Elektrik- Elektronik Mühendisliği

Permanent URI for this community

https://digitalarchive.library.bogazici.edu.tr/handle/123456789/11605

Browse

Now showing 1 - 13 of 13

2D to 3D video conversion
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2016., 2016.) Çoban Aydın, Aysun.; Acar, Burak.
Stereoscopic 3D visualisation is increasingly embedded into social life through the use of commercially available 3D-TV sets. In this work, a hybrid approach for 2D to 3D conversion is presented to produce stereoscopic 3D video automatically from 2D mono video frames. Each frame is synthesized to stereo pairs. Disparity/depth information required for 3D view is extracted from mono frame sequences based on motion and geometrical cues. Depth estimation of the scene is considered separately for background and foreground. Background geometry of the scene is determined by using geometrical cues such as vanishing point and straight lines in the image. According to this geometry, relevant information on the background depth eld of a single image is estimated to generate a canonic disparity map of the background. For foreground depth estimation, on the other hand, two approaches are presented. First approach is based on detection of moving foreground objects. A depth value is assigned to each object based on its corresponding location in the background depth map. In the second approach, background registration is applied for consecutive frames that are captured by a moving camera. By this method, disparity in foreground regions is distinguished from background disparity that leads to a distinctive 3D e ect on foreground regions. Consequently, depth/disparity information of foreground regions is combined with background canonic disparity map. According to these nal disparity maps, pixels of the original frames are shifted to generate virtual frames to enable 3D views. This work is accompanied by a subjective evaluation on the basis of user test which compare our 3D results with commercially available 3D-TV sets.
Assessing DT-MRI tractography results via sampling the fiber tract space
(Thesis (M.S.)-Bogazici University. Institute for Graduate Studies in Science and Engineering, 2011., 2011.) Yoldemir, Ahmet Burak.; Acar, Burak.
Complex neural processes in human brain are realized through a huge number of connections between neural cells. White matter tractography is the only available tool to reconstruct these anatomical connectivities non-invasively and in vivo. Following the emergence of di usion imaging, several tractography algorithms have been proposed, where the local direction of white matter ber bundles is estimated from measurements of water di usion in human brain. The goal of this thesis is to introduce a generic tractography assessment and improvement method for di usion tensor imaging (DTI) data. The proposed method takes a set of ber tracts that are generated with any tractography algorithm as the input, and allow the user to interactively assess tractography results by identifying the erroneous or inde nite regions in the DTI data along input tracts and highlighting possible branching patterns of ber bundles. By introducing alternative pathways that might have been missed by the initial tractography, given tractography results can also be improved. The technique relies on splitting the input tracts into shorter segments to prevent error accumulation, followed by sampling from the space of short tract clusters to estimate the connectivities between these short ber segments. After the connectivity values are computed, given a set of seed tracts and a connectivity threshold, the method displays the short tracts that are connected to the seed tracts with a probability higher than the given threshold in an interactive environment. Thus, the possible pathways can be investigated as a function of the connectivity threshold, highlighting the uncertainty in DTI data.
Automated electrical motor quality control via machine learning based vibration analysis
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2021., 2021.) Şentürk, Sibel.; Acar, Burak.
Industry 4.0 aims at the digital transformation to increase the reliability and capacity of production. With the integration of Sensor Analytics (SA) and Artificial Intelligence (AI) to manufacturing, the design of automated and optimized processes becomes more accessible. One of the areas where AI tools and SA are used is quality control tests of products. The main target of this thesis is to automate the quality control step based on vibration analysis by finding mechanical failures of Brushless Direct Current (BLDC) motors as an example of an AI-powered sensor analytics application. In addition, the feasibility and assessment of popular machine learning models are investigated. Two architectures are proposed to classify motors' quality. These methods are called sAIQC, Single-Stage AI-Powered Quality Control, and dAIQC, Double-Stage AI-Powered Quality Control. In the first method, motors are classified into healthy (pass) or faulty (fail), regardless of the data quality of the signal. The second proposed method is composed of two stages. The first stage makes a binary classification based on data quality, and then, the separated groups are classified at two independent classifiers in the second stage as pass or fail. Unweighted accuracy (UA), defined as the average accuracy of each class, is used as a performance metric of the classifiers. In experiments with the dataset containing 671 samples, the performance of sAIQC method was 84.9%; this performance with the dAIQC method was increased to 92.9\%. Furthermore, in experiments using big data set consisting of 25580 vibration recordings and without a data quality label, the performance of the SAIQC method is 73.5% percent. In contrast, the performance of the dAIQC method is 89.5\% percent.
