激光点云深度学习的树种识别研究

Tree Species Identification Based on Laser Point Cloud Deep Learning

针对激光雷达林业树种分类难以直接使用点云数据的问题,使用基于点云深度学习方法进行树种识别并提出PointNet-GS模型,无需将点云转为三维体素或二维图像,避免数据类型转换造成的特征丢失。以河北省塞罕坝机械林场的落叶松和白桦两个树种为研究对象。首先,将获取的点云数据进行数据预处理、单木分割,提取分割效果较好的单木作为样本;其次,将单木提取的样本进行几何下采样处理,保留更多局部特征便于网络模型学习;最后,将下采样处理的样本输入深度学习模型的网络,自动提取其高维特征进行学习,实现树种分类。实验结果表明,PointNet-GS树种分类精度达89.3%,Kappa系数为0.785,效果优于原始PointNet模型。

Aiming at the problem that it is difficult to directly use point cloud data for LiDAR forestry tree species classification,this paper uses point cloud-based deep learning methods for tree species identification and proposes the PointNet-GS model.There is no need to convert point clouds into three-dimensional voxels or two-dimensional images to avoid feature loss caused by data type conversion.This article takes two species of larch and birch from saihanba mechanical forest farm in Hebei province as the research object.Firstly,preprocess the point cloud data and segmented individual trees to extract the single trees with better segmentation effect as samples.Secondly,perform geometric down-sampling processing on the samples extracted from individual trees to retain more local features to facilitate network model learning.Finally,the down-sampling processed samples are input into the network of the deep learning model,and their high-dimensional features are automatically extracted for learning,and then the tree species classification is realized.The experimental results show that the classification accuracy of PointNet-GS tree species is 89.3%,and the Kappa coefficient is 0.785,which is better than that of the original PointNet model.