11.06.2024: Philipp Faulhammer & Stefan Herdy (Univ. Graz): Application of image-based machine learning methods for the assessment of cryptogamic covers, HS 32.01, Institut für Biologie, Bereich Pflanzenwissenschaften, Holteigasse 6, 17.00 Uhr
Cryptogamic communities (CC), consisting of lichens, bryophytes, fungi and cyanobacteria, occur in a variety of habitats all around the world. They play an important role in global biogeochemical processes, as they fix carbon and nitrogen, reduce soil erosion by water and wind, and store and release water due to their poikilohydric nature. For a quantification of these processes, it is important to estimate the overall coverage of cryptogamic communities as well as the coverage of individual organism groups. These estimations can be done efficiently on data sets of images using machine learning (ML) methods. In our talk, we present three approaches: 1. To identify and map biological soil crust communities of drylands, 2. To classify and characterize epiphytic communities in the Amazon rain forest, and 3. To distinguish and characterize species of the liverwort genus Riccia, based on the spore characteristics. We show how image datasets were recorded and prepared for the optimization of machine learning models as well as the application of these models for data analysis. We also present a new method to enable a stable image segmentation under changing conditions, e.g., due to changing camera devices and sensors over time. To facilitate the accessibility of our models, we present a Web-Application (CC-Explorer) that we are currently developing, and which can be used for the analysis of image datasets for biomonitoring.