Browsing by Author "Guney, Selda"

Now showing 1 - 20 of 24

Applications of Deep Learning Techniques to Wood Anomaly Detection
(2022) Celik, Yaren; Guney, Selda; Dengiz, Berna; Xu, J; Altiparmak, F.; Hassan, MHA; Marquez, FPG
Wood products and structures have an important place in today's industry. They are widely used in many fields. However, there are various difficulties in production systems where wood raw material is under many processes. Some difficulty and complexity of production processes result in high variability of raw materials such as a wide range of visible structural defects that must be checked by specialists on line or of line. These issues are not only difficult and biased in manual processes, but also less effective and misleading. To overcome the drawbacks of the manual quality control processes, machine vision-based inspection systems are in great of interest recently for quality control applications. In this study, the wood anomaly has been detected by using deep learning. As it will be a distinction-based method on image processing, the Convolution Neural Network (CNN), which is one of the most suitable methods, has been used for anomaly detection. In addition, it will be tried to obtain the most suitable one among different CNN architectures such as ShuffleNet, AlexNet, GoogleNet for the problem. MobileNet, SqueezeNet, GoogleNet, ShuffleNet among considered methods show promising results in classifying normal and abnormal wood products.
Automated Tuberculosis Detection Using Pre-Trained CNN and SVM
(2021) Oltu, Burcu; Guney, Selda; Dengiz, Berna; Agildere, Muhtesem
Tuberculosis (TB) is a dreadfully contagious and life-threatening disease if left untreated. Therefore, early and accurate diagnosis is critical for treatment. Today, invasive, expensive, or time-consuming tests are performed for diagnosis. Unfortunately, accurate TB diagnosis is still a major challenge. In the proposed study, a decision support system that can automatically separate normal and TB chest X-ray (CXR) images is presented for objective and accurate diagnosis. In the presented methodology, first various data augmentation methods were applied to the data set, then pre-trained networks (VGG16, MobileNet), were employed as feature extractors from augmented CXR's. Afterward, the extracted features for all images were fed into a support vector machine classifier. In training process, 5-fold cross-validation was applied. As a result of this classification, it was concluded that TB can be diagnosed with an accuracy of 96,6% and an area under the ROC curve (AUC) of 0,99.
Calibration Transfer Between E-Noses
(2014) Guney, Selda; Fernandez, Luis; Marco, Santiago; https://orcid.org/0000-0002-0573-1326; AAC-7404-2020
Electronic nose is an instrument which is composed of gas sensor array and pattern recognition unit. It is generally used for classifying, identifying or quantifying the odors or volatile organic components for these commonly used devices, calibration transfer is an important issue because of differences in each instrument, sensor drift, changes in environmental conditions or background changes. Calibration transfer is a transfer of model between different instruments which have different conditions. In this study, calibration transfer is applied to the e-noses which have different temperature conditions. Also the results of the direct standardization, piecewise direct standardization and orthogonal signal correction which are different calibration methods were compared. The results of the piecewise direct standardization method are more successful than the other methods for the dataset which is used in this study.
Classification of Canine Maturity and Bone Fracture Time Based on X-Ray Images of Long Bones
(2021) Ergun, Gulnur Begum; Guney, Selda; 0000-0002-0573-1326; 0000-0001-8469-5484
Veterinarians use X-rays for almost all examinations of clinical fractures to determine the appropriate treatment. Before treatment, vets need to know the date of the injury, type of the broken bone, and age of the dog. The maturity of the dog and the time of the fracture affects the approach to the fracture site, the surgical procedure and needed materials. This comprehensive study has three main goals: determining the maturity of the dogs (Task 1), dating fractures (Task 2), and finally, detecting fractures of the long bones in dogs (Task 3). The most popular deep neural networks are used: AlexNet, ResNet-50 and GoogLeNet. One of the most popular machine learning algorithms, support vector machines (SVM), is used for comparison. The performance of all sub-studies is evaluated using accuracy and F1 score. Each task has been successful with different network architecture. ResNet-50, AlexNet and GoogLeNet are the most successful algorithms for the three tasks, with F1 scores of 0.75, 0.80 and 0.88, respectively. Data augmentation is performed to make models more robust, and the F1 scores of the three tasks were 0.80, 0.81, and 0.89 using ResNet-50, which is the most successful model. This preliminary work can be developed into support tools for practicing veterinarians that will make a difference in the treatment of dogs with fractured bones. Considering the lack of work in this interdisciplinary field, this paper may lead to future studies.
