Debangshu Dey
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Debangshu Dey has a total of 42 co-authors. Here are the top co-authors by number of shared publications:
- Biswendu Chatterjee
Electrical Engineering Department, Jadavpur University, Kolkata, India
16 shared publications - Sugata Munshi
Jadavpur University
11 shared publications - Sovan Dalai
Electrical Engineering Department, Jadavpur University, Kolkata, India
9 shared publications Sivaji Chakravorti
Jadavpur University
8 shared publications- Biswarup Ganguly
Department of Electrical Engineering, National Institute of Technology Silchar, Silchar, Assam, India
7 shared publications - Antara Ghosh
Jadavpur University
4 shared publications
Latest Publications
- 4 November 2026
- Published by IOP Publishing in Biomedical Physics & Engineering Express
- Vol. 12 (6) , 065001
Learning discriminative time-frequency representations for EEG segment-level seizure-state classification with channel selection
Electroencephalography (EEG)-based seizure monitoring supports the diagnosis and management of epilepsy; however, high-dimensional multichannel EEG recordings impose substantial computational burden and limit suitability for portable monitoring systems. This paper proposes a reduced-channel framework for segment-level classification of EEG into ictal, preictal and interictal states. It integrates attention-guided EEG channel selection, high-resolution time-frequency analysis via the time reassigned multisynchrosqueezing Transform and concentration of frequency and time, and a hybrid CNN-BiLSTM classifier. The framework is evaluated on the CHB-MIT and Siena Scalp EEG databases under two complementary protocols: a stratified 70/30 holdout split and rigorous leave-one-subject-out (LOSO) cross-validation to assess subject independent generalization. Under holdout evaluation using a single EEG channel, the proposed framework achieves an accuracy of 99.28 ± 0.63 % , sensitivity of 99.18 % , specificity of 99.64 % , and Matthews correlation coefficient (MCC) of 0.97 on CHB-MIT, and an accuracy of 98.75 ± 0.74 % , sensitivity of 98.77 % , specificity of 99.38 % , and MCCs of 0.97 on Siena. LOSO cross-validation demonstrates subject independent generalization, exceeding 91 % accuracy on both databases. These results demonstrate that the framework achieves an effective balance among segment-level classification accuracy, computational efficiency and reduced-channel operation. Moreover, applicability for portable and wearable EEG-based seizure-state monitoring applications is investigated.
- 4 November 2026
- Published by IOP Publishing in Biomedical Physics & Engineering Express
- Vol. 12 (6) , 065001
Learning discriminative time-frequency representations for EEG segment-level seizure-state classification with channel selection
Electroencephalography (EEG)-based seizure monitoring supports the diagnosis and management of epilepsy; however, high-dimensional multichannel EEG recordings impose substantial computational burden and limit suitability for portable monitoring systems. This paper proposes a reduced-channel framework for segment-level classification of EEG into ictal, preictal and interictal states. It integrates attention-guided EEG channel selection, high-resolution time-frequency analysis via the time reassigned multisynchrosqueezing Transform and concentration of frequency and time, and a hybrid CNN-BiLSTM classifier. The framework is evaluated on the CHB-MIT and Siena Scalp EEG databases under two complementary protocols: a stratified 70/30 holdout split and rigorous leave-one-subject-out (LOSO) cross-validation to assess subject independent generalization. Under holdout evaluation using a single EEG channel, the proposed framework achieves an accuracy of 99.28 ± 0.63 % , sensitivity of 99.18 % , specificity of 99.64 % , and Matthews correlation coefficient (MCC) of 0.97 on CHB-MIT, and an accuracy of 98.75 ± 0.74 % , sensitivity of 98.77 % , specificity of 99.38 % , and MCCs of 0.97 on Siena. LOSO cross-validation demonstrates subject independent generalization, exceeding 91 % accuracy on both databases. These results demonstrate that the framework achieves an effective balance among segment-level classification accuracy, computational efficiency and reduced-channel operation. Moreover, applicability for portable and wearable EEG-based seizure-state monitoring applications is investigated.
