Abstract:Deep learning convolutional neural networks generally involve multiple-layer, forward-backward propagation machine-learning algorithms that are computationally costly. In this work, we demonstrate an alternative scheme to convolutional neural nets that reconstructs an original image from its optically preprocessed, Fourier-encoded pattern. The scheme is much less computationally demanding and more noise robust, and thus suited for high-speed and low-light imaging. We introduce a vortex phase transform with a lenslet-array to accompany shallow, dense, “small-brain” neural networks. Our single-shot coded-aperture approach exploits the coherent diffraction, compact representation, and edge enhancement of Fourier-transformed spiral phase gradients. With vortex encoding, a small brain is trained to deconvolve images at rates 5–20 times faster than those achieved with random encoding schemes, where greater advantages are gained in the presence of noise. Once trained, the small brain reconstructs an object from intensity-only data, solving an inverse mapping without performing iterations on each image and without deep learning schemes. With vortex Fourier encoding, we reconstruct MNIST Fashion objects illuminated with low-light flux ( <math display="inline"> <mn>5</mn> <mspace width="thickmathspace"></mspace> <mtext>nJ</mtext> <mrow> <mo>/</mo> </mrow> <mrow> <msup> <mtext>cm</mtext> <mn>2</mn> </msup> </mrow> </math> ) at a rate of several thousand frames per second on a 15 W central processing unit. We demonstrate that Fourier optical preprocessing with vortex encoders achieves similar accuracies and speeds 2 orders of magnitude faster than convolutional neural networks.

Optical Fourier convolutional neural network with high efficiency in image classification

Optical Convolution Based Computational Method for Low-Power Image Processing

On-chip 4F-System Based on Concave Mirrors for Optical Neural Networks

Multi-layer optical Fourier neural network based on the convolution theorem

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

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

All-optical Fourier neural network using partially coherent light

Optical Convolutional Neural Networks -- Combining Silicon Photonics and Fourier Optics for Computer Vision

Optical convolutional neural network with atomic nonlinearity

Optical Convolutional Neural Network Based on an Amplitude Modulation Spatial Light Modulator and a 4-Level Phase Plate

11 TOPS photonic convolutional accelerator for optical neural networks

Integrated Photonic FFT for Optical Convolutions towards Efficient and High-Speed Neural Networks

An on-chip photonic deep neural network for image classification

11 TeraFLOPs per second photonic convolutional accelerator for deep learning optical neural networks

Convolutional Free-space Optical Neural Networks for Image Recognition

Transformer in Optronic Neural Networks for Image Classification

High-accuracy data readout in multi-dimensional optical data storage using convolutional neural networks

Compressed Meta-Optical Encoder for Image Classification

Fourier optical preprocessing in lieu of deep learning

Integrated convolutional kernel based on two-dimensional photonic crystals

Correlated optical convolutional neural network with "quantum speedup"