Content based medical image retrieval
(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2017., 2017.) Marvasti, Neda Barzegar.; Acar, Burak.
Fast technological developments of di erent medical imaging and data collection techniques increase the expectation of more accurate interpretations and diagnoses of radiologists. However, to carefully analyze the resulting big medical data, reliable and fast systems are needed. Content-based medical image retrieval (CBMIR) is a valuable technique to assist radiologists by identifying similar images in a large archive. However, due to the huge semantic gap between low-level image features and high-level semantic features, the challenge of retrieving similar images utilizing the high-level user speci ed semantic labels, which are closer to the users understandings and interpretations, has attracted great interest from various researches. In this dissertation, an iterative search and retrieval scheme to identify similar images from a database of 3-dimensional liver computed tomography (CT) images is proposed via utilizing the combination of lesion and liver related semantic features and patients' metadata. At each retrieval iteration, the lesion related concepts are annotated in a speci c order through a proposed computer aided medical image annotation (CMIA) scheme. The proposed radiologist-in-the-loop semi-automatic CMIA is based on a Bayesian tree structured model, linked to RadLex, to exploit the inter-dependencies between concepts to update the full annotation process and to guide the radiologist to input the most critical information at each iteration. Results show the e ectiveness of this modelbased interactive annotation scheme compared to the domain-blind models, as well as its advantage in the performance of the retrieval system, where a few number of manual annotations can signi cantly boost the retrieval accuracy. Moreover, better retrieval performance is achieved by incorporating a small contribution of the non-lesion data.
Evaluation of 2D local image descriptors and feature encoding methods for depth image based object class recognition
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2014., 2014.) Kayım, Güney.; Acar, Burak.; Sankur, Bülent.
In this thesis, we have investigated the 3D object class recognition problem. We used an approach that solves this problem with the use of depth images obtained from 3D object models. In the approach we used, 3D object class recognition system is composed of two stages; training and testing. In both stages, rst, keypoints are detected from the images, and then 2D local image descriptors are built around these keypoints. This is continued by encoding local descriptors into a single descriptor. Just before this step, in training stage, a codebook is learned, and it is used for encoding local descriptors in both stages. Another extra step in training stage is, after the descriptors are encoded, for each class a binary classi er is trained. Then, these classi ers are used in testing stage. We have evaluated di erent keypoint detection methods, 2D local image descriptors and encoding methods. Then, we experimentally show their superiorities and weaknesses over each other. Our experiments clearly show the best performing keypoint detection method, local image description method and feature encoding method in the depth image domain, which are densely sampled SIFT descriptors and Fisher Vector encoding. Using di erent experimental setups yields similar results, thus the validity of the methods that are selected as best is proven.
Liver segmentation in 3D CT data
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2011., 2011.) Çimen, Serkan.; Acar, Burak.
Segmentation of liver from 3D abdominal CT data is the basis of analysis of liver which is required to aid diagnosis and treatment of liver cancer. However, common clinical practice for liver segmentation relies on manual segmentation of CT images by the help of radiologist. Generally, this procedure is tedious and time-consuming. Therefore, fast and robust and accurate methods must be devised to automate liver segmentation process. There is a vast literature for automatic, semi-automatic and interactive liver segmentation methods based on various computer vision algorithms. Each of these methods possess some limitations due to highly varying structure of liver. In this thesis, we propose a semi-automatic liver segmentation algorithm based on an e ective combination of intensity distribution modeling, probabilistic atlases (PA) and graph cuts. Major contribution of this work is twofold. First of all, a novel PA construction methodology is proposed based on convex hulls of rough initial segmentation and reference manual delineations. Secondly, a new strategy to improve implicit gray level appearance models is proposed. In addition to that, we explain how to embed PA, gray level appearance models into graph cuts. The e ectiveness of proposed algorithm was demonstrated in clinical CT images. Evaluation scores show that proposed method provides results comparable with manual segmentation of a human who has adequate training in liver segmentation.
Machine heath monitoring for cyber-physical systems
(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2020., 2020.) Aydemir, Gürkan.; Acar, Burak.
