Application of Contrastive Learning Based on ResNet34 in Unsupervised Image Classification

In response to the problems of insufficient labeled data and feature complexity in industrial scenarios, this paper proposes an unsupervised contrastive learning model framework based on ResNet34. This method combines the powerful feature extraction capability of ResNet34 with the efficient representation learning mechanism of contrastive learning. First, by introducing the contrastive loss function InfoNCE and data enhancement strategies such as random cropping, color jittering, and random flipping, the learning effect of the ResNet34 model is improved. Then, the Dropout layer is added to the output layer Averagepool to reduce overfitting. Finally, the Alyanxishe80 dataset and the self-built dataset are used for evaluation. The results indicate that the enhanced model achieves a$\mathbf{3. 4 2 \%}$and$\mathbf{0. 6 3 \%}$increase in classification accuracy compared to the original model prior to the improvements, and the loss rate is reduced by 0.044. And it shows strong generalization ability when processing complex data.