Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Explainable deep-learning detection of microplastic fibers via polarization-resolved holographic microscopy

Jan Appel, Marika Valentino, Lisa Miccio, Vittorio Bianco, Raffaella Mossotti, Giulia Dalla Fontana, Miroslav Ježek, Pietro Ferraro, Jaromír Běhal

TL;DR

The paper tackles reliable, label-free discrimination of microplastic and natural microfibers, a key environmental monitoring challenge. It introduces an explainable deep-learning pipeline that uses polarization-resolved holographic microscopy to reconstruct the Jones matrix, extract eigenpolarization-based features, and classify six microfiber classes with high accuracy. SHAP-based explainability reveals that eigenvalue-based features, particularly the ENs ratio $|EN_1/EN_2|$, dominate the predictions, and a reduced EN-focused feature set achieves substantial performance while enhancing interpretability. The approach yields $96.7\%$ validation accuracy (and $98.6\%$ on the full dataset) and demonstrates the potential of polarization fingerprints for non-destructive, scalable microplastic fiber analysis, with data openly available for validation.

Abstract

Reliable identification of microplastic fibers is crucial for environmental monitoring but remains analytically challenging. We report an explainable deep-learning framework for classifying microplastic and natural microfibers using polarization-resolved digital holographic microscopy. From multiplexed holograms, the complex Jones matrix of each fiber was reconstructed to extract polarization eigen-parameters describing optical anisotropy. Statistical descriptors of nine polarization characteristics formed a 72-dimensional feature vector for a total of 296 fibers spanning six material classes, including polyamide 6, polyethylene terephthalate, polyamide 6.6, polypropylene, cotton and wool. The designed fully connected deep neural network achieved an accuracy of 96.7 % on the validation data, surpassing that of common machine-learning classifiers. Explainable artificial intelligence analysis with Shapley additive explanations identified eigenvalue-ratio quantities as dominant predictors, revealing the physical basis for classification. An additional reduced-feature model with the preserved architecture exploiting only these most significant eigenvalue-based characteristics retained high accuracy (93.3 %), thereby confirming their dominant role while still outperforming common machine-learning classifiers. These results establish polarization-based features as distinctive optical fingerprints and demonstrate the first explainable deep-learning approach for automated microplastic fiber identification.

Explainable deep-learning detection of microplastic fibers via polarization-resolved holographic microscopy

TL;DR

The paper tackles reliable, label-free discrimination of microplastic and natural microfibers, a key environmental monitoring challenge. It introduces an explainable deep-learning pipeline that uses polarization-resolved holographic microscopy to reconstruct the Jones matrix, extract eigenpolarization-based features, and classify six microfiber classes with high accuracy. SHAP-based explainability reveals that eigenvalue-based features, particularly the ENs ratio , dominate the predictions, and a reduced EN-focused feature set achieves substantial performance while enhancing interpretability. The approach yields validation accuracy (and on the full dataset) and demonstrates the potential of polarization fingerprints for non-destructive, scalable microplastic fiber analysis, with data openly available for validation.

Abstract

Reliable identification of microplastic fibers is crucial for environmental monitoring but remains analytically challenging. We report an explainable deep-learning framework for classifying microplastic and natural microfibers using polarization-resolved digital holographic microscopy. From multiplexed holograms, the complex Jones matrix of each fiber was reconstructed to extract polarization eigen-parameters describing optical anisotropy. Statistical descriptors of nine polarization characteristics formed a 72-dimensional feature vector for a total of 296 fibers spanning six material classes, including polyamide 6, polyethylene terephthalate, polyamide 6.6, polypropylene, cotton and wool. The designed fully connected deep neural network achieved an accuracy of 96.7 % on the validation data, surpassing that of common machine-learning classifiers. Explainable artificial intelligence analysis with Shapley additive explanations identified eigenvalue-ratio quantities as dominant predictors, revealing the physical basis for classification. An additional reduced-feature model with the preserved architecture exploiting only these most significant eigenvalue-based characteristics retained high accuracy (93.3 %), thereby confirming their dominant role while still outperforming common machine-learning classifiers. These results establish polarization-based features as distinctive optical fingerprints and demonstrate the first explainable deep-learning approach for automated microplastic fiber identification.
Paper Structure (6 sections, 5 figures, 1 table)

This paper contains 6 sections, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Measurements and feature extraction
  3. Neural network architecture
  4. Classification performance
  5. Explainability
  6. Conclusions

Figures (5)

  • Figure 1: (a) Simplified sketch of experimental configuration. (b) Example of retrieved Jones matrix for PA6.6 MFF. (c) Local shape map and map orthogonal to the local shape compared to parameters of eigenpolarizations $EP_1$ and $EP_2$, respectively. (d) Histograms of the derived polarization characteristic. Black scale bars in (b) and (c) represent 100 $\mu$m lengths.
  • Figure 2: The implemented neural network with input layer (green) with dimensions corresponding to the number of features. Four hidden layers (grey) with a progressively decreasing number of neurons. All hidden layers in the network employ regularization techniques, specifically: batch normalization, dropout and additional regularization methods, see text for details.
  • Figure 3: Confusion matrix of the model trained on labeled data. (a) Evaluation on the validation data (60 labeled MFFs). (b) Evaluation on the complete dataset of training and validation data (296 labeled MFFs).
  • Figure 4: SHAP-FI values of nine polarization characteristics and their eight corresponding statistical parameters, yielding 72 SHAP-FI values. $\left|EV_1\cdot EV_2\right|$ - magnitude of the EVs inner product; $\left|EN_{1}/EN_{2}\right|$, phase$(EN_{1}/EN_{2})$, Re$(EN_{1}/EN_{2})$, Im$(EN_{1}/EN_{2})$ - modulus, phase, real part, and imaginary part of the ENs ratio; $\chi_{EP1}$, $\chi_{EP2}$ - ellipticity angles of $EP_1$ and $EP_2$; $\psi_{EP1}$-sh. - orientation of $EP_1$ relative to the local shape; $\psi_{EP2}$-sh.orth. - orientation of $EP_2$ relative to the direction perpendicular to the local shape.
  • Figure 5: Class-wise SHAP-FI values of nine polarization characteristics. The most significant polarization characteristic for synthetic fibers is the absolute value of the EN ratio, i.e. $\left|EN_{1}/EN_{2}\right|$. For natural fibers, $\chi_{EP2}$ appears to be the most important indicator for cotton and Im$(EN_{1}/EN_{2})$ for wool.