Yazar "Acir, Nurettin" seçeneğine göre listele
Listeleniyor 1 - 9 / 9
Sayfa Başına Sonuç
Sıralama seçenekleri
Öğe A complex-valued adaptive filter algorithm for system identification problem(Institute of Electrical and Electronics Engineers Inc., 2016) Mengüç, Engin Cemal; Acir, NurettinIn this study, a complex-valued adaptive filter algorithm based on Lyapunov stability theory is presented to solve a system identification problem in the complex domain. The performance of the proposed complex-valued Lyapunov adaptive filter (CLAF) algorithm is improved for the complex-valued system identification problem by integrating the LST into the filter optimization cost. The performance of the proposed algorithm is tested on a complex-valued moving average (MA) system identification problem and compared with the conventional complex-valued least mean square (CLMS) and complex-valued normalized least mean square (CNLMS) algorithms. The simulation results show that the proposed CLAF algorithm has achieved a faster convergence rate and a lower steady-state MSE performance when compared to the other algorithms. © 2015 Chamber of Electrical Engineers of Turkey.Öğe A study on the monitoring of weld quality using XGBoost with Particle Swarm Optimization(Elsevier, 2024) Avci, Adem; Kocakulak, Mustafa; Acir, Nurettin; Gunes, Emrah; Turan, SertanGas Metal Arc Welding is a joining technique with numerous uses in manufacturing. Since welding process parameters have a considerable impact on the welding quality, online monitoring systems are utilized on production lines to achieve standard welding quality with minimum welding faults. This study presents preliminary work to develop a monitoring system for defect analysis in a Gas Metal Arc Welding process. This study aims to eliminate the need for laboratory tests with a model to reduce welding quality control costs. In this study, welding data such as voltage, current, and wire feeding rate were collected during the welding process in realtime. New features were derived from the gathered data at the preprocessing stage using statistical approaches. The determination of whether the welding process is defective or flawless was made using the Extreme Gradient Boosting Algorithm. The hyperparameters of the algorithm were optimized with Particle Swarm Optimization. The accuracy value was obtained as 93.15% after repeating the conducted experiments ten times. The recall, precision, specificity, and F1 -Score values in these experiments were calculated as 97.22%, 94.75%, 72.35%, and 95.94%, respectively. The mean current value was found to be the most relevant and meaningful feature that describes the welding quality based on intensive experiments. In addition to the proposed algorithm, some other machine -learning algorithms were tested on the welding dataset. With this study, the significance of feature derivation from the acquired welding current data has been discovered to analyze welding quality.Öğe Automated vein verification using self-attention-based convolutional neural networks(Pergamon-Elsevier Science Ltd, 2023) Kocakulak, Mustafa; Avci, Adem; Acir, NurettinVein-based biometric traits have been regarded as trustworthy for biometric applications. With technical advances in deep learning, verification performance has started to be improved in these applications to increase trust level in daily life by providing usage convenience and user satisfaction. In this study, the effect of self-attention mechanism on convolutional neural networks for the performance of finger-vein and hand dorsal vein verification was investigated using an open-set protocol. To provide generalizability to the trained model, self-attention-based convolutional neural networks were used rather than existing architectures and pre-trained models. With the architecture that uses residual blocks and self-attention mechanism, a fair verification performance was suggested. Verification performance was assessed on DHVI-DB and Bosphorus hand dorsal vein datasets and SDUMLA and PolyU-F finger-vein datasets in terms of equal error rate using the distance between feature vectors through the existing and the proposed distance measures. The obtained equal error rates for hand dorsal vein datasets DHVI-DB1, DHVI-DB2, and Bosphorus are 2.17, 2.21, and 18.33, respectively and for finger-vein datasets SDUMLA and PolyU-F, are 1.65 and 10.64, respectively. Moreover, 4 different loss functions were used throughout the conducted experiments to see the discriminative ability of the proposed network for vein verification. The experimental results on these datasets indicate the potential effectiveness of the self-attention mechanism on automated vein verification.Öğe Automatic removal of ocular artefacts in EEG signal by using independent component analysis and Chauvenet criterion(Institute of Electrical and Electronics Engineers Inc., 2017) Çinar, Salim; Acir, NurettinEye movements (saccade, blink and etc.) cause artefacts in Electroencephalogram recordings. The ocular artefact can distort the EEG signals. Removal of ocular artefact is important issue in EEG signal analysis. The main task of artefact removal algorithms is to obtain cleaned EEG without losing meaningful EEG signal. The main focus of this work is to remove ocular artefact automatically by using Independent Component Analysis and Chauvenet criterion. The method is tested on real dataset. Relative error and Correlation coefficient are used for the performance test. The performance of the proposed method was Relative error= 0.273±0.148, Correlation coefficients 0.943± 0.042 in the dataset. The results show that the porposed method effectively removes ocular artefacts in EEG. © 2016 IEEE.