Abstract: This study aims to evaluate the applicability of various deep learning models in landslide susceptibility assessment and to explore pathways for transforming complex models into practical tools. Taking Yunyang District, Shiyan City, Hubei Province as the study area, ten landslide conditioning factors were selected based on geological and remote sensing data to construct the dataset. Five deep learning models—AlexNet, Inception, ResNet-101, DenseNet-201, and GoogLeNet—were constructed, trained, and validated. Model performance was evaluated using ROC curves, landslide zoning statistics, implementation complexity, and a comparative analysis with a Random Forest model. The results indicate that the GoogLeNet model exhibited the optimal performance, achieving the highest AUC value of 0.84. In the high-susceptibility zone, the landslide frequency ratio reached 48%, and the landslide density was 3.06 landslides/km². Furthermore, a deep learning-based landslide susceptibility evaluation system, integrating factor selection, model training, testing, and validation modules, was developed on the MATLAB platform. Deep learning approaches can effectively enhance the accuracy of landslide susceptibility assessment. Specifically, the GoogLeNet model achieves the best balance between predictive precision and computational efficiency.