Browsing by Author "Karakose M."
Now showing 1 - 16 of 16
Results Per Page
Sort Options
Item An Approach for Online Weight Update Using Particle Swarm Optimization in Dynamic Fuzzy Cognitive Maps(Institute of Electrical and Electronics Engineers Inc., 2018) Altundogan T.G.; Karakose M.Fuzzy cognitive maps (FCM) is a method to update a given initial vector to obtain the most stable state of a system, using a neighborhood of weights between these vectors and updating it over a series of iterations. FCMs are modeled with graphs. Neighbor weights between nodes are between-1 and 1. Nowadays it is used in business management, information technology, communication, health and medical decision making, engineering and computer vision. In this study, a dynamic FCM structure based on Particle Swarm Optimization (PSO) is given for determining node weights and online updating for modeling of dynamic systems with FCMs. Neighborhood weights in dynamic FCMs can be updated instantly and the system feedback is used for this update. In this work, updating the weights of the dynamic FCM is a PSO based approach that takes advantage of system feedback. In previous literature suggestions, dynamic FCM structure performs the weight updating process by using rule-based methods such as Hebbian. Metaheuristic methods are less complex and more efficient than rule-based methods in such optimization problems. In the developed PSO approach, the initialize vector state of the system, the weights between the vector nodes, and the desired steady state vector are taken into consideration. As a fitness function, the system has benefited from the convergence state to the desired steady state vector. As a stopping criterion for PSO, 100 ∗ n number of iteration limits have been applied for the initial vector with n nodes. The proposed method has been tested for five different scenarios with different node counts. © 2018 IEEE.Item Image Processing and Deep Neural Image Classification Based Physical Feature Determiner for Traffic Stakeholders(Institute of Electrical and Electronics Engineers Inc., 2019) Altundogan T.G.; Karakose M.Nowadays, image processing and deep learning is used in industrial and non-industrial areas. Addition to this, smart cities are very popular trend for the researchers and rd workers. In the smart city applications, researchers and rd workers present solutions about traffic, health, security and energy problems in the cities. The smart city applications for the traffic are focused on proposing solutions about detecting traffic violations, congestions, park spot suggestion, public transportations etc. We propose a solution for detecting traffic stakeholders physical features based on image processing and deep neural classification. The mentioned traffic stakeholders are automobiles, buses, trucks, trailers, motorcycles and pedestrians. We detect contours from the traffic videos which appropriate size for these traffic stakeholders then we crop these contours from the video first. Then we use the deep image classifier model for classification with detected contours. Addition to this we calculate vehicles dimensional features based on the contour size and determine colors based on HSV features. We intend with this study providing physical features to the smart city workers and researchers for using these features in their applications which controlling violations, determining statistics and the other applications like mentioned. For this reason, we provide this solution with a web service application in the future. © 2019 IEEE.Item Performance Analysis of EEG Signal Processing Based Device Control Applications(Institute of Electrical and Electronics Engineers Inc., 2019) Altundogan T.G.; Karakose M.Nowadays, many types of devices are controlled by electroselenography (EEG) signals. In the literature and in daily life, related studies with EEG controlled devices are increasing day by day. EEG based control applications are applied on many devices such as robot arm, robot, vehicle and unmanned aerial vehicle (UAV). EEG based control procedures usually involve taking, pre-processing, classifying EEG signals, and applying the resulting command to the controlled device. In this study, a performance analysis was carried out by examining the control application studies using EEG signals in the literature. In this analysis study, firstly all studies related to the subject in the literature are examined and the devices, methods, signal processing techniques and classification algorithms used in these studies are handled separately. Appropriate electrode selection for the type of device used in device control applications using EEG signals and type of interaction for command extraction from EEG signal appears to be an important step. In this respect, performance correlations between the types of EEG devices used in the literature studies and the electrode choices used in these studies were compared. Since there are a variety of preprocessing steps for EEG signals, this