A classification method and device for multi-view cross-data domain image content recognition

Abstract

The invention discloses a classification method and device for content identification of a multi-view cross data field image. The classification method has the characteristics that by utilizing the prediction consistency of semantic contents among multiple view features of the image, mutually adjusting a regression forecasting model of each view, and adjusting regression coefficients to enable the regression coefficients to simultaneously and sparsely select the image data features of different data fields, ensure that the selected features keep the most intrinsic genetic semantic structures and reduce the noise features for causing inter-domain difference as far as possible. In order to maintain the data original manifold structure of each data field, image adjustment regular terms based on original data spectrograms are introduced; the generic global structures of a cross data field are jointly maintained between a training label of an auxiliary data field and a forecasting label of a target data field for further enhancing the accuracy of the forecasting label. Compared with the existing transfer learning classification method, the classification method provided by the invention is greatly improved in both accuracy rate and stability.

Description

technical field [0001] The invention belongs to the technical field of feature selection and classification, and in particular relates to a classification method and device for multi-view and cross-data domain image content recognition. Background technique [0002] In the information age represented by massive big data, all kinds of data explode and grow exponentially, and the mining of potential value of data has become a hot spot of people's attention and research. Whether it is the Internet, mobile communications, or financial fields, our daily lives are constantly generating large amounts of data, among which classification technology is a very effective way to mine potentially useful knowledge from data. For example, Internet users need to send and receive a large number of emails every day. How to help users sort emails into categories and automatically identify spam requires accurate and effective classification technology to intelligently help users. Another exampl...

AI Technical Summary

Problems solved by technology

However, many existing algorithms ignore the use of these rich multi-view features, and have not achieved the effect of improving the performance of existing migration learning.

Images