Deep Learning-Constructed Joint Transmission-Recognition for Internet of Things

Chia-Han Lee*, Jia Wei Lin, Po Hao Chen, Yu Chieh Chang

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

5 Scopus citations


The widely deployed Internet-of-Things (IoT) devices provide intelligent services with its cognition capability. Since the IoT data are usually transmitted to the server for recognition (e.g., image classification) due to low computational capability and limited power supply, achieving recognition accuracy under limited bandwidth and noisy channel of wireless networks is a crucial but challenging task. In this paper, we propose a deep learning-constructed joint transmission-recognition scheme for the IoT devices to effectively transmit data wirelessly to the server for recognition, jointly considering transmission bandwidth, transmission reliability, complexity, and recognition accuracy. Compared with other schemes that may be deployed on the IoT devices, i.e., a scheme based on JPEG compression and two compressed sensing-based schemes, the proposed deep neural network-based scheme has much higher recognition accuracy under various transmission scenarios at all signal-to-noise ratios (SNRs). In particular, the proposed scheme maintains good performance at the very low SNR. Moreover, the complexity of the proposed scheme is low, making it suitable for IoT applications. Finally, a transfer learning-based training method is proposed to effectively mitigate the computing burden and reduce the overhead of online training.

Original languageEnglish
Article number8731945
Pages (from-to)76547-76561
Number of pages15
JournalIEEE Access
StatePublished - 1 Jan 2019


  • Internet of things (IoT)
  • JPEG
  • compressed sensing
  • deep learning
  • deep neural networks
  • joint source-channel coding
  • recognition
  • transfer learning
  • transmission

Fingerprint Dive into the research topics of 'Deep Learning-Constructed Joint Transmission-Recognition for Internet of Things'. Together they form a unique fingerprint.

Cite this