Aydin Yesildirek
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Aydin Yesildirek has a total of 31 co-authors. Here are the top co-authors by number of shared publications:
F.L. Lewis
Autom. & Robotics Res. Inst., Texas Univ., Arlington, TX, USA
9 shared publicationsK. Liu
4 shared publications- Mehmet Iscan
Department of Mechatronics Engineering, Yildiz Technical University, Turkey
3 shared publications - Ayman Hassan El-Hag
Department of Electrical and Computer Engineering, University of Waterloo, Waterloo, Canada
2 shared publications F. L. Lewis
The University of Texas at Arlington
2 shared publications- Frank L. Lewis
UTA Research Institute, The University of Texas at Arlington, Arlington, TX, USA
2 shared publications
Latest Publications
- Open Access
Improving Control Performance of Tilt-Rotor VTOL UAV with Model-Based Reward and Multi-Agent Reinforcement Learning
Tilt-rotor Vertical Takeoff and Landing Unmanned Aerial Vehicles (TR-VTOL UAVs) combine fixed-wing and rotary-wing configurations, offering optimized flight planning but presenting challenges due to their complex dynamics and uncertainties. This study investigates a multi-agent reinforcement learning (RL) control system utilizing Soft Actor-Critic (SAC) modules, which are designed to independently control each input with a tailored reward mechanism. By implementing a novel reward structure based on a dynamic reference response region, the multi-agent design improves learning efficiency by minimizing data redundancy. Compared to other control methods such as Actor-Critic Neural Networks (AC NN), Proximal Policy Optimization (PPO), Nonsingular Terminal Sliding Mode Control (NTSMC), and PID controllers, the proposed system shows at least a 30% improvement in transient performance metrics—including RMSE, rise time, settling time, and maximum overshoot—under both no wind and constant 20 m/s wind conditions, representing an extreme scenario to evaluate controller robustness. This approach has also reduced training time by 80% compared to single-agent systems, lowering energy consumption and environmental impact.
Open AccessImproving Control Performance of Tilt-Rotor VTOL UAV with Model-Based Reward and Multi-Agent Reinforcement Learning
Tilt-rotor Vertical Takeoff and Landing Unmanned Aerial Vehicles (TR-VTOL UAVs) combine fixed-wing and rotary-wing configurations, offering optimized flight planning but presenting challenges due to their complex dynamics and uncertainties. This study investigates a multi-agent reinforcement learning (RL) control system utilizing Soft Actor-Critic (SAC) modules, which are designed to independently control each input with a tailored reward mechanism. By implementing a novel reward structure based on a dynamic reference response region, the multi-agent design improves learning efficiency by minimizing data redundancy. Compared to other control methods such as Actor-Critic Neural Networks (AC NN), Proximal Policy Optimization (PPO), Nonsingular Terminal Sliding Mode Control (NTSMC), and PID controllers, the proposed system shows at least a 30% improvement in transient performance metrics—including RMSE, rise time, settling time, and maximum overshoot—under both no wind and constant 20 m/s wind conditions, representing an extreme scenario to evaluate controller robustness. This approach has also reduced training time by 80% compared to single-agent systems, lowering energy consumption and environmental impact.
- Open Access
- 25 July 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (15) , 8301
An Efficient Dropout for Robust Deep Neural Networks
Overfitting remains a major difficulty in training deep neural networks, especially when attempting to achieve good generalization in complex classification tasks. Standard dropout is often employed to address this issue; however, its uniform random inactivation of neurons typically leads to instability and insufficient performance increases. This paper proposes an upgraded regularization technique merging adaptive sigmoidal dropout with weight amplification, seeking to dynamically adjust neuron deactivation depending on weight statistics, activation patterns, and neuron history. The proposed dropout process uses a sigmoid function driven by a temperature parameter to determine deactivation likelihood and incorporates a “neuron recovery” step to restore important activations. Simultaneously, the method amplifies high-magnitude weights to select crucial traits during learning. The proposed method is tested on CIFAR-10, and CIFAR-100 datasets using four unique CNN architectures, including deep and residual-based models, to evaluate the approach. Results demonstrate that the suggested technique consistently outperforms both standard dropout and baseline models without dropout, yielding higher validation accuracy and lower, more stable validation loss across all datasets. In particular, it demonstrated superior convergence and generalization performance on challenging datasets such as CIFAR-100. These findings demonstrate the potential of the proposed technique to improve model robustness and training efficiency and provide an alternative in complex classification tasks.
