Robust and Efficient Single-Pixel Image Classificationwith Nonlinear Optics

TL;DR

This paper introduces a hybrid optical and deep learning approach for single-pixel image classification that significantly improves accuracy and robustness, especially in noisy conditions, with potential applications in lidar, image recognition, and atmospheric monitoring.

Contribution

It combines mode-selective image upconversion, Fourier transform features, and deep learning to enhance classification accuracy and robustness in single-pixel imaging.

Findings

Boosted MNIST classification accuracy from 81.25% to 99.23%.

Achieved 83% accuracy on highly contaminated images with -17 dB SNR.

Demonstrated potential for fast, robust image processing in various applications.

Abstract

We present a hybrid image classifier by mode-selective image upconversion, single pixel photodetection, and deep learning, aiming at fast processing a large number of pixels. It utilizes partial Fourier transform to extract the signature features of images in both the original and Fourier domains, thereby significantly increasing the classification accuracy and robustness. Tested on the MNIST handwritten digit images, it boosts the accuracy from 81.25% to 99.23%, and achieves an 83% accuracy for highly contaminated images whose signal-to-noise ratio is only -17 dB. Our approach could prove useful for fast lidar data processing, high resolution image recognition, occluded target identification, atmosphere monitoring, and so on.