Dataset for object detection of charcoal production sites

Abstract

This dataset provides a collection of processed airborne LiDAR data and associated annotations of historical charcoal production sites in the Erzgebirge district of Saxony, Germany for object detection. Using official files from the digital terrain model (DTM), various DTM derivatives were calculated and used as three-channel input data (hillshade, slope and sky-view factor) for a convolutional neural network. The dataset includes training, validation and test data, including annotations, as well as trained model weights and configuration files for inference. To ensure reproducibility, the suitable Conda environment for training and inference with MMDetection v2.28.2 is also provided.