Automatic Detection of Bodies of Water

Client’s Challenge

• The ports in our country serve as vital economic hubs with a commitment to ecological awareness.
• The Department of Mobility and Public Works (MOW) of the Flemish Government faces challenges in mapping the movements of water bodies crashing against port structures.
• While seawall techniques are easy to chart due to their fixed structure, mapping dynamic water movements is much more complex.

Our Solution

• Drones capture aerial photographs of port infrastructure, assisting MOW in mapping images around bodies of water. They create 3D models through photogrammetry.
• To streamline the time-consuming process of manually masking water bodies in 3D models, we developed a machine-learning model to help them.
• Our Segmantic Segmentation Model automatically detects and marks bodies of water in drone photos.
• Each pixel in the photo is marked as water (0) or non-water (1). The model achieves high accuracy, with a score of 0.985 in test jobs.

Impact Created

• The AI application significantly reduces the time employees spend marking bodies of water.
• Ensures more accurate labeling of water bodies, enhancing the quality of port infrastructure mapping.
• Enables continuous improvement of the model to perform optimally in various areas, such as rivers and canals.
• Facilitates easy uploading and processing of images through an Azure-based data pipeline, with outputs ready for client download and feedback integration.