Construction Of Forest Road Network Based On Multi-Source Data Fusion

China’s vast territory,although forest resources assets are very rich,the utilization rate is very low,forest road network information is imperfect and inaccurate is one of the important factors.At present,the identification and extraction of forest roads are relatively difficult due to the complex terrain environment and the limitation of vegetation occlusion.Therefore,it is of great significance to explore a method for timely construction and updating of forest road network.map,which can provide accurate map services for firefighters to quickly reach the fire point through navigation and tourists to plan forest tourism routes.This paper selected the state-owned forestry farm in Yuexi County,Anqing City,Anhui Province as the research area.The road was extracted using the hyperspectral remote sensing image data of Gaofen No.1 in Yuexi County,the GPS patrol trajectory data of foresters in Yuexi County,and the road network data OpenStreetMap electronic map.Based on the road information extracted from the three data sources,the multi-similarity index is used to match the same entity,and the forest road network is finally constructed.The main research results of this paper are as follows:(1)Although hyperspectral remote sensing images contain rich ground objects and spectral information,the road in remote sensing images is easily occluded by trees and other debris,which affects the connectivity of the extracted road.GPS trajectory data has the characteristics of high update frequency,wide trajectory range,low collection cost,and can extract complete road information.The commercial electronic map pays more attention to the urban road network and constructs less and updates slowly the road network in underdeveloped areas such as rural areas and forest areas.Each of the three data has its advantages and disadvantages.The combination of the three technologies can achieve complementary advantages to build a high-precision forest road network.(2)The road information of the forest region is extracted by hyperspectral remote sensing image.The convolution neural network based on the U-Net network model is used to increase the number of network layers to 5 layers,and the convolution kernel of 5*5 and the pooling kernel of 3*3 are used to improve the speed of feature extraction.Add batch standardized BN layer to improve convergence speed.Use the ELU activation function to mitigate gradient disappearance.And adjust the learning rate by setting remote sensing image forest road extraction.Results Through the improved U-Net network model,the forest road information was extracted from the high-resolution multispectral remote sensing image of Yuexi County.(3)The road information of the forest area is extracted based on the trajectory data of forest attendants.Firstly,outliers,noise points,drift points,and stagnation points are eliminated.The processed GPS trajectory data are clustered by the DBSCAN algorithm based on density clustering,and the most widely used B-spline curve is used to fit the road section and extract the road center line.The research results extracted the forest road information from the patrol data of the state-owned forestry general field in Yuexi County by the DBSCAN algorithm.(4)Multi-source forest road spatial data fusion to build forest road network.By using the matching and recognition technology of the same name entity,a multisimilarity matching method for synthesizing the buffer,distance,area,and direction of the line entity is proposed.The recall rate and matching accuracy of the line entity matching is improved to 98.98%and 99.68%.The results show that the forest road network can be constructed by combining the road information extracted from the high-resolution multispectral remote sensing image data and the trajectory data of the forest protector with the road network data of the commercial electronic map through the same name entity matching algorithm.The coverage rate can reach 90.14%,and the false detection rate is 4.02%.In summary,this paper systematically studies the construction of a forest road network based on multi-source data fusion,and the experiment proves that the scheme can effectively improve the accuracy of the forest road network.However,the forest road network constructed in this paper only contains the basic linear structure of the road and does not consider how to classify the road types extracted from high-resolution multispectral remote sensing image data and GPS trajectory data,such as forest roads,cement roads,and highways.For these problems,more factors need to be considered in the process,and the functions currently realized can only provide auxiliary references for forest road network updating.