A Deep few-shot learning algorithm for hyperspectral image classification

Abstract

For hyperspectral image classification problem of small sample, this paper proposes a depth of less sample learning algorithm, this algorithm through the simulation of the small sample classification in the process of training is to train the depth 3D convolution neural network feature extraction, the extraction of characteristic with smaller class span and large spacing between classes, more suitable for small sample classification problem, and can be used for different hyperspectral data, has better generalization ability. The trained model is used to extract the features of the target data set, and then the nearest neighbor classifier and support vector machine classifier are combined for supervised classification. Three groups of hyperspectral image data of Pavia university, Indian Pines and Salinas were used in the classification experiment. The experimental results showed that the algorithm could achieve a better classification accuracy than the traditional semi-supervised classification method under the condition of fewer training samples (only 5 marked samples were selected for each type of feature as training samples).