Estimating the failure time of the machinery that are used in the production is crucial to achieve an e cient maintenance in Industry 4:0 era. Remaining useful life (RUL) is the term that refers to the length of time in terms of the raw time intervals or usage that a machine will continue to operate before it requires a repair or replacement. Machine learning (ML), especially deep learning, provides industry practitioners with e cient tools for estimating the RUL. However, ML is far from being fully utilized, since domain knowledge is generally ignored in current studies. This thesis focuses on three main domain speci c problems in machine condition monitoring to improve the performance of ML based RUL estimation. First, RUL is ill-de ned during the healthy operation period of the machinery, hence enforcing ML with respect to a ctitious true RUL during these periods adversely a ects the overall RUL estimation accuracy. In this thesis, a system level anomaly detection triggered RUL estimation method is proposed to detect degradation onset point in sensor data to prevent ML models to estimate RUL in this period, and hence to increase the accuracy. Secondly, the operating conditions of the machines a ect their degradation pattern and related sensor measurements. Thus, the accuracy of ML based RUL estimation models decreases when the machinery operate in varying conditions. A siamese neural network based operating conditioninvariant feature extraction method is introduced to alleviate this problem. These two approaches are veri ed using a benchmark turbofan engine degradation data. Lastly, most of the ML models su er from lack of data in RUL estimation. If the data are high dimensional such as image, pro le, etc., the problem becomes more challenging. Two deep learning architectures are proposed to resolve curse of dimensionality in case of degradation data scarcity. E ciency of the proposed models is demonstrated with an infrared image data.
Multi-modal tensor representations of brain networks
(Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2021., 2021.) Durusoy, Göktekin.; Acar, Burak.
Considering the economic, social and psychological burdens of Alzheimer’s disease (AD), the most common form of dementia, it is essential to gain insight into the process and underlying mechanisms of the disease. Using structural and functional brain connectomes obtained by in-vivo MRI techniques as biomarkers is a promising approach. In this thesis, the B-Tensor structure that allows the representation of brain connectomes defined in structurally and functionally with a uni-modal and multi-modal fashion is presented. With the projection of structural connectomes onto known func tional networks, patients with AD and healthy control group are distinguished in a 7-dimensional space with a separation performance of over 90%. In addition, with the uni-modal and multi-modal tensor factorization methods, 47 patients with different levels of AD, are diagnosed with an accuracy of 77% - 100% in a 5-dimensional space. The results show that the multi-modal factorization technique performs better than the uni-modal one by successfully fusing the structural and functional networks which offer complementary information. While the neurological evaluations of the obtained sub-networks are highly consistent with previous literature, new findings regarding the progression of the disease are also recommended.
Novetly detection on streaming sensor data for IIoT applications
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2019., 2019.) Bayram, Alper.; Acar, Burak.
Assessment of present and future condition of industrial machinery is one of the core ideas that constitute Industry 4.0 paradigm. Predictive maintenance depends on integrated sensors and machine learning algorithms to achieve this assessment based on the internal parameters of machinery. This type of maintenance could save plant costs and improve eﬃciency while reducing fatal defects in machinery. It automates the maintenance process and reduces the number of periodic checks. Bearings are used in rotating machinery extensively. However, bearing faults are common and could cause time and ﬁnancial loss if they occur unexpectedly. Machine learning could be used in predictive maintenance framework to predict the health status of a bearing. Bearing fault analysis research has been traditionally conducted on its vibration signature. Due to nature of data, each bearing should be modelled separately or machine learning algorithms should be robust against environment or diﬀerent machinery settings. In the present work unsupervised novelty detection framework on streaming vi bration data is proposed. The framework is built in an unsupervised manner since each bearing is considered individually and building models for each of them is impractical. Since faulty samples are not available initially, novelty detection methods are applied on bearing degradation data. The results show that detection of bearing faults and other state changes can be made using novelty detection methods. Detection could be achieved earlier than conventional methods for some cases.
Rigid & elastic composite registration of 3D LEG T1-weighted MR images for strain analysis
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2013., 2013.) Derman, Ekberjan.; Ertüzün, Ayşın.; Acar, Burak.
Strain analysis of deformation eld is a crucial step for scienti c analysis of muscle structures in MRI data sets. They are commonly used for di erent types of data analysis such as force transmission among muscles which is one of the popular topics in recent years. Especially there is a popular trend of research concentrating on force transmission among lower leg. For successfully carrying out these types of tasks, a correct and e ective deformation eld of the MRI data must be obtained in order to get the expected corresponding strain assessments. These deformation elds are usually obtained by registration framework, in which two groups of data sets were registered to one another, one serving as moving data set, which is transformed to the other target, static data set. Formerly, in registration framework, the whole region of MRI, including bones, is elastically registered to each other. However, bone tissues are relatively hard comparing to those soft muscles, elastically registering these hard tissues, which yields non-zero deformation values among these regions, could negatively a ect the correctness of the obtained nal result. For that, we can think of keeping these bone regions as xed, and then elastically register the remaining parts to obtain a deformation eld in which there would be zero deformations in the bone regions. In this work, T1-weighted MR images of lower leg of human body in undeformed and deformed states are selected as test data. We rst segment the bone regions, keep them as xed, then perform elastic registration on the remaining parts on MR images of lower legs, and compare the obtained strain analysis of deformation eld results with those from unconstrained elastic registration method. We expect to see the nal result would be better than the one obtained by unconstrained elastic registration.
