Unsupervised Domain-Adaptive Image Classification Algorithm Incorporating Generative Adversarial Networks
Baiqiang Gan, Qiuping Dong · 2021 3rd International Conference on Machine Learning, Big Data and Business Intelligence (MLBDBI) · 2021
With the rapid development of computing power of computers, deep neural networks have achieved great success in recent years, but these successes rely on huge amounts of labeled data, which have the disadvantages of low timeliness and high cost, and thus cannot rapidly deploy deep learning models when new tasks come. Unsupervised domain adaptation techniques can take a classifier learned from a source domain rich in labeled data and use it for a target domain with only a small amount of data with labels or no labels at all, and use the knowledge learned from the source domain for a target domain with no labels at all, effectively solving the data labeling problem. However, the unsupervised domain adaptive technique suffers from the problem of target domain data without explicit labeled data and domain discrepancy, which affects the image classification performance; Therefore, this paper proposes an unsupervised domain adaptive image classification algorithm incorporating generative adversarial networks, which combines the advantages of deep generative adversarial networks and applies the trained labeled source domain depth model accurately to the unlabeled target domain. Finally, the algorithm of this paper is compared with other classical methods, and the experiments show that the efficiency and accuracy of this algorithm are higher.