- 1 August 2026
- Published by IOP Publishing in Engineering Research Express
- Vol. 8 (15) , 155315
Hexa-SymmLTP-based leakage current analysis for MOSA contamination level estimation
Surface contamination of metal oxide surge arresters (MOSAs) is a critical factor affecting the reliability and stability of power grids. While leakage current (LC) monitoring is widely adopted for MOSA condition assessment, existing methods based on global statistical or transform-domain features often fail to capture the subtle localized waveform changes associated with early-stage surface degradation. This work proposes an advanced framework for classifying surface contamination severity in MOSAs directly from time-domain LC signals, without requiring any domain transformation. The proposed Hexa-symmetric local ternary pattern (Hexa-SymmLTP) descriptor is introduced to extract texture and structural pattern features that encode local waveform transitions and asymmetric variations inherent in LC signals under different contamination levels. A Chi-square test is applied for feature selection to retain the most discriminative information, and the separability of the resulting feature space is validated through principal component analysis and t-distributed stochastic neighbor embedding. Classification is performed using the k-nearest neighbor algorithm, chosen for its ability to exploit local pattern similarity in the proposed feature space. Experimental results show that the proposed framework achieves a classification accuracy of 98.3% across five contamination levels while outperforming several established methods. The results demonstrate that direct time-domain feature extraction using Hexa-SymmLTP provides a robust, computationally efficient, and practically viable solution for MOSA contamination monitoring in real-world power system environments.
- 1 August 2026
- Published by IOP Publishing in Engineering Research Express
- Vol. 8 (15) , 155315
Hexa-SymmLTP-based leakage current analysis for MOSA contamination level estimation
Surface contamination of metal oxide surge arresters (MOSAs) is a critical factor affecting the reliability and stability of power grids. While leakage current (LC) monitoring is widely adopted for MOSA condition assessment, existing methods based on global statistical or transform-domain features often fail to capture the subtle localized waveform changes associated with early-stage surface degradation. This work proposes an advanced framework for classifying surface contamination severity in MOSAs directly from time-domain LC signals, without requiring any domain transformation. The proposed Hexa-symmetric local ternary pattern (Hexa-SymmLTP) descriptor is introduced to extract texture and structural pattern features that encode local waveform transitions and asymmetric variations inherent in LC signals under different contamination levels. A Chi-square test is applied for feature selection to retain the most discriminative information, and the separability of the resulting feature space is validated through principal component analysis and t-distributed stochastic neighbor embedding. Classification is performed using the k-nearest neighbor algorithm, chosen for its ability to exploit local pattern similarity in the proposed feature space. Experimental results show that the proposed framework achieves a classification accuracy of 98.3% across five contamination levels while outperforming several established methods. The results demonstrate that direct time-domain feature extraction using Hexa-SymmLTP provides a robust, computationally efficient, and practically viable solution for MOSA contamination monitoring in real-world power system environments.
- 11 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Dielectrics and Electrical Insulation
- Vol. PP (10709878) , 1
Improved Localization of Partial Discharge Sources Using Powered Superlets
Reliable detection and localization of partial discharges (PDs) in transformers are critical for ensuring insulation integrity and extending equipment lifetime. Conventional acoustic and time–frequency (TF) analysis methods such as the Short-Time Fourier Transform, Continuous Wavelet Transform, Stockwell Transform often face limitations under low signal-to-noise ratio (SNR). This paper introduces the Powered Superlet Transform (PST), a novel TF representation that employs a powered growth schedule for the number of cycles in wavelet sets, resulting in increased inter location distance in TF plots and thus increased accuracy in localization. The PST is applied to acoustic signals measured from multiple PD sources in a controlled transformer tank setup. Subsequently, feature extraction through TF methods and classification through Support Vector Machine is performed. Comparative analysis with established TF techniques and acoustic localization methods demonstrates that PST is the most effective technique in terms of accuracy while both PST and Adaptive Superlet Transform (ASLT) provides superior noise robustness in terms of reduced localization error. Experimental results under varying SNR conditions show that PST can reliably localize PD sources with accuracy of 78.8% and approximate RMS localization errors 11.857 cm at SNRs as low as 2.5 dB, while established methods are less accurate. These findings highlight PST as a promising tool for advanced acoustic PD monitoring and condition assessment of high-voltage transformers.
