2018 26th Signal Processing and Communications Applications Conference (SIU), 2018
Facial expression recognition is a popular computer vision subject that has many applications suc... more Facial expression recognition is a popular computer vision subject that has many applications such as humancomputer interaction and behavior analysis. As for many computer vision problems, lighting and contrast differences increase the difficulty of the problem. Especially the non-planar structure of the face, protruding regions such as the nose and chin and recessed regions such as eye sockets cause variations in lighting. Another problem with facial expression recognition problems is that the multi-scale detection methods do not align the faces accurately. This leads to comparing features that are extracted from different facial regions, which degrades performance. FHOG features are a contrast-sensitive variation of histogram oriented gradients (HOG) features, which perform well at object detection applications. In this study, the performance of FHOG features at facial expression recognition is investigated. Additionally, aligning with respect to the facial landmarks is proposed to prevent performance degradation due to misalignment. The proposed method is shown to deliver 93% accuracy in facial expression recognition in the extended Cohn-Kanade dataset.
2018 26th Signal Processing and Communications Applications Conference (SIU), 2018
Recommender systems are becoming increasingly important to propose personalized recommendations f... more Recommender systems are becoming increasingly important to propose personalized recommendations for individual users and businesses. In the literature, the proposed recommender systems algorithms focus on improving the accuracy of the recommendation, other important factors affecting the quality of the recommendation are usually overlooked, such as the diversity of recommendation list that presented to the user. In this study, a recommender system algorithm was developed to generate more diverse recommendations and to calculate the accuracy of the recommendation with different comparison techniques, so it is aimed to present a recommendation list to the user's with the balance of recommendation accuracy-diversity. We studied on the currently well-used real data sets and recommendation algorithms that use different optimization techniques, it has been observed that the diversity of recommendation has consistently increased the gain in system accuracy.
2016 24th Signal Processing and Communication Application Conference (SIU), 2016
Brains of patients with dementia show physical differences according to disease types and phases.... more Brains of patients with dementia show physical differences according to disease types and phases. Physical characteristics of brains such as cortical thickness and volumes of some parts have a significant effect on determining the type of the disease. Magnetic resonance imaging devices create visual files which contain patient information appropriate to the medical imaging standards. Using image processing techniques, numerical expressions of patients' brains can be extracted via these files. By means of using these numeric values with classification methods, patients can be classified. In this study, samples having three diseases: Alzheimer's disease, vascular dementia and fron to temporal dementia are used. After extracting cortical surface area, thickness and volume features, samples are classified successfully with artificial neural networks due to feature selection.
Machine learning algorithms builds a model based on train data which is assumed as number of inst... more Machine learning algorithms builds a model based on train data which is assumed as number of instances between different classes are nearly equal. In real world problems usually data sets are unbalanced and this can cause seriously negative effect on built model. Researches on imbalance data sets focus on over-sampling minority class or under-sampling majority class and recently several methods has been purposed which modified support vector machine, rough set based minority class oriented rule learning methods, cost sensitive classifier perform good on imbalanced data set. Although these methods provides a balanced train set artificially, in some real world problems sense of error can be vital since cost of false negative error is expensive than false-positive error. For instances, during classification of satellite image for diseased tree classification, naturally most of trees in a forest is expected to be healthy. Classification algorithm is said to be effective whether critical information is not to be lost. One of the reason why tree’s become diseased in forest is inspect epidemic. Whether classification system could not detect wilted tree, it is not only cause to dry the tree but also possibility to transmission of disease will still contain by insect which can spread. Therefore main goal of this work is minimizing false negative errors. In this work, pre-processing methods for imbalance data sets which divert classification results as minimize false negative error, is discussed.
