KNN Based Denoising Algorithm for Photon-Counting LiDAR: Numerical Simulation and Parameter Optimization Design

Abstract

Photon-counting LiDAR can obtain long-distance, high-precision target3D geographic information, but extracting high-precision signal photons from background noise photons is the key premise of photon-counting LiDAR data processing and application. This study proposes an adaptive noise filtering algorithm that adjusts parameters according to the background photon count rate and removes noise photons based on the local mean Euclidean distance. A simulated photon library that provides different background photon count rates and detection probabilities was constructed. It was then used to fit the distribution relationship between the background photon count rate and the average KNN (K-Nearest Neighbor) distance (k = 2–6) and to obtain the optimal denoising threshold under different background photon count rates. Finally, the proposed method was evaluated by comparing it with the modified density-based spatial clustering (mDBSCAN) and local distance-based statistical methods. The experimental results show that various methods are similar when the background noise rate is high. However, at most non-extreme background photon count rate levels, the F of this algorithm was maintained between 0.97–0.99, which is an improvement over other classical algorithms. The new strategy eliminated the artificial introduction of errors. Due to its low error rates, the proposed method can be widely applied in photon-counting LiDAR signal extraction under various conditions.