Abstract: Corrosion is a common form of damage in steel structures. If it is not discovered and repaired in time， it will affect the safety of the entire structure. This paper takes the surface corrosion of steel structures as the research object and proposes a method for detecting surface corrosion of steel structures based on computer vision by using drones to obtain high-definition images. This method obtains high-definition corrosion images through drones， constructs a corrosion dataset， and trains and improves the YOLOv8 model to achieve high-precision corrosion target detection. The U_Net model is used to segment the detected corrosion areas， enabling the calculation of the corrosion area. The improved YOLOv8 network model is based on the YOLOv8-OBB model. The deep convolutional layers of the backbone network of the YOLOv8-OBB model are pruned， and the Contextual Anchor Attention （CAA） module and the Receptive Field Attention convolution （RFAConv） are introduced to bring the shallow feature information of the backbone network to the neck network， thereby achieving feature fusion. This method was verified by experiments， resulting in a 3.8% improvement in precision （P）， a 5% increase in recall （R）， and a 3.5% enhancement in mean Average Precision （mAP） for corrosion detection. Meanwhile， the number of parameters was reduced by 63.2%， while floating-point operations increased by 11.1×10⁹， and frames per second （FPS） decreased by 17.47.