2011 IEEE 19th Signal Processing and Communications Applications Conference (SIU), 2011
... Bu örneklere benzer olarak zincir kodlarının kullanılmasıyla geliştirilen çeşitli karakter ta... more ... Bu örneklere benzer olarak zincir kodlarının kullanılmasıyla geliştirilen çeşitli karakter tanıma uygulamaları da bulunmaktadır [13, 14]. ... Sarkar, eğrilerin poligonal yaklaşımlarını bulmak için zincir kodları üzerinde anlamlı noktalar tespit etmiştir [17]. ...
2018 6th International Symposium on Digital Forensic and Security (ISDFS), 2018
Advances in machine learning technologies have provided that malicious programs can be detected b... more Advances in machine learning technologies have provided that malicious programs can be detected based on static and dynamic features. Moreover, a crowded set of studies throughout literature indicates that malware detection can be handled with remarkable accuracy rate once converted into image domain. To realize this, some image based techniques have been developed together with feature extraction and classifiers in order to discover the relation between malware binaries in grayscale color representation. With a similar way, we have contributed the CNN features to overcome the malware detection problem. Findings of experimental research support that the malware types can be classified with 85% accuracy rate when applying the machine learning system on 36 (including benign type) malware families consisting of 12,279 malware samples. Moreover, we have achieved the 99% accuracy rate when conducting and experiment on 25 families having 9, 339 malware samples.
Early fault detection and real-time condition monitoring systems have become quite significant fo... more Early fault detection and real-time condition monitoring systems have become quite significant for today’s modern industrial systems. In a high volume of manufacturing facilities, fleets of equipment are expected to operate uninterrupted for days or weeks. Any unplanned interruptions to equipment uptime could jeopardize manufacturers’ cycle time, capacity, and, most significantly, credibility for their customers. With the help of smart manufacturing technologies, companies have started to develop and integrate fault detection and classification systems where end-to-end constant monitoring of equipment is facilitated, and smart algorithms are adapted for the early generation of fault alarms and classification. This paper proposes a generic real-time fault diagnosis and condition monitoring system utilizing edge artificial intelligence (edge AI) and a data distributor open source middleware platform called FIWARE. The implemented system architecture is flexible and includes interfaces...
Chlorine content is one of the most important parameters in Refuse Derived Fuels (RDFs) used as a... more Chlorine content is one of the most important parameters in Refuse Derived Fuels (RDFs) used as a fuel in cement kilns. The main problem with the use of RDF is that chlorine in the waste weakens the cement, increases the risk of corrosion in the kiln and forms toxic gas emissions. Alternative fuels containing high amounts of chlorine, such as plastic waste should be used in limited quantities with the quality of the kiln used and the cement being should be preserved by preparing the appropriate RDF mixture. Analyses conducted on the samples taken before the RDF is given to the furnace are time consuming and costly. Therefore, in this study, the aim is to present a more efficient solution to classify by using chlorine analysis results with hyperspectral imaging and a deep learning model study. For this purpose, a model was created using validated laboratory results and spectral data from samples, the model was tested on a prototype conveyor belt, and was implemented using an online early warning system for high chlorine concentrations. The chlorine content of the RDF samples used in the study ranged from 0.10% to 1.41%, with an average of 0.27%. According to the results, the accuracy, precision, Recall and F1 Score related to the early warning system were found to be 0.8909, 0.8889, 0.8889, 0.8889, respectively. In addition, chlorine measurements were performed at 200, 500 and 1000 mm/s belt speeds and accuracy values of 78.39%, 76.35% and 69.94 %, respectively were obtained.
Deep learning (DL) techniques have been gaining ground for intelligent equipment/process fault di... more Deep learning (DL) techniques have been gaining ground for intelligent equipment/process fault diagnosis applications. However, employing DL methods for such applications comes with its technical challenges. The DL methods are utilized to extract features from raw data automatically, which leads up to its own complications in data preprocessing and/or feature engineering phases. Moreover, another difficulty arises when DL methods are employed utilizing single type of sensor data as the performance of a fault diagnosis application is hindered. To address these issues, we propose utilization of a deep residual network-based multi-sensory data fusion method. The method is established on time-frequency images obtained by short-time Fourier transform to diagnose machine faults. The experimental results demonstrate that the proposed model combining different types of measured signals can diagnose bearing conditions on machines more effectively compared to a single type of measured signal in terms of diagnostic accuracy.
