Dakar Dataset with Anomalies for Convolutional Autoencoders for Data Compression and Anomaly Detection in Small Satellite Technologies Dakar Dataset

This dataset contains aerial satellite imagery of Dakar, Senegal, acquired during the Copernicus Sentinel-2 mission. It includes 70 normal images and 2 anomalous images corresponding to Saharan dust storm events. The dataset was used to develop and evaluate a convolutional autoencoder (CAE) for dual-purpose tasks: image compression and anomaly detection in small satellite technologies. Each image folder is named using the YYYYMMDD format and contains two subfolders representing spatial resolutions of 20m and 60m. Within these, the RGB bands (Bands 2, 3, and 4) are stored separately in .jp2 format. The dataset was preprocessed using QGIS and includes: Band selection (RGB) Patch extraction (256×256 pixels) Normalization and datatype conversion Metadata preservation The CAE model trained on this dataset achieved a compression ratio of 8 and demonstrated effective anomaly detection capabilities using the Structural Similarity Index Measure (SSIM). Anomalous images, including real dust storm events and synthetic pixel defects, were successfully identified based on reconstruction quality.