Simple Implementation of Terrain Classification Models via Fully Convolutional Neural Networks

This paper introduces a simple implementation of three versions (large, medium, and small) of terrain multi-classification models using Fully Convolutional Neural Networks (FCNNs) for imagery data. The proposed methodology involves labeled and unlabeled data collection from European Space Agency (ESA) WorldCover and Sentinel-2 MultiSpectral Instrument (MSI) on the Google Earth Engine, compressing datasets into Tensorflow records format with 9 diverse terrain types, and handling Google Cloud training computations. There were prepared different dataset portions of 10 megabytes, 200 megabytes, and around a gigabyte files. The experimental results demonstrate the effectiveness of the CNN-based approach, achieving a tolerable 71% accuracy of the Terrain Classification Model (TCM) and robust classification performance. The simplicity and efficiency of the proposed method make it suitable for real-world applications requiring reliable and fast terrain classification.