Despite being a challenging research field with many unresolved problems, recommender systems are... more Despite being a challenging research field with many unresolved problems, recommender systems are getting more popular in recent years. These systems rely on the personal preferences of users on items given in the form of ratings and return the preferable items based on choices of like-minded users. In this study, a graph-based recommender system using link prediction techniques incorporating similarity metrics is proposed. A graph-based recommender system that has ratings of users on items can be represented as a bipartite graph, where vertices correspond to users and items and edges to ratings. Recommendation generation in a bipartite graph is a link prediction problem. In current literature, modified link prediction approaches are used to distinguish between fundamental relational dualities of like vs. dislike and similar vs. dissimilar. However, the similarity relationship between users/items is mostly disregarded in the complex domain. The proposed model utilizes user-user and ...
This paper proposes a Quaternion-based link prediction method, a novel representation learning me... more This paper proposes a Quaternion-based link prediction method, a novel representation learning method for recommendation purposes. The proposed algorithm depends on and computation with Quaternion algebra, benefiting from the expressiveness and rich representation learning capability of the Hamilton products. The proposed method depends on a link prediction approach and reveals the significant potential for performance improvement in top-N recommendation tasks. The experimental results indicate the superior performance of the approach using two quality measurements – hits rate, and coverage on the Movielens and Hetrec datasets. Additionally, extensive experiments are conducted on three subsets of the Amazon dataset to understand the flexibility of this algorithm to incorporate different information sources and demonstrate the effectiveness of Quaternion algebra in graph-based recommendation algorithms. The proposed algorithms obtain comparatively higher performance, they are improve...
2018 26th Signal Processing and Communications Applications Conference (SIU), 2018
Facial expression recognition is a popular computer vision subject that has many applications suc... more Facial expression recognition is a popular computer vision subject that has many applications such as humancomputer interaction and behavior analysis. As for many computer vision problems, lighting and contrast differences increase the difficulty of the problem. Especially the non-planar structure of the face, protruding regions such as the nose and chin and recessed regions such as eye sockets cause variations in lighting. Another problem with facial expression recognition problems is that the multi-scale detection methods do not align the faces accurately. This leads to comparing features that are extracted from different facial regions, which degrades performance. FHOG features are a contrast-sensitive variation of histogram oriented gradients (HOG) features, which perform well at object detection applications. In this study, the performance of FHOG features at facial expression recognition is investigated. Additionally, aligning with respect to the facial landmarks is proposed to prevent performance degradation due to misalignment. The proposed method is shown to deliver 93% accuracy in facial expression recognition in the extended Cohn-Kanade dataset.
2018 26th Signal Processing and Communications Applications Conference (SIU), 2018
Recommender systems are becoming increasingly important to propose personalized recommendations f... more Recommender systems are becoming increasingly important to propose personalized recommendations for individual users and businesses. In the literature, the proposed recommender systems algorithms focus on improving the accuracy of the recommendation, other important factors affecting the quality of the recommendation are usually overlooked, such as the diversity of recommendation list that presented to the user. In this study, a recommender system algorithm was developed to generate more diverse recommendations and to calculate the accuracy of the recommendation with different comparison techniques, so it is aimed to present a recommendation list to the user's with the balance of recommendation accuracy-diversity. We studied on the currently well-used real data sets and recommendation algorithms that use different optimization techniques, it has been observed that the diversity of recommendation has consistently increased the gain in system accuracy.
