SAR image scene classification method based on online gradient lifting

The invention discloses an SAR image scene classification method based on online gradient lifting, and the method specifically comprises the steps: adding an embedding layer at the end of a CNN, dividing the embedding layer into a plurality of sub-embedding parts, and adding a classifier after each sub-embedding part; in the training process, a classifier is trained through online gradient lifting, meanwhile, a regression device is added between every two sub-inserts, the regression devices are used for mapping embedded features of different sub-inserts to the same feature space, and a metric function based on Pearson correlation coefficients is learned in the space. The regression device is ignored during testing, and prediction of each classifier is integrated through a weighted voting method. According to the SAR image scene classification method based on online gradient lifting provided by the invention, the CNN can focus on learning complex SAR scene images, and in addition, the CNN can learn more divers