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. A novel binary-weighted optical computing implementation with spatial multiplexing and aggregation (B-OSMA) scheme achieves high-efficiency image sensing. Coupling with a light-weighted digital discriminator, B-OSMA demonstrates good performances that are competitive to full-precision electric computing on MNIST and Fashion-MNIST datasets, with greatly reduced storage consumption and promoted resilience against systematic errors and ambient noise. image

Position-robust Optronic Convolutional Neural Networks Dealing with Images Position Variation

Optronic Convolutional Neural Networks of Multi-Layers with Different Functions Executed in Optics for Image Classification.

Optical Convolution Based Computational Method for Low-Power Image Processing

Countering Variations and Thermal Effects for Accurate Optical Neural Networks.

Training Optronic Convolutional Neural Networks on an Optical System Through Backpropagation Algorithms.

OP-FCNN: an Optronic Fully Convolutional Neural Network for Imaging Through Scattering Media.

OACNNs: Orientation adaptive convolutional neural networks

Optical-electronic Hybrid Fourier Convolutional Neural Network Based on Super-Pixel Complex-Valued Modulation

Optronic Convolutional Neural Network for SAR Target Recognition.

Orientation-Aware Deep Neural Network for Real Image Super-Resolution.

High-accuracy optical convolution unit architecture for convolutional neural networks by cascaded acousto-optical modulator arrays

Optimization of CNN through Novel Training Strategy for Visual Classification Problems

Compressed Meta-Optical Encoder for Image Classification

Efficient and Robust Homography Estimation Using Compressed Convolutional Neural Network

Orthogonal Decomposition Network for Pixel-Wise Binary Classification

A Novel Robust Algorithm for Position and Orientation Detection Based on Cascaded Deep Neural Network

An Optical Frontend for a Convolutional Neural Network

Prob-POS: A Framework for Improving Visual Explanations from Convolutional Neural Networks for Remote Sensing Image Classification

Image optical processing based on convolutional neural networks in sports video recognition simulation

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

A Self-Attention Armed Optronic Transformer in Imaging Through Scattering Media