Resource‐Saving and High‐Robustness Image Sensing Based on Binary Optical Computing

Abstract Computational imaging, as a novel technology utilizing encoded image acquisition, relies on intelligent decoding methods for effective image restoration and sensing. Optical computing‐based decoders can efficiently process and extract features from pre‐sensor information, reducing the computational burden on digital computers. However, mainstream parallel optical neural network (ONN) architectures based on wavefront propagation typically possess complex network structures and high‐precision parameters, which pose challenges in terms of precise fabrication and system calibration, as well as sensitivity to signal‐to‐noise ratios. In this work, a binary‐weighted optical computing engine is proposed with spatial multiplexing and aggregation (B‐OSMA), a large‐scale passive ONN implementation that achieves high‐efficiency image sensing. Employing B‐OSMA as an optical decoder, demonstrated image categorizing from 2% compressive is experimented sampling with 92.0% and 83.8% accuracy on MNIST and fashion‐MNIST datasets, respectively, approaching the performance of full‐precision electronic computing while reducing storage requirements by 97%. Compared to conventional ONNs with analog weights, the B‐OSMA exhibits enhanced resilience against systematic errors and ambient noise. This work represents a significant advancement towards practical applications of optical computing in image sensing.