STUDY ON AUTOMATIC CLASSIFICATION OF GRAVEL BEACH SEDIMENTS USING MACHINE LEARNING

Abstract

 A machine learning model classifying orth-mosaic images of a gravel beach created from UAV-SfM/MVS survey into "gravel", "drifting", "vegetation", and "block" was constructed and examined in terms of the characteristics and usefulness. From investigation, it was found that the discrimination accuracy can be improved by reducing the size of the filter layer and increasing the number of filter layers. The visualization using Grad-CAM showed that the indiscrimination of “gravel” and “drifting” was caused by a few points of interest. Additionally, the machine with the trained model can work well even for images that are not used for training.