Early and reliable detection of pneumonia from chest X-rays is essential for improving patient outcomes, especially in resource-limited healthcare settings. In this study, we present DenseXplainNet, a lightweight deep learning framework built on DenseNet121 with a custom convolutional head for binary classification of chest radiographs. Unlike conventional multi-stage fine-tuning approaches, our method uses a single-pass training strategy that balances efficiency with accuracy, making it well-suited for GPU-constrained environments. To enhance interpretability, we integrated Grad-CAM and LIME for visual explanations, enabling clinicians to understand the decision-making process of the model. The proposed system achieved a test accuracy of 96.79% and an AUC of 0.9823, demonstrating strong discriminatory power between normal and pneumonia cases. Remained consistently high across both classes, confirming robustness of the Model. These findings suggest that DenseXplainNet offers a balanced solution that combines computational efficiency, high accuracy, and interpretability.