Rajdeep Chatterjee

Dr. Rajdeep Chatterjee

orcid

Research Focus & Metrics

Academic Career Length

13Years (2014-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
10 February 2021 in IEEE Transactions on Consumer Electronics
Verified
Journal Article
Closed access
Rajdeep ChatterjeeRajdeep Chatterjee
Rajdeep Chatterjee
SMSaptarshi Mazumdar
Saptarshi Mazumdar
R. Simon SherrattR. Simon Sherratt
R. Simon Sherratt
RHRohit Halder
Rohit Halder
TMTanmoy Maitra
Tanmoy Maitra
DGDebasis Giri
Debasis Giri
Artificial Intelligence (AI) based Speech Emotion Recognition (SER) has been widely used in the consumer field for control of smart home personal assistants, with many such devices on the market. However, with the increase in computational power, connectivity, and the need to enable people to live in the home for longer though the use of technology, then smart home assistants that could detect human emotion will improve the communication between a user and the assistant enabling the assistant of offer more productive feedback. Thus, the aim of this work is to analyze emotional states in speech and propose a suitable method considering performance verses complexity for deployment in Consumer Electronics home products, and to present a practical live demonstration of the research. In this paper, a comprehensive approach has been introduced for the human speech-based emotion analysis. The 1-D convolutional neural network (CNN) has been implemented to learn and classify the emotions associated with human speech. The paper has been implemented on the standard datasets (emotion classification) Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS) and Toronto Emotional Speech Set database (TESS) (Young and Old). The proposed approach gives 90.48%, 95.79% and94.47% classification accuracies in the aforementioned datasets. We conclude that the 1-D CNN classification models used in speaker-independent experiments are highly effective in the automatic prediction of emotion and are ideal for deployment in smart home assistants to detect emotion.

1 September 2019 in Future Generation Computer Systems
Verified
Journal Article
Closed access
Rajdeep ChatterjeeRajdeep Chatterjee
Rajdeep Chatterjee
TMTanmoy Maitra
Tanmoy Maitra
SK Hafizul IslamSK Hafizul Islam
SK Hafizul Islam
Mohammad Mehedi HassanMohammad Mehedi Hassan
Mohammad Mehedi Hassan
AAAtif Alamri
Atif Alamri
Giancarlo FortinoGiancarlo Fortino
Giancarlo Fortino

27 April 2023 in IEEE Systems Journal
Verified
Journal Article
Closed access
DGDibyajyoti Guha
Dibyajyoti Guha
Rajdeep ChatterjeeRajdeep Chatterjee
Rajdeep Chatterjee
BSBiplab Sikdar
Biplab Sikdar
This article proposes a deep generative model for anomaly detection in unsupervised power grid data. One-class classifier-based methods suffer from performance degradation when training data contain anomalous samples. Due to the temporal characteristics in most of the power grid datasets, we explore a long short-term memory-variational autoencoder-based deep generative model that can tolerate the moderate presence of anomalous data during training instead of standard data. This work demonstrates the advantage of reconstruction-based methods over clustering-based methods. As part of the reparameterization of the latent layer, a method is proposed by employing wavelet decomposition of the wavelet coefficients found from the high and medium frequency representations of the input time-series data. For further improvement, we have incorporated a $\log \text{cos}h$ -based cost function instead of the traditional consideration of the $L_{2}$ norm-based cost function. The numerical results demonstrate an improvement of performance metrics, such as AUC by 0.1–0.2 of our method over other benchmark methods. The transient stability threshold ( $\delta$ ) is an important system parameter in the performance assessment of power grid systems. Through time domain simulations, it has been shown that $\delta =0.3$ obtains optimal accuracy for transient stability assessment in the IEEE NewEngland-39 bus.

