To improve the landslide detection accuracy in remote sensing images under complex backgrounds, a landslide detection method for remote sensing images based on an improved mask region-based convolutional network (Mask RCNN) was proposed. First, the backbone network in the Mask RCNN model was replaced with a residual network 101 (ResNet101), and additional modules including the convolutional block attention module (CBAM), the path aggregation feature pyramid network (PAFPN), and a cascade detector were integrated to construct a landslide detection model based on remote sensing images; Then, the model was trained using a remote sensing landslide dataset; Finally, the trained model was used to perform detection and segmentation experiments on the test images.The results show that, compared with the original Mask RCNN model, the improved model increases the box average precision from 80.2% to 83.7%, and the mask average precision from 79.1% to 81.1%. The overall inference time remains virtually unchanged. The improved Mask RCNN model exhibits high detection accuracy and real-time processing capability, serving as a key technical support for landslide disaster early warning.