稳健性(进化)
绝缘体(电)
计算机科学
计算机视觉
人工智能
可靠性工程
工程类
目标检测
运动模糊
实时计算
训练集
运动检测
模拟
电子工程
工作(物理)
异步通信
数据建模
电力传输
作者
Lei Jin,S F Chen,Y J Zhang
出处
期刊:
[SPIE]
日期:2026-05-13
卷期号:: 30-30
摘要
With advances in object detection technology, unmanned aerial vehicles (UAVs) have been widely adopted for insulator inspection and defect identification. Defective insulators can cause serious power outages, highlighting the need for reliable detection. However, existing public datasets, such as the Chinese Power Line Insulator Dataset (CPLID), are limited in scale and lack images of insulators and their defects under diverse environmental conditions. This limitation constrains the performance of detection models in real-world scenarios. To address the challenges of scarce training data and insufficient environmental diversity, this paper systematically develops three data augmentation algorithms: motion blur synthesis, fog synthesis, and rain synthesis. These algorithms are designed to simulate complex inspection environments and effectively expand the training dataset. Using CPLID as a base, this work creates an augmented version named CPLID-Aug. The proposed data augmentation methods significantly improve the accuracy and robustness of insulator defect detection models, offering an effective solution for reliable inspection in complex conditions.
科研通智能强力驱动
Strongly Powered by AbleSci AI