基于图像的石化储罐罐顶锈迹识别算法  

Image Based Algorithm for Identifying Rust on the Top of Petrochemical Storage Tanks

作　　者：傅钰江 王若琳 隋顾磊 李雪 石磊 陈博 Fu Yujiang;Wang Ruoin;Sui Guei;Li Xue;Shi Lei;Chen Bo(SINOPEC(Dalian)Research Institute of Petroleum and Petrochemicals Co.,Ltd.,Liaoning,Dalian,116045)

机构地区：[1]中石化(大连)石油化工研究院有限公司,辽宁大连116045

出　　处：《安全、健康和环境》2024年第4期7-13,共7页Safety Health & Environment

基　　金：国家重点研发计划项目(2022YFB3305900),流程制造资源能源生产计划决策软件与工业应用。

摘　　要：大型石化储罐罐区的日常巡检工作繁重,无人机拍摄可有效覆盖储罐罐顶。针对罐顶锈蚀问题,建立储罐罐顶图像数据集,开发基于形状导向的罐体分割和分块集成锈迹识别算法。实验结果表明:①形状导向优化策略保证95%以上的罐顶分割准确度;②分块集成将锈迹识别的准确度提高到80%以上;③与深层网络相比,浅层的分割网络更适用于锈迹识别和罐顶分割。将边缘特征优化和分块集成有机结合,解决了小目标锈迹在整幅图像难以识别的问题,有效提升了锈迹识别的准确度,适合石化储罐锈迹识别领域的实际应用。The daily inspection work of large oil stor-age tank areas is heavy,and the UAV shooting can effectively cover the roof of the storage tank.To ad-dress the issue of tank top rust,the image data set of tank top was established,and the shape-based tank segmentation and block integrated rust recognition al-gorithm were developed.The experimental results showed that,①The shape oriented optimization strat-egy can guarantee more than 95%accuracy of tank top segmentation.②Block integration improved the accuracy of rust identification to more than 80%.③Compared with the deep network,the shallow net-work was more suitable for rust identification and tank top segmentation.The combination of edge feature optimization and block integration solved the problem that small target rust was difficult to recognize in the whole image,effectively improved the accuracy of rust recognition,and was suitable for practical appli-cation in the field of petrochemical storage tank rust recognition.

关 键 词：石化储罐 锈迹识别 图像分割 边缘特征优化 分块集成 锈蚀评价 

分 类 号：TE65[石油与天然气工程—油气加工工程] TP391.4[自动化与计算机技术—计算机应用技术]