HRT-YOLO: a hybrid ratio-tunable adaptive framework for infrared power equipment detection

Abstract

To address challenges of limited computational resources on drones and interference from low resolution and background noise in infrared electrical equipment detection, this study proposes a hybrid ratio-tunable you only look once (HRT-YOLO) adaptive framework. Inspired by RepViT, the lightweight RepViTBlock (LRepViTBlock) was proposed, enabling the decoupling of the token-channel mixer, improving computational efficiency, and enhancing the model’s learning ability. A hybrid ratio-tunable convolution (HRTConv) was proposed, enabling computation-accuracy trade-offs through channel allocation factor α. A C3K2 module incorporating HRTConv was proposed, effectively reducing the computational complexity while maintaining negligible accuracy loss. When α is 0.75, evaluations on thermal imagery show HRT-YOLO achieves 95.5% mAP@0.5 and 78.6% mAP@0.5:0.9 with 4.7 giga floating-point operations per second (GFLOPS), reducing computations by 25.4% versus baseline while outperforming YOLOv5 by 6.8% in mAP@0.5:0.9. Furthermore, HRT-YOLO achieves rapid adjustment of model capacity through scaling factor α modulation, achieving computational-accuracy trade-offs for diverse situation requirements.

Keywords

A

References

[1] REN S, HE K, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. Advances in neural information processing systems, 2015, 28.

[2] Redmon J, Divvala S, Girshick R, et al. You only look once: unified, real-time object detection. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, June 27–30, 2016, Las Vegas, NV, USA, 2016, New York: IEEE, 779-788[C]

[3] Wang C Y, Bochkovskiy A, Liao H Y M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, June 18–22, 2023, Vancouver, BC, Canada, 2023, New York: IEEE, 7464-7475[C]

[4] Wang C Y, Yeh I H, Mark Liao H Y. YOLOv9: learning what you want to learn using programmable gradient information. European Conference on Computer Vision, September 29–October 4, 2024, Milan, Italy, 2024, Cham: Springer Nature Switzerland, 1-21[C]

[5] Wang A, Chen H, Liu L, et al. YOLOv10: real-time end-to-end object detection. Advances in neural information processing systems, 2024, 37: 107984-108011. J]

[6] Carion N, Massa F, Synnaeve G, et al. End-to-end object detection with transformers. European Conference on Computer Vision, August 23–28, 2020, Glasgow, UK, 2020, Cham: Springer International Publishing, 213-229[C]

[7] ZHU X, SU W, LU L, et al. Deformable DETR: deformable transformers for end-to-end object detection[EB/OL]. (2020-10-08) [2026-01-23]. https://arxiv.org/abs/2010.04159.

[8] Zhao Y, Lv W, Xu S, et al. DETRs beat YOLOs on real-time object detection. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, June 17–21, 2024, Seattle, WA, USA, 2024, New York: IEEE, 16965-16974[C]

[9] Yuan H, Chen X, Wang Y, et al. State detection of electrical equipment based on infrared thermal imaging technology. Pattern Recognition and Computer Vision: 2nd Chinese Conference, November 8–11, 2019, Xi’an, China, 2019, Cham: Springer International Publishing, 251-260. C]

[10] Li J M, Han B, Jiang M S. Anomaly monitoring and early warning of electric moped charging device with infrared image. Optoelectronics letters, 2025, 213136141. J]

[11] Lei T, Lv F, Liu J, et al. Research on fault detection algorithm of electrical equipment based on neural network. Mathematical problems in engineering, 2022, 202219015796[J]

[12] Cong S, Pu H, Wang X, et al. Application of improved YOLOv5 in infrared image recognition of electrical equipment. 2023 8th Asia Conference on Power and Electrical Engineering, April 25–27, 2023, Chongqing, China, 2023, New York: IEEE, 1836-1843[C]

[13] Hu J, Wan W, Qiao P, et al. Power insulator defect detection method based on enhanced YOLOV7 for aerial inspection. Electronics, 2025, 143408. J]

[14] Li D, Yang P, Zou Y. Optimizing insulator defect detection with improved DETR models. Mathematics, 2024, 12101507. J]

[15] Gao S, Ruan Y, Hong Q, et al. Infrared thermal image fault detection based on YOLOV3-L. IEEE International Conference on Artificial Intelligence and Computer Applications, June 24–26, 2022, Dalian, China, 2022, New York: IEEE, 175-178[C]

[16] Li J, Xu Y, Nie K, et al. PEDNet: a lightweight detection network of power equipment in infrared image based on YOLOv4-tiny. IEEE transactions on instrumentation and measurement, 2023, 72: 1-12[J]

[17] Tao H, Paul A, Wu Z. Infrared image detection and recognition of substation electrical equipment based on improved YOLOv8. Applied sciences, 2024, 151328. J]

[18] Wang A, Chen H, Lin Z, et al. RepViT: revisiting mobile CNN from ViT perspective. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, June 17–21, 2024, Seattle, WA, USA, 2024, New York: IEEE, 15909-15920[C]

[19] Ding X, Zhang X, Ma N, et al. RepVGG: making VGG-style convnets great again. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, June 19–25, 2021, Nashville, TN, USA, 2021, New York: IEEE, 13733-13742[C]

[20] Li H, Li J, Wei H, et al. Slim-neck by GSConv: a lightweight-design for real-time detector architectures. Journal of real-time image processing, 2024, 21362. J]

[21] Han K, Wang Y, Tan Q, et al. Ghostnet: more features from cheap operations. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, June 13–19, 2020, Seattle, WA, USA, 2020, New York: IEEE, 1580-1589[C]

[22] Tang Y, Han K, Guo J, et al. GhostNetv2: enhance cheap operation with long-range attention. Advances in neural information processing systems, 2022, 35: 9969-9982. J]

[23] LIU Z, HAO Z, HAN K, et al. Ghostnetv3: exploring the training strategies for compact models[EB/OL]. (2024-04-17) [2026-01-23]. https://arxiv.org/abs/2404.11202.

Rights & permissions

Tianjin University of Technology