Research on signal extraction algorithm and point cloud image joint reconstruction algorithm for 256 × 128 large-array GM-APD active laser imaging based on Poisson weight gain

For 256 × 128 Geiger-mode APD (GM-APD) single-photon imaging under low pulse energy, the high trigger probability at the initial position of ranging gate, combined with dark counts, produces a pronounced noise peak at the initial position of ranging gate. By masking subsequent signal returns, this peak introduces outliers and degrades the contour and detail of the reconstructed point cloud. It also causes the foreground and background trigger rates to converge due to the low SNR, which invalidates simple intensity thresholding. We propose a Poisson-weight-gain (PWG) signal extraction method that exploits Poisson statistics in time-histogram data to suppress noise peak at the initial position of ranging gate and enhance true echoes. A Point Cloud Clustering and Morphological Filtering (PCCMF) collaborative denoising strategy is further introduced to refine the initial point cloud by integrating DBSCAN density clustering with morphological filtering, removing discrete outliers while preserving target geometry. Based on the two methods described above, a Poisson Weight Gain-Point Cloud Clustering and Morphological Filtering (PWG-PCCMF) serial joint algorithm framework is constructed. Experiments on multi-range building-group scenes using a 256 × 128 active imaging system demonstrate effective background suppression and high-quality reconstruction when SNR > 0.066, outperforming peak extraction method, cross-neighborhood, matched filtering, and their hybrid baselines.