C. K. Yogesh
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
C. K. Yogesh has a total of 19 co-authors. Here are the top co-authors by number of shared publications:
- S. Vatchala
School of Computer Science and Engineering, Vellore Institute of Technology, Chennai, India
2 shared publications Aashish Vinod Arul
School of Computer Science and Engineering, Vellore Institute of Technology, Chennai, India
1 shared publicationAbdul Hamid Adom
1 shared publicationAnirudh Vinodh
School of Computer Science and Engineering, Vellore Institute of Technology, Chennai, India
1 shared publication- Arpan Singh
Vellore Institute of Technology Chennai Tamil Nadu India
1 shared publication Chawki Berkai
1 shared publication
Latest Publications
- Open Access
- 13 March 2026
- Published by Wiley in Engineering Reports
- Vol. 8 (3)
Human Eye Visualization With an Attention Mechanism: Enhancing Flower Detection and Classification With Visual Attention and Photometric Principles Inspired by Human Vision
In this paper, a visual attention network (VANet) that draws inspiration from the complex workings of the human visual system is presented. The human eye possesses an extraordinary capacity to selectively concentrate on particular features within a scene while adjusting to varying light conditions. The proposed model replicates this ability by focusing on essential spatial and channel‐specific aspects of the input image. This is accomplished through an advanced architecture that incorporates attention mechanisms within residual learning blocks, which enhances the model's aptitude for recognizing crucial details while lessening the impact of irrelevant information. To further increase the effectiveness of the model, particularly in situations with changing illumination, a Luminance Enhancement Module is integrated. This module is intended to dynamically modify and improve brightness and contrast, enabling the network to operate efficiently in different environmental settings. By emulating how humans respond to variations in light, we strive to develop a more resilient and reliable visual perception system. This cutting‐edge model not only seeks to mimic the mechanical functions associated with human sight but also improves the vision tasks performed by machines, thereby expanding their use in fields like robotics and automated systems. Furthermore, we introduce a novel method that combines the functionalities of VANet with the human visual system with attention mechanism (HEVAM) for the detection and classification of flowers with remarkable accuracy. The synergy of these approaches results in precise image detection and classification, establishing a new benchmark for performance in botanical recognition tasks. These innovations propel the advancement of improved machine vision systems, enabling them to better understand and interact with their surroundings. The system proposed in this study demonstrated an accuracy of above 95% in the Flower Recognition (5 classes), Flowers dataset (16 classes), Oxford‐17 Flower (17 classes), and Oxford‐102 Flower (102 classes), surpassing the benchmark results.
- 13 March 2026
- Published by Wiley in Engineering Reports
- Vol. 8 (3)
Open AccessHuman Eye Visualization With an Attention Mechanism: Enhancing Flower Detection and Classification With Visual Attention and Photometric Principles Inspired by Human Vision
In this paper, a visual attention network (VANet) that draws inspiration from the complex workings of the human visual system is presented. The human eye possesses an extraordinary capacity to selectively concentrate on particular features within a scene while adjusting to varying light conditions. The proposed model replicates this ability by focusing on essential spatial and channel‐specific aspects of the input image. This is accomplished through an advanced architecture that incorporates attention mechanisms within residual learning blocks, which enhances the model's aptitude for recognizing crucial details while lessening the impact of irrelevant information. To further increase the effectiveness of the model, particularly in situations with changing illumination, a Luminance Enhancement Module is integrated. This module is intended to dynamically modify and improve brightness and contrast, enabling the network to operate efficiently in different environmental settings. By emulating how humans respond to variations in light, we strive to develop a more resilient and reliable visual perception system. This cutting‐edge model not only seeks to mimic the mechanical functions associated with human sight but also improves the vision tasks performed by machines, thereby expanding their use in fields like robotics and automated systems. Furthermore, we introduce a novel method that combines the functionalities of VANet with the human visual system with attention mechanism (HEVAM) for the detection and classification of flowers with remarkable accuracy. The synergy of these approaches results in precise image detection and classification, establishing a new benchmark for performance in botanical recognition tasks. These innovations propel the advancement of improved machine vision systems, enabling them to better understand and interact with their surroundings. The system proposed in this study demonstrated an accuracy of above 95% in the Flower Recognition (5 classes), Flowers dataset (16 classes), Oxford‐17 Flower (17 classes), and Oxford‐102 Flower (102 classes), surpassing the benchmark results.
- Open Access
- 21 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 72442-72455
Implementation for Lightweight Deep Learning for Anomaly Detection and Denoising on Gravitational Waves
As gravitational wave astronomy has advanced, the need for effective and quick signal processing has never been more critical. New detectors such as Laser Interferometer Gravitational-Wave Observatory (LIGO) produces huge volumes of data, which poses a significant challenge to identify genuine astrophysical events amidst transient noise. With this in view, this study proposes a new, efficient deep-learning architecture designed to process gravitational wave signals. Motivated by the WaveNet model, our method uses dilated convolutions to precisely model long-term dependencies in the data ensuring that subtle characteristics are captured. In addition, higher-order recurrent layers like Long Short-Term Memory(LSTM) networks are also used to precisely model temporal characteristics so that accuracy is preserved with enhanced anomaly detection and noise removal. Our experimental validation shows that this method has achieved a Root Mean Square Error (RMSE) of 0.00228, a Residual Signal Ratio (RSR) of 0.08798, a Peak Signal-to-Noise Ratio (PSNR) of 33.34 dB, and an average volume error of 0.03965. These performance metrics show the framework’s ability to operate in real time with minimal computational overhead is a key requirement given the massive datasets involved in gravitational wave science. Also, by an explicit consideration of initial and boundary conditions, our scheme exhibits not only stability for the current missions but also scalability for future detection missions, like Laser Interferometer Space Antenna (LISA), allowing these missions to enhance their sensitivity and reliability. Our study demonstrates the development of state-of-the-art deep learning methods to surmount the specific obstacles concerning gravitational wave detection, paving the way for real-time processing of astrophysical data and an improved understanding of the Universe.
