Zero sample classification method based on dual-triple deep metric learning network

Abstract

The invention relates to a zero sample classification method based on a double-triple deep metric learning network, which comprises the following steps of: inputting semantic features of samples intoa mapping network, and outputting the semantic features to a visual space; in a visual space, selecting a pair of semantic features and visual features belonging to the same category to form a positive sample pair, then selecting a semantic feature different from the positive sample pair to form a triple, and inputting the triple into a semantic-guided triple network; selecting a pair of semanticfeatures and visual features belonging to the same category are selected to form a positive sample pair, then a visual feature different from the positive sample pair in category to form a triple, andinputting the triple into a visual guidance triple network; inputting the output of the semantic-guided triple network and the output of the visual-guided triple network into a double-triple loss function for calculation; and finally, classifying the test samples by using a nearest neighbor classifier. The structure is easy to realize, the training method is simpler, the training parameters are fewer, and training can still be carried out under the condition that computer hardware equipment is poorer.

Description

technical field [0001] The invention relates to a zero-sample classification method. In particular, it concerns a zero-shot classification method based on a dual triplet deep metric learning network. Background technique [0002] In recent years, deep learning has been vigorously developed because of its outstanding performance on large-scale dataset classification tasks. However, a prominent problem is becoming more and more obvious. It takes a lot of manpower and material resources to obtain labeled data, and in some cases it is even extremely difficult to obtain, such as images of endangered animals. Compared with deep learning that requires a large amount of labeled data, humans only need a small number of samples, or even no samples to perform recognition tasks. For example, when a person already knows a cat and tells him that a tiger has a "king" pattern on its forehead, he is very likely to recognize the tiger. Inspired by this ability of human beings, scholars beg...

AI Technical Summary

Problems solved by technology

However, such methods do not make full use of the difference and connection between the visual features and semantic features of samples.

Images