Study On The Vegetation Height Prediction Model Based On Photogrammetry And Machine Learning

Aerial photogrammetry is an important method to gain basic geographic information and also one of the major ways for terrain mapping.As visual remote sensing is adopted by Aerial Photogrammetry,its signals cannot penetrate through the plant canopy and reach the earth’s surface,and can only generate Digital Surface Model(DSM)rather than accurate topography terrain under forest and vegetation height.Even through stereoscopic viewing,it is difficult to compute the height accurately.With the basis of Aerial photogrammetry,Digital Surface Model(DSM)and Digital Orthophoto Map(DOM),adopt machine learning,make use of Random Forest For Regression(RF-R)and Gradient Boosting Decision Tree(GBDT),generate vegetation height model to ensure its reliability and adoptability and provide a new idea and method for researching vegetation modeling.It mainly includes the following works:(1)Select appropriate feature factors and set down relevant conditions for evaluation.With the Digital Orthophoto Map(DOM)and Digital Surface Model(DSM),generated from Aerial photogrammetry,as the basic data,combined its features with grids division and the height controlling points,spectral feature factors and geometric feature factors amid node-based local mesh are elected.And then,the connection between feature factors and vegetation height controlling points is evaluated by using the correlation coefficient,which will be used to select follow-up feature factors.(2)For high accuracy of height modeling,vegetation height modeling based on Random Forest For Regression(RF-R)is generated.Through combining the basic principles and analyzing features of Random Forest For Regression(RF-R),set appropriate plans and procedures for modeling to figure out its adoptability and reliability for generating the forecast model.First of all,create vegetation height forecast model based on Random Forest For Regression(RF-R),which is on the data basis of Digital Orthophoto Map(DOM)and Digital Surface Model(DSM),to compute the vegetation height.Afterwards,creating Random Forest For Regression(RF-R)model,which consists of feature factors and vegetation controlling points,optimize model parameters and feature factors for improving the accuracy of forecast model.As relevant research shows,besides a certain amount of error of original data,the forecast model is highly accurate which can reach the meter level,and compared with the support vector machine(SVM)modeling method,its model accuracy has advantages.(3)For high accuracy and efficiency of height modeling,vegetation height modeling based on Gradient Boosting Decision Tree(GBDT)is generated.The basic principles and features of Gradient Boosting Decision Tree(GBDT)are integrated to investigate the adoptability reliability and efficiency of Gradient Boosting Decision Tree(GBDT)for generating vegetation height forecast model.Through vegetation forecast modeling based on Gradient Boosting Decision Tree(GBDT),to get the forecast results with high accuracy.As relevant research shows,besides a certain amount of error of original data,the accuracy of forecast model can reach sub meter through optimizing the parameter and feature factors;and compared with support vector machine(SVM)and Random Forest For Regression(RF-R),it has more superiority in accuracy and efficiency.In conclusion,through combining the data in machine learning from Random Forest For Regression(RF-R),Gradient Boosting Decision Tree(GBDT)and Aerial photogrammetry,the way of machine learning for vegetation height modeling is provided.And the forecast model for the vegetation height is generated and its practicability and adoptability is justified,which provides new ideas and ways for vegetation height modeling in the forest.