Tree Species Classification Using Optimized Features Derived from Light Detection and Ranging Point Clouds Based on Fractal Geometry and Quantitative Structure Model

Tree species classification is a ubiquitous task in the forest inventory field. Only directly measured feature vectors have been applied to most existing methods that use LiDAR technology for tree classification. As result, it difficult obtain satisfactory performance. To solve this challenge, authors of paper developed two new kinds vectors, including fractal geometry-based and quantitative structural model (QSM)-based vectors. In terms geometry, both parameters were extracted as reflecting how architecture distributed three-dimensional space. QSM, ratio length change radius different branches reduce vector dimensionality explore valuable dimension reduction was conducted using regression (CART). Five hundred sixty-eight individual trees with five selected evaluating performance The experimental results indicate Fagus sylvatica achieved highest overall accuracy, which 98.06%, while Quercus petraea obtained lowest 96.65%. Four other classical supervised learning adopted comparison. comparison result indicates proposed method outperformed four no matter accuracy indicator adopted. relevant method, eight also performed much better. This QSM-based can effectively improve