| China has the largest area of planted forests in the world,and planted forests play a very crucial role in environmental protection and forest management.The dynamic and accurate measurement of forest biomass,especially the characteristics of forest trees and the forest environment,is an important part of modern forest management and provides the basis for better ecological and economic benefits of forest areas.Traditionally,it is time-consuming and labor-intensive to rely on manual tools for forest area survey,and the rapid and accurate measurement of forest biomass parameters is a hot research topic in the field of forest area exploration in the world nowadays,and the intelligent sensing of forest carbon stock is also a difficult point in current forestry research.In this paper,we developed a Li DAR-based forest under-canopy data collection system,proposed a laser point cloud slicing and segmentation method based on computational geometry and deep neural network,and realized the extraction of under-canopy tree features(diameter at breast height,tree height and crown width)and slope features.The main research contents and results of this paper are as follows.(1)Aiming at the inconvenience of directly acquiring fine data of complete forest areas by scanning methods such as ground-based Li DAR(TLS),a portable backpack acquisition device is developed,which is combined with a mobile cart acquisition device in forest areas to realize the acquisition of fine point cloud data of complete forest areas and under the canopy,and point cloud data of Qipo forest,Jiufeng experimental forest and Xiaotangshan nursery were collected.(2)To fully exploit the elevation information of forest scenes,this paper proposes a method for regional slicing of forest land laser point clouds based on local elevation features,which avoids the problem of incomplete point clouds of trees at the edges of the blocks caused by slicing according to a fixed range or point volume.The results showed that the number of incorrectly segmented trees was reduced by 86.2% and 82.4% compared with the range and point volume slicing methods,respectively.(3)The point clouds obtained from the under-canopy data collection device in the forest area have Li DAR reflection intensity information.Based on this feature,a modified Point Net incorporating laser reflection intensity information is proposed in this paper to realize the segmentation of different objects such as trees and ground,which is compared with the traditional mathematical morphology segmentation algorithm,cloth simulation filter segmentation algorithm and Point Net method by experiment,and the recognition accuracy for the test point The recognition accuracy of the cloud set reached 0.793,which is significantly better than the traditional algorithm.(4)PLS and MLS in forest areas are of poor quality compared with TLS point clouds,and it is difficult to extract the under-canopy tree features(diameter at breast height,tree height and crown width)and slope features by traditional methods,this paper uses the clustering method to segment the single wood point cloud information from the forest point clouds,extract the wood crown width and tree height feature parameters,and construct a digital elevation model(DEM)based on the ground point cloud data to extract the under-canopy slope features.In the DBH extraction,a chest diameter extraction method based on multi-layer progressive point cloud circle fitting is proposed,which is suitable for accurate chest diameter extraction scenarios with sparse point clouds.In this paper,we use the slicing method based on local features to reduce the difficulty of point cloud processing in the whole forest area,optimize the effect of forest point cloud segmentation by fusing laser reflection intensity information,and supplement the sub-canopy fine feature point cloud data for forest intelligent perception,which provides effective technical support for accurate prediction of forest biomass and forest carbon stock in the later stage. |
PDF Full Text Request