study provides comparisons based on EEG signal preprocessing techniques. Artificial neural networks (ANN), support vector machines (SVM) and K nearest neighbours (Knn) are used to classify the works in the literature. In this study, comparative studies based on classification methods used in literature studies are also included. As a result, in this study, the studies in the literature for the device control using the EEG signal are examined, compared, interpreted and evaluated, and the points to be considered in the designs to be performed in this area are given. © 2018 IEEE.Item A New Deep Neural Network Based Dynamic Fuzzy Cognitive Map Weight Updating Approach(Institute of Electrical and Electronics Engineers Inc., 2019) Altundogan T.G.; Karakose M.Fuzzy cognitive maps are a soft computational technique with artificial neural network nature expressed by graphs used to model complex systems. Fuzzy cognitive maps are used in areas like health, business, energy, computer science, etc. in the literature. It is used to solve problems in many areas. Fuzzy cognitive maps have a non-dynamic method of weight determination and updating in their classical applications. This is usually done by subjecting the expert opinions to fuzzy membership functions. In some studies, the weights obtained by using expert data are improved by utilizing historical data of the system modeled. In some studies, weight determination and updating process is realized by using intuitive optimization methods without benefiting from expert opinions. Here, fuzzy cognitive maps using a different weight matrix for different input values are considered dynamic. In this study, we propose a dynamic fuzzy cognitive map structure that performs the weight update process with deep learning. Here, the cognitive map weights determined in the data set used for deep neural network training are determined by a genetic optimization-based method using past system data. The proposed new DFCM structure has been tested on two different scenarios. In addition, the performance comparison with deep artificial neural network models dealing with the same scenarios was performed. When the experimental and comparative results are examined, the performance of the proposed method is quite satisfactory. © 2019 IEEE.Item Multiple Object Tracking with Dynamic Fuzzy Cognitive Maps Using Deep Learning(Institute of Electrical and Electronics Engineers Inc., 2019) Altundogan T.G.; Karakose M.Object tracking is the process of matching objects detected on image sequences onto image frames. There are different types of object tracking applications used for different scenarios. For example, if a single object is being traced on an image, this is a single object tracking application. Tracking multiple objects on an image is called multiple object tracking. Fuzzy cognitive maps, on the other hand, form the model of a system by using the features of a system and the relationships between these features. Here, the single object tracking process is a matching problem, so FCM assumes a classifier role. In conventional operations, FCMs use the same weight matrix for all initial concept values. This can reduce the performance of the solution that the FCM produces for the problem it tackles. The FCM structure we use here takes advantage of the dynamic learning of FCM weights with deep learning. The study was tested on different image sequences and the performance of the proposed method were very satisfactory. © 2019 IEEE.Item Genetic Algorithm Based Fuzzy Cognitive Map Concept Relationship Determination and Sigmoid Configuration(Institute of Electrical and Electronics Engineers Inc., 2020) Altundogan T.G.; Karakose M.Fuzzy cognitive maps take the features of a problem or a system as a concept to solve a problem or to model real world systems. Here, the positive or negative relationships between the concepts of fuzzy cognitive map are expressed with fuzzy weights, and the proposed solution or cognitive construction of the system model is completed. Previously, many different perspectives have been proposed in the literature for FCM weighting. Some of them utilize an expert group to determine the relationships between concepts, some using Hebbian-based methods, some using population-based meta-heuristic optimization methods, and some FCMs use different supervised computational intelligence methods. For FCMs that use supervised learning techniques in weight determination and updating, it is not possible to provide a data set with system-related conceptual relationships. Data sets for systems or problems generally contain the values that the concept values take over time. In this study, it is aimed to bring the cognitive weights corresponding to the concept values of the raw data sets of the systems to be expressed with FCM by a genetic algorithm based method. The method tested on different scenarios has a very high performance. © 2020 IEEE.Item A Noise Reduction Approach Using Dynamic Fuzzy Cognitive Maps for Vehicle Traffic