Optical Neural Networks for Holographic Image Recognition (Invited Paper)

• DOI: 10.2528/pier22092907
• Publication Date: 2022-11-27T20:15:27Z
• Authors: Yiming Feng; Junru Niu; Yiyun Zhang; Yixuan Li; Hongsheng Chen; Haoliang Qian
• Keywords: 0303 health sciences; 03 medical and health sciences

Inspired by neural networks based on traditional electronic circuits, optical (ONNs) show great potential in terms of computing speed and power consumption.Though some progress has been made devices schemes, ONNs are still a long way from replacing generalizability.Here, we present complex network (cONN) for holographic image recognition, within which high-speed parallel operating unit matrices is proposed, targeting the real-imaginary-splitting column splitting.Based proposed cONN, have numerically demonstrated training-recognition process our cONN images converted handwritten digit datasets, achieving an accuracy 90% back-propagation algorithm.Our training-verification integrated architecture will enrich further development applications on-chip photonic matrix computing.