Classification of Different Objects with Artificial Neural Networks Using Electronic Nose
(2015) Ozsandikcioglu, Umit; Atasoy, Ayten; Guney, Selda; 0000-0002-5397-6301; 0000-0003-1188-2902; 0000-0002-0573-1326; AAR-4368-2020; HJH-3630-2023; AAC-7404-2020
In this paper; an e-nose with low cost which consisting of 8 different gas sensors was used and with this e-nose 9 different odors ((mint, lemon, egg, rotten egg, angelica root, nail polish, naphthalene, rose water, and acetone) was classified. This 9 different odor are classified with Artificial Neural Networks and by using different activation functions, and then the successes of the classification were compared with each other. The maximum success of the testing data is obtained with 100% accuracy rate by using logsig activation function in hidden layer and tansig activation function in output layer. In conclusion; using the chemical database containing the odor of the different objects, distinct odors were shown to be classified correctly.
Classification of Human Movements by Using Kinect Sensor
(2023) Acis, Busra; Guney, Selda; https://orcid.org/0000-0001-6683-0005; https://orcid.org/0000-0002-0573-1326; HDM-2942-2022
In recent years, studies have been carried out to classify human movements in many areas such as health and safety. To classify human movements, image processing methods have also started to be used in recent years. With the help of learning-based algorithms, human posture can be defined in the images obtained by various imaging methods. The predecessor methods of these classification algorithms are machine learning and deep learning. In addition, in recent years, the use of sensors that can detect human joints in perceiving human posture has also increased. The Kinect sensor, developed by Microsoft, is one of the most frequently used sensors because it is not wearable and can detect joints with infrared rays and transfer this information directly to the computer via USB connection. This study used a dataset called CAD60 that included real-time human posture information and images obtained using a Microsoft Kinect sensor, which is available in the literature. This dataset contains data that includes different movements/postures of different people. Within the scope of this study, the performances of these algorithms were obtained by using classification algorithms with the MATLAB program and these performances were compared. The classification algorithms have been used to try to improve the results by using different architectures. When raw data is used, classification accuracy is obtained as 72.60% with one of the machine learning methods, the Cosine K-Nearest Neighbor method. With the feature selection method, this success value has been increased to 74.18%. In addition, when classified by the Support Vector Machines method after the feature extraction process using the Long Short Term Memory method from the deep network architectures, which is the method proposed in this study, the accuracy rate was increased to 98.95%. The best method of classifying human posture was investigated by using different methods and a method was proposed by comparing it with the literature.
Comparative Study for Tuberculosis Detection by Using Deep Learning
(2021) Karaca, Busra Kubra; Guney, Selda; Dengiz, Berna; Agildere, Muhtesem
Tuberculosis (TB) is an infectious disease which becomes a significant health problem worldwide. Many people have been affected by this disease owing to deficiency of treatment and late or inaccuracy of diagnosis. Therefore, accurate and early diagnosis is the very major solution to checking and preventing the disease. A chest x-ray is a main diagnostic tool used to diagnose tuberculosis. This diagnostic method is limited by the availability of radiologists and the experience and skills of radiologists in reading x-rays. To overcome such a challenge, a computer-aided diagnosis (CAD) system is supposed for the radiologist to interpret chest x-ray images easily. In this study, a CAD system based upon transfer learning is developed for TB detection using Montgomery Country chest x-ray images. We used the VGG16, VGG19, DenseNet121, MobileNet, and InceptionV3 pre-trained CNN models to extract features automatically and used the Support Vector Machine (SVM) classifier to the detection of tuberculosis. Furthermore, data augmentation techniques were applied to boost the performance results. The proposed method performed the highest accuracy of 98.9% and area under the curve (AUC) of 1.00, respectively, with the DenseNet121 on augmented images.
