Noise Filtering Post-Processing Method For High Signal-to-Noise Ratio Single Photon Imaging Based On Clustering Algorithm

Single photon imaging lidar based on TCSPC technology uses SPAD single photon detectors to detect signal light with high sensitivity and can detect very weak diffuse reflected signal light,but as the imaging distance increases,the detector receives less signal light,and the data is more background noise.More of the data is background noise.Therefore,in order to obtain higher imaging quality,the signal light needs to be extracted from the noise.To address the above problems,according to the data characteristics of single-photon imaging lidar,this paper proposes a scheme based on clustering algorithm to filter out single-photon imaging noise.First,the theoretical analysis of the distribution characteristics of the imaging data points and the characteristics of the detection target,present the histogram pre-processing method of the data.Simulate The target noise data of plane,surface and step type,and select Gaussian mixture clustering and DBSCAN algorithms as the noise filtering methods in this paper,and combined with the SOR algorithm to filter out small-scale noise.Secondly,the effects of clustering parameters and data scale on the noise filtering effect of GMM algorithm and DBSCAN algorithm are studied by simulation,and the signal-to-noise ratio,filtering accuracy and cross-merge ratio are used as criteria to compare and analyze the noise filtering effect of the algorithms before and after optimization.Finally,a single-photon imaging experimental system is built to obtain a set of experimental data,and the GMM algorithm and DBSCAN algorithm are used to filter the data before and after optimization to verify the feasibility of the clustering algorithm to filter the single-photon imaging noise.Compared with direct cumulative histogram peak imaging,the clustering algorithm can ensure high signal to noise ratio,high noise filtering accuracy and high intersection and parallel ratio.