In the project initiative, out team embarked on a task to enhance the LIFT GIS software’s capabilities through the integration of an advanced automated land classification system. Utilizing the AI4Copernicus service, we developed deep neural network models tailored for build-up, rural, water, and forest land types detection, ensuring minimal user input for maximal output efficiency.
Our focus on user experience paid off, with users simply needing to define the area and select the land type, while the program handled the rest. The results were displayed as separate layers for each outcome, facilitating an in-depth analysis of the predicted areas.
Testing the models against real-world data, particularly in the Ljubljana municipality and its surrounding regions, revealed the build-up model’s exceptional performance. However, it also highlighted the rural and forest models need for improvement, presenting opportunities for future advancements.
The project, despite minor deviations and initial delays, stayed resilient, achieving all deliverables and milestones, including the creation of annotated satellite images, trained models, and comprehensive reports. The integration within the Analysis tool of the LIFT software proved beneficial, especially for insurance companies, showcasing the project’s practical applicability and success.