Skeletal muscle deformation analysis using diffusion tensor magnetic resonance imaging
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2013., 2013.) Akyazı, Pınar.; Acar, Burak.; Yücesoy, Can A.
Skeletal muscles are highly organized tissues formed of ber bundles packed together. Muscle bers have distinct orientations which makes them a favorable subject for di usion tensor imaging (DTI) based analyses. DTI provides in vivo measures revealing the structural characteristics of tissues based on di usion anisotropies of water molecules within structures. Local ber orientations can be extracted for deformation analysis of the spatial distribution of di usion and strain characteristics along ber directions. This work aims to present a framework for the assessment of local strain and di usion anisotropy changes as skeletal muscles of human subjects (n=3) become deformed by moving from a exed con guration (150 knee angle) to an extended con guration (180 knee angle). Changes between the di usion anisotropy indices and strain coe cients along ber tracts between the tibialis anterior muscle ends are computed, visualized and modeled to account for heterogeneous changes in the microstructure resulting from deformation. Results are indicators of e ects of myofascial force transmission on human muscles in vivo, including local di erences between sarcomere length changes (maximal lengthening and shortening equals 34.62% and -33.78%, respectively) and di usivity changes in the proximo-distal direction as well as in the transverse plane. The demonstrated methodology also provides an image processing toolbox for the thorough analysis of skeletal muscles. Final results presented here can have clinical implications by contributing to explaining and improving the treatment options of movement limitations.
Spectral analysis of cancer biomarkers in human serum using a custom portable analyzer
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2019., 2019.) Oktay, Kaan.; Acar, Burak.
In this study, we systematically evaluate metabolites and proteins in blood to develop a pipeline to identify potential biomarkers for breast cancer risk. Our aim is to identify a group of molecules, which can be targeted in the design of portable and low-cost blood biomarker detection devices. We obtained plasma samples from women who were cancer free and women who were cancer free at the time of blood collection but developed breast cancer later. Potential prognostic biomarkers for breast cancer risk were extracted from plasma metabolomics and proteomics data using statistical and discriminative power analyses. One of these biomarkers was validated by inde pendent biological measurements. These biomarkers can be used to develop low-cost screening methods towards early diagnosis and hence decreased mortality due to breast cancer. As a low-cost biomarker screening solution, we proposed a Fabry-Perot (FP) interferometer-based spectroscopic biomarker analysis setup to quantitatively detect speciﬁc molecules in their environments. We used commercial Fabry-Perot interferom eters and their evaluation board, which takes the output of the interferometers and process it to make the measurements readable and interpretable. For the surface on which we drop our samples for the measurements, we used Calcium Fluoride (CaF2) windows. In addition to hardware parts, we used a reconstruction method on computer generated spectra to overcome the inherited blurring problem of our interferometers.
Structural brain connectome embedding for Alzheimer's disease
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2019., 2019.) Gamgam, Gurur.; Acar, Burak.
Neurodegenerative diseases are known to alter brain connectivity. Alzheimer’s Disease (AD) is the most common one among these diseases. Although, many re searches have been made to understand AD, there are still more to explore about the complicated nature of AD. To solve these mysteries, features extracted from connec tomes are widely used. Following the poor speciﬁcity of global connectome features, more recently focus has been shifted towards substructures as potential biomarkers. A new model, inspired by the Deepwalk, is proposed to represent these substructures in this thesis. The model treats each individual connectome as a unique graph and learns nodal embeddings per connectome by means of a random walk and a neural network approach. The learned nodal embeddings are used as latent representations of local connectivity and their discriminative power is assessed in SVM based leave-one-out ex periments over a cohort of 91 individuals. Promising results were obtained for AD-SCI / AD-MCI / MCI-SCI / AD-MCI-SCI classiﬁcation tasks. Apart from classiﬁcation, such latent representations of local connectivity may serve as an appropriate space to deﬁne the continuum of neurodegenerative disease progression temporally and spatially which means nodal embeddings can be utilized for monitoring disease progression

Browse

Browsing Elektrik- Elektronik Mühendisliği by Author "Acar, Burak."

Results Per Page

Sort Options