Published on May 13, 2024, 5:24 am

Seresco has successfully concluded its research and development project aimed at enhancing cartographic production services through the application of artificial intelligence (AI) techniques on high-resolution satellite images of territories and areas. This innovative approach has enabled the rapid and accurate interpretation of the evolution and features of specific geographical areas. Known as A4Geo, this project is part of Seresco’s strategic plan to enhance the procedures for obtaining geoinformation. Co-funded by the Government of Spain and the European Union through Red.es and FEDER funds, the project has reached its final phase after 20 months of execution.

Led by Seresco’s Cartography and Cadastre department in collaboration with the Technology and Innovation division, along with support from the University of Oviedo, the project has resulted in the development of a tool for automatically detecting changes between two orthoimages (corrected aerial or satellite images eliminating perspective and tilt effects) taken at the same spatial location but at different time points. The primary goal is to identify changes between these images to streamline updates to cartographic databases.

This innovative tool was created through training convolutional neural networks (CNNs), which are capable of analyzing image pairs post-training to generate a change mask that pinpoints geographic differences between the two time points. The outcome is a robust, scalable tool with high generalization capabilities that significantly advances change detection across various geographical features like appearance, disappearance, or modifications while filtering out irrelevant changes such as shadows, vehicles, facade tilts, vegetation alterations, water bodies, or livestock movements.

The implementation of this tool greatly enhances efficiency in updating cartographic databases by automating change detection tasks that previously required manual inspection of orthoimages. By focusing efforts solely on areas where automatic changes are detected, substantial time savings are achieved compared to traditional methods.

Utilizing a custom dataset consisting of Sentinel satellite images, PNOA aerial images, and labeled examples created by Seresco’s team combined with deep learning techniques training and evaluation resulted in models achieving close to 90% accuracy.

The newly developed tool has been integrated into MapStorm – a web-based GIS software developed by Seresco offering various tools for visualizing and editing geoinformation from multiple sources. MapStorm now includes a change detection service for orthoimages streamlining the process for end-users providing an intuitive way to initiate change detection processes and view results effortlessly.

Share.

Comments are closed.