Lung cancer remains the leading cause of cancer-related mortality worldwide, making timely and accurate diagnosis essential for optimizing patient prognosis and treatment strategies. Identifying pathological subtypes in histopathological images serves as a crucial assessment method that significantly influences treatment decisions. However, pathologists encounter substantial challenges when conducting manual evaluations due to the inherent complexity of these images, along with variability introduced by the subjective nature of their assessments. This study addresses these challenges by presenting a novel multi-level SwinTransformer-based automatic classification network designed to classify four tumor subtypes in lung cancer histopathology images effectively. Our approach incorporates a specifically designed multi-level fusion module that captures pertinent features from various levels of pathological images, thereby enhancing classification performance. Through extensive experimentation, we demonstrate that our model achieves outstanding accuracy and robustness across all four tumor subtypes. The results indicate the effectiveness of our method and its potential in assisting oncologists with rapid and efficient early diagnosis. By facilitating precise predictions for different lung tumor subtypes, this research offers a powerful diagnostic tool, ultimately aiming to improve patient outcomes and streamline the evaluation process in clinical settings. Our work underscores the significant role of advanced machine learning techniques in enhancing diagnostic accuracy in oncology and sets the stage for future research into automated pathology solutions.