- 21 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 72442-72455
Open AccessImplementation for Lightweight Deep Learning for Anomaly Detection and Denoising on Gravitational Waves
As gravitational wave astronomy has advanced, the need for effective and quick signal processing has never been more critical. New detectors such as Laser Interferometer Gravitational-Wave Observatory (LIGO) produces huge volumes of data, which poses a significant challenge to identify genuine astrophysical events amidst transient noise. With this in view, this study proposes a new, efficient deep-learning architecture designed to process gravitational wave signals. Motivated by the WaveNet model, our method uses dilated convolutions to precisely model long-term dependencies in the data ensuring that subtle characteristics are captured. In addition, higher-order recurrent layers like Long Short-Term Memory(LSTM) networks are also used to precisely model temporal characteristics so that accuracy is preserved with enhanced anomaly detection and noise removal. Our experimental validation shows that this method has achieved a Root Mean Square Error (RMSE) of 0.00228, a Residual Signal Ratio (RSR) of 0.08798, a Peak Signal-to-Noise Ratio (PSNR) of 33.34 dB, and an average volume error of 0.03965. These performance metrics show the framework’s ability to operate in real time with minimal computational overhead is a key requirement given the massive datasets involved in gravitational wave science. Also, by an explicit consideration of initial and boundary conditions, our scheme exhibits not only stability for the current missions but also scalability for future detection missions, like Laser Interferometer Space Antenna (LISA), allowing these missions to enhance their sensitivity and reliability. Our study demonstrates the development of state-of-the-art deep learning methods to surmount the specific obstacles concerning gravitational wave detection, paving the way for real-time processing of astrophysical data and an improved understanding of the Universe.
- Open Access
- 27 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 28029-28041
Multi-Modal Biometric Authentication: Leveraging Shared Layer Architectures for Enhanced Security
In this study, we introduce a novel multi-modal biometric authentication system that integrates facial, vocal, and signature data to enhance security measures. Utilizing a combination of Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), our model architecture uniquely incorporates dual shared layers alongside modality-specific enhancements for comprehensive feature extraction. The system undergoes rigorous training with a joint loss function, optimizing for accuracy across diverse biometric inputs. Feature-level fusion via Principal Component Analysis (PCA) and classification through Gradient Boosting Machines (GBM) further refine the authentication process. Our approach demonstrates significant improvements in authentication accuracy and robustness, paving the way for advanced secure identity verification solutions.
- 27 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 28029-28041
Open AccessMulti-Modal Biometric Authentication: Leveraging Shared Layer Architectures for Enhanced Security
In this study, we introduce a novel multi-modal biometric authentication system that integrates facial, vocal, and signature data to enhance security measures. Utilizing a combination of Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), our model architecture uniquely incorporates dual shared layers alongside modality-specific enhancements for comprehensive feature extraction. The system undergoes rigorous training with a joint loss function, optimizing for accuracy across diverse biometric inputs. Feature-level fusion via Principal Component Analysis (PCA) and classification through Gradient Boosting Machines (GBM) further refine the authentication process. Our approach demonstrates significant improvements in authentication accuracy and robustness, paving the way for advanced secure identity verification solutions.
- 1 March 2017
- Published by Elsevier in Expert Systems with Applications
- Vol. 69, 149-158
A new hybrid PSO assisted biogeography-based optimization for emotion and stress recognition from speech signal
- 1 March 2017
- Published by Elsevier in Expert Systems with Applications
- Vol. 69, 149-158
A new hybrid PSO assisted biogeography-based optimization for emotion and stress recognition from speech signal
- 1 December 2014
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 916-921
Fractal feature based detection of muscular and ocular artifacts in EEG signals
Electroencephalogram (EEG) is used to measure the bioelectric potential on the brain scalp. The recorded EEG signal can have different types of artifacts and the interpretation of a noisy EEG signal is difficult. In this research work, a simple method is proposed to minimize the artifacts present in the EEG signals recorded while perceiving a pure tone. The recorded EEG signal can contain artifacts, such as movement artifacts, muscle contraction artifacts and saturation artifacts. In the proposed method, fractal dimension based features with different interval length and time-domain based energy features were extracted from the EEG signals with and without simulated noise. Using the extracted features, neural network models were developed to classify the EEG signal as a normal or a noisy signal. Further, the performance of the model is also evaluated in terms of classification rate. From the results, it is observed that the neural network model developed with the combined fractal dimension features of interval length 2,3,4,5 and 6 with frame size 128 has the highest classification accuracy of 95.5%.
- 1 December 2014
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 916-921
Fractal feature based detection of muscular and ocular artifacts in EEG signals
Electroencephalogram (EEG) is used to measure the bioelectric potential on the brain scalp. The recorded EEG signal can have different types of artifacts and the interpretation of a noisy EEG signal is difficult. In this research work, a simple method is proposed to minimize the artifacts present in the EEG signals recorded while perceiving a pure tone. The recorded EEG signal can contain artifacts, such as movement artifacts, muscle contraction artifacts and saturation artifacts. In the proposed method, fractal dimension based features with different interval length and time-domain based energy features were extracted from the EEG signals with and without simulated noise. Using the extracted features, neural network models were developed to classify the EEG signal as a normal or a noisy signal. Further, the performance of the model is also evaluated in terms of classification rate. From the results, it is observed that the neural network model developed with the combined fractal dimension features of interval length 2,3,4,5 and 6 with frame size 128 has the highest classification accuracy of 95.5%.