- 11 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Dielectrics and Electrical Insulation
- Vol. PP (10709878) , 1
Improved Localization of Partial Discharge Sources Using Powered Superlets
Reliable detection and localization of partial discharges (PDs) in transformers are critical for ensuring insulation integrity and extending equipment lifetime. Conventional acoustic and time–frequency (TF) analysis methods such as the Short-Time Fourier Transform, Continuous Wavelet Transform, Stockwell Transform often face limitations under low signal-to-noise ratio (SNR). This paper introduces the Powered Superlet Transform (PST), a novel TF representation that employs a powered growth schedule for the number of cycles in wavelet sets, resulting in increased inter location distance in TF plots and thus increased accuracy in localization. The PST is applied to acoustic signals measured from multiple PD sources in a controlled transformer tank setup. Subsequently, feature extraction through TF methods and classification through Support Vector Machine is performed. Comparative analysis with established TF techniques and acoustic localization methods demonstrates that PST is the most effective technique in terms of accuracy while both PST and Adaptive Superlet Transform (ASLT) provides superior noise robustness in terms of reduced localization error. Experimental results under varying SNR conditions show that PST can reliably localize PD sources with accuracy of 78.8% and approximate RMS localization errors 11.857 cm at SNRs as low as 2.5 dB, while established methods are less accurate. These findings highlight PST as a promising tool for advanced acoustic PD monitoring and condition assessment of high-voltage transformers.
- Open Access
- 27 April 2026
- Published by Elsevier in SSRN Electronic Journal
Acoustic Partial Discharge Localization Using Enhanced Time–Frequency Representation and Multi-Sensor Deep Learning
This paper presents an advanced deep learning framework for accurate localization of incipient partial discharge (PD) sources in a power equipment based on acoustic emission. In this regard, an experimental setup was configured to emulate single- and multiple-location partial-discharge activity. Five acoustic sensors (at five different wall sides) were employed to capture acoustic emission from PD sources during PD activity. The acoustic signals were analyzed in joint time-frequency space, and their representation was improved through the Tuned Multisynchrosqueezing Transform. A multi-sensor fusion deep neural network was configured to process time-frequency spectrograms corresponding to each acoustic sensor PD data simultaneously and predict PD source location. The deep neural network was trained on a large database and tested utilizing a cross-dataset learning strategy. Ablation study revealed superior performance of the proposed framework under a noisy background, suggesting it’s potential in real-life PD localization in power equipment.
- 27 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessAcoustic Partial Discharge Localization Using Enhanced Time–Frequency Representation and Multi-Sensor Deep Learning
This paper presents an advanced deep learning framework for accurate localization of incipient partial discharge (PD) sources in a power equipment based on acoustic emission. In this regard, an experimental setup was configured to emulate single- and multiple-location partial-discharge activity. Five acoustic sensors (at five different wall sides) were employed to capture acoustic emission from PD sources during PD activity. The acoustic signals were analyzed in joint time-frequency space, and their representation was improved through the Tuned Multisynchrosqueezing Transform. A multi-sensor fusion deep neural network was configured to process time-frequency spectrograms corresponding to each acoustic sensor PD data simultaneously and predict PD source location. The deep neural network was trained on a large database and tested utilizing a cross-dataset learning strategy. Ablation study revealed superior performance of the proposed framework under a noisy background, suggesting it’s potential in real-life PD localization in power equipment.
