Scattering-induced entropy boost for highly-compressed optical sensing and encryption
Image classification often relies on a high-quality machine vision system with a large view field and high resolution, demanding fine imaging optics, heavy computational costs, and large communication bandwidths between an image sensor and a computing unit. Here, we report a novel image-free sensing framework for resource efficient image classification where the required number of measurements can be reduced by up to two orders of magnitude. In the proposed framework of single-pixel detection, the optical field from a target is first scattered by an optical diffuser and then two-dimensionally modulated by a spatial light modulator. The optical diffuser simultaneously serves as a compressor and an encryptor for the target information, effectively narrowing the view field and improving the system's security. The one-dimensional sequence of intensity values, measured with time-varying patterns on the spatial light modulator, is then used to extract semantic information based on end-to-end deep learning. The proposed sensing framework is shown to provide over 95 percentage accuracy with the sampling rate of 1 percentage and 5 percentage, respectively, for the classification of MNIST dataset and the recognition of Chinese license plate, which was up to 24 percentage more efficient compared with the case without an optical diffuser. The proposed framework represents a significant breakthrough in realizing high-throughput machine intelligence for scene analysis, with low-bandwidth, low cost, and strong encryption.
PDF Abstract
MNIST
Fashion-MNIST
