Graph-Based Active Learning for Surface Water and Sediment Detection in Multispectral Images
Jul 1, 2023·,,,·
1 min read
Bohan Chen
Kevin Miller
Andrea L. Bertozzi
Jon Schwenk
Abstract
We develop a graph active learning pipeline (GAP) to detect surface water and in-river sediment pixels in satellite images. The active learning approach is applied within the training process to optimally select specific pixels to generate a hand-labeled training set. Our method obtains higher accuracy with far fewer training pixels than both standard and deep learning models. According to our experiments, our GAP trained on a set of 3270 pixels reaches a better accuracy than the neural network method trained on 2.1 million pixels.
Type
Publication
IGARSS 2023 - 2023 IEEE International Geoscience and Remote Sensing Symposium
The graph active learning pipeline (GAP) treats multispectral pixels as nodes in a similarity graph and uses graph Laplace learning to distinguish land, surface water, and in-river sediment. An acquisition function identifies the pixels whose expert labels are expected to improve the classifier most, directly targeting the expensive step of building a hand-labeled remote-sensing dataset.
On the RiverPIXELS imagery, GAP trained with only 3,270 selected pixels surpasses the reported neural-network baseline trained on roughly 2.1 million pixels. The result demonstrates that graph geometry and targeted labeling can provide strong segmentation performance when dense annotation is impractical.