Multiple-view D²NNs array: realizing robust 3D object recognition

Abstract

As an optical-based classifier of the physical neural network, the independent diffractive deep neural network ( D 2 N N ) can be utilized to learn the single-view spatial featured mapping between the input lightfields and the truth labels by preprocessing a large number of training samples. However, it is still not enough to approach or even reach a satisfactory classification accuracy on three-dimensional (3D) targets owing to already losing lots of effective lightfield information on other view fields. This Letter presents a multiple-view D 2 N N s array (MDA) scheme that provides a significant inference improvement compared with individual D 2 N N or Res- D 2 N N by constructing a different complementary mechanism and then merging all base learners of distinct views on an electronic computer. Furthermore, a robust multiple-view D 2 N N s array (r-MDA) framework is demonstrated to resist the redundant spatial features of invalid lightfields due to severe optical disturbances.

Multiview Fusion 3D Target Information Perception Model in Nighttime Unmanned Intelligent Vehicles

“…Among them, target information perception technology is the underlying module of unmanned technology and the basic guarantee for safe vehicle driving [5]. Through unmanned driving technology based on 3D target information perception model with multiview fusion, we can effectively control many unstable human factors, such as drunk driving and fatigue driving [6].…”

QuATON: Quantization Aware Training of Optical Neurons

“…optical processors equivalent to neural networks-have already shown great promise in a variety of applications. For instance, Diffractive networks 1 , made of a cascade of passive diffractive layers, could perform all-optical classification 1 , all-optical quantitative phase imaging (QPI) 2,3 , optical logic operations 4 , spatiotemporal signal processing 5 , saliency segmentation 6 , and 3D object detection 7 . Similarly, learnable optical Fourier processors have shown to be capable of all-optical quantitative phase imaging 3 , medical image processing 8,9 , optical image encryption 10 , and image classification 11 .…”

Time‐Lapse Image Classification Using a Diffractive Neural Network

“…Diffractive networks can achieve universal linear transformations, [ 7–9 ] and various applications using diffractive processors have been demonstrated such as object classification, pulse processing, imaging through random diffusers, hologram reconstruction, quantitative phase imaging, class‐specific imaging, super‐resolution image display, all‐optical logic operations, beam shaping, and orbital angular momentum mode processing, among others. [ 10–30 ]…”