Improving YOLOv5 with Attention Mechanism for Detecting Boulders from Planetary Images

It is of great significance to apply the object detection methods automatically detect boulders from planetary images and analyze their distribution. This contributes selection candidate landing sites understanding geological processes. paper improves state-of-the-art method YOLOv5 with attention mechanism designs a pyramid based approach images. A new feature fusion layer has been designed capture more shallow features small boulders. The modules implemented by combining convolutional block module (CBAM) efficient channel network (ECA-Net) are also added into highlight information that contribute boulder detection. Based on Pascal Visual Object Classes 2007 (VOC2007) dataset which widely used for evaluations we constructed Bennu asteroid, evaluation results have shown improvements increased performance 3.4% in precision. With improved method, extracts several layers different resolutions large detects scales layers. We applied proposed asteroid. distribution asteroid analyzed presented.