Detection Model and Correction Method for Quadrant Detector-Based Computational Ghost Imaging System

Quadrant detector (QD) is a widely adopted position sensor. By adopting ghost imaging, the four-channel outputs of the detector can be multiplexed to position and image the target simultaneously. However, the lens defocus and detector blind area distort the detected laser intensities, which would increase positioning error and reduce reconstruction image quality. In this research, the detection model is set up and analyzed based on the light spot distribution and detector characteristics. A neural network (NN)-based fitting method is proposed to directly predict spot position and correct total light intensities with detector outputs. The network is constructed and trained with simulation data. The prediction accuracy and generalization performance are verified by numerical simulations. The effectiveness of the proposed method is demonstrated in the experimental system. The positioning error of the proposed method decreases by 98.0% compared to the classic algorithm, and the signal-to-noise ratio of the reconstructed image increases by 31.94 dB. The proposed scheme has the potential to improve positioning accuracy and imaging quality of QD-based detection systems, such as radar and guidance applications.