- Open Access
- 27 April 2026
- Published by Elsevier in SSRN Electronic Journal
Acoustic Partial Discharge Localization Using Enhanced Time–Frequency Representation and Multi-Sensor Deep Learning
This paper presents an advanced deep learning framework for accurate localization of incipient partial discharge (PD) sources in a power equipment based on acoustic emission. In this regard, an experimental setup was configured to emulate single- and multiple-location partial-discharge activity. Five acoustic sensors (at five different wall sides) were employed to capture acoustic emission from PD sources during PD activity. The acoustic signals were analyzed in joint time-frequency space, and their representation was improved through the Tuned Multisynchrosqueezing Transform. A multi-sensor fusion deep neural network was configured to process time-frequency spectrograms corresponding to each acoustic sensor PD data simultaneously and predict PD source location. The deep neural network was trained on a large database and tested utilizing a cross-dataset learning strategy. Ablation study revealed superior performance of the proposed framework under a noisy background, suggesting it’s potential in real-life PD localization in power equipment.
- 27 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessAcoustic Partial Discharge Localization Using Enhanced Time–Frequency Representation and Multi-Sensor Deep Learning
This paper presents an advanced deep learning framework for accurate localization of incipient partial discharge (PD) sources in a power equipment based on acoustic emission. In this regard, an experimental setup was configured to emulate single- and multiple-location partial-discharge activity. Five acoustic sensors (at five different wall sides) were employed to capture acoustic emission from PD sources during PD activity. The acoustic signals were analyzed in joint time-frequency space, and their representation was improved through the Tuned Multisynchrosqueezing Transform. A multi-sensor fusion deep neural network was configured to process time-frequency spectrograms corresponding to each acoustic sensor PD data simultaneously and predict PD source location. The deep neural network was trained on a large database and tested utilizing a cross-dataset learning strategy. Ablation study revealed superior performance of the proposed framework under a noisy background, suggesting it’s potential in real-life PD localization in power equipment.
- 23 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (3) , 4860-4871
Pose Under Covers: In-Bed Human Pose Estimation Using Multisensor Image Fusion and Graph Convolutional Network
Advancements in computer vision provide critical insights for various medical use cases, including human behavioral monitoring. Challenges of occlusions and the scarcity of labelled datasets, especially for privacy preserving modalities like long-wave infrared (LWIR) imaging, necessitate innovative approaches for in-bed pose estimation. This study introduces a robust deep learning framework that incorporates novel multimodal feature fusion methodology, generative model, and graph convolutional network (GCN) for in-bed pose estimation. The Simultaneously-collected multimodal Lying Pose (SLP) dataset has been used for performance evaluation of this reported methodology where three modalities, including LWIR, depth, and pressure images, are used to train the model. This model effectively handles modality gaps and adapts to real-world occluded conditions. This work underscores the potential of leveraging multimodal data from privacy-preserving image domains and ability of GCNs to capture spatial joint dependencies for human pose estimation. Extensive evaluations demonstrate the model’s ability to achieve high accuracy of 98.46% at PCKh@0.5 across different cover conditions, outperforming state-of-the-art in-bed pose estimation models on SLP dataset.
- 23 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (3) , 4860-4871
Pose Under Covers: In-Bed Human Pose Estimation Using Multisensor Image Fusion and Graph Convolutional Network
Advancements in computer vision provide critical insights for various medical use cases, including human behavioral monitoring. Challenges of occlusions and the scarcity of labelled datasets, especially for privacy preserving modalities like long-wave infrared (LWIR) imaging, necessitate innovative approaches for in-bed pose estimation. This study introduces a robust deep learning framework that incorporates novel multimodal feature fusion methodology, generative model, and graph convolutional network (GCN) for in-bed pose estimation. The Simultaneously-collected multimodal Lying Pose (SLP) dataset has been used for performance evaluation of this reported methodology where three modalities, including LWIR, depth, and pressure images, are used to train the model. This model effectively handles modality gaps and adapts to real-world occluded conditions. This work underscores the potential of leveraging multimodal data from privacy-preserving image domains and ability of GCNs to capture spatial joint dependencies for human pose estimation. Extensive evaluations demonstrate the model’s ability to achieve high accuracy of 98.46% at PCKh@0.5 across different cover conditions, outperforming state-of-the-art in-bed pose estimation models on SLP dataset.
