Neural cellular automata improve land cover map boundaries in satellite images
Multiclass Boundary Refinement with Neural Cellular Automata for Remote Sensing Land Cover Segmentation
Computer Vision and Pattern Recognition
Summary
Maps made by computers to show different land types from satellite pictures often get their borders slightly wrong. The authors developed a method called Boundary Refinement Neural Cellular Automata that looks at these maps and corrects the edges step by step. This method works by updating the probabilities that each pixel belongs to a certain land type, especially improving tricky borders like those around irregular woodlands. It made the border accuracy better by a small but measurable margin. This shows computer-made maps can be sharpened to reflect boundaries more precisely.
What this means in practice
- •For environmental monitoring teams: Improve land cover maps from satellite images for better tracking of forest and vegetation boundaries.
- •For urban planning departments: Refine maps of land use edges to support more accurate urban and regional planning decisions.
Authors
Pascoassis Souza Santos Meira, Gilson Antonio Giraldi, Rafael Duarte Coelho dos Santos
Abstract
Semantic segmentation of remote sensing imagery often places class boundaries a few pixels away from where they truly lie, and this position error is concentrated in the thin band measured by Boundary IoU (BIoU). We treat boundary refinement as the iterated application of a small learned local rule: the Boundary Refinement Neural Cellular Automaton (BRNCA), that starts from the class probabilities of a frozen, pre-trained segmenter and corrects them cell by cell. We generalise the binary rNCA to the multiclass case by keeping the visible channels on the probability simplex through an Aitchison (log-space) update and by adding free hidden channels. The best configuration improves the mean BIoU over the base segmenter by 2.36 points (from 46.63 to 48.99). The gain is concentrated on the irregular woodland class (+8.5 points BIoU); this suggests that position error is correlated with the fractal dimension of the class boundaries.