- 25 July 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (15) , 8301
Open AccessAn Efficient Dropout for Robust Deep Neural Networks
Overfitting remains a major difficulty in training deep neural networks, especially when attempting to achieve good generalization in complex classification tasks. Standard dropout is often employed to address this issue; however, its uniform random inactivation of neurons typically leads to instability and insufficient performance increases. This paper proposes an upgraded regularization technique merging adaptive sigmoidal dropout with weight amplification, seeking to dynamically adjust neuron deactivation depending on weight statistics, activation patterns, and neuron history. The proposed dropout process uses a sigmoid function driven by a temperature parameter to determine deactivation likelihood and incorporates a “neuron recovery” step to restore important activations. Simultaneously, the method amplifies high-magnitude weights to select crucial traits during learning. The proposed method is tested on CIFAR-10, and CIFAR-100 datasets using four unique CNN architectures, including deep and residual-based models, to evaluate the approach. Results demonstrate that the suggested technique consistently outperforms both standard dropout and baseline models without dropout, yielding higher validation accuracy and lower, more stable validation loss across all datasets. In particular, it demonstrated superior convergence and generalization performance on challenging datasets such as CIFAR-100. These findings demonstrate the potential of the proposed technique to improve model robustness and training efficiency and provide an alternative in complex classification tasks.
- 16 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-10
A Hybrid Model for 3-D Gunshot Localization Using Muzzle Blast Sound Only
Humans and animals possess the ability to roughly estimate the distance and direction of specific sound sources through their auditory organs. This study introduces an innovative hybrid model for 3D gunshot sound localization, drawing inspiration from human sound localization models and leveraging the strengths of both microphone arrays and deep learning techniques. The proposed system employs a six-microphone array with co-centered orthogonal placement to capture the sounds of gunshot muzzle blasts. The methodology involves first estimating azimuth and elevation angles using an approximate closed-form Direction-of-Arrival (DoA) formula based on Time-Difference-of-Arrival (TDoA) values obtained from the Generalized Cross-Correlation with Phase Transform (GCC-Phat) algorithm. Sub-sequently, Mel Spectrograms are derived from the captured sound data and fed into a Convolutional Neural Network (CNN) for distance estimation. Ultimately, the direction and distance estimations are integrated to achieve the 3D localization of the gunshot source. The proposed approach stands as a notable contribution to the literature, relying solely on the sound of a muzzle blast for localization with a single microphone array. The average distance estimation error in the range of 50-500 m is 6.87%, demonstrating an improvement compared to existing systems with an error rate of approximately 16%. This study is pioneering in its application of deep learning trained on a dataset of actual explosion sounds for distance estimation in gunshot localization systems.
- 16 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-10
A Hybrid Model for 3-D Gunshot Localization Using Muzzle Blast Sound Only
Humans and animals possess the ability to roughly estimate the distance and direction of specific sound sources through their auditory organs. This study introduces an innovative hybrid model for 3D gunshot sound localization, drawing inspiration from human sound localization models and leveraging the strengths of both microphone arrays and deep learning techniques. The proposed system employs a six-microphone array with co-centered orthogonal placement to capture the sounds of gunshot muzzle blasts. The methodology involves first estimating azimuth and elevation angles using an approximate closed-form Direction-of-Arrival (DoA) formula based on Time-Difference-of-Arrival (TDoA) values obtained from the Generalized Cross-Correlation with Phase Transform (GCC-Phat) algorithm. Sub-sequently, Mel Spectrograms are derived from the captured sound data and fed into a Convolutional Neural Network (CNN) for distance estimation. Ultimately, the direction and distance estimations are integrated to achieve the 3D localization of the gunshot source. The proposed approach stands as a notable contribution to the literature, relying solely on the sound of a muzzle blast for localization with a single microphone array. The average distance estimation error in the range of 50-500 m is 6.87%, demonstrating an improvement compared to existing systems with an error rate of approximately 16%. This study is pioneering in its application of deep learning trained on a dataset of actual explosion sounds for distance estimation in gunshot localization systems.