Camera Images(Institute of Electrical and Electronics Engineers Inc., 2020) Altundogan T.G.; Karakose M.Noise is a generic term for data loss or corruption due to hardware or software causes on the signal. Since the images are two-dimensional signals, there are noises in this type of signal due different reasons. In addition, fuzzy cognitive maps (FCM) have a structure based on a graph theory that can produce many probing solutions today. Fuzzy cognitive maps can provide their iterations as static (fixed neighborhood values) or dynamic (variable neighborhood values) depending on the solution, which belong to interested problem. In this study, a method is presented using fuzzy cognitive maps for noise reduction in images and mean filter, which is a widely used method for noise reduction. The proposed method provide to minimize the loss of data in the noise reduction process with the average filter. In this work, FCM takes noisy and average filtered noisy image masks and accepts each pixel value in these masks as nodes. Then we update the neighborhood weights between these nodes in each iteration. The developed method has been tested primarily with different images and the performance obtained only by the method in which the average filter is applied is quite high. Then, the proposed method was tested on images of traffic monitoring systems taken from vehicle cameras. The results obtained are very successful. © 2020 IEEE.Item Cracked Wall Image Classification Based on Deep Neural Network Using Visibility Graph Features(Institute of Electrical and Electronics Engineers Inc., 2021) Altundogan T.G.; Karakose M.Visibility graphs are graphs created by making use of the relations of objects with each other depending on their visibility features. Today, visibility graphs are used quite frequently in signal processing applications. In this study, cracked and non-cracked wall images taken from a dataset were classified by a deep neural network depending on the visibility graph properties. In the proposed method, firstly, histograms of the images are obtained. The resulting histogram is then expressed by visibility graphs. A feature vector of each image is created with the maximum clique and maximum degree features of the obtained visibility graphs. Then, deep neural network training is performed with the feature vectors created. The classification success of the proposed method on images separated for testing is 99%. © 2021 IEEE.Item Identifying Clinical Characteristics of Hypoparathyroidism in Turkey: HIPOPARATURK-NET Study(Springer, 2022) Konca Degertekin C.; Gogas Yavuz D.; Pekkolay Z.; Saygili E.; Ugur K.; Or Koca A.; Unubol M.; Topaloglu O.; Aydogan B.I.; Ozdemir Kutbay N.; Hekimsoy Z.; Yilmaz N.; Balci M.K.; Tanrikulu S.; Aydogan Unsal Y.; Ersoy C.; Omma T.; Keskin M.; Yalcin M.M.; Yetkin I.; Soylu H.; Karakose M.; Yilmaz M.; Karakilic E.; Piskinpasa H.; Batman A.; Akbaba G.; Elbuken G.; Tura Bahadir C.; Kilinc F.; Bilginer M.C.; Turhan Iyidir O.; Canturk Z.; Aktas Yilmaz B.; Sayiner Z.A.; Eroglu M.Hypoparathyroidism is an orphan disease with ill-defined epidemiology that is subject to geographic variability. We conducted this study to assess the demographics, etiologic distribution, treatment patterns and complication frequency of patients with chronic hypoparathyroidism in Turkey. This is a retrospective, cross-sectional database study, with collaboration of 30 endocrinology centers located in 20 cities across seven geographical regions of Turkey. A total of 830 adults (mean age 49.6 ± 13.5 years; female 81.2%) with hypoparathyroidism (mean duration 9.7 ± 9.0 years) were included in the final analysis. Hypoparathyroidism was predominantly surgery-induced (n = 686, 82.6%). The insulting surgeries was carried out mostly due to benign causes in postsurgical group (SG) (n = 504, 73.5%) while patients in nonsurgical group (NSG) was most frequently classified as idiopathic (n = 103, 71.5%). The treatment was highly dependent on calcium salts (n = 771, 92.9%), calcitriol (n = 786, 94.7%) and to a lower extent cholecalciferol use (n = 635, 76.5%) while the rate of parathyroid hormone (n = 2, 0.2%) use was low. Serum calcium levels were most frequently kept in the normal range (sCa 8.5–10.5 mg/dL, n = 383, 46.1%) which might be higher than desired for this patient group. NSG had a lower mean plasma PTH concentration (6.42 ± 5.53 vs. 9.09 ± 7.08 ng/l, p < 0.0001), higher daily intake of elementary calcium (2038 ± 1214 vs. 1846 ± 1355 mg/day, p = 0.0193) and calcitriol (0.78 ± 0.39 vs. 0.69 ± 0.38 mcg/day, p = 0.0057), a higher rate of chronic renal disease (9.7% vs. 3.6%, p = 0.0017), epilepsy (6.3% vs. 1.6%, p = 0.0009), intracranial calcifications (11.8% vs. 7.3%, p < 0.0001) and cataracts (22.2% vs. 13.7%, p = 0.0096) compared to SG. In conclusion, postsurgical hypoparathyroidism is the dominant etiology of hypoparathyroidism in Turkey while the nonsurgical patients have a higher disease burden with greater need for medications and increased risk