Abstract: Objectives: Landslide susceptibility assessment can provide scientific basis and technical support for disaster prevention and mitigation. In Ankang City, the unique topography of "three mountains enclosing two rivers" results in a varied landscape and complex geological structures. Additionally, factors such as engineering projects, land use, and heavy rainfall contribute to the frequent occurrence of landslides. In response to the challenges of inaccurate landslide prediction caused by using single-scale feature extraction in different levels of network structures, and the high resource demands and computational costs due to model complexity, the lightweight network based on multi-scale feature learning (MFL-LN) for landslide susceptibility assessment is proposed. Methods: Combining the geological environment characteristics of landslide development in the study area, the Spearman correlation analysis was conducted on evaluation factors. This analysis led to the identification of 13 evaluation factors across four categories: human activity factors (distance to roads, distance to settlements), hydrological factors (annual rainfall, distance to rivers), topographical factors (elevation, slope, aspect, plan curvature, profile curvature, aspect variability, slope variability, and landform), and vegetation factor (normalized difference vegetation index). These factors were used to establish a landslide susceptibility evaluation factor system for the study area. To ensure the randomness and rationality of negative sample selection, distance constraints were applied, such as maintaining a minimum distance of 500 meters between hazard and non-hazard points, as well as between non-hazard points. Nonhazard units were randomly selected to match the number of hazard units, forming a balanced sample dataset. The MFL-LN method employed the deep feature enhancement module (DFEM) to construct the network architecture, reducing the number of parameters and computational load to achieve model lightweighting. Additionally, it dynamically adjusted the attention to different channel features, enhancing the model's ability to perceive and utilize important features. Subsequently, the multi-scale feature fusion module (MSFF) was employed to integrate the features extracted through down-sampling in the encoding phase, enabling information interaction on the fused deep features and adaptively adjusting the feature weights. It can effectively capture both global and local spatial characteristics of landslides and compensate for the feature extraction loss caused by down-sampling, thereby improving the model's predictive capability for landslide susceptibility. Results: An ablation study was conducted on the MFL-LN method, and it was compared with the support vector machine (SVM) model, multi-dimensional convolutional neural network coupled (Multi-CNN) model, depthwise separable convolutional neural network (DS-CNN) model, multi-scale convolutional neural network (MSCNN) model, and lightweight separable factorized convolution network (SFCNet) model. The accuracy of the susceptibility evaluation models was analyzed from three perspectives: landslide susceptibility zoning maps, landslide hazard point density zoning statistics, and evaluation model accuracy. The hazard point density for very low and very high susceptibility areas in the MFL-LN model was 0.009 and 0.557, respectively, ensuring that the hazard point density is lower in very low susceptibility areas and highest in very high susceptibility areas. Additionally, the area under the receiver operating characteristic (ROC) curve was the highest at 0.837. The model was analyzed for lightweight optimization based on parameter count and floating-point operations. The parameter count of the MFL-LN model was 0.081, and the floating-point operation count was 41.102. This effectively reduced the model training costs and improved the accuracy of landslide susceptibility zoning. Conclusions: By selecting Ankang City in Shaanxi Province as the study area and analyzing the historical landslide hazard data, the MFL-LN model was constructed utilizing the DFEM and MSFF with an encoder-decoder structure to evaluate landslide susceptibility. This approach improved computational efficiency while maintaining the model's performance, making the landslide prediction model more efficient and practical.