Classification method and device based on learning image content recognition in cross-data field subspace

Abstract

The invention discloses a classification method and device based on learning the image content recognition in a cross-data field subspace. The method comprises the following steps: extracting characteristics of various target images to be classified and converting image data into numeric data available for classification; inputting data characteristics of various target image data fields and auxiliary data fields as well as generic label data of corresponding auxiliary field data, and establishing a spectrogram adjacent graph; establishing a mathematical model for the image data, label information and adjacent graphs; according to the mathematical model, updating a constructing coefficient of a base vector in the subspace, new representation characteristics of the data in the subspace, classifier coefficients and quasi labels predicted on target field image data; and predicting classification labels of the target field image data and classifying the target image data through the classification labels. The classification method disclosed by the invention improves the accuracy and the stability significantly compared with a current transfer learning classification technology.

Description

technical field [0001] The invention belongs to the technical field of transfer learning and classification, in particular to a classification method and device for image content recognition based on learning cross-data domain subspaces. Background technique [0002] With the advent of the era of big data, various forms of data have grown exponentially. How to extract useful knowledge and information from these massive data has become a hot spot for data mining and machine learning researchers in industry and academia. Among many data mining and machine learning technologies, classification technology has very important application value. For example, anomaly detection in video data, spam classification, and object recognition in images all require good classifiers. It is worth noting that the training of classifiers requires labeled data, but people often encounter the problem of insufficient labeled data. For example, in the application of learning classifiers for onlin...

AI Technical Summary

Problems solved by technology

In the problem of knowledge transfer learning from the auxiliary domain to the target domain, the data in the auxiliary data domain and the target data domain have differences in feature distribution, and traditional classification techniques often fail.

Images