Deep Learning

Object Detection and Semantic Segmentation for the Analysis of Complex Biointerfaces

Scematic of micro- and macrofouler detection using deep neural networks

We promote the utilization of modern deep learning methods for the automated analysis of complex surfaces from (short-term) field tests. These are valuable tools for the quantitative and unbiased evaluation of novel coatings. For the reliable detection of microfoulers in the early stages of biofouling, we established a fast neural network counting diatoms in difficult environments with a human-like precision on fluorescence microscopy images.[1] For the detailed determination of the sophisticated macrofouler distribution in later biofouling stages, we employed advanced semantic segmentation[2] of common but highly diverse field panel photos with unique precision.

Students working on this project: Lutz Krause

[1] Fully Convolutional Neural Network for Detection and Counting of Diatoms on Coatings after Short-Term Field Exposure, Lutz M. K. Krause, Julian Koc, Bodo Rosenhahn, and Axel Rosenhahn, Environmental Science & Technology 2020 54 (16), 10022-10030, DOI: 10.1021/acs.est.0c01982

[2] Semantic Segmentation for Fully Automated Macrofouling Analysis on Coatings after Field Exposure, Lutz M. K. Krause, Emily Manderfeld, Patricia Gnutt, Louisa Vogler, Ann Wassick, Kailey Richard, Marco Rudolph, Kelli Z. Hunsucker, Geoffrey W. Swain, Bodo Rosenhahn, and Axel Rosenhahn (in review)

 

Dataset and Models for Semantic Segmentation of Macrofouling Organisms

Different fouler images with the respective ground truth

This is the official website for the dataset and models used in our publication „Semantic Segmentation for Fully Automated Macrofouling Analysis on Coatings after Field Exposure“ by Lutz M. K. Krause, Emily Manderfeld, Patricia Gnutt, Louisa Vogler, Ann Wassick, Kailey Richard, Marco Rudolph, Kelli Z. Hunsucker, Geoffrey W. Swain, Bodo Rosenhahn, and Axel Rosenhahn. Also, a preprint version of our paper is available which does not contain any changes that have been made during the peer-review process.

You can download the dataset here and trained models here. Currently, only the validation data is publicly available. If you need more data for your research please contact the authors. Consider our GitHub page for detailed information about using the dataset and trained models. If our paper helps your research, please cite it in your publication.