Flow Factory

Detaillierte Waldüberwachung mittels skalierbarer KI-Methoden

Managing and conserving forest ecosystems in Europe and worldwide is an indispensable component of climate adaptation and climate change mitigation strategies. Precise and up-to-date information about the health and the carbon balance of forests are, hence, critical to assess the current state of forests, trigger appropriate countermeasures against forest loss, and develop improved management strategies. Advances in both Earth observation and artificial intelligence have paved the way for the automation of forest monitoring using satellite time series data, including optical, radar, and LiDAR measurements. The forest maps produced by today's approaches, however, are still often limited to coarse resolutions and/or to relatively small spatial areas. To overcome those limitations, the AI4Forest project brings together experts in artificial intelligence, applied mathematics, computer science, spatial remote sensing, and climate change. AI4Forest strives for both conceptually novel AI methods for forest monitoring as well as for scalable AI methods that allow to process large amounts of data efficiently and at low cost. The resulting techniques will facilitate the generation of detailed forest maps at a very high spatial and temporal resolution for the whole European continent and the entire world, including tree species identification down to the level of individual trees.

Monitoring Changes in Big Satellite Data via Massively-Parallel Artificial Intelligence

The remote sensing field witnesses an explosion in the amount of available data, with petabytes of data being gathered by single satellites every year. Such data allow the identification of fine details in the landscape and the recent breakthroughs in artificial intelligence (AI) facilitate application areas such as agricultural monitoring, infrastructure management, mapping forest development, and many others. Applying AI models on a global scale can become extremely time-consuming with analyses potentially taking weeks, months, or even years. This project aims at the development of highly-efficient parallel implementations for AI methods that allow to detect and monitor “changes” visible in time series satellite data. Collaboration with the University of Copenhagen (Cosmin Oancea and Marcos Vaz Sallies). Supported by the Independent Research Fund Denmark (DFF).

DeepCrop: Quantification of Carbon Stocks via Deep Learning

Recent technological developments in deep learning and drone-borne Lidar scanners pave the way for constraining the uncertainty inherent to quantify and project ecosystems' carbon (C) stocks. With a rising demand for biomass, DeepCrop aims to precisely measure above ground biomass and to estimate C sinks in croplands and forests. The ambition is to bridge expertise of experimental and computer scientists to develop novel tools for the automated processing of Lidar data utilizing deep learning and drones. Joint work with the University of Copenhagen (Katerina Trepekli, Thomas Friborg, Christian Igel). This project is, in part, supported by the Villum Foundation and the Data+ program of the University of Copenhagen.

Verbund - KI: Energie- und ressourceneffiziente künstliche Intelligenz für moderne IoT-Anwendungen

The Internet of Things (IoT) offers us more and more opportunities to connect different devices and exchange data. Accordingly, the number of IoT-enabled end devices such as smartphones, smartwatches or fitness trackers is increasing. Many of these devices have microcontrollers or sensors, such as cameras or motion sensors, with which they exchange data via cloud services using AI processes. However, both the devices themselves and the AI applications consume resources in the form of materials and energy. It is important to reduce this resource consumption - especially for IoT applications that are used in the environmental sector.