An efficient neural network for low sampling computational ghost imaging based on EMNIST training

Abstract

In this paper, we propose a new network to achieve low-sampling computational ghost imaging. The proposed neural network that combines ResNeXt, eHoloNet and spatial attention mechanism is efficient for object reconstruction just based on EMNIST training, which is much simpler and time-saving. The generalization of the neural network is verified with multi-slits as well as complex object. Both simulated and experimental results show that the proposed network can give an effective reconstruction at 0.7% sampling rate. The neural network in this work is of great significance to ghost imaging in a wider application scenarios, such as real-time imaging and dynamic detection of motion object.