Öğe Benchmarking domain adaptation for LiDAR-based 3D object detection in autonomous driving(Springer London Ltd, 2025) Balim, Mustafa Alper; Hanilci, Cemal; Acir, NurettinThe generalization capability of 3D object detection models is crucial for ensuring robust perception in autonomous driving systems. While state-of-the-art models such as Voxel R-CNN, PV-RCNN, and CenterPoint have demonstrated strong performance on publicly available datasets (e.g., KITTI, Waymo, and nuScenes). In this study, we conduct a comprehensive benchmark evaluation. We introduce two custom datasets: (i) a real-world dataset collected using KARSAN's autonomous minibus equipped with a 128-channel LiDAR sensor under diverse traffic conditions, and (ii) a simulated dataset generated using the AWSIM simulation platform, capturing over five hours of synthetic driving data with virtual LiDAR sensors. Our results indicate that 3D object detection performance is highly dataset-dependent, as no single model achieves superior results across all datasets and metrics. Cross-dataset evaluation highlights the challenges of domain mismatch, which causes significant performance degradation when models are tested on our custom datasets, particularly in the synthetic domain. To mitigate these effects, we explore six domain adaptation techniques and demonstrate that their application substantially improves model performance. Bi3D, SESS, and Uni3D outperform UDA, CLUE, and ST3D, yielding more robust generalization across both real-world and simulated environments. These findings shed light on the potential of domain adaptation to improve model performance across domain shifts, despite the ongoing challenges in achieving consistent outcomes across all environments.Öğe Classification of motor imagery signals by convolutional neural network for BCI applications(Institute of Electrical and Electronics Engineers Inc., 2019) Balim, Mustafa Alper; Acir, NurettinElectroencephalography (EEG) signals have been using for clinical purposes for many years. However, studies on the use of EEG signals in brain computer interface (BBA) applications are increasing. It is possible to control machines using only mental activities, especially for patients with limited mobility. Motor imagery signals (MIS) which are formed as a result of the imagination of moving a limb are one of the most common signal used for this purpose. In this study, it is aimed to classify MIS signals with Convolutional Neural Network by using BCI-IV 2b dataset. As a result, higher (%75,7) performance was obtained with lower number of parameters compared to similar previous studies. © 2019 IEEE.Öğe Dynamic ROI Extraction for Palmprints using MediaPipe Hands(Ieee, 2022) Kocakulak, Mustafa; Acir, NurettinHand-based biometric traits have been widely used in recognition systems. Dynamic region of interest extraction is an important preprocessing step for these systems to avoid recognition performance degradation. In this study, a dynamic region of interest extraction method that can be used for palm vein, palmprint, and dorsal hand vein has been proposed using Google's MediaPipe Hands framework. Since 3 biometric traits focus on nearly the same region that contains biometric information on the images, this study aims to show that the proposed extraction method can be utilized for these traits on mobile biometric applications. This method has been implemented on IIT Delhi Touchless Palmprint Database and 93% accuracy was obtained. The average processing time per image for ROI extraction was recorded as 2.64 seconds. With this study, a paradigm for future studies on hand biometrics has been created and the required processing time for a dynamic extraction has been reduced considerably.Öğe Enhancement of finger vein images using Gabor filter(Institute of Electrical and Electronics Engineers Inc., 2018) Kocakulak, Mustafa; Acir, NurettinIn this study, rather than locating a fixed region of interest directly on finger vein images, spatial filtering is applied and a texture-based edge detection method is used to give stable results. Koschmieder's Law, which eliminates the scattering effect of light on these images, is applied to the designated region of interest through white balancing process. After this step, Gabor filter bank was created in different scales and orientations. These bank elements were convolved with various images and Gabor filter application was completed. In this study, by applying Gabor filter to the images enhanced by Koschmieder's Law, it is verified that the biometric information extraction of a person is facilitated. © 2018 IEEE.Öğe Motor Imagery Signal Classification Using Constant-Q Transform for BCI Applications(European Signal Processing Conference, EUSIPCO, 2021) Balim, Mustafa Alper; Hanilçi, Cemal; Acir, NurettinElectroencephalography (EEG) signals have been using for brain-computer interface applications for the last two decades. Motor imagery (MI) signals are one of the EEG signal types formed by imagining a limb's movement. Recently with the help of deep neural networks (DNN) for classifying MI signals using time-frequency (TF) features, considerable performance improvement has been reported. This paper proposes using a well-known TF representation technique called Constant-Q Transform (CQT) for the MI signal classification. Experiments conducted on BCI IV 2b dataset with DNN classifier using CQT spectrogram show that CQT outperforms traditional short-time Fourier transform (STFT) representation.