- Open Access
- 18 June 2024
- Published by Wiley in International Journal for Numerical Methods in Biomedical Engineering
- Vol. 40 (8) , e3838
An intelligent aortic valve model for complete cardiac cycle
The aortic valve (AV) is crucial for cardiovascular (CV) hemodynamic, impacting cardiac output (CO) and left ventricular volumetric flow rate (LVQ). Its nonlinear behavior challenges standard LVQ prediction methods as well as CO one. This study presents a novel approach for modeling the AV in the CV system, offering an improved method for estimating crucial parameters like LVQ across various AV conditions, including aortic stenosis (AS). The model, based on AV channel length during the entire cardiac phase, introduces a time‐varying AV resistance (TV‐AVR) parameterized by the pressure ratio across the AV and LVQ, enabling the simulation of both healthy and AS‐related conditions. To validate this model, in vitro measurements are compared using a hybrid mock circulatory loop device. An unconventional use of a convolutional neural network (CNN) corrects the model's estimates, eliminating the need for labeled datasets. This approach, incorporating real‐time learning and transforming 1‐D CV signals into 2‐D tensors, significantly improves the accuracy of LVQ measurements, achieving an error rate of less than 3.41 ± 4.84% for CO in healthy conditions and 2.83 ± 1.35% in AS cases—a 33.13% enhancement over linear diode models. These results underscore the potential of this approach for enhancing the diagnosis, prediction, and treatment of AV diseases. The key contributions of the proposed method encompass nonlinear TV‐AVR estimation, investigation of transient CV responses, prediction of instantaneous CO, development of a flexible framework for noninvasive measurements integration, and the introduction of an adjustable resistance model using an extended Kalman filter (EKF) and CNN combination, all without requiring labeled data.
- 18 June 2024
- Published by Wiley in International Journal for Numerical Methods in Biomedical Engineering
- Vol. 40 (8) , e3838
Open AccessAn intelligent aortic valve model for complete cardiac cycle
The aortic valve (AV) is crucial for cardiovascular (CV) hemodynamic, impacting cardiac output (CO) and left ventricular volumetric flow rate (LVQ). Its nonlinear behavior challenges standard LVQ prediction methods as well as CO one. This study presents a novel approach for modeling the AV in the CV system, offering an improved method for estimating crucial parameters like LVQ across various AV conditions, including aortic stenosis (AS). The model, based on AV channel length during the entire cardiac phase, introduces a time‐varying AV resistance (TV‐AVR) parameterized by the pressure ratio across the AV and LVQ, enabling the simulation of both healthy and AS‐related conditions. To validate this model, in vitro measurements are compared using a hybrid mock circulatory loop device. An unconventional use of a convolutional neural network (CNN) corrects the model's estimates, eliminating the need for labeled datasets. This approach, incorporating real‐time learning and transforming 1‐D CV signals into 2‐D tensors, significantly improves the accuracy of LVQ measurements, achieving an error rate of less than 3.41 ± 4.84% for CO in healthy conditions and 2.83 ± 1.35% in AS cases—a 33.13% enhancement over linear diode models. These results underscore the potential of this approach for enhancing the diagnosis, prediction, and treatment of AV diseases. The key contributions of the proposed method encompass nonlinear TV‐AVR estimation, investigation of transient CV responses, prediction of instantaneous CO, development of a flexible framework for noninvasive measurements integration, and the introduction of an adjustable resistance model using an extended Kalman filter (EKF) and CNN combination, all without requiring labeled data.
