A Self-Supervised Hierarchical Clustering Network For Multiple Change Detection In Multitemporal Hyperspectral Images

There are many change detection (CD) algorithms using multitemporal Hyperspectral Images (HSIs) to recognize changes on the Earth's surface. Supervised models trained by large amount of labeled data can reach high accuracy. However, it is hard to annotate pixel-level labels manually. Spectral heterogeneity makes it difficult to detect multiple changes in unsupervised fashion. Thus, a Self-Supervised Hierarchical Clustering (SSHC) network is proposed for multiple change detection in HSIs. Experiments on two data sets show that SSHC performs better than some benchmarks and state-of-the-art methods for multiple CD.