The steep natural terrain in Hong Kong is susceptible to shallow, small to medium-sized landslides induced by rainfall. These landslides usually involve failures occurring within the top one to two meters of the surface mantle. Terrain features manifested on the ground surface, such as historical landslides, rock outcrops, tension cracks, depressions could affect the terrain's susceptibility to landsliding in future. Identification of these features by conventional means involves substantial resources for Aerial Photo Interpretation (API) and field mapping. A pilot study was carried out to explore the potential of using deep learning to enhance the efficiency in identifying rock outcrops at a territory-wide scale with a view to improving the landslide susceptibility analysis. A methodology of combining Convolutional Neural Network and remote sensing techniques has been developed to derive a very high-resolution map of rock outcrops exposed on the natural terrain throughout the entire Hong Kong territory. The developed algorithm considers the spatial relationship and texture of the surrounding pixels as well as the spectral signature of the pixel of remote sensing imageries. The identification of rock outcrops has been conducted using the orthophotos acquired in years 2012 and 2015, with the aid of SPOT satellite images acquired in year 2015 and airborne LiDAR data acquired in year 2010, and resulting in a five-meter spatial resolution rock outcrops map. The results were promising when validated with the results mapped by engineering geologists using API on the same sets of orthophotos. The developed algorithm provides an alternative for leveraging the balance between spectral and spatial resolution, and for mapping the natural surface features and enhancing the landslide susceptibility analysis.