- Open Access
- 27 October 2023
- Published by Hindawi Limited in Applied Bionics and Biomechanics
- Vol. 2023, 1-20
A New Cardiovascular Mock Loop Driven by Novel Active Capacitance in Normal and Abnormal Conditions
The hybrid mock circulatory loop (hMCL) serves as a crucial hemodynamic simulation tool, offering exceptional flexibility, controllability, and reproducibility for investigating the mechanisms underlying cardiovascular diseases (CVD) in a controlled environment, circumventing the limitations of live organism studies. This paper introduces a novel design and control strategy for hMCL, introducing a novel left ventricle volumeelastance (LVVE) equation that unifies the autoregulation of the FrankStarling mechanism (FSM) with left ventricle contractility (LVC). LVVE establishes a dynamic link between left ventricular volume (LVV) and LVC, inherently satisfying the regulatory relationship between left ventricular pressure (LVP) and LVV through a mathematical equation. For the first time, LVVE integration significantly enhances the physiological relevance of hMCL by faithfully replicating FSM responses across diverse conditions, including aortic stenosis (AS), variations in systemic vascular resistance (SVR), and heart rate (HR) variations. Furthermore, this study introduces the stability proofs for the discrete closed-loop hMCL, enabling real-time proportional valve control through discrete feedback linearizationan innovative departure from conventional methods. Notably, FSM emulation is achieved by tracking reference maximum and minimum LVV values, eliminating the reliance on predefined functions or existing data, such as the maximum LV elastance value. Rigorous experimental validation, encompassing numerical simulations and comparative analyses with prior research, attests to the precision and efficacy of the proposed hMCL in faithfully replicating both normal and abnormal CV conditions. Significantly, the hMCL demonstrates that increasing HR enhances LVC while maintaining physiological pressures; however, this increase in LVC corresponds with a decrease in LVV, in alignment with human data and FSM principles. Crucially, the coupling mechanism between the FSM and LVC yields results of enhanced physiological fidelity, significantly advancing the hMCLs utility in physiological research. Moreover, the hMCLs capacity to simulate critical cardiovascular scenarios, including AS, SVR fluctuations, and HR variations, underscores its versatility and substantial potential for investigating complex CV dynamics.
- 27 October 2023
- Published by Hindawi Limited in Applied Bionics and Biomechanics
- Vol. 2023, 1-20
Open AccessA New Cardiovascular Mock Loop Driven by Novel Active Capacitance in Normal and Abnormal Conditions
The hybrid mock circulatory loop (hMCL) serves as a crucial hemodynamic simulation tool, offering exceptional flexibility, controllability, and reproducibility for investigating the mechanisms underlying cardiovascular diseases (CVD) in a controlled environment, circumventing the limitations of live organism studies. This paper introduces a novel design and control strategy for hMCL, introducing a novel left ventricle volumeelastance (LVVE) equation that unifies the autoregulation of the FrankStarling mechanism (FSM) with left ventricle contractility (LVC). LVVE establishes a dynamic link between left ventricular volume (LVV) and LVC, inherently satisfying the regulatory relationship between left ventricular pressure (LVP) and LVV through a mathematical equation. For the first time, LVVE integration significantly enhances the physiological relevance of hMCL by faithfully replicating FSM responses across diverse conditions, including aortic stenosis (AS), variations in systemic vascular resistance (SVR), and heart rate (HR) variations. Furthermore, this study introduces the stability proofs for the discrete closed-loop hMCL, enabling real-time proportional valve control through discrete feedback linearizationan innovative departure from conventional methods. Notably, FSM emulation is achieved by tracking reference maximum and minimum LVV values, eliminating the reliance on predefined functions or existing data, such as the maximum LV elastance value. Rigorous experimental validation, encompassing numerical simulations and comparative analyses with prior research, attests to the precision and efficacy of the proposed hMCL in faithfully replicating both normal and abnormal CV conditions. Significantly, the hMCL demonstrates that increasing HR enhances LVC while maintaining physiological pressures; however, this increase in LVC corresponds with a decrease in LVV, in alignment with human data and FSM principles. Crucially, the coupling mechanism between the FSM and LVC yields results of enhanced physiological fidelity, significantly advancing the hMCLs utility in physiological research. Moreover, the hMCLs capacity to simulate critical cardiovascular scenarios, including AS, SVR fluctuations, and HR variations, underscores its versatility and substantial potential for investigating complex CV dynamics.
