Cross-domain image classification method based on structural feature enhancement and class center matching

Abstract

The invention provides a cross-domain image classification method based on structural feature enhancement and class center matching. The method comprises the following steps: acquiring a source domain image and a target domain image; constructing a visual feature extractor and a structural feature extractor to extract initial visual features and initial structural features of the image; obtaining a source domain image enhancement feature and a target domain image enhancement feature based on the initial visual feature and the initial structural feature of the image, and performing class center matching on the target domain image by using the source domain image enhancement feature and the target domain image enhancement feature to obtain a pseudo label of the target domain image; training a visual feature extractor, a structural feature extractor and a classifier by using a source domain image with a real label and a target domain image with a pseudo label; and utilizing the trained visual feature extractor, the structure feature extractor and the classifier to obtain a classification result of the target domain image to be classified. According to the method, the visual features and the structural features of the images are comprehensively considered, the class center matching method is used for carrying out false label endowing operation, and the classification precision of the cross-domain images is improved.

Description

technical field [0001] The present invention relates to the technical field of image classification, and more specifically, to a cross-domain image classification method based on structural feature enhancement and class center matching. Background technique [0002] At present, some people at home and abroad have begun to research and explore cross-domain image classification methods. Currently, the features used in cross-domain image classification methods are generally visual features, while structural features are ignored. Classification does not work well on images. At the same time, most of the current methods directly assign pseudo-labels to the target domain images, and then use these pseudo-labeled target domain images to train the network. However, this method of directly assigning pseudo-labels cannot guarantee the accuracy of pseudo-labels, so that wrong pseudo-labels introduce error information to the classifier, resulting in poor classification effect of the cl...

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the defect that the above-mentioned prior art considers a single feature when classifying cross-domain images, resulting in low classification accuracy, the present invention provides a method based on structural feature enhancement and class The cross-domain image classification method of center matching, comprehensively considers the visual features and structural features of the image when classifying cross-domain images, and uses the class center matching method to assign pseudo-labels to the target domain images to be classified, which improves the classification accuracy of cross-domain images

Images