Transmission Line Key Components and Defects Detection Based on Meta-Learning

• Published in: IEEE transactions on instrumentation and measurement 2024, Vol.73, p.1-13
• Main Authors: Dong, Chao, Zhang, Ke, Xie, Zhiyuan, Wang, Jiacun, Guo, Xiwang, Shi, Chaojun, Xiao, Yangjie
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Cascade region-aware fusion (RAF) heads; Clamps; Deep learning; Defects; Feature extraction; Inspection; Insulators; Machine learning; meta-learning; Object detection; Power transmission lines; Queries; RAF module; Training; transmission line object detection; Transmission lines; two-stage training strategy
• ISSN: 0018-9456; 1557-9662
• DOI: 10.1109/TIM.2024.3403202

The detection of key components with defects in transmission lines is a critical task in maintaining a power system's stability. Deep learning (DL) can play an important role in the detection. However, due to limited samples of defect components, DL methods can easily suffer from overfitting in model training. To address this issue, we propose a novel meta-learning-based model. This model effectively integrates query features with support features, enabling the identification of objects in query images belonging to the same category as the support images. It uses a region-aware fusion (RAF) module to transform support images into RA vectors to guide the detection network by customizing the allocation of support information to local regions of query images. In addition, a two-stage fine-tuning training strategy is developed to leverage the majority of data to assist the minority, alleviating overfitting during small-sample training and reducing the data gap between new and base classes. The experimental results demonstrate that our proposed model outperforms Faster region-based convolutional neural network (Faster RCNN) under a 30-shot setting, achieving a higher mean average precision (mAP) with a significant improvement of 40.9%.