Pneumonia is a common acute respiratory infection, typically caused by bacteria, viruses, or fungi, and is one of the leading causes of death in children under five years old. Early detection of pneumonia can effectively enable treatment measures, thereby reducing the mortality rate of pediatric pneumonia. Chest X-rays (CXR) are the primary medical imaging method of detecting pediatric pneumonia. However, the substantial volume of image readings imposes significant work pressure on radiologists, leading to potential diagnostic errors. To enhance the efficiency of detecting pediatric pneumonia using CXR in clinical practice, this paper proposes a lightweight classification algorithm based on deep learning. The algorithm employs lightweight convolutional modules and attention mechanism to deeply extract high level features from CXR images. Additionally, a dual-branch down-sampling module is designed to reduce the dimensionality of feature maps while preserving local details. Finally, a multi-scale feature fusion module aggregates global contextual information, enhancing the model's ability to identify pneumonia-infected regions. The experimental results indicated that our method can quickly and accurately detect pediatric pneumonia, achieving accuracy and F1 score of 98.63% and 99.07% respectively on the test set.