Hyperspectral wetland image classification method based on graph capsule neural network

The invention discloses a hyperspectral wetland image classification method based on a graph capsule neural network, and the method comprises the following steps: S1, generating an adversarial domain adaptive frame, carrying out the learning feature transformation, and enabling a source domain sample and a target domain sample of a hyperspectral wetland image to carry out the feature matching; s2, constructing a map capsule neural domain adaptive network structure, extracting domain invariant features and domain related features, discovering migratable features and performing cross-domain sharing; and S3, designing two classifiers of a coupling structure, training the two classifiers by using the source domain sample, maximizing the classification difference of the target domain sample, and realizing accurate classification of the hyperspectral wetland image by identifying the classification boundary. According to the method, transferable knowledge is discovered, cross-domain sharing is realized, effective di