基于逻辑语义判别的调峰电源设备漏油图像识别  

Oil Leakage Image Recognition of Peak Regulation Power Equipment Based on Logical Semantic Discrimination

作　　者：宋倩 祁炜雯 方冰 范强 郑慧娟 胡飞 SONG Qian;QI Wei-wen;FANG Bing;FAN Qiang;ZHENG Hui-juan;HU Fei(Water and Innovation Department,State Grid Corporation of China,Beijing 100031,China;Shaoxing Power Supply Company of State Grid Zhejiang Electric Power Co.,Ltd.,Shaoxing 310000,China;State Grid Electric Power Research Institute,Nanjing 211106,China)

机构地区：[1]国网水新部,北京100031 [2]国网浙江省电力有限公司绍兴供电公司,浙江绍兴310000 [3]国网电力科学研究院有限公司,江苏南京211106

出　　处：《水电能源科学》2023年第5期199-202,210,共5页Water Resources and Power

基　　金：国家电网有限公司总部科技项目(5700-202140381A-0-0-00)。

摘　　要：针对调峰电源设备漏油图像识别中的难点问题,将逻辑规则判别引入图像识别领域,提出一种全新的图像识别方法。基于直方图均衡化技术增强原始图像,利用Mask RCNN网络初步获取储油装置、地面、疑似油污区域的位置和轮廓信息。根据这些信息判断物体间位置关系,并结合逻辑表达式判断疑似区域是否属于漏油区域。基于现场拍摄的调峰电源设备图像进行实例分析,结果表明该方法能有效克服漏油区域识别中的难点问题,大幅提升了识别准确率。To address the challenges in oil leakage image recognition of peak regulation power equipment,a new method is proposed by introducing logical rule discrimination strategy into the task of image recognition.The technology of histogram equalization is adopted to improve original image.Then Mask RCNN network is introduced to obtain the preliminary location and contour information of storage device,ground and suspected oil area.Based on the above information,positional relationships between objects are judged,and logical expressions are adopted to determine the oil leakage area.Example analysis is conducted based on filed images of peak regulation power equipment.The results indicate that the proposed framework solves the problems in oil leakage area recognition,largely boosting model performance.

关 键 词：调峰电源设备 漏油图像识别 物体间关系识别 逻辑语义判别 

分 类 号：TM712[电气工程—电力系统及自动化]