply phase information from SAR images and to measure ground surface deformation. However, given the current state of technology, the quality of InSAR data can be distorted by several factors, such as image co-registration, interferogram generation, phase unwrapping, and geocoding. Image co-registration aligns two or more images so that the same pixel in each image corresponds to the same point of the target scene. Super-Resolution (SR), on the other hand, is the process of generating high-resolution (HR) images from a low-resolution (LR) one. SR influences the co-registration quality and therefore could potentially be used to enhance later stages of SAR image processing. Our research resulted in two major contributions towards the enhancement of SAR processing. The first one is a new learning-based SR model that can be applied with SAR, and similar applications. A second major contribution is utilizing the devised model for improving SAR co-registration and InSAR interferogram generation, together with methods for evaluating the quality of the resulting images. In the case of phase unwrapping, the process of recovering unambiguous phase values from a two-dimensional array of phase values known only modulo $2\pi$ rad, our research produced a third major contribution. This third major contribution is the finding that quantum annealers can resolve problems associated with phase unwrapping. Even though other potential solutions to this problem do currently exist - based on network programming for example - network programming techniques do not scale well to larger images. We were able to formulate the phase unwrapping problem as a quadratic unconstrained binary optimization (QUBO) problem, which can be solved using a quantum annealer. Since quantum annealers are limited in the number of qubits they can process, currently available quantum annealers do not have the capacity to process large SAR images. To resolve this limitation, we developed a novel method of recursively partitioning the image, then recursively unwrapping each partition, until the whole image becomes unwrapped. We tested our new approach with various software-based QUBO solvers and various images, both synthetic and real. We also experimented with a D-Wave Systems quantum annealer, the first and only commercial supplier of quantum annealers, and we developed an embedding method to map the problem to the D-Wave 2000Q_6, which improved the result images significantly. With our method, we were able to achieve high-quality solutions, comparable to state-of-the-art phase-unwrapping solvers. / Graduate

Diploma thesis "Classification of UAV hyperspectral images using deep learning methods" focuses on the classification methods, namely convolutional neural networks (CNN), of hyperspectral (HS) images. Based on a thorough literature review, a comprehensive overview on CNN utilisation in remote sensing is assembled as a basis for identifying suitable methods for the specific task of this thesis. Two methods with an open solution in programming language Python were selected - Capsule Network and U-Net. The main aim of this work is to verify the suitability of chosen methods for the classification of hyperspestral images. The images were acquired by sensors with high spatial resolution carried by a UAV over Krkonoše Mts. tundra. Important step was to prepare input HS data (54 bands, 9cm) to have suitable form for entering the network. Not all the required results were achieved due to the complexity of the Capsule Network architecture. The U-Net method was used in purpose of comparing and verifying the results. Accuracies retrieved from the U-Net overcome results achieved by traditionally used machine learning methods (SVM, ML, RF, etc). Overall accuracy for U-Net was higher than 90% where other mentioned methods did not get over 88%. Especially classes block fields and dwarf pine achieved higher...