TinyDef-DETR:An Enhanced DETR Detector for UAV Power Line Defect Detection

2025-09-07

AI Summary

To address the challenge of detecting fine, ambiguous power line defects under complex backgrounds in UAV-based inspection, this paper proposes a lightweight DETR architecture. It replaces conventional stepwise downsampling with lossless spatial-to-depth downsampling to preserve fine-grained details; introduces edge-aware convolution to enhance boundary features; designs a cross-stage dual-domain (spatial–frequency) multi-scale attention module to strengthen global context and local cue integration; and proposes Focaler-Wise-SIoU regression loss, specifically optimized for hard small-object localization. Evaluated on the CSG-ADCD dataset, our method significantly outperforms mainstream detectors—especially achieving notable mAP gains on the small-object subset—while maintaining low computational overhead. Further experiments on VisDrone confirm its strong generalization capability across diverse aerial detection scenarios.

Technology Category

• Computer Vision: Object Detection & Categorization
• Machine Learning: Hardware-aware ML
• Search and Optimization: Learning to Search

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Automated inspection of transmission lines using UAVs is hindered by the difficulty of detecting small and ambiguous defects against complex backgrounds. Conventional detectors often suffer from detail loss due to strided downsampling, weak boundary sensitivity in lightweight backbones, and insufficient integration of global context with local cues. To address these challenges, we propose TinyDef-DETR, a DETR-based framework designed for small-defect detection. The method introduces a stride-free space-to-depth module for lossless downsampling, an edge-enhanced convolution for boundary-aware feature extraction, a cross-stage dual-domain multi-scale attention module to jointly capture global and local information, and a Focaler-Wise-SIoU regression loss to improve localization of small objects. Experiments conducted on the CSG-ADCD dataset demonstrate that TinyDef-DETR achieves substantial improvements in both precision and recall compared to competitive baselines, with particularly notable gains on small-object subsets, while incurring only modest computational overhead. Further validation on the VisDrone benchmark confirms the generalization capability of the proposed approach. Overall, the results indicate that integrating detail-preserving downsampling, edge-sensitive representations, dual-domain attention, and difficulty-adaptive regression provides a practical and efficient solution for UAV-based small-defect inspection in power grids.

Problem

• Detecting small power line defects in complex UAV backgrounds
• Overcoming detail loss from strided downsampling in detectors
• Integrating global context with local cues for defect detection

Innovation

• Stride-free space-to-depth module for lossless downsampling
• Edge-enhanced convolution for boundary-aware feature extraction
• Cross-stage dual-domain attention for global-local integration