Study On Forest Stand Volume Estimation Based On UAV Image

Stand volume refers to the total volume of all standing trees in the stand,and is an important indicator of forest resource investigation.Field surveys and remote sensing estimation are two main methods to obtain forest stocks.However the calculation of stock volume based on field surveys is time-consuming and labor-intensive,while through satellite remote sensing images if of poor timeliness and low accuracy.On the other hand,the application of drone in calculation of stock volume has the advantages of facile in operation,low cost,and fast maneuvering response,is intensively researched.Besides,compared with the traditional satellite remote sensing,which are easily affected by weather conditions such as clouds and fog,have long re-visit cycle and untimely emergency response.The image data of drone is of high resolution,high-efficiency,and great accuracy in estimating forest stock volume.In this paper,Hunan Hengshan Forest Ecological Positioning Observation and Research Station is taken as the research area,and the experimental sample plot in the research area is taken as the research object.The field surveys are used to obtain the survey data of the tree's DBH and tree height,and the image data of the experimental plot is obtained by drone.A high-precision three-dimensional model is obtained by combining the preprocessed experimental sample UAV image data and field survey data;Afterwards,object-oriented multi-scale segmentation and GIS spatial analysis technology was used to extract the canopy information based on the preprocessed results,thus obtaining the four stand factors,that is the forest average stand crown width,forest canopy density,forest stand height and plant number density;Then,three types of stand stock volume estimates models were constructed by using the extracted stand factors as independent variables and the measured stock volume as the dependent variable.Finally,the best fitting model was determined by model test and comparative analysis.The main findings are as follows:(1)UAV data acquisition and preprocessing.The UAV digital photogrammetry processing platform Agisoft PhotoScan is used for preprocessing the image data collected by the DJI UAV and the elevation data of the experimental samples.The high resolution Digital orthophoto model,high-precision digital surface model and digital elevation model are constructed through multi-view 3D reconstruction and image external orientation elements.(2)Extract canopy information.Based on high-precision digital orthophotos,using object-oriented image analysis methods and multi-scale segmentation technology,set the appropriate segmentation scale,compactness and shape factor for multi-scale segmentation of the image;use spectrum,texture and shape features as classification The indicators are classified,and the crown information is effectively extracted,and the F measure reaches 0.73.(3)e xtract stand factors.Based on the preprocessing results of UAV image data and the results of single tree canopy segmentation,extract the parameters of average stand width,stand canopy density,stand average tree height and plant number density of the experimental plots.Using field survey data as verification data,the overall accuracy of the plots is:the average stand width of the stand is 85.24%,the stand density of the stand is 89.91%,the average height of the stand is 85.78%;the plant density is 79.86%(4)Construct a stock volume estimation model.Taking the measured accumulation volume as the dependent variable and the extracted stand factors as independent variables,three accumulation volume models were constructed:multiple linear regression model,partial least squares regression model and BP neural network.It is the partial least squares regression model that selects the best fitting model according to the test coefficients such as the determination coefficient and error analysis.The determination coefficient is 0.778 and the relative error is 18.93%.The second is the multiple linear regression model,the determination coefficient is 0.641,The relative error is 21.16%;the last is the BP neural network,the determination coefficient is 0.641,and the relative error is 39.68%.In general,the stock volume estimation model is constructed by extracting the modeling factors using remote sensing methods based on the high-resolution image data of the UAV.This method can quickly and accurately obtain the forest stock volume and the obtained results provide a variety of reference methods in estimation of sub-accumulation using UAV,and are good guidance in forestry monitoring and survey data acquisition.