Semi-supervised image classification based on deep clustering and pre-trained models

With the rapid development of deep learning in the field of image classification, supervised learning relies on a large amount of labeled data to achieve excellent classification performance. However, the acquisition of labeled data is time-consuming and expensive, which makes traditional supervised methods prone to overfitting and insufficient generalization when data is limited. To address these challenges, semi-supervised learning (SSL) combines a small amount of labeled data with a large amount of unlabeled data to improve model performance. This paper proposes a new semi-supervised image classification method that achieves collaborative optimization by combining clustering and classification algorithms in the latent feature space. Based on the self-supervised pre-training model, this method makes full use of its effective public feature information, combines supervised classification loss with unsupervised clustering loss, and uses the intrinsic structure of the data to improve feature representation and classification ability. By introducing clustering in classification learning, this method effectively utilizes unlabeled data to help the model find better decision boundaries and more discriminative feature representations, thereby achieving better image classification performance under limited labeled data conditions. Compared with other state-of-the-art algorithms, this method performs well on datasets such as Cifar10, Cifar100, and STL-10.