Abstract: Research Objective Rapid and accurate early identification of landslides is essential for effective landslide hazard prevention and mitigation. Traditional approaches based on field geological surveys and visual interpretation are typically time-consuming and labor-intensive. Research Methods In recent years, deep learning algorithms have been increasingly applied in regional landslide detection due to their efficiency and automation capabilities. This study proposes a regional landslide detection approach using the Mask Region-Based Convolutional Neural Network (Mask R-CNN) to distinguish landslide areas from non-landslide zones and to accurately classify various types of landslides. The effectiveness of the proposed method was validated using data from Sai Kung East Country Park in Hong Kong. Research Results Results show that the proposed method not only enables effective automatic detection of landslides across large areas, but also quantifies the accuracy of landslide bounding boxes through confidence scores. By integrating Digital Terrain Model (DTM) data with satellite imagery, the approach significantly improve the distinction between landslides and man-made structures (e.g., roads), improving identification accuracy. Conclusion This approach offers a valuable technical approach for regional cluster landslide identification.