The field of remote sensing and image segmentation is moving towards more accurate and efficient methods for analyzing and understanding complex data. Recent developments have focused on leveraging multimodal information, such as spectral and spatial features, to improve classification performance. Additionally, there is a growing interest in self-supervised learning methods that can learn dense representations for patches and reduce the need for expensive pixel-level labeling. Another trend is the use of curriculum learning strategies to train lightweight models for onboard satellite deployment, which can minimize the transmission of redundant or low-value data. Noteworthy papers include: CWSSNet, which achieved state-of-the-art performance in hyperspectral image classification, and SlotSAR, which disentangled target representations from background clutter in SAR images. IG-CAM also achieved state-of-the-art performance in weakly supervised semantic segmentation, and SPAM introduced a versatile framework for segmenting images into accurate yet regular superpixels. CMTSSL demonstrated consistent gains in downstream segmentation tasks using lightweight architectures for onboard satellite deployment.