2016 24th Signal Processing and Communication Application Conference (SIU), 2016
Brains of patients with dementia show physical differences according to disease types and phases.... more Brains of patients with dementia show physical differences according to disease types and phases. Physical characteristics of brains such as cortical thickness and volumes of some parts have a significant effect on determining the type of the disease. Magnetic resonance imaging devices create visual files which contain patient information appropriate to the medical imaging standards. Using image processing techniques, numerical expressions of patients' brains can be extracted via these files. By means of using these numeric values with classification methods, patients can be classified. In this study, samples having three diseases: Alzheimer's disease, vascular dementia and fron to temporal dementia are used. After extracting cortical surface area, thickness and volume features, samples are classified successfully with artificial neural networks due to feature selection.
Machine learning algorithms builds a model based on train data which is assumed as number of inst... more Machine learning algorithms builds a model based on train data which is assumed as number of instances between different classes are nearly equal. In real world problems usually data sets are unbalanced and this can cause seriously negative effect on built model. Researches on imbalance data sets focus on over-sampling minority class or under-sampling majority class and recently several methods has been purposed which modified support vector machine, rough set based minority class oriented rule learning methods, cost sensitive classifier perform good on imbalanced data set. Although these methods provides a balanced train set artificially, in some real world problems sense of error can be vital since cost of false negative error is expensive than false-positive error. For instances, during classification of satellite image for diseased tree classification, naturally most of trees in a forest is expected to be healthy. Classification algorithm is said to be effective whether critical information is not to be lost. One of the reason why tree’s become diseased in forest is inspect epidemic. Whether classification system could not detect wilted tree, it is not only cause to dry the tree but also possibility to transmission of disease will still contain by insect which can spread. Therefore main goal of this work is minimizing false negative errors. In this work, pre-processing methods for imbalance data sets which divert classification results as minimize false negative error, is discussed.
2011 IEEE 19th Signal Processing and Communications Applications Conference (SIU), 2011
... Bu örneklere benzer olarak zincir kodlarının kullanılmasıyla geliştirilen çeşitli karakter ta... more ... Bu örneklere benzer olarak zincir kodlarının kullanılmasıyla geliştirilen çeşitli karakter tanıma uygulamaları da bulunmaktadır [13, 14]. ... Sarkar, eğrilerin poligonal yaklaşımlarını bulmak için zincir kodları üzerinde anlamlı noktalar tespit etmiştir [17]. ...
2018 6th International Symposium on Digital Forensic and Security (ISDFS), 2018
Advances in machine learning technologies have provided that malicious programs can be detected b... more Advances in machine learning technologies have provided that malicious programs can be detected based on static and dynamic features. Moreover, a crowded set of studies throughout literature indicates that malware detection can be handled with remarkable accuracy rate once converted into image domain. To realize this, some image based techniques have been developed together with feature extraction and classifiers in order to discover the relation between malware binaries in grayscale color representation. With a similar way, we have contributed the CNN features to overcome the malware detection problem. Findings of experimental research support that the malware types can be classified with 85% accuracy rate when applying the machine learning system on 36 (including benign type) malware families consisting of 12,279 malware samples. Moreover, we have achieved the 99% accuracy rate when conducting and experiment on 25 families having 9, 339 malware samples.
Early fault detection and real-time condition monitoring systems have become quite significant fo... more Early fault detection and real-time condition monitoring systems have become quite significant for today’s modern industrial systems. In a high volume of manufacturing facilities, fleets of equipment are expected to operate uninterrupted for days or weeks. Any unplanned interruptions to equipment uptime could jeopardize manufacturers’ cycle time, capacity, and, most significantly, credibility for their customers. With the help of smart manufacturing technologies, companies have started to develop and integrate fault detection and classification systems where end-to-end constant monitoring of equipment is facilitated, and smart algorithms are adapted for the early generation of fault alarms and classification. This paper proposes a generic real-time fault diagnosis and condition monitoring system utilizing edge artificial intelligence (edge AI) and a data distributor open source middleware platform called FIWARE. The implemented system architecture is flexible and includes interfaces...
