Głębokie uczenie w procesie teledetekcyjnej interpretacji przestrzeni geograficznej – przegląd wybranych zagadnień

Abstract

The use of machine learning (ML) and deep learning (DL), especially deep convolutional neural networks (DCNN) in image processing and interpretation is currently a widely discussed topic among representatives of actively developing remote sensing and geoinformation scientific community. This article is an attempt to systematize the knowledge of DL in its supportive role in the aerial and satellite imagery interpretation of geographical space. The target audience of this overview are geographers who would like to enrich their research with methods based on artificial neural networks. The text presents main concepts and methods of DL along with example tasks that can be completed with their help i.e.: semantic segmentation, classification, augmentation of research dataset and feature engineering. The description of each task category was enriched with a use case and a literature review, thus making it possible to take the first step towards applying the specified technique in future research. The article conclusion includes a discussion on new directions and opportunities of applying DL in the discipline of Earth and environmental sciences.