- Open Access
- 14 October 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13 (20) , 11297
Closed-Form DoA Solution for Co-Centered Orthogonal Microphone Arrays Based on Multilateration Equations
This study proposes a closed-form direction-of-arrival (DoA) solution derived from multilateration equations for microphone arrays of co-centered and orthogonal pairs. The generalized cross-correlation phase transform (GCC-Phat) algorithm is used to obtain the time-difference-of-arrival (TDoA) values. Simulation studies have shown the success of our proposed method compared to existing DoA methods in the literature by varying the sampling frequency of the sound signal, inter-microphone distances, and the source distance. The results from the simulation are validated by the measurements from our experiments. Our proposed solution gives better results than the far-field solution against the angle error, which is more pronounced at incidence angles smaller than 15°. These angle errors, which approach 3° using the far-field method, are reduced to less than 0.5 degrees using our proposed solution. Our solution also gives more stable results against TDoA measurement errors. Our proposed solution achieves a 66% improvement for azimuth angle and 5.88% improvement for elevation angle compared to the simulation results in the absence of TDoA measurement error, outperforming the far-field approach. When normally distributed sampling error is added to TDoA measurements, with a standard deviation of three samples, our proposed solution achieves a 41% improvement for azimuth angle and a 5.44% improvement for elevation angle. In our field measurements, an absolute mean error of 0.94 degrees was observed with our proposed method for azimuth angle. It is shown to be a more stable and faster solution method for real-time applications.
- 14 October 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13 (20) , 11297
Open AccessClosed-Form DoA Solution for Co-Centered Orthogonal Microphone Arrays Based on Multilateration Equations
This study proposes a closed-form direction-of-arrival (DoA) solution derived from multilateration equations for microphone arrays of co-centered and orthogonal pairs. The generalized cross-correlation phase transform (GCC-Phat) algorithm is used to obtain the time-difference-of-arrival (TDoA) values. Simulation studies have shown the success of our proposed method compared to existing DoA methods in the literature by varying the sampling frequency of the sound signal, inter-microphone distances, and the source distance. The results from the simulation are validated by the measurements from our experiments. Our proposed solution gives better results than the far-field solution against the angle error, which is more pronounced at incidence angles smaller than 15°. These angle errors, which approach 3° using the far-field method, are reduced to less than 0.5 degrees using our proposed solution. Our solution also gives more stable results against TDoA measurement errors. Our proposed solution achieves a 66% improvement for azimuth angle and 5.88% improvement for elevation angle compared to the simulation results in the absence of TDoA measurement error, outperforming the far-field approach. When normally distributed sampling error is added to TDoA measurements, with a standard deviation of three samples, our proposed solution achieves a 41% improvement for azimuth angle and a 5.44% improvement for elevation angle. In our field measurements, an absolute mean error of 0.94 degrees was observed with our proposed method for azimuth angle. It is shown to be a more stable and faster solution method for real-time applications.
- Open Access
- 9 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 111909-111926
Modeling Transient Cardiovascular Hemodynamics With Physiological Conscious Autoencoder
Current biomedical research relies primarily on in silico studies to model complex systems like cardiovascular hemodynamics. However, for comprehensive validation of mathematical models, real in vitro experiments are indispensable. This paper introduces a framework that bridges these approaches through the hybrid mock circulatory loop (hMCL), offering precise control, flexibility, and reproducibility. This innovation enables the investigation of cardiovascular disease mechanisms in a controlled setting, overcoming the limitations of live organism studies. The framework employs a modified autoencoder with a partially guided latent space, incorporating physiological insights into a deep neural network. It leverages time-delayed cardiovascular signals, including pressures, flow rates, and unmeasurable cardiovascular system (CVS) parameters, to estimate critical parameters like aortic and mitral resistance, systemic resistance, and left ventricle elastance. The autoencoder’s loss function is tailored to predict these parameters, enhancing the understanding of cardiovascular dynamics. The study utilizes in silico data to train the model and validates it through in vitro tests using a hybrid mock loop device, yielding a remarkable accuracy of less than 5.7% error in replicating CVS signals. Furthermore, the framework demonstrates the adaptability of CVS variables to perturbations in closed-loop conditions and exhibits a diagnostic model with an impressive 98.55% F1 score for classifying cardiovascular disease severity. This research significantly advances the field by modifying the autoencoder to include physiological signals, introducing a novel loss function, developing a structured network, presenting a diagnostic model, and proposing an innovative approach for generating transient responses in cardiovascular hemodynamics, validated through in vitro and in silico experiments.
