Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Robust Cross-Domain Adaptation in Texture Features Transferring for Wood Chip Moisture Content Prediction

Abdur Rahman, Mohammad Marufuzzaman, Jason Street, Haifeng Wang, Veera G. Gude, Randy Buchanan

TL;DR

This work tackles robust wood chip MC prediction across heterogeneous sources by leveraging texture features extracted from RGB images. It introduces AdaptMoist, a domain-adversarial framework that learns domain-invariant texture representations to transfer knowledge between sources without target labels, guided by an AMI-based model-saving callback. Empirical results show state-of-the-art single-source accuracy (0.95 with combined texture features) and a 23-point improvement in cross-domain accuracy (0.80 avg vs 0.57 without adaptation), with Haralick features offering the strongest transferability. The approach promises real-time, cross-source MC estimation for wood-chip industries and highlights future avenues including deeper models and raw-image integration for further gains.

Abstract

Accurate and quick prediction of wood chip moisture content is critical for optimizing biofuel production and ensuring energy efficiency. The current widely used direct method (oven drying) is limited by its longer processing time and sample destructiveness. On the other hand, existing indirect methods, including near-infrared spectroscopy-based, electrical capacitance-based, and image-based approaches, are quick but not accurate when wood chips come from various sources. Variability in the source material can alter data distributions, undermining the performance of data-driven models. Therefore, there is a need for a robust approach that effectively mitigates the impact of source variability. Previous studies show that manually extracted texture features have the potential to predict wood chip moisture class. Building on this, in this study, we conduct a comprehensive analysis of five distinct texture feature types extracted from wood chip images to predict moisture content. Our findings reveal that a combined feature set incorporating all five texture features achieves an accuracy of 95% and consistently outperforms individual texture features in predicting moisture content. To ensure robust moisture prediction, we propose a domain adaptation method named AdaptMoist that utilizes the texture features to transfer knowledge from one source of wood chip data to another, addressing variability across different domains. We also proposed a criterion for model saving based on adjusted mutual information. The AdaptMoist method improves prediction accuracy across domains by 23%, achieving an average accuracy of 80%, compared to 57% for non-adapted models. These results highlight the effectiveness of AdaptMoist as a robust solution for wood chip moisture content estimation across domains, making it a potential solution for wood chip-reliant industries.

Robust Cross-Domain Adaptation in Texture Features Transferring for Wood Chip Moisture Content Prediction

TL;DR

This work tackles robust wood chip MC prediction across heterogeneous sources by leveraging texture features extracted from RGB images. It introduces AdaptMoist, a domain-adversarial framework that learns domain-invariant texture representations to transfer knowledge between sources without target labels, guided by an AMI-based model-saving callback. Empirical results show state-of-the-art single-source accuracy (0.95 with combined texture features) and a 23-point improvement in cross-domain accuracy (0.80 avg vs 0.57 without adaptation), with Haralick features offering the strongest transferability. The approach promises real-time, cross-source MC estimation for wood-chip industries and highlights future avenues including deeper models and raw-image integration for further gains.

Abstract

Accurate and quick prediction of wood chip moisture content is critical for optimizing biofuel production and ensuring energy efficiency. The current widely used direct method (oven drying) is limited by its longer processing time and sample destructiveness. On the other hand, existing indirect methods, including near-infrared spectroscopy-based, electrical capacitance-based, and image-based approaches, are quick but not accurate when wood chips come from various sources. Variability in the source material can alter data distributions, undermining the performance of data-driven models. Therefore, there is a need for a robust approach that effectively mitigates the impact of source variability. Previous studies show that manually extracted texture features have the potential to predict wood chip moisture class. Building on this, in this study, we conduct a comprehensive analysis of five distinct texture feature types extracted from wood chip images to predict moisture content. Our findings reveal that a combined feature set incorporating all five texture features achieves an accuracy of 95% and consistently outperforms individual texture features in predicting moisture content. To ensure robust moisture prediction, we propose a domain adaptation method named AdaptMoist that utilizes the texture features to transfer knowledge from one source of wood chip data to another, addressing variability across different domains. We also proposed a criterion for model saving based on adjusted mutual information. The AdaptMoist method improves prediction accuracy across domains by 23%, achieving an average accuracy of 80%, compared to 57% for non-adapted models. These results highlight the effectiveness of AdaptMoist as a robust solution for wood chip moisture content estimation across domains, making it a potential solution for wood chip-reliant industries.
Paper Structure (19 sections, 1 theorem, 13 equations, 6 figures, 38 tables)

This paper contains 19 sections, 1 theorem, 13 equations, 6 figures, 38 tables.

Table of Contents

  1. Introduction
  2. Material and Methods
  3. Wood Chip Dataset Acquisition
  4. Texture Feature Extraction
  5. Haralick Features
  6. FOS
  7. FPS
  8. GLRLM
  9. LBP
  10. Predictive Models & Evaluation Metrics
  11. Proposed AdaptMoist Model
  12. Baselines for Domain Adaptation
  13. Implementation Details
  14. Results and Discussions
  15. Performance of Machine Learning Methods on Individual Data
  16. ...and 4 more sections

Key Result

Theorem 1

For a hypothesis $h\in \mathcal{H}$ and two domain $\mathcal{D}_s$ and $\mathcal{D}_t$, we have where $\epsilon _t(h)$ and $\epsilon _s(h)$ represent the error for target and source domains, respectively. The third term on the right side is the difference between the labeling functions across the two domains, which is expected to be negligible. The second term represents the $L1$ divergence me

Figures (6)

  • Figure 1: Wood chip acquisition method from multiple sources.
  • Figure 2: Wood chip data collection station: (a) Data collection setup, (b) Uniform white lights, (c) Wood chips in the container to capture images (Reproduced from Rahman et al. rahman2025moistnet with permission).
  • Figure 3: Sample images from the wood chip dataset. S1, S2, and S3 correspond to three different sources, and S2-B1 and S2-B2 represent two batches from S2. Dry, medium, and wet designations represent the three moisture content ranges.
  • Figure 4: Proposed AdaptMoist method to integrate unsupervised domain adaptation for predicting wood chip MC classes. Texture features $\hat{\mathbf{x}}_s$ and $\hat{\mathbf{x}}_t$ are initially extracted from the source and target raw images $\mathbf{x}_s$ and $\mathbf{x}_t$, respectively. These features serve as the input to a feature extractor network $F$, which produces the domain invariant features $\mathbf{f}$. A label classifier network $G$ and a domain discriminator network $D$ then take $\mathbf{f}$ as input and predict the moisture class $\hat{y}$ and $\hat{d}$, respectively. This framework learns domain-invariant feature $\mathbf{f}$ with the help of a gradient reversal layer in the domain discriminator $D$.
  • Figure 5: Confusion matrices of Haralick and combined features with the voting classifier on datasets from different sources and batches. Since we conducted k-fold cross-validation, confusion matrices for only the first folds have been illustrated. D, M, and W indicate dry, medium, and wet classes, respectively.
  • ...and 1 more figures

Theorems & Definitions (1)

  • Theorem 1