| The investigation and monitoring of forest resources is the basis for forest management,it is the basic guarantee for the sustainable development of forest resources to find out and implement the quantity and quality of forest resources.However,the traditional method of artificial ground investigation of forest resources is time-consuming and inefficient,and it is difficult to adapt to the requirements of modem forestry.For this,the traditional acquisition method of forest resource information will be improved in this study.Applying 3D imaging on the forest resources investigation,the 3D images were constructed from the ground,UAV and ZY-3 satellite scales,the information extraction ability of 3D images in different scales about forest resources survey was discussed.The aim of this study were to reduce the cost and improve the survey efficiency,it provided the theoretical basis for the realization of informationize,automation,intelligence and three-dimensional of forest resources survey,and provided good technical support to promote the rapid development of forestry.The main conclusions are as follows:1)On the information extraction about forest resource with ground 3D imaging:Using common digital camera,it obtained the stereo images in study area by double film photography,analyzed stereo image measurement using tree testing software compiled,and realized the acquisition of tree diameter,visual tree height and tree volume in forest resource survey.By comparing with the measured data,the determination coefficient of DBH,visual height and volume of standing tree were 0.816,0.827and 0.734 respectively.Method was proposed to obtain the 3D point cloud in order to extract tree forest resource survey parameter using intelligent mobile phone.Pix 4D software is used to carry out the feature matching and forward intersection of the acquired image and the GPS coordinates of automatic recording of the mobile phone.Restore the three-dimensional of forest.Using 3D coordinates acquisited,The extraction of DBH,tree height and other forest parameters were realized,comparing with and fitting the measured result,the determination coefficient of tree height and DBH were 0.903 and 0.884 respectively,and the coefficient of breast diameter determination in forest observation plots was 0.817.2)On the information extraction about forest resource with UAV 3D imaging:based on the fixed wing UAV remote sensing system designed and constructed,the flight test was performed on the Laotudingzi forest farm of Benxi in Liaoning province.Under the condition of ensuring the course overlap of 80%and the lateral overlap of 70%,the UAV equipped with SONY ILCE-7R digital camera,photographed and imaged the research area of forest farm,and successfully accessed to high-definition digital images and POS data.The digital terrain model(DTM),digital orthophoto map(DOM),and digital elevation model(DEM)of Laotudingzi forest farm were obtained with a spatial resolution of 0.14 meter by processing the UAV images using PixelGrid software.Moreover,the terrain factors of Laotudingzi forest farm were obtained based on DEM,and the subcompartment division of Laotudingzi forest farm was performed under the condition of tree species and controlling the terrain line(ridge line and valley line)with object oriented segmentation method.The method of supervised and unsupervised classification was used to extract forest land information.The results showed that the highest extraction accuracy of maximum likelihood method was 75%,and the lowest extraction accuracy of minimum distance classification was 65.89%.The tree crown was splited with the method of artificial visual interpretation,oriented object information-extraction with the eCognition software and improved watershed method.Based on the artificial visual interpretation of crown segmentation,combining with the ground survey data,the optimal DBH-crown width and DBH-crown width and tree height models were obtained through fitting.,The best two model of coniferous forest were D=8.0424e0.13IIK,Ik and D=0.569K+1.048H-1.156 respectively,that of broad leaved forest was D=-3.016+6.578K-0.243K2 and D=2.673K+1.011H-2.052 respectively.The tree height extracted by canopy model and canopy shadow method were compared and fitted with actual value,the determination coefficients were 0.776 and 0.892,respectively.The Voronoi and Delaunay triangulation network graph were constructed to realize automatic extraction of forest spatial structure information based on the forest position,tree species and DBH of UAV image,The size ratio,mixed degree and square degree were calculated fast and compared with the measured data.According to UAV image,the eight edges micro-sample method was proposed to extract the characteristic parameter of stand forest,the relevant experiments were carried out combined with field survey data and images,the data from eight shape micro-sample method and 50 angle gauge sampling survey methods were compared and fitted,we found that the determination coefficient of stand average tree height,DBH,stem density and volume were 0.685,0.514,0.807 and 0.782,respectively.The results showed that the eight edges micro-sampling method could be used to extract the stand characteristic parameters from UAV images under the condition of low canopy density.3)On the information extraction about forest resource with ZY-3 satellite imaging:Based on the front and rear view image of resources satellite three line array of Jiufeng Forest farm,DSM was generated after construction of stereo images,relative orientation,epipolar image generation and absolute orientation by the DEM extraction module in the software ENV1 5.3.The 1:5000 topographic map was scanned and vectorized,and DEM was generated by interpolation using Delaunay triangulation network method.The vegetation canopy height model generated based on DSM and DEM,the accuracy comparison test of vegetation canopy height model and the average height of subcompartment in two survey data was done.The results showed that the vegetation canopy height model in general reflected the vegetation height distribution of Jiufeng Forest.Compared with the average height of 103 pieces of ground angle gauge survey data,the determination coefficient is 0.2701,the fitting precision is low.The terrain factors were extracted using the data of DEM topographic map obtained from vectorization,combining with the texture and spectral factor extracted from the multi spectral data and panchromatic data.The forest volume estimation research of Jiufeng has been carried out using machine learning method to consider spectrum,texture and terrain factor.Taking the 46 factors of texture,topography and spectral factors as independent variables and the measured data of ground angle gauge as dependent variables,and then the partial correlation analysis was done.The 12 factors with higher correlation coefficient by F test were selected as the modeling factors,and the model of forest volume estimation was established using the BP neural network,particle swarm optimization least squares support vector machine and random forest three methods of machine study,The results show that:in the three constructed forest volume models,for the random forest volume model,all of determine coefficient of the coniferous forest,broad-leaved forest and mixed forest were greater than 0.879,RMSE were all less than 6.4536m3/hm2,all the determination coefficients of prediction model were greater than 0.808,all the RMSE were less than 6.4562m3/hm2.Therefore,it indicated that the modeling and prediction accuracy of the random forest inventory model were significantly higher than those of other two models.Based on the random forest volume estimation model,the overall estimation of forest volume in Jiufeng Forest was conducted.It found that the total volume of Jiufeng Forest was 73850.52m3,the largest local forest volume was mainly in the middle west of forest farm,the forest volume of the north was low.In summary,we mainly study the methods of generating stereo image pairs and the methods of extracting forest information of three different scales of ground,UAV and ZY-3 satellite in this paper,.The stereo image pairs constructed by ground photogrammetry can be used to extract the information of height,DBH and volume of single tree,and the stereo image pairs constructed by UAV photogrammetry could be used to generate DSM,DEM and DOM.The information extraction about Forest resources survey at single tree or stand level was realized,such as information extraction of crown width,plant number density,canopy density and stand average height,The forest land information extraction at forest level and subcompartment division were done.Based on the forward and backward data provided by ZY-3 satellite,combined with the DEM data obtained from the topographic map vectorization,the canopy height model can be constructed to estimate the height of the forest canopy.The random forest of machine learning method combined with spectral factor,texture factor,terrain factor of remote sensing image can be used to improve estimating accuracy of forest volume,Taking account of the advantages and disadvantages of 3D images in forest resources survey,It can greatly improves survey efficiency of forest resources. |
PDF Full Text Request