<p>Accurate and efficient landslide susceptibility assessments in the Sichuan-Tibet region are critical for disaster prevention and mitigation in railroad and highway infrastructure projects. To address the overemphasis of specific factors caused by data channel merging in current convolutional neural network (CNN)-based models, this study proposes a multi-branch data fusion model for landslide susceptibility mapping. The model integrates multi-source remote sensing data and uses deep CNNs to extract semantic information from evaluation factors by incorporating a multi-branch network and adaptive weighting mechanism, thereby enabling precise susceptibility assessments. The southeastern Sichuan-Tibet transportation corridor of the Qinghai-Tibet Plateau was selected as the study area for validation. Comparative and ablation experiments were conducted using the random forest (RF), support vector machine (SVM), shallow CNN, and ResNet101 models. The proposed multi-branch network model outperformed the existing models, achieving higher accuracy, precision, recall, F1-score, area under the curve (AUC), and frequency ratio accuracy, with values of 0.88, 0.89, 0.92, 0.89, 0.92, and 0.97, respectively. The study area was divided into four subregions based on geological conditions to explore the relationships among environmental driving factors. The results indicate that landslide risks in the Sichuan-Tibet region are significantly influenced by topography, rainfall, and geology, posing major transportation infrastructure threats. These findings validate the effectiveness of the proposed model in assessing landslide susceptibility across diverse geographic conditions. This study offers a reference for landslide mechanisms and guidance for disaster management and infrastructure planning.</p>