Chlorine content is one of the most important parameters in Refuse Derived Fuels (RDFs) used as a... more Chlorine content is one of the most important parameters in Refuse Derived Fuels (RDFs) used as a fuel in cement kilns. The main problem with the use of RDF is that chlorine in the waste weakens the cement, increases the risk of corrosion in the kiln and forms toxic gas emissions. Alternative fuels containing high amounts of chlorine, such as plastic waste should be used in limited quantities with the quality of the kiln used and the cement being should be preserved by preparing the appropriate RDF mixture. Analyses conducted on the samples taken before the RDF is given to the furnace are time consuming and costly. Therefore, in this study, the aim is to present a more efficient solution to classify by using chlorine analysis results with hyperspectral imaging and a deep learning model study. For this purpose, a model was created using validated laboratory results and spectral data from samples, the model was tested on a prototype conveyor belt, and was implemented using an online early warning system for high chlorine concentrations. The chlorine content of the RDF samples used in the study ranged from 0.10% to 1.41%, with an average of 0.27%. According to the results, the accuracy, precision, Recall and F1 Score related to the early warning system were found to be 0.8909, 0.8889, 0.8889, 0.8889, respectively. In addition, chlorine measurements were performed at 200, 500 and 1000 mm/s belt speeds and accuracy values of 78.39%, 76.35% and 69.94 %, respectively were obtained.
Deep learning (DL) techniques have been gaining ground for intelligent equipment/process fault di... more Deep learning (DL) techniques have been gaining ground for intelligent equipment/process fault diagnosis applications. However, employing DL methods for such applications comes with its technical challenges. The DL methods are utilized to extract features from raw data automatically, which leads up to its own complications in data preprocessing and/or feature engineering phases. Moreover, another difficulty arises when DL methods are employed utilizing single type of sensor data as the performance of a fault diagnosis application is hindered. To address these issues, we propose utilization of a deep residual network-based multi-sensory data fusion method. The method is established on time-frequency images obtained by short-time Fourier transform to diagnose machine faults. The experimental results demonstrate that the proposed model combining different types of measured signals can diagnose bearing conditions on machines more effectively compared to a single type of measured signal in terms of diagnostic accuracy.
Despite being a challenging research field with many unresolved problems, recommender systems are... more Despite being a challenging research field with many unresolved problems, recommender systems are getting more popular in recent years. These systems rely on the personal preferences of users on items given in the form of ratings and return the preferable items based on choices of like-minded users. In this study, a graph-based recommender system using link prediction techniques incorporating similarity metrics is proposed. A graph-based recommender system that has ratings of users on items can be represented as a bipartite graph, where vertices correspond to users and items and edges to ratings. Recommendation generation in a bipartite graph is a link prediction problem. In current literature, modified link prediction approaches are used to distinguish between fundamental relational dualities of like vs. dislike and similar vs. dissimilar. However, the similarity relationship between users/items is mostly disregarded in the complex domain. The proposed model utilizes user-user and ...
This paper proposes a Quaternion-based link prediction method, a novel representation learning me... more This paper proposes a Quaternion-based link prediction method, a novel representation learning method for recommendation purposes. The proposed algorithm depends on and computation with Quaternion algebra, benefiting from the expressiveness and rich representation learning capability of the Hamilton products. The proposed method depends on a link prediction approach and reveals the significant potential for performance improvement in top-N recommendation tasks. The experimental results indicate the superior performance of the approach using two quality measurements – hits rate, and coverage on the Movielens and Hetrec datasets. Additionally, extensive experiments are conducted on three subsets of the Amazon dataset to understand the flexibility of this algorithm to incorporate different information sources and demonstrate the effectiveness of Quaternion algebra in graph-based recommendation algorithms. The proposed algorithms obtain comparatively higher performance, they are improve...
Uploads
Papers by Kemal özkan