Exploiting spatiotemporal metadata, such as geolocation and image acquisition time, can significantly improve the performance of semi-supervised learning models on remote sensing image classification tasks.
