Automatic detection of agricultural field boundaries and seeded acres using super-resolution Satellite Earth Observation data and deep learning

DigiFarm is a Norwegian agricultural technology startup, established in 2019, focused on developing deep neural network (AI) models based on super-resolved Satellite Earth Observation (SatEO) data. The objective is to automatically detect agricultural field boundaries and planted areas from in-season vegetation. This paper presents the conceptual and technical overview of our work conducted under a GPU allocation grant on the LUMI supercomputer (Proposal ID: EHPC-AI-2024A01-080), awarded through the EuroHPC “AI and Data-Intensive Applications Access call”. We detail the development of a novel deep neural network that intersects deep learning and computer vision to delineate agricultural fields. The model, which includes 12 land cover classification outputs, achieves an Intersection over Union (IoU) of over 0.94. This automated, low-cost, and scalable approach provides a significant improvement over existing manual digitization methods, which are time-consuming, resource-intensive, and costly. Our solution reduces the time-to-result from months to hours and increases the accuracy of detected field boundaries by 12-15% compared to existing Cadastral data. This represents a breakthrough for the agricultural industry, which has historically struggled with fragmentation and a lack of digital adoption.