of complications than the postsurgical patients. © 2021, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.Item EEG Signal Classification with Deep Neural Networks using Visibility Graphs(Institute of Electrical and Electronics Engineers Inc., 2022) Altundogan T.G.; Karakose M.EEG signals are data presented by collecting electrical activities in the brain at a certain frequency. Today, applications using the EEG signal are implemented in many fields such as medicine, computer science, robotic. Visibility Graphs, on the other hand, are graphs where certain points are associated according to their visibility features in order to perform mapping and operations in areas such as robotics. Visibility Graphs are also used today to express signals. In this study, the EEG signals are expressed with visibility graphs after certain pre-processing. Then, the classification of the obtained graph depending on the clique and degree features was carried out by using deep artificial neural networks. EEG signals have a very noisy nature, and complex pre-processing and feature extractions are used in applications using EEG signals. In the proposed method, EEG signals are subjected to very simple pre-processing and classified with a 95% success rate. © 2022 IEEE.Item LSTM Encoder Decoder Based Text Highlight Abstraction Method Using Summaries Extracted by PageRank(Institute of Electrical and Electronics Engineers Inc., 2023) Altundogan T.G.; Karakose M.Automatic highlighting from texts is an abstractive summarization problem that is frequently focused on in natural language processing. In encoder-decoder architectures, developed for abstractive summarization, as the size of the input array increases, the learning ability of the architecture becomes difficult. To solve this problem, the focus is on minimizing this disadvantage of encoder - decoder architectures by using the Attention mechanism. In this study, we used an LSTM encoder - decoder with an attention mechanism to perform the highlight abstraction process. In addition, we used an extractive summarization step as a preprocess to increase the learning ability of the encoder - decoder architecture and reduce the input text size. We preferred the PageRank method in the extractive summarization process here. In the PageRank method, sentence vectors were extracted by using Glove embeddings to calculate similarities of text sentences. The proposed approach achived the extractive summarization by 67.6% and abstractive summarization by 59.6% in ROUGE-1 score. © 2023 IEEE.Item Sentiment Analysis for Patient Reviews in Hospitals by CNN and LSTM Neural Networks Using Pretrained Word Embeddings(Institute of Electrical and Electronics Engineers Inc., 2023) Altundogan T.G.; Karakose M.; Yilmazer S.; Hanoglu E.; Demirel S.Medical reviews of patients are very important for the medical management departments and sentiment analysis is one of the most popular application areas of Natural Language Processing. In this study, we use and compare different neural architectures for sentiment analysis of patient reviews about hospitals. We developed four neural models to classify the patient review as positive or negative. First, the data retrieved from an online platform were preprocessed. Then, before the neural training, Skipgram word embeddings were carried out for transfer learning. Finally, training was performed. A model which we trained has only fully connected dense layers. One of the trained models includes LSTM and fully connected layers. One of them includes CNN and fully connected layers. One model has CNN, LSTM and fully connected layers. After the training phases our best two neural models (LSTM-CNN and LSTM) have achieved sentiment classification with over 85% performance. © 2023 IEEE.Item BART Fine Tuning based Abstractive Summarization of Patients Medical Questions Texts(Institute of Electrical and Electronics Engineers Inc., 2023) Altundogan T.G.; Karakose M.; Tokel O.Today, many people get counseling about their problems by interviewing doctors or communicating via online platforms. With neural architectures such as Transformer achieving high-performance results in the field of natural language processing, the use of these approaches has become quite common in solving many natural language processing problems in the medical field. In this study, a method using BART (Bidirectional Auto-Regressive Transformer) neural architecture is proposed for abstractive summarization of questions asked by patients to doctors. In the proposed method, the pretrained BART neural architecture is retrained using a dataset consisting of questions asked by patients to doctors and summaries of these questions. The evaluation of the summary questions obtained was carried out with the ROUGE metric and compared with other approaches in the literature from different perspectives. When the comparative results are examined, the ROUGE performance of our approach is higher than 92% of other studies that use abstractive medical summary. © 