object-based sample augmentation

Augmenting seabed samples to sharpen sediment maps in the Sea of Japan

A single gravel sample in the Sea of Japan used to be nearly invisible to seabed mapping. By propagating those sparse, field-validated labels across SLIC superpixels of calibrated backscatter and bathymetry, this study raised the rare-class F1-score from 0.036 to 0.436. The resulting gravel, sand, mud map recovers the coarse substrates that pixel-based classifiers miss, strengthening the baseline data needed for benthic habitat mapping and marine spatial planning.