Insulator Defect Detection Method Based on Model Parameter Migration BAAC-YOLOv11

Abstract

To enhance the accuracy and operational applicability of transmission line insulator defect detection systems, this study proposes a hybrid framework integrating model parameter transfer learning with an optimized YOLOv11 architecture. First, A parameter transfer framework was engineered to mitigate cross-domain performance degradation in insulator defect detection, incorporating an adversarial domain discriminator with dual-branch architecture for simultaneous image-level texture analysis and instance-level feature alignment. This module was trained using gradient reversal layers to enforce domain-invariant feature learning while preserving discriminative defect patterns; Second, taking advantage of the advantages of YOLOv11 model being lighter and easier to deploy on resource constrained devices compared to other versions, as well as its friendly application scenario for detecting insulator defects in outdoor transmission lines, a bidirectional feature pyramid network (BiFPN) is adopted as a new multi-scale feature extraction network, and attention mechanism and anchor box clustering method are introduced to improve the YOLOv11 model and further enhance its performance. An anchor box clustering YOLOv11 algorithm based on BiFPN and Attention mechanism (BAAC-YOLOv11) was proposed. Through simulation experiments on the artificial sample generation set of insulator defects, the results show that under harsh conditions such as foggy weather, this method has good transfer recognition effect on the insulator defect detection model, verifying the effectiveness of this method. © 2025 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.