The Comparison of Estimation Algorithms for Mobile Robot Navigation
(2016) Guney, Selda; Bilen, Murat; 0000-0002-0573-1326; AAC-7404-2020
In this study, a robot with different maneuvras is followed with different estimation algorithms. The mobile robot has acted first linear, then maneuver and finally linear again. It's speed is constant through the way. Standard Kalman Filter, Adaptive Kalman Filter, Extended Kalman Filter and Interacting Multiple Model consist of multiple model Kalman Filter combined of linear and non-linear model are used to follow the act of the robot. The results of these estimations are compared with each other. Multiple model Kalman Filter is the best estimation algorithm among them for this motion model.
Comparison of Expectation Maximization and Naive Bayes Algorithms in Character Recognition
(2016) Guney, Selda; Cakar, Ceyhun; 0000-0002-0573-1326; AAC-7404-2020
Statistical character recognition methods are used very common nowadays in the character recognition. A certain number of features are extracted from characters recognized for the recognition of the character. The classification is performed with the recognition of these features. Extracted features can be considered as input signal of a prediction system. Thus proposed methods for estimation can be used for the recognition. In this study, digit characters are classified with expectation maximization and Naive Bayes methods over Gaussian mixture models. These two methods are compared with each other.
Daphnet Freezing Recognition with Gait Data by Using Machine Learning Algorithms
(2020) Guney, Selda; Boluk, Busra
The aim of this study was to test the success of the data set obtained by a wearable health assistant developed for the symptom of freezing (FOG) in gait of Parkinson's patients and to increase the success of the system. The system was tested with different machine learning methods to measure the success of the wearable health assistant system. For all patients (ten patients), the highest success value was obtained and the mean sensitivity and specificity values of the system were calculated and compared with the results obtained in the literature review. In the literature, mean sensitivity and specificity were 73.1% and 81.6%, respectively; In this study, mean sensitivity and specificity were 91.9% and 71.14%, respectively. In order to better analyze the success of the system, two patients with successful and unsuccessful results were selected for the data set in line with the results obtained in the literature review. The success of the system was tested by using different machine learning methods on the data sets of two patients. Finally, the successes obtained by feature extraction methods were tried to be increased. Among the different machine learning methods on the data sets used for patient 8 and patient 3, the most successful method was obtained by combining the models (ensemble). The highest achievement value obtained by attribute extraction methods was obtained when PCA was applied. However, the success value obtained with raw data could not be increased. All results are tabulated and presented.
A Deep LSTM Approach for Activity Recognition
(2019) Guney, Selda; Erdas, Cagatay Berke; 0000-0002-0573-1326; 0000-0003-3467-9923; AAC-7404-2020
Since 1990s, activity recognition effectual field in machine learning literature. Most of studies that relevant activity recognition, use feature extraction method to achieve higher classification performance. Moreover, these studies mostly use traditional machine learning algorithms for classification. In this paper, we focus on a deep (Long Short Term Memory) LSTM neural network for feature free classification of seven daily activities by using raw data that collected from three-dimensional accelerometer. Based on the results, the proposed deep LSTM approach can classify raw data with high performance. The results show that the proposed deep LSTM approach achieved 91.34, 96.91, 88.78, 87.58 as percent classification performance in terms of accuracy, sensitivity, specificity, F-measure respectively.
Discrimination of Different Fish Species by E-nose
(2015) Guney, Selda; Atasoy, Ayten; 0000-0003-1188-2902; 0000-0002-0573-1326; HJH-3630-2023; AAC-7404-2020
The aim of this study is to distinguish three different types of fish which are anchovy, horse mackerel and whiting by an electronic nose. Generally, the electronic noses are composed of three units. These are sensor unit which has 8 metal oxide sensors in this study, an electronic unit and a pattern recognition unit. In the pattern recognition unit, signal preprocessing, feature extraction and classification stages are performed. For distinguishing different fish species, different feature extraction methods and classification methods are compared with each other. Then the best combination of feature extraction and classification method is selected and applied to the fish database.
