As one of the severe respiratory diseases that endanger human health, the diagnosis of pneumonia poses significant challenges in both developed and developing countries. Traditional X-ray diagnosis methods suffer from accuracy and speed among radiologists, necessitating reliable automated tools with strong classification capabilities. In this study, we propose a modified model named “ResNet-PPMMamba”, which combines a ResNet network, a pooling pyramid module and a Mamba architecture based on state-space models to improve the accuracy of pneumonia detection from chest X-rays, in order to improve the accuracy of pneumonia detection from chest X-rays. Our model combines CNN’s local feature extraction capability, multi-scale context modeling and efficient global relationship reasoning. Using the Kermany dataset, our model achieves high accuracies of 94.87% and 93.91% with ResNet-50 and ResNet-101 as the basic backbones, respectively, demonstrating significant performance improvements and less parameters in image classification tasks. Ablation experiments demonstrate that both the pooling pyramid module and the Mamba module are indispensable for model performance.