2023 IEEE.Item Reconstructing Artistic Heritage: Style Transfer in Ottoman Miniature Paintings Using Pre-Trained CNN with PSNR Based Image Similarity(Institute of Electrical and Electronics Engineers Inc., 2024) Altundogan T.G.; Karakose M.Deep learning techniques, inspired by the workings of the human brain, are enhancing the performance of many productive computer science applications in the field of culture and art. In this study, pre-trained CNNs were used to transfer styles on works of art that have a unique style, such as Ottoman miniatures. The styles of different Ottoman miniatures were transferred to other paintings that are not works of art or works of art with this method. The new images obtained were evaluated using PSNR and SSIM metrics to be compared with similar studies in the literature. The results showed that this study performed in parallel with other similar studies. This study presents a potential method that can contribute to research in the field of transferring styles on works of art such as Otto man miniatures. © 2024 IEEE.Item Transformer Based Multimodal Summarization and Highlight Abstraction Approach for Texts and Speech Audios(Institute of Electrical and Electronics Engineers Inc., 2024) Altundogan T.G.; Karakose M.; Tanberk S.Multimodal summarization is a kind of summarization application in which its inputs and/or outputs can be in different data types like text, video, and audio. In this study, a new approach based on fine tuning of different pre-trained transformers was developed for abstractive and extractive summarization of audio and text data. In the proposed method, abstractive and extractive summaries of text data are provided only as text, while extractive summaries of audio data are presented as both text and audio data. Abstractive summaries of the audio data are presented as text only. Transformers with text2text input-output relationship were used in both extractive and abstractive summarization processes of the proposed method. For the training and inference processes of audio this type of data to be handled in transformers, an ASR step was followed before the summarization step. The experimental results obtained were given in detail and compared with similar approaches in the literature. As a result of the comparison, it was seen that the proposed method achieved better performance than similar prior approaches. © 2024 IEEE.Item Dynamic Fuzzy Cognitive Maps-Based Crowd Analysis Using Time Series Obtained From Video Processing(Institute of Electrical and Electronics Engineers Inc., 2025) Goktug Altundogan T.; Karakose M.; Yaman O.; Tanberk S.; Mert F.; Egemen Yilmaz A.Fuzzy cognitive maps treat the components of a problem or system expressed as fuzzy concepts and model the system with the relationships between these concepts. We predicted that FCM' s approach to calculating with these relationships could perform multivariate time-series forecasting with high performance. However, especially in real-world systems and related data sets, numerical values that clearly express the relationships between time series elements are not included, and this is a challenge. Another challenge that FCMs have for these problems is that real-world systems are highly dynamic structures and these relationships have variable properties in different situations. We evaluated these challenges as the main motivation factor and developed a GA-based method to determine system relationships for different states of the system. Then, in order to dynamically handle these relationships determined for different states on FCM, we took advantage of neural architectures that take the initial concept vectors as input and calculate the relationships between these concepts. In order to evaluate the performance of the time-series forecasting approach we developed, we performed time-series forecasting on two different scenarios using an artificially generated data set and a benchmark data set containing real-world data. In this way, we saw that the time series modeling performance of our proposed system is over 95%. FCMs perform the calculations they have made until the system becomes stable. This allows time series analyses to be performed not only depending on time but also depending on the steady state. Therefore, using this capability of the approach we developed to model time series formed from crowd analysis data obtained with video analytics is quite suitable in terms of providing the contribution points we present in this article. In this context, we integrated our FCM-based time series forecasting approach to two different video analytics scenarios by applying two different crowd analysis approaches we developed within the scope of this study. The success of our proposed method is over 95% both for performing for forecasting the time series obtained as a result of crowd analysis. Beside, crowd analysis approaches developed in this study have similar performance to state of the art approaches in the literature. © 2013 IEEE.