Estimation of Concentration Values of Different Gases Based on Long Short-Term Memory by Using Electronic Nose
(2021) Bakiler, Hande; Guney, Selda; 0000-0002-0573-1326
An electronic nose (e-nose) is commonly used in different areas. In the e-nose studies, one of the most important subjects is the estimation of the different concentration values of different gases. An accurate estimation of gas concentrations plays a very important role in sensitive issues such as disease detection. This study has been carried out to increase the classification and regression successes of concentration values of four different gases detected by 4 metal oxide gas sensors. The different methods are used to compare the success of the classification of the concentration levels and the success of the estimation of concentration values of these all gases. In order to realize these classification and regression processes, first a preprocessing and a feature extraction steps were applied to the raw data. The focus of this study is to increase the success achieved in classification and regression by performing the feature extraction using the proposed method. In the proposed method, "Fully Connected Layer" of Long Short-Term Memory networks was used as a feature extraction. Then, these extracted features were used. The results of the proposed method are compared the other traditional methods. It was observed that there was an improvement in both the classification and regression results with the proposed method. The highest accuracy rate in the classification were obtained in the Support Vector Machine method with 90.8% and in the regression problem, the best mean square errors were obtained with Gaussian Process Regression by using the proposed method.
Fault Detection System For Paper Cup Machine Based On Real-Time Image Processing
(Başkent Üniversitesi Mühendislik Fakültesi, 2024-03-31) Aydin, Alaaddin; Guney, Selda
In the production of paper cups in industrial factories, it is tried to print high quality cups with less waste loss with the help of sensors and heating resistances mounted on the paper cup machine. In this study, a system that detects faulty products based on image processing and removes it by controlling the machine with servo motors, asynchronous motors and programmable logic controller (PLC) is designed. For fault product detection, classification has been performed using real-time Haarcascade algorithm and You Only Look Once (YOLO) algorithm which is a deep learning methods, and real-time object detection has been carried out using the OpenCv library. With this study, an effective faulty product detection and removing hardware system was realized by adapting artificial intelligence algorithms to a machine used in industry. Based on the results, a whole system can be applied to systems that involve removing a faulty product from a band in any production, packaging etc. facility is proposed. A hardware consisting of servo motors, asynchronous motors and PLC was designed to separate faulty cups from the existing paper cup production machine in this study. Then, a data set composed of 1068 images was created with images taken from the camera for faulty and faultless paper cups. Using this dataset, the effect of different deep learning methods on performance in the real-time system has been examined and successful results have been obtained. The optimal outcome was achieved, yielding a real-time application accuracy rate of 90.8% through the utilization of the Yolov5x architecture.
Fish Freshness Testing with Artificial Neural Networks
(2015) Atasoy, Ayten; Ozsandikcioglu, Umit; Guney, Selda; 0000-0002-0573-1326; AAC-7404-2020
In this work, with the use of an electronic nose which has 8 metal oxide gas sensors and was set up at Karadeniz Technical University, a fish freshness system was designed. There are 7 classes (1, 3, 5, 7, 9, 11, 13 day for fish storage) for classification and to perform classification process, Artificial Neural Networks was used in this work. To increase the classification success, Artificial Neural Network architecture, activation functions and input data obtained from different feature extraction method was changed, the storage condition is very important factor for fish freshness and fishes used in this study were stored at fish market conditions. In this study to determine the classification success, 5-Fold Cross Validation method was used and the maximum success rate was obtained as 98.94 %.
Freshness Classification of Horse Mackerels with E-Nose System Using Hybrid Binary Decision Tree Structure
(2020) Guney, Selda; Atasoy, Ayten; 0000-0002-0573-1326
The aim of this study is to test the freshness of horse mackerels by using a low cost electronic nose system composed of eight different metal oxide sensors. The process of freshness evaluation covers a seals of seven different classes corresponding to 1, 3, 5, 7, 9, 11, and 13 storage days. These seven classes are categorized according to six different classifiers in the proposed binary decision tree structure. Classifiers at each particular node of the tree are individually trained with the training dataset. To increase success in determining the level of fish freshness, one of the k-Nearest Neighbors (k-NN), Support Vector Machines (SVM), Linear Discriminant Analysis (LDA) and Bayes methods is selected for every classifier and the feature spaces change in every node. The significance of this study among the others in the literature is that this proposed decision tree structure has never been applied to determine fish freshness before. Because the freshness of fish is observed under actual market storage conditions, the classification is more difficult. The results show that the electronic nose designed with the proposed decision tree structure is able to determine the freshness of horse mackerels with 85.71% accuracy for the test data obtained one year after the training process. Also, the performances of the proposed methods are compared against conventional methods such as Bayes, k-NN, and LDA.
