Enhancing Tree Species Classification: Insights from YOLOv8 and Explainable AI Applied to TLS Point Cloud Projections

High accuracy achieved with interpretable AI: The study demonstrates that YOLOv8 models can classify tree species with 96% accuracy while remaining interpretable, showing that strong performance and explainability are not mutually exclusive in AI systems.

AI decisions are traceable to specific features: Using explainable AI techniques (Finer-CAM), researchers identified exactly which parts of tree images the model uses to make decisions—primarily tree crowns for most species and stems for others—enabling verification that the AI relies on biologically meaningful characteristics rather than spurious patterns.

Explainability reveals species-specific classification strategies: Different tree species are distinguished by different structural features (crown shape vs. stem characteristics), insights that would remain hidden without explainable AI and could inform both model improvement and scientific understanding of tree morphology.

Framework enables systematic AI auditing: The paper presents a replicable methodology for analyzing what deep learning models "see" when making decisions, providing a practical approach that other organizations could adopt to verify their AI systems are making decisions for the right reasons.

Responsible AI requires looking inside the "black box": This work demonstrates why explainability matters in practice—without understanding which features drive classifications, developers cannot confidently deploy AI systems or identify potential biases or errors in model reasoning.