Small sample image classification method and device based on global and local content matching

The invention discloses a small sample image classification method and device based on global and local content matching, and the method comprises the steps: obtaining a natural image sample set, extracting image features in the natural image sample set, obtaining global features and local features, and converting the text description corresponding to an image into a high-dimensional feature vector; selectively combining discriminative local features, and constructing local representation by using local feature fusion; matching the local representation with the global feature, and constructing a robust classifier; and training the robust classifier, and carrying out image classification by using the trained robust classifier. According to the method, local contents of an image are concerned, and a local structure is designed by selectively combining the local contents to construct robust local representation, so that a classifier is helped to concern details of the image; two classifiers are combined, local d