| Due to factors such as natural environment,the accuracy of navigation in forest areas is low,which makes it difficult for personnel and equipment to reach the work areas in the forest quickly using existing navigation technology,thereby affecting the efficiency of forest management.Accurately extracting the accessible areas in forests is an important part of building a forest navigation model,which can promote efficient forest management and is of great significance for the development of forests.The key to constructing a forest navigation model lies in accurately extracting the accessible areas in forest regions.In this paper,we focus on the Shuangdao Forest Farm and use ground and unmanned aerial vehicle(UAV)multi-source remote sensing data for data fusion,exploring methods for accurately extracting the accessible areas in forests to provide data support for precise navigation of forest equipment and personnel.(1)The data acquisition for the experimental area was carried out using a DJI Matrice300 drone equipped with a visible light P1 sensor and a L1 Li DAR sensor,as well as a hand-held RTK surveying instrument from XAG.In 2021,visible light data of the experimental area was obtained,while in 2022,both visible light and Li DAR data as well as ground data were collected.We spliced and position-corrected the two-year remote sensing data to generate ortho-images,point clouds,and other data.(2)Att U2-Net,U2-Net,and U-Net neural network models were used to extract the non-crown accessible areas in forests.We used data augmentation methods to expand the dataset of visible light images of forest areas and trained and verified the models.The results show that the U2-Net model with the added attention mechanism has better segmentation results,accurately extracting the non-crown accessible areas in forests,with an accuracy of 0.9168,a Dice coefficient of 0.8236,and a recall rate of 0.8835.(3)Point cloud segmentation algorithms can extract the accessible areas in the crown of forests by combining the coordinates of trees and their diameters,providing data support for establishing a model of accessible areas in forests.Single-tree segmentation of point cloud data was performed,and the accuracy of the tree coordinates was evaluated.The results show that the single-tree segmentation method based on point clouds has the best segmentation effect.In low-density forest areas,the recall rate was 0.893,the accuracy was0.925,and the F-score was 0.908.In medium-density forest areas,the recall rate was 0.768,the accuracy was 0.915,and the F-score was 0.835.In high-density forest areas,the recall rate was 0.655,the accuracy was 0.878,and the F-score was 0.750.(4)A multi-source data fusion algorithm was used to integrate the results of extracting the accessible areas in forests,and the accessible areas in forests model was established based on parameters such as elevation and slope,calculating the accessibility weight of forest regions.The A* algorithm was used for path planning in the model,and the accuracy of the A* path was verified by comparing it with the actual accessible areas and path coordinates.The results show that the number of turns in the A* planned path increased by68% compared to the actual path,the path length increased by 6.18%,the A* path was within the range of the actual accessible areas in forests,and the average slope of the path was 32°.Building a model of accessible areas in forests based on multi-source data can meet the navigation requirements of forest equipment and personnel,and the method of extracting paths of accessible areas in forests based on multi-source data provides data support for the precise navigation of forest equipment and personnel. |
PDF Full Text Request