Image-free single-pixel classifier using feature information measurement matrices

Abstract

Single-pixel imaging-free (SPIF) technology is a novel object classification method that projects a small set of measurement matrix patterns onto the target and then analyzes the reflected light intensity to achieve imaging-free classification of the target. However, in the existing shallow-learning methods, the information interaction between the measurement matrices and the target is irrelevant. This results in the captured light intensity information often carrying random target features, which makes the classification models constructed from this information less accurate. This study proposes a new SPIF scheme that effectively mitigates this issue by utilizing feature information extracted from prior datasets to build the measurement matrices. Simulation experiments and actual tests demonstrate that our method achieves higher recognition accuracy than classical measurement matrices at the same sampling rate and shows more stable feature extraction capabilities in disturbed environments.