Small data cross-domain action recognition method based on double-chain deep double-flow network

Abstract

The invention discloses a small data cross-domain action recognition method based on a double-chain deep double-flow network. Efficient action recognition of other small-scale data sets based on a source domain data set is achieved. Meanwhile, a data set CDSAR used on the cross-domain action recognition task is provided. The method comprises the following specific steps: (1) preprocessing video; (2) building a model based on a double-chain deep double-flow network; (3) constructing a target function based on the double-chain deep double-flow network; (4) adopting a small data cross-domain action recognition method based on a double-chain deep double-flow network. Efficient action recognition can be carried out on other data sets only with a small number of samples based on the known data set, the problems that the data size of the target data set is small and data distribution among different data sets is inconsistent can be effectively solved, and the convergence speed of the method is high.

Description

technical field [0001] The invention belongs to the technical field of computer vision and pattern recognition, and relates to a small data cross-domain action recognition method based on a double-chain deep double-stream network, which can reduce the difference in data distribution between different data sets and effectively suppress over-fitting caused by a small amount of training data On the combination problem, the validity of the model is verified on the video action dataset with a small amount of data; Background technique [0002] In recent years, the video representation learning method based on deep learning has made great progress, and the feature representation has become more and more robust; the classic methods are: Two-Stream (two-stream convolutional neural network): the basic principle is to video sequence The dense optical flow is calculated every two frames to obtain the dense optical flow sequence (ie temporal information), and then the convolutional netw...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to solve the problem of action recognition of small data volume target data sets for the cross-domain task of action recognition. The common method is that the cross-domain task cannot be effectively solved, and the second is that the model is easy to overfit when training the small data target data set. , providing an efficient small data cross-domain action recognition method based on a double-chain deep double-stream network;

[0006] The training set and test set of common action recognition methods are divided from the same data set, which cannot effectively solve the cross-domain problem; the present invention solves the problem of inconsistent data distribution between different data sets, and effectively suppresses the The overfitting problem caused by the small amount of training data uses the action information of the source domain to identify the target domain action. This method has a fast convergence speed, thus providing help for cross-domain action recognition problems on small-scale data sets;

Images