In the wake of the COVID-19 outbreak, swift and precise classification of lung CT images becomes paramount. We introduce a novel model named CDenseNet by embedding the CBAM attention mechanism into DenseNet. Specifically, CDenseNet emphasizes salient feature information across the network. Furthermore, the replacement of ReLU with the Swish activation function in deep Dense Block modules bolsters feature expressivity. Comparative results on two public datasets confirms the superiority of CDenseNet over several state-of-the-art models, showcasing its potential for diverse COVID-19 CT image classifications.