基于LiDAR点云特征优选的麦秸秆堆精准识别OA

To address the problem of precise identification of wheat straw stacks in light detection and ranging(LiDAR)point clouds in an agricultural area of Gansu Province,this paper collected measured airborne LiDAR point cloud data from a specific region in Gansu Province and constructed a dedicated extraction dataset containing non-uniformly distributed wheat straw stacks under complex farmland scenarios.With the PointNet++deep learning framework as the core,the influence of point cloud fea-ture combinations on the semantic segmentation performance of ground objects was investigated.By comparing the segmenta-tion effects of four groups of feature input schemes(coordinates,coordinates+intensity,coordinates+total number of returns,and coordinates+intensity+total number of returns),the results show that the feature input of coordinates+total number of returns performs best in the wheat straw stack extraction task,and its segmentation accuracy is significantly better than that of other feature combinations.This scheme achieves an extraction intersection over union(IoU)of 64.9%for wheat straw stacks and a mean intersection over union(mIoU)of 75.78%for overall ground objects,which can meet the project requirements for wheat straw stack extraction in large-area scenarios.This paper provides an effective idea for agricultural waste monitoring,and future exploration is needed for lightweight deployment schemes and model architecture optimization.