Zero-sample image recognition algorithm and system based on auto-encoder

Abstract

The invention relates to the technical field of image recognition, and discloses a zero-sample image recognition algorithm and system based on an auto-encoder, and the algorithm comprises the steps: extracting a feature vector of a known sample in a source domain through employing a pre-trained Arc-SENet network; learning a projection matrix and a decoding projection matrix of the source domain inthe source domain by using a preset auto-encoder; projecting the attribute of the unknown sample to the feature space by using a preset auto-encoder in the target domain; associating the auto-encoderof the source domain with the encoder of the target domain, and obtaining an attribute matrix and a feature matrix of unknown samples in the target domain through iterative computation; and label prediction is performed on the unknown sample by using two modes of forward label prediction and reverse label prediction, and if label results obtained by the two modes are the same, the obtained labelis an image sample recognition result. The invention further provides a system of the zero-sample image recognition algorithm based on the auto-encoder. According to the invention, image recognition is realized.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to a zero-sample image recognition algorithm and system based on an autoencoder. Background technique [0002] Image recognition is one of the main research directions of artificial intelligence. With the vigorous development of artificial intelligence, image recognition has been widely used in artificial intelligence fields such as defect detection, unmanned driving, and medical diagnosis. The current research on image recognition is mainly aimed at the classification of images, but with the rapid development of social networks and social labeling systems, new labels and concepts are constantly emerging, followed by the need for people to use these new labels to mark images . However, the existing supervised learning methods require a large amount of labeled information for effective classification, so they cannot be applied to the situation where there is no labeled inf...

AI Technical Summary

Problems solved by technology

Therefore, if the projection matrix learned in the source domain is directly applied to the target domain, it may lead to the problem of inaccurate attribute prediction of samples of unknown classes.

Images