The identification of regional loess seismic landslides provides fundamental data for landslide disaster risk control. In this study, deep learning is applied to automatically identify and classify typical seismic landslides in loess areas of China using remote sensing image data. First, a database of loess seismic landslides was used to support the visual interpretation of remote sensing images for selecting landslide and nonlandslide samples. This database comprises landslides investigated by the seismic landslide research team from the Institute of Disaster Prevention in Gansu Province and the Ningxia Hui Autonomous Region. Subsequently, the GoogLeNet network model was employed to automatically identify and classify loess seismic landslides and nonlandslides. Finally, the classification and recognition accuracies of the model were evaluated to assess its performance in identifying seismic landslides in loess areas. The results show that this method achieves high identification accuracy and efficiency for loess seismic landslides, and that key landslide areas can be rapidly determined from remote sensing images. Therefore, the proposed method enables rapid assessment of similar landslide-prone regions and provides technical support for large-scale investigations of landslide disasters.