Heart Disease Prediction by Using Machine Learning Algorithms
(2020) Erdogan, Alperen; Guney, Selda
Nowadays, one of the most important illness is heart disease which cause of mostly patients dead. Medical diagnosis of heart diseases is very difficult. While heart diseases are diagnosed medically, they can be confused with other diseases that show same symptoms such as chest pain, shortness of breath, palpitations and nausea. This makes it difficult to diagnose heart diseases medically. In this study, the presence of heart diseases was determined by using machine learning algorithms. In this study, the data obtained from the patients were weighted according to their effects on the success rate. In this study, a method is proposed for determine weight coefficient. According to proposed method's results, 86,90% success was achieved with 13 different features obtained from the patients.
Human Activity Recognition by Using Different Deep Learning Approaches for Wearable Sensors
(2021) Erdas, Cagatay Berke; Guney, Selda; 0000-0003-3467-9923
With the spread of wearable sensors, the solutions to the task of activity recognition by using the data obtained from the sensors have become widespread. Recognition of activities owing to wearable sensors such as accelerometers, gyroscopes, and magnetometers, etc. has been studied in recent years. Although there are several applications in the literature, differently in this study, deep learning algorithms such as Convolutional Neural Networks, Convolutional LSTM, and 3D Convolutional Neural Networks fed by Convolutional LSTM have been used in human activity recognition task by feeding with data obtained from accelerometer sensor. For this purpose, a frame was formed with raw samples of the same activity which were collected consecutively from the accelerometer sensor. Thus, it is aimed to capture the pattern inherent in the activity and due to preserving the continuous structure of the movement.
Investigation of Harmonic Estimation Using Imperalist Competitive Algorithm Method
(2015) Ertugrul, Emre; Guney, Selda; 0000-0002-0573-1326; AAC-7404-2020
Recently, some new methods and algorithms have been started to use as an alternative for harmonic estimation instead of Fourier Transform based traditional algorithms. Because, it is inevitable to look for alternative solutions for harmonic estimation problems due to the existing limitations of Fourier Transform based algorithms. The Imperialist Competitive Algorithm (ICA) investigated in this study is a social based intuitive optimization algorithm. The advantages of the evolutionary approximations selected by the nature have been suggested in genetic algorithms and its derivations in order to be useful in optimization area. The animal behaviors have been concluded as partial swarm and ant colony optimization algorithms. Recently, the animal behaviors simulate the human social behaviors and guide us to solve some engineering problems.
Machine Learning-Based Weather Prediction With Radiosonde Observations
(JOURNAL OF THE FACULTY OF ENGINEERING AND ARCHITECTURE OF GAZI UNIVERSITY, 2024-07-16) Gogen, Eralp; Guney, Selda
From the past to the present, weather forecasting holds significant importance for humanity. The precise execution of weather forecasting enables the implementation of precautions against natural disasters such as floods, tsunamis, etc., thereby minimizing the adverse effects that may arise. In this study, weather prediction is conducted using Radiosonde data. Within this prediction, estimations for both the highest and lowest temperatures are made employing machine learning algorithms. Unlike previous temperature prediction studies in the literature, a three-year dataset of Radiosonde observations is utilized. This dataset, measured at intervals of 1mbar up to an altitude of 40 km from the ground, allows for a more accurate modeling of the atmosphere compared to other studies in the literature. In this model, predictions for the highest and lowest temperatures for the next day are made. In this stage, the effects of normalization, feature extraction, or selection on the results are analyzed, and the most suitable model for prediction is determined. The software, implemented in the MATLAB environment, compares different regression methods. As a result of these analyses, utilizing the Gaussian Process Regression (GPR) method, the highest temperature prediction for the next day is achieved with the highest accuracy, with a mean square root deviation of 1.2. Using the same method, the lowest temperature prediction is made with a mean square root deviation ratio of 2.4. The results indicate more successful temperature predictions compared to studies in the literature.