Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Instance camera focus prediction for crystal agglomeration classification

Xiaoyu Ji, Chenhao Zhang, Tyler James Downard, Zoltan Nagy, Ali Shakouri, Fengqing Zhu

TL;DR

This work tackles crystal agglomeration classification in 2D microscopy, where depth-induced defocus can mislead interpreters. It introduces an integrated pipeline combining a Yolo-focus per-instance focus predictor with a Yolo-segmentation model, using a contrast focus metric—computed from neighboring focus levels—to distinguish agglomerated from non-agglomerated crystals. Ablation studies show that a Contrast 2 strategy, which emphasizes depth differences among neighbors, yields superior separation and accuracy, outperforming Mask R-CNN and Yolov12 baselines on ammonium perchlorate and sugar crystal datasets. The approach improves both instance segmentation and agglomeration classification, offering a robust tool for microscopic crystallization analysis and depth-aware interpretation of 2D images.

Abstract

Agglomeration refers to the process of crystal clustering due to interparticle forces. Crystal agglomeration analysis from microscopic images is challenging due to the inherent limitations of two-dimensional imaging. Overlapping crystals may appear connected even when located at different depth layers. Because optical microscopes have a shallow depth of field, crystals that are in-focus and out-of-focus in the same image typically reside on different depth layers and do not constitute true agglomeration. To address this, we first quantified camera focus with an instance camera focus prediction network to predict 2 class focus level that aligns better with visual observations than traditional image processing focus measures. Then an instance segmentation model is combined with the predicted focus level for agglomeration classification. Our proposed method has a higher agglomeration classification and segmentation accuracy than the baseline models on ammonium perchlorate crystal and sugar crystal dataset.

Instance camera focus prediction for crystal agglomeration classification

TL;DR

This work tackles crystal agglomeration classification in 2D microscopy, where depth-induced defocus can mislead interpreters. It introduces an integrated pipeline combining a Yolo-focus per-instance focus predictor with a Yolo-segmentation model, using a contrast focus metric—computed from neighboring focus levels—to distinguish agglomerated from non-agglomerated crystals. Ablation studies show that a Contrast 2 strategy, which emphasizes depth differences among neighbors, yields superior separation and accuracy, outperforming Mask R-CNN and Yolov12 baselines on ammonium perchlorate and sugar crystal datasets. The approach improves both instance segmentation and agglomeration classification, offering a robust tool for microscopic crystallization analysis and depth-aware interpretation of 2D images.

Abstract

Agglomeration refers to the process of crystal clustering due to interparticle forces. Crystal agglomeration analysis from microscopic images is challenging due to the inherent limitations of two-dimensional imaging. Overlapping crystals may appear connected even when located at different depth layers. Because optical microscopes have a shallow depth of field, crystals that are in-focus and out-of-focus in the same image typically reside on different depth layers and do not constitute true agglomeration. To address this, we first quantified camera focus with an instance camera focus prediction network to predict 2 class focus level that aligns better with visual observations than traditional image processing focus measures. Then an instance segmentation model is combined with the predicted focus level for agglomeration classification. Our proposed method has a higher agglomeration classification and segmentation accuracy than the baseline models on ammonium perchlorate crystal and sugar crystal dataset.
Paper Structure (13 sections, 3 figures, 2 tables)

This paper contains 13 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Yolo-segmentation Model
  4. Yolo-focus Model
  5. Contrast Focus Measurement
  6. Dataset
  7. Experiments
  8. Experimental setting
  9. Quantitative Comparison
  10. Agglomeration Classification and Segmentation Accuracy
  11. Instance Camera Focus Distribution
  12. Qualitative Comparison
  13. Conclusion

Figures (3)

  • Figure 1: Overview of our proposed agglomeration classification method. Yolo-Segmentation model predicts instance segmentation masks and the agglomeration classification is based on Yolo-focus model predictions and contrast focus measurement.
  • Figure 2: Normalized average contrast focus comparison. Class1 and class2 represent non-agglomerated and agglomerated crystals. The normalized average contrast focus is calculated as the mean contrast focus of each method divided by the maximum contrast focus. From left to right are Laplacian variance Sub1993 on entire mask and mask contour, Brenner Brenner1976, Reblur crete07 and our methods. A larger difference between class 1 and class 2 indicates better separation of focus levels.
  • Figure 3: Visual comparison of three example images. Red circles highlight out-of-focus crystals in (a)-(c). Blue color indicates non-agglomerated and yellow color indicates agglomerated.