面向电力生产精细化风险解译的高度相似防护工具智能检测技术研究

Research on Intelligent Detection Technology of Highly Similar Protection Tools for Refined Risk Interpretation of Power Production

电力生产通常面临高低电压、强弱电流等复杂工作环境转换,不同作业场景有严格的防护工具使用标准,因此,研究生产作业过程防护工具的精细辨识对保障人员安全及电网安全意义重大。已有研究可实现安全帽、工作服等基础着装类检测,而实际生产中存在大量形态高度相似的实体防护工具,如绝缘手套与线手套、绝缘杆与验电杆等。为此,该文提出一种基于深度代表性度量学习的相似防护工具智能检测方法。将目标类别特征学习转换为以差异化表达不同目标特征距离为目的的嵌入式空间特征学习,得到表征不同目标的深度代表性特征向量,通过计算未知目标与代表性特征向量的距离进行类别判断,最后以现场图像进行试验验证。试验结果表明:所提方法实现了对形态相似防护工具的特征差异表达和精准辨识,相比于常见目标检测模型具有更优越的辨识性能,从而提高电力生产安全风险辨识的精细化水平。

Power production is usually faced with complex working environment conversion such as high and low voltage,strong and weak current,and different operation scenarios have strict standards for the use of protection tools.Therefore,it is of great significance to study the fine identification of protection tools in the production process to ensure the safety of personnel and even the safe operation of power grid.Existing methods can realize the detection of basic clothing such as hats and work clothes.However,there are physical protection tools with highly similar shapes in actual production,such as insulating gloves and cotton gloves,insulating poles and testing poles.Therefore,this paper proposes an intelligent detection method of similar protection tools based on deep representative metric learning.The target category feature learning is transformed into embedded spatial feature learning to express the feature distances of different targets,to obtain the deep representative feature vectors representing different targets.By calculating the distance between the unknown target and the representative feature vectors to realize protection tools identification.Finally,the experimental verification is carried out with the images collected in the field.The results show that proposed method realizes the feature difference expression and accurate identification of similar protection tools,and has superior identification performance compared with common target detection models,thus improving the refined level of power production safety risk identification.