Tree classification with fused mobile laser scanning and hyperspectral data.

Area of Science:

• Forestry and Remote Sensing
• Geospatial Data Analysis
• Ecology

Background:

• Accurate tree species identification is crucial for urban forest management and ecological studies.
• Traditional methods often rely on field surveys or aerial imagery, which can be time-consuming and limited in detail.
• Mobile sensing technologies offer new possibilities for detailed forest inventories.

Purpose of the Study:

• To evaluate the effectiveness of fused mobile laser scanning (MLS) and hyperspectral data for tree classification.
• To compare the performance of combined sensor data against individual spatial and spectral data sources.
• To assess classification accuracy for distinguishing coniferous vs. deciduous trees and identifying individual tree species.

Main Methods:

• Collected simultaneous MLS (point cloud) and hyperspectral data using the Finnish Geodetic Institute Sensei system in an urban garden.
• Manually identified 168 individual trees of 23 species for ground truth.
• Performed tree classification using spatial data only, spectral data only, and fused data.
• Evaluated classification accuracy for broad categories (coniferous/deciduous) and specific species.

Main Results:

• Fused MLS and hyperspectral data achieved 95.8% accuracy for coniferous/deciduous separation and 83.5% for species identification.
• Single-sensor data yielded lower accuracies: MLS (90.5% and 65.4%) and hyperspectral (90.5% and 62.4%).
• The combined data significantly outperformed individual sensor data in both classification tasks.

Conclusions:

• Mobile collected fused spatial and hyperspectral data offer a significant advantage over single-sensor approaches for tree classification.
• This methodology provides a robust and accurate tool for detailed urban forest inventories and ecological assessments.
• The findings highlight the potential of integrated mobile sensing for advancing forest research and management.