- 9 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 111909-111926
Open AccessModeling Transient Cardiovascular Hemodynamics With Physiological Conscious Autoencoder
Current biomedical research relies primarily on in silico studies to model complex systems like cardiovascular hemodynamics. However, for comprehensive validation of mathematical models, real in vitro experiments are indispensable. This paper introduces a framework that bridges these approaches through the hybrid mock circulatory loop (hMCL), offering precise control, flexibility, and reproducibility. This innovation enables the investigation of cardiovascular disease mechanisms in a controlled setting, overcoming the limitations of live organism studies. The framework employs a modified autoencoder with a partially guided latent space, incorporating physiological insights into a deep neural network. It leverages time-delayed cardiovascular signals, including pressures, flow rates, and unmeasurable cardiovascular system (CVS) parameters, to estimate critical parameters like aortic and mitral resistance, systemic resistance, and left ventricle elastance. The autoencoder’s loss function is tailored to predict these parameters, enhancing the understanding of cardiovascular dynamics. The study utilizes in silico data to train the model and validates it through in vitro tests using a hybrid mock loop device, yielding a remarkable accuracy of less than 5.7% error in replicating CVS signals. Furthermore, the framework demonstrates the adaptability of CVS variables to perturbations in closed-loop conditions and exhibits a diagnostic model with an impressive 98.55% F1 score for classifying cardiovascular disease severity. This research significantly advances the field by modifying the autoencoder to include physiological signals, introducing a novel loss function, developing a structured network, presenting a diagnostic model, and proposing an innovative approach for generating transient responses in cardiovascular hemodynamics, validated through in vitro and in silico experiments.
- Open Access
- 11 February 2023
- Published by MDPI AG in Sustainability
- Vol. 15 (4) , 3340
An Autonomous Marine Mucilage Monitoring System
Mucilage bloom is a current issue, especially for countries in the Mediterranean Basin, due to economic activities and ecological effects. The main causes are increased nutrient load due to organic and industrial pollution in the sea, global warming, and meteorological conditions at a level that can trigger mucilage bloom. It is important to take permanent measures to combat the increased nutrient load causing mucilage. However, there are various actions that can be performed during the mucilage bloom period, especially the collection of mucilage on the sea surface. Surface vehicles can be used to monitor and collect mucilage on the sea surface. The aim of this study is to design an autonomous marine mucilage monitoring system for systems such as unmanned surface vehicles (USV). We suggest monitoring the risky Marmara Sea continuously and recording some of the key parameters using a USV. The onboard solution proposed in this study has an architect based on a three-tier mucilage monitoring system. In the first tier, the sea surface is scanned with camera(s) in a certain radius in real time. When mucilage-candidate areas are determined, the vehicle is directed to this region autonomously. In the second tier, seawater in the region is measured in real time with some onboard sensors, pH level, conductivity, and dissolved oxygen level. The third tier is where real samples at three different depths are collected (if possible) for detailed posterior lab analysis. We have compared image processing, CNN (ResNet50), kNN, SVM, and FFNN approaches and have shown that the accuracy of our proposed mucilage classification method offers better and more promising performance.
- 11 February 2023
- Published by MDPI AG in Sustainability
- Vol. 15 (4) , 3340
Open AccessAn Autonomous Marine Mucilage Monitoring System
Mucilage bloom is a current issue, especially for countries in the Mediterranean Basin, due to economic activities and ecological effects. The main causes are increased nutrient load due to organic and industrial pollution in the sea, global warming, and meteorological conditions at a level that can trigger mucilage bloom. It is important to take permanent measures to combat the increased nutrient load causing mucilage. However, there are various actions that can be performed during the mucilage bloom period, especially the collection of mucilage on the sea surface. Surface vehicles can be used to monitor and collect mucilage on the sea surface. The aim of this study is to design an autonomous marine mucilage monitoring system for systems such as unmanned surface vehicles (USV). We suggest monitoring the risky Marmara Sea continuously and recording some of the key parameters using a USV. The onboard solution proposed in this study has an architect based on a three-tier mucilage monitoring system. In the first tier, the sea surface is scanned with camera(s) in a certain radius in real time. When mucilage-candidate areas are determined, the vehicle is directed to this region autonomously. In the second tier, seawater in the region is measured in real time with some onboard sensors, pH level, conductivity, and dissolved oxygen level. The third tier is where real samples at three different depths are collected (if possible) for detailed posterior lab analysis. We have compared image processing, CNN (ResNet50), kNN, SVM, and FFNN approaches and have shown that the accuracy of our proposed mucilage classification method offers better and more promising performance.
