Prethodno priopćenje

Geometric approximation of pollen grains using ellipse packing optimization

Deni Vale ; Istarsko veleučilište u Puli, Pula, Hrvatska

Sažetak

This paper presents a new methodology for pollen grain detection that solves the problem of
subsegmentation in dense clusters. Instead of classical segmentation at the pixel level, the problem is
defined as a geometric approximation of a binary mask by a set of parametric ellipses. Initial positioning
of candidate centers is performed using a distance transformation, which significantly narrows the
search space. The main phase involves the optimization of a multi-criteria objective function that
balances mask coverage and the number of objects while penalizing mutual overlaps. Redundancy in
detections is removed by applying a non-maximum suppression (NMS) procedure adapted to geometric
primitives. Preliminary validation on willow, angelica and maple samples showed high accuracy of the
method even in conditions where the boundaries between grains visually disappeared. While standard
blob detection shows significant inadequacy in saturated regions, the proposed method achieves
results comparable to the advanced Cellpose system. Although Cellpose was originally intended for
cell and bacterial segmentation, its application to pollen confirms the competitiveness of the proposed
model in the extraction of biometric parameters. This approach lays a solid foundation for further
automation and increasing the objectivity of melissopalynological analysis.

Ključne riječi

pollen grain detection, object counting, morphological processing, multi-criteria model evaluation, geometric approximation by ellipses

Hrčak ID:

352239

URI

https://hrcak.srce.hr/352239

Datum izdavanja:

15.10.2026.

Podaci na drugim jezicima: hrvatski

Posjeta: 0 *