GCN is Graph Convolutional Network, a newly developed deep learning framework combining the advantages of traditional CNN and graph-based representation learning.
By defining graph convolution filters and introducing graph topological information represented by adjacent matrix and graph Laplacian, GCN is applicable to arbitrarily structured non-Euclidean domain and thus capable of handling irregular real-world data via graph structure modeling.
For hyperspectral imagery in the domain of culture heritage artifacts, we apply GCN on image classification for the Selden Map, a medieval artifact imaged in the collections at the Bodleian Library of Oxford University.