A Few-Shot Defect Detection Method for Transmission Lines Based on Meta-Attention and Feature Reconstruction

In tasks of transmission line defect detection, traditional object detection algorithms are ineffective, with few training samples defective components. Meta-learning uses multi-task learning as well fine-tuning to learn common features in different tasks, which has the ability adapt new quickly, shows good performance few-shot and generalization tasks. For this reason, we proposed a method (Meta PowerNet) Meta-attention RPN Feature Reconstruction Module for lines based on meta-learning. First, stage region proposal, proposal network (Meta-Attention Region Proposal Network, MA-RPN) is designed fuse support set query filter noise anchor boxes. addition, it focus subtle texture smaller-sized objects by fusing low-level from set. Second, meta-feature construction stage, meta-learner feature reconstruction module core capture defect-related channels. The experimental results show that under condition, there only 30 various types component defects. achieves 72.5% accuracy defects, significant improvement compared other mainstream detection. Meanwhile, MA-RPN paper can be used meta-learning models universally.