Integrating Machine Learning and OBIA for Vegetation Classification in Archived Grayscale Aerial Imagery

Abstract

This study explores the use of machine learning models to classify water, vegetation, and non-vegetation land cover types in archived grayscale aerial imagery. The input images are segmented using a superpixel algorithm, and the resulting segments are mapped to expert-provided reference data. The region-based and patch-based approaches are evaluated using artificial neural networks and convolutional neural networks, respectively. The region-based method achieves an average accuracy of 0.83, while the patch-based method reaches 0.79. Although the patch-based method shows slightly lower overall accuracy, it significantly improves recall rates, particularly for the water and non-vegetation classes.