基于先验知识辅助聚类的粗-精输电线路多金具检测
Coarse-fine Transmission Line Multi-fitting Detection Based on Prior Knowledge Auxiliary Clustering
翟永杰 1郭聪彬 1陈年昊 1王璐瑶 1王乾铭 1赵文清1
作者信息
- 1. 华北电力大学控制与计算机工程学院,河北省 保定市 071003
- 折叠
摘要
为了解决输电线路多金具检测任务中存在的小目标和密集遮挡问题并充分利用金具高分辨率航拍图像的信息优势,提出基于先验知识辅助聚类的粗-精输电线路多金具检测方法.首先,通过粗检测模块实现对输电线路高分辨率航拍图像的初步感知.接着,通过先验知识指导结构场景子区域选取模块中聚类算法半径的确定,以自适应聚类出合适的子区域.最后,设计精检测模块充分利用高分辨率航拍图像中的关键信息,进行金具的精确感知,并融合粗检测结果以实现由粗到精的金具识别.经实验证明,基于先验知识辅助聚类的粗-精输电线路多金具检测模型比之基线模型准确率提高了11.3%,对其中小目标金具和密集遮挡金具检测准确率的提高尤为明显.
Abstract
To solve the problem of tiny-size objects and dense occlusion in the transmission line multi-fitting detection task and make full use of the information of fitting high-resolution aerial images,a coarse-fine transmission line multi-fitting detection method based on prior knowledge auxiliary clustering is proposed.First,the coarse detection module is utilized to realize the preliminary perception of the high-resolution aerial images of transmission lines.Then,the determination of clustering algorithm's radius in structural scene sub-region selection module is guided through the prior knowledge for adaptively clustering the appropriate sub-regions.Finally,a fine detection module is designed to fully utilize key information from high-resolution aerial images for precise perception of fittings,and to realize the recognition of fittings from coarse to a finer level by fusing the coarse detection results.The experimental results show that the accuracy of the coarse-fine transmission line multi-fitting detection method based on prior knowledge auxiliary clustering is 11.3%higher than that of the baseline model.Especially,the detection accuracy of tiny-size fittings and dense occlusion fittings is improved obviously.
关键词
输电线路/金具/航拍图像/深度学习/目标检测/粗-精检测Key words
transmission lines/fitting/aerial images/deep learning/object detection/coarse-fine detection引用本文复制引用
基金项目
国家自然科学基金项目(U21A20486)
出版年
2024
国家科技期刊平台
NSTL
万方数据