Since May 2021 dida works together with the Remote Sensing Methodology Chair (LMF) of the Technical University of Munich (TUM) on a project "PreTrainAppEO" funded by the German Federal Ministry for Economic Affairs and Energy (BMWI). The project is granted with an amount of 199k EUR and is designed for a period of two years. The name "PreTrainAppEO" stands for Pre-Training Applicability in Earth Observation.
Earth observation and remote sensing have become an important part of, for example, transportation and utility sectors, as well as earth sciences and meteorology, due to technological advances in space and imaging technologies and sensors. At the same time, the use of machine learning (ML) methods in the analysis and interpretation of image and sensor data is becoming increasingly important. This is also very evident from some of our projects: ASMSpotter, CropClass, Estimation of solar roof capacity, 4D Urban Insights, DWD Clouds.
Although ML approaches are characterized by very high performance and accuracy, their performance is limited to the specific use case and training dataset. As a result, the training effort and the acquisition of large datasets must be repeated for each new use case, leading to high costs in application development. Unfortunately - in contrast to already established pre-trained AI models such as ResNet or VGG, which are based on the generic image dataset ImageNet - there are currently no pre-trained AI models available for satellite data or Earth observation applications.
All the more we are happy to be a part of "PreTrainAppEO". The goal of this project is to make the use of AI in the field of Earth observation and remote sensing more attractive and efficient by developing a methodology that uses the approach of pre-trained AI models to achieve generalizability to different standard applications in this field.