ANALYZING LAND COVER AND LAND USE CHANGES DETECTION IN THE BRAHMAPUTRA RIVER BASIN

Travis Frink, University of Rhode Island

Abstract

Three alterations are investigated: The addition of a feature pyramid network to improve small object recognition, a loss weighting strategy dependent on object poses, and the addition of a transformer for feature comparisons.As a result increased performance was achieved for both the FPN and loss weighting alterations while the transformer alterations slightly underperformed the original model.