基于互信息的机载LiDAR点云自动滤波处理特征选择

Feature Selection of Airborne LiDAR Point Cloud Automatic Filtering Processing Based on Mutual Information

随着机器学习的发展,在LiDAR滤波处理中找到一个适用于所有地形的高效高精度滤波算法是我们一直追求的目标。现在,越来越多的学者致力于提取多个特征进行滤波计算。在选择的特征中难免出现冗余等现象,影响计算效率和滤波精度。本文基于机载激光雷达特有的高程精度优势,采用高程互信息作为测度去判断所选择特征的优劣。经过实验,验证了本文所用方法能够有效剔除自动滤波选特征中的冗余特征。

With the development of machine learning,it is a goal we have been pursuing to find a high-efficiency and high-precision filtering algorithm for all terrain in Li DAR filtering. More and more scholars are now working on extracting many features to carry out filtering calculation. In the selected features often appear redundant phenomena such as the impact of computing efficiency and filtering accuracy. Based on the unique elevation accuracy of airborne Li DAR,height mutual information is taken as a measure to judge the merits of the selected features. Experiments show that the proposed method can effectively eliminate the redundant features in the selected features.