The invention relates to a small sample image recognition method based on deep learning. The method comprises the following steps of 1, dividing a training set; 2, generating a noise image; 3, pre-training a prototype space discrimination network; 4, training a deception image generation network; 5, training a prototype space discrimination network; 6, repeating the step 4 and the step 5 for crossiteration training until preset iteration times are reached or the accuracy is not improved any more; 7, performing image category recognition. According to the method, on the premise that a trainedmodel is not changed, by means of a few labeled samples of each class, through generalization of the rare classes, new classes which are not seen in the training process are recognized, extra trainingis not needed, and the image recognition accuracy is high.