Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Analysis of 3D Urticaceae Pollen Classification Using Deep Learning Models

Tijs Konijn, Imaan Bijl, Lu Cao, Fons Verbeek

TL;DR

This work tackles automated pollen classification in the Urticaceae family using full 3D z-stack microscopy and deep learning. It systematically compares multiple 3D architectures (ResNet3D, MobileNetV2 3D, SwinTransformer 3D) and 2D-anchored object-detection backbones (RetinaNet, Faster R-CNN) adapted to 3D inputs, with a preprocessing step that selects the central focal layer via edge strength. The best result comes from a pre-trained ResNet3D backbone achieving a F1-score of approximately 0.981, highlighting the value of 3D spatial information and pretrained weights, while other models offer competitive performance with different trade-offs. The study underscores the potential for 3D CNN-based pollen monitoring while noting limitations due to domain mismatch in pretrained weights and suggesting future domain-specific pretraining and broader taxonomic coverage to enhance generalizability and utility for environmental health applications.

Abstract

Due to the climate change, hay fever becomes a pressing healthcare problem with an increasing number of affected population, prolonged period of affect and severer symptoms. A precise pollen classification could help monitor the trend of allergic pollen in the air throughout the year and guide preventive strategies launched by municipalities. Most of the pollen classification works use 2D microscopy image or 2D projection derived from 3D image datasets. In this paper, we aim at using whole stack of 3D images for the classification and evaluating the classification performance with different deep learning models. The 3D image dataset used in this paper is from Urticaceae family, particularly the genera Urtica and Parietaria, which are morphologically similar yet differ significantly in allergenic potential. The pre-trained ResNet3D model, using optimal layer selection and extended epochs, achieved the best performance with an F1-score of 98.3%.

Analysis of 3D Urticaceae Pollen Classification Using Deep Learning Models

TL;DR

This work tackles automated pollen classification in the Urticaceae family using full 3D z-stack microscopy and deep learning. It systematically compares multiple 3D architectures (ResNet3D, MobileNetV2 3D, SwinTransformer 3D) and 2D-anchored object-detection backbones (RetinaNet, Faster R-CNN) adapted to 3D inputs, with a preprocessing step that selects the central focal layer via edge strength. The best result comes from a pre-trained ResNet3D backbone achieving a F1-score of approximately 0.981, highlighting the value of 3D spatial information and pretrained weights, while other models offer competitive performance with different trade-offs. The study underscores the potential for 3D CNN-based pollen monitoring while noting limitations due to domain mismatch in pretrained weights and suggesting future domain-specific pretraining and broader taxonomic coverage to enhance generalizability and utility for environmental health applications.

Abstract

Due to the climate change, hay fever becomes a pressing healthcare problem with an increasing number of affected population, prolonged period of affect and severer symptoms. A precise pollen classification could help monitor the trend of allergic pollen in the air throughout the year and guide preventive strategies launched by municipalities. Most of the pollen classification works use 2D microscopy image or 2D projection derived from 3D image datasets. In this paper, we aim at using whole stack of 3D images for the classification and evaluating the classification performance with different deep learning models. The 3D image dataset used in this paper is from Urticaceae family, particularly the genera Urtica and Parietaria, which are morphologically similar yet differ significantly in allergenic potential. The pre-trained ResNet3D model, using optimal layer selection and extended epochs, achieved the best performance with an F1-score of 98.3%.

Paper Structure

This paper contains 12 sections, 3 figures, 4 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. METHOD
  4. Data collection
  5. Preprocessing
  6. Training
  7. RESULTS
  8. Optimal Number of layers
  9. Increasing Amount of Epochs
  10. Comparison of Deep Learning Models
  11. DISCUSSION AND CONCLUSION
  12. FUTURE WORK

Figures (3)

  • Figure 1: A z-stack example of pollen grain. The z-stack contains 20 z-layers, with z-layer 1 on top left and z-layer 20 on bottom right.
  • Figure 2: Example of edge detection on pollen grain. Figure on the left side is the normal one layer image of a pollen grain. On the right side is the result of edge detection.
  • Figure 3: These graphs represent the pre-trained ResNet3D model with 6 optimal layers and 50 epochs. On the test set this model achieved a F1-score of 0.982 and a loss of 0.053.