基于改进YOLOv8级联架构的水工金属结构锈蚀检测与等级判定方法OA

To mitigate the subjectivity of rust annotation and enhance annotation efficiency,this paper proposes a rust detection method based on a YOLOv8 cascaded architecture to rapidly implement deep learning-based corrosion detection for hydraulic engineering metal structures.To enhance accurate detection of rust target,the MobileViTv3 module is integrated into YOLOv8n,resulting in a modified model named YOLOv8-vit.Based on YOLOv8-vit,we further propose YOLOv8-vit-cls for rust grade learning and classification.This network leverages the pretrained parameters of YOLOv8-vit to efficiently learn features of different rust grades.Finally,a cascaded architecture combining YOLOv8-vit and YOLOv8-vit-cls is constructed to perform both rust detection and grade classification of hydraulic metal structures.