- 1 April 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-4
Hand gesture movement classification based on dynamically structured neural network
As one of the complex biomedical signals, electromyography represents the electrical activation of the muscle providing useful information about gesture movements. A dynamically structured neural network is proposed to be used as an efficient tool to classify such movement patterns through some EMG signals generated from hand/arm movements with unknown complexity in real-time. The dynamically structured neural network is trained to establish input/output mapping while learning its optimal structure. The proposed architecture's performance is compared to various size statically structured neural network configurations. Over two folds of improvements are observed in terms of accuracy and computation time when four common hand gestures (flexion, extension, fist and rest) tested both in the training and validation data sets. Additionally, the robust performance was obtained by reducing an optimal number of EMG channels. After completing the extensive tests, the present algorithm showed the same excellent classification performance of EMG signals obtained from subjects even including those not trained before. It was concluded that the suggested algorithm offers a powerful classification solutions for hand gesture movements in real-time applications and promises to be a viable methodology for future expansions.
- 1 April 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-4
Hand gesture movement classification based on dynamically structured neural network
As one of the complex biomedical signals, electromyography represents the electrical activation of the muscle providing useful information about gesture movements. A dynamically structured neural network is proposed to be used as an efficient tool to classify such movement patterns through some EMG signals generated from hand/arm movements with unknown complexity in real-time. The dynamically structured neural network is trained to establish input/output mapping while learning its optimal structure. The proposed architecture's performance is compared to various size statically structured neural network configurations. Over two folds of improvements are observed in terms of accuracy and computation time when four common hand gestures (flexion, extension, fist and rest) tested both in the training and validation data sets. Additionally, the robust performance was obtained by reducing an optimal number of EMG channels. After completing the extensive tests, the present algorithm showed the same excellent classification performance of EMG signals obtained from subjects even including those not trained before. It was concluded that the suggested algorithm offers a powerful classification solutions for hand gesture movements in real-time applications and promises to be a viable methodology for future expansions.
- 1 November 2015
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
Designing and modeling of a point absorber wave energy converter with hydraulic power take-off unit
In this project, the primary aim is to produce optimum parameters for electric power generation via renewable sea wave energy for the Turkish sea coastlines. The modular system is composed of wave actuation mechanism, hydraulic system and generator. This system is used to model and compute the optimal parameters but also monitor the Turkish coastline characteristics. A hydrodynamic model based optimum PTO drives the generator that are further connected to other similar units to construct a wave energy farm. A test bench is created to mimic the operation of wave actuation in lab environment. This unit drives hydraulic system that can generate mechanical power to excite a generator shaft. Optimal wave actuation mechanism parameters suitable to our coastlines have been calculated. With these aims, the system designed on the basis of the mechanism that based on point absorber buoy. Initial design and hydrodynamic simulations in MATLAB/Simulink is given.
- 1 November 2015
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
Designing and modeling of a point absorber wave energy converter with hydraulic power take-off unit
In this project, the primary aim is to produce optimum parameters for electric power generation via renewable sea wave energy for the Turkish sea coastlines. The modular system is composed of wave actuation mechanism, hydraulic system and generator. This system is used to model and compute the optimal parameters but also monitor the Turkish coastline characteristics. A hydrodynamic model based optimum PTO drives the generator that are further connected to other similar units to construct a wave energy farm. A test bench is created to mimic the operation of wave actuation in lab environment. This unit drives hydraulic system that can generate mechanical power to excite a generator shaft. Optimal wave actuation mechanism parameters suitable to our coastlines have been calculated. With these aims, the system designed on the basis of the mechanism that based on point absorber buoy. Initial design and hydrodynamic simulations in MATLAB/Simulink is given.