雨天变电站设备智能视觉检测算法与实时实现技术

Detection of Substation Equipment in Rainy Days and Real-Time Implementation

视频的变电站户外设备监控是保障电力系统安全运维的重要手段,但传统的目视监控效率低、人力成本高,因此无人化智能视觉检测得到广泛关注。变电站户外监控设备在采集图像时,不可避免会遭受雨天等恶劣天气影响,导致设备检测性能严重下降。针对雨天场景下的变电站设备检测给出了软硬件系统解决方案,主要围绕雨天变电站设备视觉检测算法和嵌入式AI实时部署展开研究。首先,提出一种基于细节增强的图像去雨网络(DERN),利用雨条的高频特性作为注意力权重指导去雨,进一步将其与YOLOv4目标检测算法进行联合优化,从而实现对雨天稳健的变电站设备检测算法;基于昇腾310 AI处理器构建嵌入式边缘端智能计算系统,从算子优化、多线程并行等方面提升计算性能。实验表明,在雨天场景下变电站设施设备目标检测精度从实施前的53.4%提高到82.1%,1080P视频运行速率达到30 f/s,验证了所提算法和系统的有效性,有效提升了变电站设备全天候视觉智能监控能力。

Video substation outdoor equipment monitoring is an important means to ensure the safe operation and maintenance of power system,but the traditional visual monitoring has low efficiency and high labor cost,so unmanned intelligent visual detection has attracted extensive attention.When collecting images,outdoor monitoring equipment in substation will inevitably be affected by bad weather such as rainy days,resulting in serious degradation of equipment detection performance.The hardware and software system solutions for substation equipment detection in rainy scenarios are presented.The visual detection algorithm of substation equipment in rainy day and the real-time deployment of embedded AI are studied.Detail enhancement residual network(DERN)is proposed,which uses the high-frequency characteristics of rain bars as the attention weight to guide the rain removal,and further optimizes it jointly with YOLOv4 target detection algorithm,so as to realize a robust substation equipment detection algorithm in rainy days.Based on Ascend 310 AI processor,an embedded edge intelligent computing system is built to improve the computing performance from the aspects of operator optimization and multi-threaded parallelism.The experiment shows that the object detection accuracy of substation facilities and equipments is improved from 53.4%before implementation to 82.1%in rainy scenarios,and the frame rate of 1080 P video reaches 30 f/s,which verifies the effectiveness of the algorithm and system,and effectively improves the all-weather visual intelligent monitoring ability of substation equipment.