Feature-Supervised Compressed Sensing for Microwave Imaging Systems

Abstract

Traditional compressed sensing (CS) applications use sparse information for downsampling but ignore overall system objectives such as feature extraction. This article jointly designs the sensing and feature-extraction process to improve the efficiency of the microwave imaging systems in time, storage, and feature extraction. A feature-supervised CS (FsCS) is proposed for the cases where not all data contribute to feature extraction. Compared with the traditional spatial–spectral sweep and CS solutions, a feature constraint is added in designing the CS measurement matrix. More efficient sensing and feature extraction are achieved, because only the data contributing to feature extraction are sampled and reconstructed. To improve the time efficiency of CS reconstruction, an aligned spatial–spectral sensing (ASSS) is involved in FsCS to enable joint reconstruction. The proposed scheme is validated in an open-ended waveguide imaging system for low-energy impact damage feature detection. The experimental results demonstrate one order of magnitude improvement in time and two orders of improvement in the data compression ratio compared with the state-of-the-art method, while preserving the interested feature. This article can inspire the joint sensing-processing designs for more intelligent industrial processes.