摘要
使用YOLOv3深度网络模型,针对航拍图像中绝缘子检测的准确性问题进行研究,提出了一种分解聚合算法。为解决目标的错检、漏检等问题,将目标分解成多个存在交集的可变型部件,并对其进行检测。在保证子目标检测精度与速度的前提下,利用各部件之间相交区域的特征及含义,对其进行聚合并重新定义,使检测到的目标区域更准确。由于群体性目标中包含的可变因素过多,原算法无法准确定义,提出的改进方法则可根据必需部件对其进行检测,同时为单独的子目标找出它所隶属的整体,通过多级标签对其进行更深刻意义上的描述。以COCO数据集为例,对比算法改进前后的检测效果。实验结果表明,该方法显著提高了目标检测的准确性,解决了漏检、错检等问题。
In this paper, the depth network model of YOLOv3 is adopted to study the accuracy of insulator detection in aerial photography images, and the decomposition and aggregation algorithm is proposed. In order to solve the problem of wrong check and missing check, the target is decomposed into several intersecting variable parts and its detection is conducted. On the premise of ensuring the detection accuracy and speed of sub-targets, the features and meanings of intersecting areas between components are used to aggregate and redefine them, thus making the detected target area more accurate. The original algorithm cannot define it accurately because there are too many variables in the group objective, the improved method proposed in this paper can detect it according to the required parts, and find out the whole of its membership for the individual sub-targets, and it is described in a more profound sense through multi-level tags. Taking COCO data set as an example, the detection effect is compared before and after the improvement of the algorithm, and the experimental results show that this method can improve the accuracy of target detection and solve the problems of missed detection and wrong detection.
作者
高强
廉启旺
Gao Qiang;Lian Qiwang(Institute of Electrical and Electronic Engineering, North China Electric Power University,Baoding 071000,Hebei, China)
出处
《电测与仪表》
北大核心
2019年第5期119-123,共5页
Electrical Measurement & Instrumentation
关键词
YOLOv3
目标检测
分解
聚合
YOLOv3
target detection
decomposition
aggregation
