7. July 2020 - 8. July 2020
This special GEOBIA track will be a 1.5 days conference within the multi-faceted GI_Week, taking place from July 7 – 10, 2020. As continuation of seven successful GEOBIA conferences since its launch in 2006, this special event will return “home” – to the place where it all began.
The primary goal of Earth observation (EO) is to map, analyse, and monitor the status and dynamics of complex Earth system phenomena. Through the spatial lens and ‘imaging’ reality, we shall obtain quantifiable indicators and other objective information for improved decision-making and status reporting. Within EO applications, object-based image analysis (OBIA) has become a key methodological enabler to address spatial, topological, and hierarchical properties of image objects with an underlying conceptual model close to the way how we perceive (and mentally organise) reality.
Recent advances in big EO data management and analytics, including data cube solutions, massive data storage and access infrastructure, has pushed for higher grounds in operational image processing and analysis-ready data; but often re-emphasizing per-pixel analysis. Machine learning, in particular deep learning using CNNs, can also employ scale and neighbourhood, while the performance largely depends on quality and amount of samples. Hybrid approaches may foster the combined use of knowledge-based and (physical) model-based approaches and machine learning. On finer levels of elementary image objects, where sufficient high-quality samples are abundant, ML-strategies may outperform the tedious process of explicit object descriptors and segmentation tuning; complementary, when generating composite objects on higher organisational levels, where samples are sparse and clear models and definitions exist, a knowledge-based direct modelling of object properties and relationships may be more cost- and time-efficient.
Topics include, but are not limited, to:
- Spatio-temporal concepts in image understanding
- Multi-scale image analysis
- Object-specific quality indicators
- Hybrid approaches (combining machine learning and machine teaching)
- State of play of OBIA methods and tools in open source and commercial software environments
- Best practice OBIA in operational solutions (e.g. Copernicus info services, SDG indicators, etc.)
- Big EO data – back to pixels?