显著性特征重构下的海上钻井作业场景潜在安全风险视觉图像识别算法  

Visual Image Recognition Algorithm for Potential Safety Risks in Offshore Drilling Operation Scenarios Based on Significant Feature Reconstruction

作　　者：覃建宇 刘宇沛 邓文杨 王天昊 黄泽彬 QIN Jianyu;LIU Yupei;DENG Wenyang;WANG Tianhao;HUANG Zebin(Shenzhen Branch,CNOOC China Limited,.Shenzhen,Guangdong 841000,China;CNOOC Research Institute Ltd.,Beijing 100022,China)

机构地区：[1]中海石油(中国)有限公司深圳分公司,广东深圳841000 [2]中海油研究总院有限责任公司,北京100022

出　　处：《计算技术与自动化》2025年第1期19-24,共6页Computing Technology and Automation

摘　　要：在海上钻井作业场景潜在安全风险分析过程中,基于几何和纹理特征识别视觉图像包含的潜在安全风险,在复杂场景下会受到背景噪声干扰,使得识别结果F1值较低。因此,提出了显著性特征重构下的海上钻井作业场景潜在安全风险视觉图像识别算法。通过现场监控平台采集海上钻井作业场景视觉图像,利用改进直方图均衡算法实现视觉图像增强处理。建立图像显著性矩阵获取图像感兴趣区域,将小面积区域去除后针对该区域提取显著性特征,依托于自适应特征重构金字塔结构完成显著性特征重构。以基于区域的卷积神经网络为核心构建识别模型,将显著性特征重构结果输入模型中进行学习,输出海上钻井作业场景潜在安全风险识别结果。实验结果表明:当前算法应用后得出的识别结果F1值保持在0.93以上,充分体现了该识别方式的优越性。In the process of analyzing potential safety risks in offshore drilling operations,geometric and texture features are mainly used to identify potential safety risks contained in visual images.In complex scenarios,background noise may interfere with the recognition results,resulting in lower F1 values.Therefore,a visual image recognition algorithm for potential safety risks in offshore drilling operation scenarios based on saliency feature reconstruction is proposed.Collect visual images of offshore drilling operation scenes through on-site monitoring platforms,and use improved histogram equalization algorithm to achieve visual image enhancement processing.Establish an image saliency matrix to obtain the region of interest in the image.After removing small areas,extract saliency features for that area,and rely on the adaptive feature reconstruction pyramid structure to complete saliency feature reconstruction.Construct a recognition model based on a region based convolutional neural network as the core,input the reconstruction results of salient features into the model for learning,and output the identification results of potential safety risks in offshore drilling operation scenarios.The experimental results show that the recognition result F1 value obtained by the current algorithm after application remains above 0.93,fully reflecting the superiority of this recognition method.

关 键 词：显著性特征重构 视觉图像 海上钻井作业 潜在安全风险 风险识别 特征提取 

分 类 号：TU714[建筑科学—建筑技术科学]