- 20 November 2025
- Published by Elsevier in Electric Power Systems Research
- Vol. 253
Multisynchrosqueezing transform aided transfer learning based approach for diagnosis of single and mixed power quality events
- 20 November 2025
- Published by Elsevier in Electric Power Systems Research
- Vol. 253
Multisynchrosqueezing transform aided transfer learning based approach for diagnosis of single and mixed power quality events
- 10 November 2025
- Published by Springer Nature in Signal, Image and Video Processing
- Vol. 19 (16) , 1-11
Cross-correlation aided feature extraction for automated diagnosis of phonocardiogram signals using attention based Bi-LSTM network
- 10 November 2025
- Published by Springer Nature in Signal, Image and Video Processing
- Vol. 19 (16) , 1-11
Cross-correlation aided feature extraction for automated diagnosis of phonocardiogram signals using attention based Bi-LSTM network
- Top Cited Papers
- 13 June 2025
- Published by Elsevier in Electric Power Systems Research
- Vol. 248
Recent trends and open challenges in acoustic partial discharge signal denoising techniques: A review
- 13 June 2025
- Published by Elsevier in Electric Power Systems Research
- Vol. 248
Top Cited PapersRecent trends and open challenges in acoustic partial discharge signal denoising techniques: A review
- 27 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 26 (4) , 4517-4527
An Attention Deep Learning Framework-Based Drowsiness Detection Model for Intelligent Transportation System
Drivers’ drowsiness has been considered one of the prime reasons for accidents and road fatalities. Drowsiness may be caused by sleep disorders resulting in unusual mental and health conditions that have detrimental effects on human lives. This article aims to present an attention deep learning (DL) framework for drivers’ drowsiness monitoring for an intelligent transportation system. The proposed imaging system, comprising an Infrared-Cut camera embedded in a microcomputer, has been employed for capturing both day and night mode images for automated detection of drivers’ drowsiness. The frames captured are preprocessed and fed to the proposed attention DL framework based on “you only look once” version 3 (YOLOv3) for eye region detection followed by eye state classification and interpretation. Feature extraction has been carried out via a convolutional neural network module, and multiscale fusion along with the non-maximum suppression method has been applied to detect and classify the eye region of the drivers for monitoring drowsiness. Moreover, the eye region has been interpreted via a classification activation map using the proposed attention module. Experimental evaluations reveal the efficacy of the proposed system on our acquired dataset and two benchmark datasets. The proposed drowsiness detection device and system can possess good potential by increasing safety in an advanced driver assistance system.
- 27 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 26 (4) , 4517-4527
An Attention Deep Learning Framework-Based Drowsiness Detection Model for Intelligent Transportation System
Drivers’ drowsiness has been considered one of the prime reasons for accidents and road fatalities. Drowsiness may be caused by sleep disorders resulting in unusual mental and health conditions that have detrimental effects on human lives. This article aims to present an attention deep learning (DL) framework for drivers’ drowsiness monitoring for an intelligent transportation system. The proposed imaging system, comprising an Infrared-Cut camera embedded in a microcomputer, has been employed for capturing both day and night mode images for automated detection of drivers’ drowsiness. The frames captured are preprocessed and fed to the proposed attention DL framework based on “you only look once” version 3 (YOLOv3) for eye region detection followed by eye state classification and interpretation. Feature extraction has been carried out via a convolutional neural network module, and multiscale fusion along with the non-maximum suppression method has been applied to detect and classify the eye region of the drivers for monitoring drowsiness. Moreover, the eye region has been interpreted via a classification activation map using the proposed attention module. Experimental evaluations reveal the efficacy of the proposed system on our acquired dataset and two benchmark datasets. The proposed drowsiness detection device and system can possess good potential by increasing safety in an advanced driver assistance system.