An Asbestos Fiber Detection Technique Utilizing Image Processing Based on Dispersion Color

Kuniaki Kawabata, Yuta Komori, Taketoshi Mishima, Hajime AsamaParticulate 2009 · DOI

TL;DR

This paper introduces an automated image processing technique to detect and count asbestos fibers, building upon the established dispersion staining method. By analyzing color changes induced by polarization and employing image alignment, the system aims to reduce the labor-intensive and time-consuming nature of manual microscopic analysis, offering a more efficient and objective approach to asbestos identification.

What Problem Does This Paper Address?

Asbestos is a naturally occurring mineral valued for its strength, heat resistance, and insulating properties, leading to its widespread use in construction. However, inhaling dispersed asbestos fibers can cause severe, long-latency health issues like mesothelioma and lung cancer. Current methods for identifying asbestos in building materials primarily rely on X-ray diffraction analysis and visual inspection under a microscope, with dispersion staining being a key visual technique. While effective, manual visual inspection is extremely labor-intensive, time-consuming, and prone to human error, limiting the number of samples that can be analyzed daily. This paper addresses the need for a more efficient, objective, and automated method to support asbestos fiber analysis, particularly for identifying fibers within complex construction material samples.

Methods

The core of the proposed method lies in leveraging the optical properties of asbestos fibers when subjected to polarized light. The technique involves:

  1. Image Acquisition: Two images of the same sample are captured using a phase-contrast dispersion microscope, with the polarization angle changed between captures. This exploits the phenomenon where asbestos fibers exhibit distinct color dispersion patterns under polarized light.
  2. Image Alignment: To accurately compare the two images, a robust image alignment algorithm is employed. This process involves converting images to grayscale and comparing pixel brightness values, along with their surrounding pixels, to correct for translational shifts (vertical and horizontal). Median values are used to mitigate the impact of outliers caused by slight variations in reflected light.
  3. Candidate Fiber Identification: After alignment, the system calculates the sum of squared differences in RGB values between corresponding pixels in the two images. Pixels exceeding a defined distance threshold ($t_d$) are binarized, highlighting potential asbestos fibers. A low threshold ($t_d=100$) is used to maximize detection, even at the risk of some false positives.
  4. Aspect Ratio Calculation: For identified candidate fibers, an aspect ratio is calculated. This involves rotating the binarized fiber image in increments and enclosing it in a bounding rectangle to determine the maximum aspect ratio. This step is crucial for distinguishing harmful fibers, which are typically defined by a minimum aspect ratio.
  5. Harmful Fiber Detection: Candidate asbestos fibers with an aspect ratio of 2.0 or greater are considered to be potentially harmful and are labeled for counting. This threshold was determined through experimental analysis to balance detection failures and erroneous detections.

Datasets

The study utilized commercial samples of chrysotile, amosite, and crocidolite asbestos. Prepared samples were fabricated for observation. The paper lists the "Original images used for the experiment" in Table 4, detailing the image type and the asbestos fiber count for each sample type (chrysotile, amosite, crocidolite), with counts of 78, 108, and 76, respectively, across 39, 54, and 38 sets of images.

Key Insights and Ideas

  • Dispersion Color + Polarization = Detection: The fundamental insight is that the combination of dispersion staining (color dispersion in an immersion liquid) and polarized light creates unique, detectable color shifts in asbestos fibers. This optical phenomenon is the primary basis for distinguishing asbestos from other materials.
  • Image Alignment is Critical: Accurately aligning images taken under slightly different polarization conditions is paramount. The paper highlights that even minor pixel misalignments can lead to significant errors in detecting color differences, underscoring the importance of their robust alignment algorithm.
  • Aspect Ratio as a Harmfulness Indicator: The aspect ratio of a fiber is not just a shape descriptor but a critical parameter for identifying "harmful" asbestos fibers. The research meticulously investigates how different aspect ratio thresholds affect detection accuracy and erroneous detections, leading to the selection of an optimal threshold.
  • Trade-off Between Detection and False Positives: The study explicitly addresses the challenge of setting the distance threshold ($t_d$) for candidate fiber identification. A lower threshold increases the likelihood of detecting actual asbestos fibers but also increases the number of false positives (erroneous detections). This highlights a common challenge in image analysis where sensitivity and specificity must be balanced.

Key Results

The proposed technique was evaluated on samples of chrysotile, amosite, and crocidolite. The statistical analysis of the experimental results showed:

  • Low Detection Failures: The average number of asbestos fiber detection failures per image was approximately 0.4 for chrysotile and amosite, and approximately 0.2 for crocidolite.
  • Moderate Erroneous Detections: The average number of incorrectly identified asbestos fiber droplets was approximately 1.2 for chrysotile and amosite, and approximately 4.0 for crocidolite.
  • Distinguishing Fibers: The method successfully distinguished between asbestos and other fibrous materials by considering both shape and dispersion color differences. For example, in Figure 6, droplets other than crocidolite were detected by color tone but were filtered out by the aspect ratio calculation.

Limitations and Future Work

  • Static Images: The current technique is designed for static microscopic images. The authors acknowledge that applying this to real-time detection using a video camera would require further improvements.
  • Aspect Ratio Influence: The paper notes that the aspect ratio calculation can be influenced by color tone differences and reflected light, potentially leading to smaller calculated aspect ratios than the true ones.
  • Crocidolite Erroneous Detections: While detection failures were low for crocidolite, the average number of erroneous detections was higher (around 4.0) compared to chrysotile and amosite. This is attributed to a higher number of crocidolite droplets per image and potentially weaker reflected light making distinction harder.
  • Future Work: The authors explicitly state that future work will focus on applying the technique to real-time detection using video cameras.

Summary updated

Abstract

Abstract In an asbestos qualitative analysis, the major methods are X-ray diffraction analysis and visual observation by operators using a microscope. In particular, a major method of visual evaluation is dispersion staining. In the usual visual observation process, the operators check the asbestos fibers in the view of the microscope and count the number of asbestos fibers. The method presented here attempts to detect asbestos fibers in the images taken by a microscope. The dispersion staining method identifies asbestos fibers with color dispersion from an immersion liquid combined with polarization. The presented method employs color changes of asbestos fibers associated with polarization. Specifically, candidate asbestos fibers are identified using the changes of dispersion colors and also position-matching between two images. The performance of the method has been evaluated by comparing its results with the results obtained by a human expert. Keywords: asbestosdispersion colorimage processingqualitative analysis Acknowledgments The authors give thanks to Takeda-Rika Co. and Japan Testing Center for Construction Materials (JTCCM) for their kind contribution to this research. This research was supported by the research program of Ministry of the Environment of Japan (Nos. K1920, K2061).
BibTeX
@inproceedings{kawabata2009asbestos,
  author = {Kuniaki Kawabata and Yuta Komori and Taketoshi Mishima and Hajime Asama},
  title = {{An Asbestos Fiber Detection Technique Utilizing Image Processing Based on Dispersion Color}},
  booktitle = {Particulate},
  year = {2009},
  doi = {10.1080/02726350902776259},
}
Figure