Open Access

ARTICLE

A Cost-Sensitive Transformer-Based Network for Small Defect Detection in Power Transmission Line

Congcong Ma^1,2, Wenqing Zhao^1,3, Feifei Fu^1,2, Jiaqi Mi^4,*

1 Department of Computer, North China Electric Power University, Baoding, China
2 Engineering Research Center of Intelligent Computing for Complex Energy Systems, Ministry of Education, Baoding, China
3 Hebei Key Laboratory of Knowledge Computing for Energy & Power, Baoding, China
4 College of Artificial Intelligence, Nankai University, Tianjin, China

* Corresponding Author: Jiaqi Mi. Email:

Computers, Materials & Continua 2026, 89(2), 74 https://doi.org/10.32604/cmc.2026.085854

Received 19 May 2026; Accepted 06 August 2026; Issue published 15 September 2026

Abstract

Small object detection is a common and challenging task in transmission line inspection scenarios. Existing one-stage and two-stage object detection methods in this scenario are still constrained by extremely small object scale and limited feature representation capability, resulting in suboptimal performance in small object detection. To address these issues, this paper proposes a Cost-Sensitive Transformer-based Network for small object detection. First, a Transformer-based backbone is designed, coupled with a neck that integrates Feature Pyramid and Path Aggregation structures, enabling enhanced multi-scale feature extraction and fusion. Second, a hybrid-domain attention-based decoupled detection head is proposed, where an independent localization confidence branch enhances bounding box precision for small objects and reduces missed detections. Third, a cost-sensitive loss is designed to mitigate multi-dimensional imbalance in small object detection tasks. Extensive experiments are conducted on a proprietary transmission line inspection dataset and the COCO benchmark. The proposed method achieves a mean Average Precision (mAP) of 92.7% on the proprietary dataset, outperforming the state-of-the-art by 0.7%, and attains an APS of 35.9% on COCO, showing a significant improvement over the baseline. These results demonstrate the effectiveness and generalization capability of the proposed method for small object detection in transmission line inspection tasks.

---

Keywords

---