LAUSR

Compressed sensing (CS) is attractive in wireless multimedia sensor networks (WMSNs) because it achieves sampling-compression and confidentiality protection simultaneously. Moreover, applications in WMSNs emphasize the requirement to process real-time data rapidly. However, the reconstruction of CS has high memory and computational overhead, which prevents it from being used online on resource-limited edge devices. In order to effectively analyze the received CS measurements, a low-overhead compressive analysis framework based on linear transformation is proposed. The linear transformation is used to fast construct a proxy that approximates the original signal from CS measurements for inferences online, and the transformation matrix is built by collaborative learning off-line. In addition, we introduce a hardware-friendly binary matrix to update the measurement matrix to ensure the security of the sampling framework. Both theoretical and experimental analyses validate the low overhead and confidentiality of the proposed compressive analysis framework.