Research on the single-photon imaging algorithm based on the adaptive feature fusion residual network

Single-photon avalanche diodes possess photon-level sensitivity in signal detection and are currently widely used for three-dimensional imaging in various complex scenarios. However, the detection efficiency of single-photon detectors is low, and noise such as ambient light interference and detector dark counts can interfere with effective signals, posing significant challenges for accurately predicting depth images. Therefore, distinguishing effective signals from noise in sparse photon data imposes higher demands on single-photon imaging algorithms. In this paper, we propose a residual network based on adaptive feature fusion for efficient photon imaging from highly noisy data. This network improves the conventional residual network and simultaneously incorporates an attention learning mechanism and an adaptive feature fusion module, which enhance the representation of effective signals and achieve efficient complementary fusion of shallow detail features and deep semantic features, demonstrating significant advantages in generating reliable and accurate depth maps. Comprehensive experiments conducted on simulated datasets and real-world collected data show that our network maintains superior imaging performance under different signal-to-background ratio conditions.