One of the most catastrophic natural catastrophes is landslides since they typically follow other major disasters, further harming already damaged systems. In order to mitigate the impact of landslides, it is crucial to be able to quickly and accurately identify potential landslide locations. One way to achieve this is by using high-spatial-resolution remote-sensing image technology, which allows emergency rescue teams to quickly scan large areas and identify potential landslide locations. This research focuses on landslide detection from high-resolution optical satellite images using the YOLOv7, a single-stage object detection algorithm which is known for its fast and accurate object detection capabilities. The dataset from Bijie landslide is utilised to train and evaluate the model. In addition, an Efficient Channel Attention (ECA) module is introduced in YOLOv7, which increases the model’s capacity to focus on relevant features in the image. The findings demonstrate that the attention mechanism improved YOLOv7 performed better than other state-of-the-art models with an F1-score of 96.7% and a mAP score of 97.3%.