Sensornet A Scalable And Lowpower Deep Convolutional Neural Network For Multimodal Data Classification

• doi: 10.1109/TCSI.2018.2848647
• title: SensorNet: A Scalable and Low-Power Deep Convolutional Neural Network for Multimodal Data Classification
• publisher: IEEE
• isbn:
• issn: 1558-0806
• rank: 3523
• access_type: LOCKED
• content_type: Journals
• abstract: This paper presents SensorNet which is a scalable and low-power embedded deep convolutional neural network (DCNN), designed to classify multimodal time series signals. Time series signals generated by different sensor modalities with different sampling rates are first converted to images (2-D signals), and then DCNN is utilized to automatically learn shared features in the images and perform the classification. SensorNet: 1) is scalable as it can process different types of time series data with variety of input channels and sampling rates; 2) does not need to employ separate signal processing techniques for processing the data generated by each sensor modality; 3) does not require expert knowledge for extracting features for each sensor data; 4) makes it easy and fast to adapt to new sensor modalities with a different sampling rate; 5) achieves very high detection accuracy for different case studies; and 6) has a very efficient architecture which makes it suitable to be deployed at Internet of Things and wearable devices. A custom low-power hardware architecture is also designed for the efficient deployment of SensorNet at embedded realtime systems. SensorNet performance is evaluated using three different case studies including physical activity monitoring, stand-alone tongue drive system, and stress detection, and it achieves an average detection accuracy of 98%, 96.2%, and 94% for each case study, respectively. We implement SensorNet using our custom hardware architecture on Xilinx FPGA (Artix-7) which on average consumes 246-μJ energy. To further reduce the power consumption, SensorNet is implemented using application-specified integrated circuit at the post layout level in 65-nm CMOS technology which consumes approximately 8x lower power compared to the FPGA implementation. In addition, SensorNet is implemented on NVIDIA Jetson TX2 SoC (CPU + GPU) and compared to TX2 single-core CPU and GPU implementations, FPGA-based SensorNet obtains 15× and 4× improvement in energy consumption.
• article_number: 8419763
• pdf_url: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8419763
• html_url: https://ieeexplore.ieee.org/document/8419763/
• abstract_url: https://ieeexplore.ieee.org/document/8419763/
• publication_title: IEEE Transactions on Circuits and Systems I: Regular Papers
• conference_location:
• conference_dates:
• publication_number: 8919
• is_number: 8566198
• publication_year: 2019
• publication_date: Jan. 2019
• start_page: 274
• end_page: 287
• citing_paper_count: 64
• citing_patent_count: 0
• download_count: 2582
• insert_date: 20180725
• index_terms:
  • ieee_terms:
    • Time series analysis
    • Feature extraction
    • Computer architecture
    • Convolution
    • Biomedical monitoring
    • Hardware
    • Field programmable gate arrays
  • author_terms:
    • Multimodal time series
    • deep neural networks
    • energy efficiency
    • FPGA
    • low power
    • embedded systems
    • classification
  • dynamic_index_terms:
    • Neural Network
    • Scalable
    • Deep Convolutional Neural Network
    • Sensor Net
    • Multimodal Data Classification
    • Time Series
    • Time-series
    • Energy Consumption
    • Energy Expenditure
    • Sampling Rate
    • Low Power
    • Signal Processing
    • Deep Neural Network
    • Detection Accuracy
    • Time Series Data
    • Time-series Data
    • Power Consumption
    • Average Accuracy
    • Internet Of Things
    • Internet-of-Things
    • Wearable Devices
    • Wearable Technology
    • Input Channels
    • Variety Of Channels
    • Diverse Channels
    • Convolutional Layers
    • Fully-connected Layer
    • Fully Connected Layer
    • Number Of Filters
    • Rectified Linear Unit Activation
    • Shaping Filter
    • Convolution Operation
    • Results Of Experiments
    • Feature Maps
    • Action Recognition
• authors:
  • Author Name: Ali Jafari
    Affiliation: Computer Science and Electrical Engineering Department, University of Maryland Baltimore County, Baltimore, MD, USA
    Author URL: https://ieeexplore.ieee.org/author/37085622965
    ID: 37085622965
    Order: 1
    Author Affiliations:
    • Computer Science and Electrical Engineering Department, University of Maryland Baltimore County, Baltimore, MD, USA
  • Author Name: Ashwinkumar Ganesan
    Affiliation: Computer Science and Electrical Engineering Department, University of Maryland Baltimore County, Baltimore, MD, USA
    Author URL: https://ieeexplore.ieee.org/author/37086036676
    ID: 37086036676
    Order: 2
    Author Affiliations:
    • Computer Science and Electrical Engineering Department, University of Maryland Baltimore County, Baltimore, MD, USA
  • Author Name: Chetan Sai Kumar Thalisetty
    Affiliation: Computer Science and Electrical Engineering Department, University of Maryland Baltimore County, Baltimore, MD, USA
    Author URL: https://ieeexplore.ieee.org/author/37086547203
    ID: 37086547203
    Order: 3
    Author Affiliations:
    • Computer Science and Electrical Engineering Department, University of Maryland Baltimore County, Baltimore, MD, USA
  • Author Name: Varun Sivasubramanian
    Affiliation: Computer Science and Electrical Engineering Department, University of Maryland Baltimore County, Baltimore, MD, USA
    Author URL: https://ieeexplore.ieee.org/author/37086543406
    ID: 37086543406
    Order: 4
    Author Affiliations:
    • Computer Science and Electrical Engineering Department, University of Maryland Baltimore County, Baltimore, MD, USA
  • Author Name: Tim Oates
    Affiliation: Computer Science and Electrical Engineering Department, University of Maryland Baltimore County, Baltimore, MD, USA
    Author URL: https://ieeexplore.ieee.org/author/37711900900
    ID: 37711900900
    Order: 5
    Author Affiliations:
    • Computer Science and Electrical Engineering Department, University of Maryland Baltimore County, Baltimore, MD, USA
  • Author Name: Tinoosh Mohsenin
    Affiliation: Computer Science and Electrical Engineering Department, University of Maryland Baltimore County, Baltimore, MD, USA
    Author URL: https://ieeexplore.ieee.org/author/37392295000
    ID: 37392295000
    Order: 6
    Author Affiliations:
    • Computer Science and Electrical Engineering Department, University of Maryland Baltimore County, Baltimore, MD, USA

Image Sensor

• sensor_type: CMOS
• resolution: not specified
• dynamic_range: not specified
• pixel_size: not specified
• dark_current: not specified

Optical Data

• focal_length: not specified
• aperture: not specified
• field_of_view: not specified
• distortion: not specified

Performance Metrics

Applications & Benefits

• cell_imaging: CMOS image sensors are used in cell imaging applications among others.
• benefits: Enable digital imaging in smartphones, medical devices, and automotive systems.

Supporting Organizations

• supported_by: not specified

Manuscript Details

• publication_date: Jan. 2019

Relevancy Score

• score: 8
• missing_fields:
  • resolution
  • dynamic_range
  • pixel_size
  • dark_current
  • focal_length
  • aperture
  • field_of_view
  • distortion
  • frame_rate
  • signal_to_noise_ratio
  • sensitivity
  • shutter_speed
  • power_consumption
  • noise
  • supported_by
  • publication_date

Processed JSON Filename

request_2e2beec0-1e44-4619-97c7-60fc48adf6c6-sensornet_a_scalable_and_lowpower_deep_convolutional_neural_network_for_multimodal_data_classification.json