Remote sensing image scene classification method of dynamic routing capsule network with viewpoint consciousness

The invention provides a remote sensing image scene classification method of a dynamic routing capsule network with viewpoint consciousness. The method comprises the following steps: firstly, randomly dividing a scene data set into a training set and a test set in proportion; preprocessing the images in the data set; inputting the training set image into a spatial transformation network, learning transformation of foreground object viewpoint consciousness of the training set, obtaining an attitude matrix with object-observer viewpoint information, inputting the attitude matrix into a convolutional capsule layer and a dynamic route learned by sub-concepts for training, extracting capsule feature representation with invariable viewpoints, and obtaining output of a category capsule layer; and taking the category capsule with the highest prediction probability as the prediction category of the scene. And finally, inputting a test image into the capsule network model to obtain a classification result. According to