Pollen Recognition

Automated Recognition of Allergenic Pollen: Grass, Birch and Mugwort

Graduation project in collaboration with Dr. Emile A. Hendriks
Information and Communication Theory Group, Technical University Delft.

Chun Chen, Berend Stoel

Background

Approximately five percent of the Dutch population sometimes suffer from hay fever (pollinose). This allergic reaction is caused by pollen from different species of plants (predominantly grass) and trees. When the symptoms (stuffy nose, running eyes, sometimes shortness of breath) become serious, patients can use their drugs or take other measures (i.e. stay indoors). In order to help these patients taking these measures and support the medical doctor in diagnosing hay fever, every week a so-called pollen forecast is given, which is broadcasted on Radio 1.
GrasTo formulate this forecast, samples are taken from the air with a type of vacuum cleaner, which is located on the roof of the LUMC. This vacuum cleaner hoovers the outside air along a thin strip of cellulose, which is covered by a layer of Vaseline, in which small particles are captured. Subsequently, the pollen are being counted manually with the use of a light microscope (see Figure 1). To speed up this procedure, it would be advantageous if these pollen would be detected automatically from the images and if the pollen would be categorized automatically.

Goals

  • To develop methods for calculating features of the different pollen from the microscopic images
  • To determine which features are the most discriminative for distinguishing grass, birch and mugwort, as a pilot study
  • To evaluate different pattern recognition methods

Approach

  • Pollen images were acquired from 254 purified samples of grass, birch and mugwort 
  • Pollen were detected in the images, through simple thresholding
  • Three types of features were measured:
    • shape features (binary)
    • statistical gray-invariant features (gray)
    • specific features (pore & colpus) (see Figure 2-4)
  • Parameter settings were optimized, and classifiers were trained
  • In a cross-validation study a classification rate of 97.2% was obtained

Status

Finished

Publication

Chun C, Hendriks EA, Duin RPW, Reiber JHC, Hiemstra PS, de Weger LA, Stoel BC, Feasibility study on automated recognition of allergenicpollen: grass, birch and mugwort, Aerobiologia 22:275–284, 2006

Gallery

Figure 1
Figure 1. Example pollen image.

Figure 2
Figure 2. Pore detection in grass.

Figure 3
Figure 3. Pore detection in birch.

Figure 4
Figure 4. Colpi detection in mugwort

Contact

For further information, please contact:
B.C. Stoel, PhD.
Division of Image Processing
Department of Radiology, 1-C2S
Leiden University Medical Center
P.O. Box 9600
2300 RC Leiden
The Netherlands
Tel. +31 (0)71 526 1911
Fax. +31 (0)71 526 6801
e-mail: B.C.Stoel@lumc.nl