Feature Augmented Hybrid CNN for Stress Recognition Using Wrist-based Photoplethysmography Sensor

Rashid, Nafiul; Chen, Liming; Dautta, Manik; Jimenez, Abel; Tseng, Peter; Faruque, Mohammad Abdullah Al

doi:10.1109/embc46164.2021.9630576

Cited by 20 publications

(18 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Authors in [6] explored different combinations of chest and wrist sensors across a variety of models. The results for three deep learning methods are also shown [7]- [9]. For our three selected branch classifiers, the softand hard-voting methods are applied, showing performance improvements compared to the individual branch classifiers for both 3-class and 2-class classification.…”

Section: B Experimental Resultsmentioning

confidence: 90%

See 1 more Smart Citation

SELF-CARE: Selective Fusion with Context-Aware Low-Power Edge Computing for Stress Detection

Rashid¹,

Mortlock²,

Faruque³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

Detecting human stress levels and emotional states with physiological body-worn sensors is a complex task, but one with many health-related benefits. Robustness to sensor measurement noise and energy efficiency of low-power devices remain key challenges in stress detection. We propose SELF-CARE, a fully wrist-based method for stress detection that employs context-aware selective sensor fusion that dynamically adapts based on data from the sensors. Our method uses motion to determine the context of the system and learns to adjust the fused sensors accordingly, improving performance while maintaining energy efficiency. SELF-CARE obtains state-of-theart performance across the publicly available WESAD dataset, achieving 86.34% and 94.12% accuracy for the 3-class and 2-class classification problems, respectively. Evaluation on real hardware shows that our approach achieves up to 2.2× (3-class) and 2.7× (2-class) energy efficiency compared to traditional sensor fusion.

show abstract

Section: B Experimental Resultsmentioning

confidence: 90%

“…Preprocessing is used over the raw, unfiltered sensor data by applying various filters (e.g., band-pass filters or lowpass filters) to the input data to reduce sensor noises and more easily extract important features. The preprocessing performed over each sensing modality follows that in [9].…”

Section: A Preprocessingmentioning

confidence: 99%

SELF-CARE: Selective Fusion with Context-Aware Low-Power Edge Computing for Stress Detection

Rashid¹,

Mortlock²,

Faruque³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Pereira et al (2019) applied CNN in the segmentation of brain tumor images [ 19 ]. The specific locations in the convolutional layer were merged to create a new image [ 20 , 21 ]. The optimized CNN algorithm was utilized, and RIU-Net was applied to segment CT image.…”

Section: Discussionmentioning

confidence: 99%

“…The model was optimized by deep learning [ 22 ]. The optimized model was defined as the RIU-net network, and it was then equipped with SE module, to enhance the feature extraction image [ 20 ]. The BM3D algorithm was compared with DnCNN and Cascaded CNN, and it was concluded that the BM3D algorithm had better denoising effects.…”

Section: Discussionmentioning

confidence: 99%

Optimized Deconvolutional Algorithm‐based CT Perfusion Imaging in Diagnosis of Acute Cerebral Infarction

Chen

Bai

Shu

et al. 2022

Contrast Media & Molecular Imaging

View full text Add to dashboard Cite

To apply deconvolution algorithm in computer tomography (CT) perfusion imaging of acute cerebral infarction (ACI), a convolutional neural network (CNN) algorithm was optimized first. RIU-Net was applied to segment CT image, and then equipped with SE module to enhance the feature extraction ability. Next, the BM3D algorithm, Dn CNN, and Cascaded CNN were compared for denoising effects. 80 patients with ACI were recruited and grouped for a retrospective analysis. The control group utilized the ordinary method, and the observation group utilized the algorithm proposed. The optimized model was utilized to extract the feature information of the patient’s CT images. The results showed that after the SE module pooling was added to the RIU-Net network, the utilization rate of the key features was raised. The specificity of patients in observation group was 98.7%, the accuracy was 93.7%, and the detected number was (1.6 ± 0.2). The specificity of patients in the control group was 93.2%, the accuracy was 87.6%, and the detected number was (1.3 ± 0.4). Obviously, the observation group was superior to the control group in all respects (P < 0.05). In conclusion, the optimized model demonstrates superb capabilities in image denoising and image segmentation. It can accurately extract the information to diagnose ACI, which is suggested clinically.

show abstract

“…Os resultados mostram que o modelo de rede neural densa proposta obteve taxas de acurácia de 95,21% e F1-score de 94,24% para classificação binária. Rashid et al (2021) propõem o uso de uma arquitetura de rede profunda híbrida que usa tanto extração de características manuais quanto extração automática realizada por uma rede de convolução (CNN) usando apenas o sinal BVP da base de dados WESAD. Os experimentos mostram que o modelo proposto atingiu acurácia de 88,56% e F1-score de 86,18% na classificação binária.…”

Section: Trabalhos Relacionadosunclassified

Um Modelo de Aprendizado Profundo Multimodal para Classificação de Estresse Utilizando Sinais Obtidos por Dispositivos Vestíveis de Pulso

Medeiros¹,

Cunha²,

Santos³

et al. 2022

Anais Do XXII Simpósio Brasileiro De Computação Aplicada À Saúde (SBCAS 2022)

View full text Add to dashboard Cite

A recente proliferação dos dispositivos vestíveis (e.g. smartphones e smartwatches) e sua capacidade de medir vários sinais fisiológicos criam uma oportunidade de monitoramento contínuo e discreto do estado de estresse de um indivíduo. Tendo isso em vista, este trabalho propõe um novo modelo para classificação de estresse baseado em uma rede de convolução usando dados coletados por dispositivos de pulso. Os resultados obtidos, utilizando a validação leave-one-subject-out (LOSO), alcançaram a taxa média de 95,67% de acurácia, sendo um valor superior à maioria dos trabalhos já existentes na literatura.

show abstract

Feature Augmented Hybrid CNN for Stress Recognition Using Wrist-based Photoplethysmography Sensor

Cited by 20 publications

References 15 publications

SELF-CARE: Selective Fusion with Context-Aware Low-Power Edge Computing for Stress Detection

SELF-CARE: Selective Fusion with Context-Aware Low-Power Edge Computing for Stress Detection

Optimized Deconvolutional Algorithm‐based CT Perfusion Imaging in Diagnosis of Acute Cerebral Infarction

Um Modelo de Aprendizado Profundo Multimodal para Classificação de Estresse Utilizando Sinais Obtidos por Dispositivos Vestíveis de Pulso

Contact Info

Product

Resources

About