12 June 2020 in SN Computer Science
Verified
Journal Article
Open access
PDF file
RHRohit Halder
Rohit Halder
Rajdeep ChatterjeeRajdeep Chatterjee
Rajdeep Chatterjee
In this paper, a lightweight computational model has been introduced for the better classification of violent and non-violent activities. Nowadays, the occurrences of violent activities get increased in public places. It has various social and economic reasons behind the growth of violent actions. Therefore, the government agencies and public administrators need to check such incidents using smart surveillance. Deep learning-based an efficient violent activity detection model can help the authorities in detecting a violent activity in real-time. The evaluated results can be henceforth sent to store and analyze the captured video to automate the crime monitoring system. Convolutional Neural Network-based Bidirectional LSTM has been used to detect violent activities and also compared with other existing approaches. Our proposed model gives 99.27%, 100% and 98.64% classification accuracies for the widely used standard Hockey Fights, Movies and Violent-Flows video datasets, respectively.

15 September 2016
Verified
Conference Paper
Closed access
Rajdeep ChatterjeeRajdeep Chatterjee
Rajdeep Chatterjee
TBTathagata Bandyopadhyay
Tathagata Bandyopadhyay
DSDebarshi Kumar Sanyal
Debarshi Kumar Sanyal
This paper examines the quality of feature set obtained from Wavelet based Energy-entropy with variation of scale and wavelet type. Here motor imagery of left-right hand movement classification problem has been studied. Elliptic bandpass filters are used to discard unwanted signals and also to extract alpha & beta rhythms. We have implemented wavelet-based energy-entropy with three level of decomposition in combination with ten wavelet types (Daubechies). We want to identify the best pair of level of decomposition and wavelet type for EEG based motor imagery classifications. We have verified our study with three classifiers- Naïve Bayes, Multi layered Perceptron and Support Vector Machine. The classifiers performance for best wavelet decomposition level is analyzed using evaluation metrics such as accuracy, F-measure and area under ROC.

12 April 2023 in IEEE Access
Verified
Journal Article
Open access
Rajdeep ChatterjeeRajdeep Chatterjee
Rajdeep Chatterjee
ACAnkita Chatterjee
Ankita Chatterjee
MPManas Ranjan Pradhan
Manas Ranjan Pradhan
BABiswaranjan Acharya
Biswaranjan Acharya
Tanupriya ChoudhuryTanupriya Choudhury
Tanupriya Choudhury
Violence, in any form, is a disgrace to our civilized world. Nevertheless, even in modern days, violence is an integral part of our society and causes the deaths of many innocent lives. One of the conventional means of violence is using a firearm. Firearm-related death is currently a global phenomenon. It is a threat to society and a challenge to law enforcement agencies. A significant portion of such crimes happens in semi-urban areas or cities. Nowadays, CCTV-based surveillance is widely used by governments and private organizations for monitoring and prevention. However, human-based monitoring requires a significant amount of person-hours as a resource and is prone to mistakes. On the other hand, automated smart surveillance for violent activities is more suitable for scale and reliability. The paper’s main focus is to showcase that deep learning-based techniques can be used in combination to detect firearms (particularly guns). This paper uses different detection techniques, such as Faster Region-Based Convolutional Neural Networks (Faster RCNN) and the latest EfficientDet-based architectures for detecting guns and human faces. An ensemble (stacked) scheme has improved the detection performance to identify human faces and guns at the post-processing level using Non-Maximum Suppression, Non-Maximum Weighted, and Weighted Boxes Fusion techniques. This paper has empirically discussed the comparative results of various detection techniques and their ensembles. It helps the police to gather quick intelligence about the incident and take preventive measures at the earliest. Also, the same method can be used to identify social-media videos for gun-based content detection. Here, theWeighted Boxes Fusion-based ensemble detection scheme provides mean average precision 77.02%, 16.40%, 29.73% for the mAP 0.5 , mAP 0.75 and mAP [0.500.95] , respectively. The results achieve the best performance among all the experimented alternatives. The model has been rigorously tested with unknown test images and movie clips. The obtained ensemble schemes are satisfactory and consistently improve over primary models.