Compressive Sensing based Asymmetric Semantic Image Compression for Resource-constrained IoT system
TimeWednesday, July 13th3:50pm - 4:10pm PDT
Location3000, Level 3
Design of Cyber-physical Systems, Cloud Computing and IoT
DescriptionThe widespread application of deep learning has made machine-to-machine semantic communication more intelligent. But the resource-constrained IoT systems cannot provide enough computation capacity for the great amount of semantic recognition demand. Thus, in this paper, we propose Compressed Sensing based Deep Semantic Image Compression (CS-DSIC) for resource-constrained IoT systems, which consists of a lightweight encoder on the front device, and an iterative deep decoder offloaded at the server. We further consider a task-oriented scenario and optimize CS-DSIC for the semantic recognition tasks. The experiment results show that CS-DSIC achieves considerable rate-distortion, rate-accuracy trade-off, and low encoding complexity over prevailing codecs.