Publication detail

Improvement of Microbial Colony Segmentation using Synthetic Data

CICATKA, M. BURGET, R LANCOS, J

Original Title

Improvement of Microbial Colony Segmentation using Synthetic Data

Type

article in a collection out of WoS and Scopus

Language

English

Original Abstract

This paper presents an image processing pipeline for the generation of synthetic agar plate images. The generator superimposes patches of microbial colonies onto the images of empty agar plates. Data acquisition and processing of both the empty agar plate images and colony patches is described. A testing dataset was created which consists of real-life agar plate images with various microbes cultivated on various agars. The performance of the generator is evaluated by trained U-Net models with training and validation dataset enlarged by 33 %, 66 %, 100 %, 150 % and 200 % of artificially created data. The best performance (0.729 F1 score and 0.656 Jaccard index) was achieved by a model trained on 150 % of additional synthetic data. However, all models trained with more than 66 % of the artificially generated data showed comparable performance, with only small differences in F1 score and Jaccard index.

Keywords

agar plates, synthetic data generator, segmentation, microbial colonies

Authors

CICATKA, M.; BURGET, R; LANCOS, J

Released

25. 4. 2023

Publisher

Vysoké učení technické v Brně, Fakulta elektrotechniky a komunikačních technologií

Location

Brno

Pages count

5

URL

BibTex

@inproceedings{BUT184373,
  author="Michal {Čičatka} and Radim {Burget} and Jan {Láncoš}",
  title="Improvement of Microbial Colony Segmentation using Synthetic Data",
  year="2023",
  pages="5",
  publisher="Vysoké učení technické v Brně, Fakulta elektrotechniky a komunikačních technologií",
  address="Brno",
  url="https://www.eeict.cz/download"
}