Single-pixel fast-moving object classification based on optical-electronical hybrid neural network （Invited）

Successive classification of fast-moving objects is significant in various fields. However, due to the limited data transmission bandwidth and data storage space, it is challenging to perform fast-moving object classification based on scene photography for a long duration. Inspired by single-pixel imaging and combined with deep learning, a single-pixel fast-moving object classification method based on optical-electronic hybrid neural network was proposed. The proposed method had no need to acquire the images of objects, but obtained the feature information for classification directly by using spatial light modulating and single-pixel detecting. Thus, the massive image data produced by the image-based classification for a long duration was avoided. As part of the neural network, the single-pixel detecting connected optical computing and electronic computing seamlessly, an optical-electronic hybrid neural network for object classification was constructed. The proposed method in classifying fast-moving handwritten digits on a rotating disk was experimentally demonstrated, which passed through the field of view successively. The experiment confirmed that the classification ability of the proposed method had exceeded human vision.