文献类型：期刊文献

中文题名：基于YOLO算法的金属表面腐蚀图像识别分析系统

英文题名：Image Recognition and Analysis System for Metal Surface Corrosion Based on YOLO Algorithm

作者：胡杰珍[1,2];杨靖荣[1];邓培昌[2,3];蓝文杰[1];钟声昊[1]

机构：[1]广东海洋大学机械工程学院,广东湛江524088;[2]广东海洋大学湛江市海洋工程及装备腐蚀与防护重点实验室,广东湛江524088;[3]广东海洋大学化学与环境学院,广东湛江524088

年份：2025

卷号：58

期号：9

起止页码：124

中文期刊名：材料保护

外文期刊名：Materials Protection

基金：广东省自然科学基金(2021A15150129)。

语种：中文

中文关键词：腐蚀图像;YOLO算法;腐蚀监测;识别分析系统

外文关键词：corrosion image;YOLO algorithm;corrosion monitoring;identification and analysis system

中文摘要：金属表面腐蚀识别分析系统是基于腐蚀形貌分析腐蚀情况的金属腐蚀监测设备的核心组成部分,开发金属表面腐蚀识别分析系统对金属腐蚀监测技术的发展具有重要意义。使用F1-Score和mAP评估方法,从YOLO v5,v6,v7和v84个版本算法中优选出适用于金属表面腐蚀识别分析系统的YOLOv8模型。通过数据清洗、数据增强、XML注释、边界框标注等方法和步骤对金属表面腐蚀图像原始数据集进行处理,形成计算机深度学习训练数据集,编写了导入图像和视频处理、YOLO模型加载、计算设备选择以及数据集中类别名称处理等模块对应的程序。经计算机深度学习,基于YOLOv8的金属表面腐蚀识别分析系统训练损失和验证损失下降,精度和召回率提高,mAP值逐渐上升,模型具备较好的泛化能力。将该模型应用于实际发生的金属装备腐蚀检测中可以发现,模型识别分析效果较为理想,将该模型应用于实际工况下金属表面腐蚀图像的识别与分析,能准确识别腐蚀发生位置,进行准确的腐蚀分类。

外文摘要：The metal surface corrosion identification and analysis system is the core component of metal corrosion monitoring equipment that analyzes corrosion conditions based on corrosion morphology.The development of this system is of great significance for the development of metal corrosion monitoring technology.In this study,the YOLO v8 model was selected as the optimal model for the metal surface corrosion identification and analysis system from four versions of YOLO(v5,v6,v7 and v8)using F1-score and mean Average Precision(mAP)evaluation methods.The original dataset of metal surface corrosion images was processed through data cleaning,data augmentation,XML annotation and bounding box labeling to form a deep learning training dataset for computer.Programs corresponding to modules such as image and video import processing,YOLO model loading,computing device selection and class name processing in the dataset were developed.After computer deep learning training,the YOLO v8-based metal surface corrosion identification and analysis system demonstrated decreasing training and validation losses,improved precision and recall and a gradually increasing mAP value,and the model exhibited good generalization ability.When this model was applied to the detection of corrosion on actual metal equipment,the model exhibited favorable recognition and analysis performance.Furthermore,when applied to the recognition and analysis of metal surface corrosion images under real operating conditions,the model accurately identified corrosion locations and realized accurate corrosion classification.