BarkXAI: A Lightweight Post-Hoc Explainable Method for Tree Species Classification with Quantifiable Concepts

TL;DR

The paper tackles the challenge of explainability for texture-dominant bark-based tree species classification by introducing BarkXAI, a lightweight post-hoc XAI that uses operator perturbations to reflect global visual features and surrogate models to quantify concept-importance without external concept-image datasets. It defines global visual features (Color, Texture, Shape, Groove/Surface), introduces commutative operators (CUCO) and a pipeline for perturbing and evaluating images, and employs surrogate models to reveal inter-concept relationships. The authors demonstrate that BarkXAI achieves superior alignment with human intuition (as measured by Kendall's $\tau$) compared with TCAV and Llama3.2 across 21 species, highlighting its practical utility for transparent bark-based classification in forestry. The approach offers a practical, domain-conscious path to interpretable texture-based vision, with potential limitations tied to operator design and subjective concept mappings.

Abstract

The precise identification of tree species is fundamental to forestry, conservation, and environmental monitoring. Though many studies have demonstrated that high accuracy can be achieved using bark-based species classification, these models often function as "black boxes", limiting interpretability, trust, and adoption in critical forestry applications. Attribution-based Explainable AI (XAI) methods have been used to address this issue in related works. However, XAI applications are often dependent on local features (such as a head shape or paw in animal applications) and cannot describe global visual features (such as ruggedness or smoothness) that are present in texture-dominant images such as tree bark. Concept-based XAI methods, on the other hand, offer explanations based on global visual features with concepts, but they tend to require large overhead in building external concept image datasets and the concepts can be vague and subjective without good means of precise quantification. To address these challenges, we propose a lightweight post-hoc method to interpret visual models for tree species classification using operators and quantifiable concepts. Our approach eliminates computational overhead, enables the quantification of complex concepts, and evaluates both concept importance and the model's reasoning process. To the best of our knowledge, our work is the first study to explain bark vision models in terms of global visual features with concepts. Using a human-annotated dataset as ground truth, our experiments demonstrate that our method significantly outperforms TCAV and Llama3.2 in concept importance ranking based on Kendall's Tau, highlighting its superior alignment with human perceptions.

Figures (7)

• Figure 1: Illustration of Attribution-Based Methods on a Bark Image (Trained and Evaluated with MobileNetV2) and domain expert explanation. While attribution-based methods are widely used in bark vision model explanation and similar fields, they cannot reveal global visual features used by domain experts.
• Figure 2: Pipeline of BarkXAI in Explaining Bark Vision Models
• Figure 3: Ten operations demonstration on Black cherry (Prunus serotina).
• Figure 4: Feature Importance of Operators on All Testing Images
• Figure 5: Examples of Explanations Generated using BarkXAI(LR). Feature importance values are calculated with softmax performed on slope values. The relative importance of inferred concepts are calculated based on feature importance.