基于深度学习和视觉语义关系的防振锤滑移目标检测  

Target detection of vibration damper movement based on deep learning and visual semantic relations

作　　者：安金鹏 王晓春 AN Jinpeng;WANG Xiaochun(Xinjiang Information Industry Co.,Ltd.,Urumqi 830011,China)

机构地区：[1]新疆信息产业有限责任公司,乌鲁木齐830011

出　　处：《智能计算机与应用》2022年第9期183-188,共6页Intelligent Computer and Applications

摘　　要：为了精确定位防振锤滑移异常,保障输电线路安全运行,在输电线路巡检过程中,通常使用视觉检测的方法,对防振锤滑移异常及相关部件进行识别。但输电线路场景复杂多变,常规目标检测算法对防振锤滑移异常检测的精度较差,无法满足实际检测需求。因此,本文提出一种基于深度学习和视觉语义关系的防振锤滑移目标检测方法。该方法依据空间上下文信息,判断所检测目标间的视觉语义关系;联合Cascade R-CNN目标检测算法,并利用制定的相应判别规则及约束算法,实现防振锤滑移异常判别。实验结果表明,与常规目标检测算法相比,本文方法对防振锤滑移异常目标识别更加有效,在输电线路巡检中具有较高的理论价值与可观应用前景。In order to accurately locate the abnormal sliding of the vibration damper to ensure the safe operation of the transmission line, visual inspection method is usually used to identify the abnormal sliding of the vibration damper and the related components during the inspection process of the transmission line. However, the target detection algorithm has poor accuracy in the detection of vibration damper slip anomalies in the experiment and cannot meet the actual detection needs. Therefore, in this paper, a vibration damper slip target detection method based on deep learning and visual semantic relationship is proposed. This method judges the visual semantic relationship between the detected targets according to the spatial context information. It combines the Cascade R-CNN target detection algorithm and uses the corresponding discrimination rules and constraint algorithms developed to realize the vibration damper slip anomaly discrimination. The experimental results show that compared with the conventional target detection algorithm, the method proposed in this paper is more effective for the abnormal target recognition of vibration damper slippage, and has higher theoretical value and broad application prospects in the inspection of transmission lines.

关 键 词：防振锤 深度学习 Cascade R-CNN 视觉语义关系 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]