Attention-Guided Autoencoder Fusion for Insulator Defect Detection Using UAV Transmission-Line Imaging

2026-06-08 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

AE-YOLO, an Attention-Guided AutoEncoder-Enhanced YOLO framework, addresses challenges in automated high-voltage transmission-line insulator defect detection using UAV imagery. It integrates lightweight bottleneck autoencoders within a Feature Pyramid Network-Path Aggregation Network (FPN-PAN) neck to preserve anomaly-sensitive information during multi-scale feature fusion. Convolutional Block Attention Modules (CBAM) enhance feature discrimination, and a variance-maximizing autoencoder regularization strategy encourages diverse, defect-discriminative latent representations. The network trains with a unified objective combining focal loss, CIoU loss, and autoencoder regularization. During inference, Weighted Boxes Fusion (WBF) combines predictions from YOLOv8, YOLOv10, and YOLO11, with an autoencoder-guided confidence boosting mechanism. On the Insulator-Defect Detection dataset, AE-YOLO with an EfficientNetV2 backbone achieved 95.10% mAP@0.5, 96.40% precision, and 93.80% recall, surpassing the strongest YOLO baseline by 5.0 points in mAP@0.5 and 6.7 points in recall.

Key takeaway

For Machine Learning Engineers developing UAV-based inspection systems, AE-YOLO offers a robust solution for critical defect detection. You should consider integrating attention-guided autoencoders into your YOLO-based frameworks to significantly improve recall for small, rare anomalies like pollution-flashover. This approach enhances sensitivity to subtle defects, crucial for preventing costly infrastructure failures and improving grid reliability.

Key insights

Integrating autoencoders and attention into YOLO enhances UAV-based insulator defect detection, especially for small, rare anomalies.

Principles

• Anomaly-aware feature preservation improves recall.
• Variance-maximizing regularization diversifies latent codes.
• Ensemble inference boosts robustness for rare defects.

Method

AE-YOLO uses a configurable attention-enhanced backbone, an FPA neck with bottleneck autoencoders for anomaly-aware feature refinement, a multi-scale detection head, and an ensemble inference module with AE-guided confidence boosting and WBF.

In practice

• Implement bottleneck autoencoders in FPN-PAN necks.
• Apply CBAMs for feature discrimination.
• Combine focal, CIoU, and AE regularization losses.

Topics

• Insulator Defect Detection
• UAV Inspection
• YOLO Frameworks
• Autoencoders
• Attention Mechanisms
• Feature Pyramid Networks
• Object Detection

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• Attention-Guided Autoencoder Fusion for Insulator Defect Detection Using UAV Transmission-Line Imaging — AE-YOLO integrates autoencoders and attention mechanisms into YOLO for robust UAV-based insulator defect detection, improving mAP and recall.
• Edge-Constrained UAV Small-Object Detection with P2 Enhancement and Quantum-Inspired Lightweight Structure Search — Combining a P2 high-resolution branch with QIEA structure search enhances UAV small-object detection under edge constraints.
• YOLO — You Only Look Once — A decade of real-time vision — YOLO transformed object detection into a single regression problem for real-time performance.
• Visual Prompting Meets Feature Reconstruction-Based Anomaly Detection with Dual-Teacher Supervision — Robust anomaly detection in varied real-world conditions is achieved by combining visual prompting, adaptive teacher models, and synthetic data.
• A.R.I.S.: Automated Recycling Identification System for E-Waste Classification Using Deep Learning — A.R.I.S. uses YOLOx for real-time e-waste classification, significantly improving material recovery efficiency.
• YOLOv26: An Object Detector Built for Real-Time Deployment — YOLOv26 prioritizes edge deployment through architectural simplification and hardware-friendly optimizations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.