Algorithmic evaluation study demonstrates 90% accuracy in classifying tree species in an urban environment, highlighting an effective framework for automated municipal forestry.
Key Points
To develop a machine learning pipeline that optimizes hyperparameters and selects optimal features to distinguish deciduous and coniferous tree species using urban LiDAR point cloud data.
Segmented individual tree crowns from LiDAR data in Fatih, Istanbul, via hierarchical rule-based classification and mean shift clustering, validated with field reference data.
Optimized Random Forest algorithm hyperparameters using a grid search approach.
Assessed feature importance using Mean Decrease in Gini to isolate eight optimal classification features in custom Python code.
The optimized Random Forest algorithm achieved a 90% overall classification accuracy using the top eight selected features.
The automated pipeline successfully classified tree crowns into deciduous and coniferous classes across the urban test area.