Citation

Rahman, Aviv Yuniar and Zakaria, Zuhaina and Lim, Siow Chun (2025) Hybrid Deep Learning for Text Detection in Power Grids: YOLOv8 & Fast R-CNN. In: 2025 Multimedia University Engineering Conference, MECON 2025, 21 July 2025 - 23 July 2025, Cyberjaya, Malaysia.

Text 5.pdf - Published Version Restricted to Repository staff only Download (1MB)

Abstract

Power distribution networks (PDNs) require accurate schematic interpretation for efficient operation and maintenance. Manual text detection is labor-intensive and error-prone, necessitating automated solutions. This paper presents a hybrid deep learning framework combining YOLOv8 and Fast R-CNN for text detection in PDN schematics. YOLOv8 performs rapid region proposals, while Fast R-CNN refines localization accuracy. To improve robustness, three ensemble strategies—Soft Voting, Hard Voting, and Weighted Average Voting—are employed. Evaluated on a real-world PLN dataset, the hybrid model consistently outperforms individual detectors in precision, recall, and mean average precision (mAP). Hard Voting achieves the best balance, with mAP@50 of 0.8950 and latency of 48.63 ms. The framework effectively handles occlusions, font variations, and cluttered backgrounds, surpassing OCR-based methods. It demonstrates strong potential for scalable schematic interpretation and automated grid monitoring. Future research will focus on edge deployment optimization via quantization and pruning, benchmarking against detectors such as CRAFT, DBNet, and EAST, and integration with advanced OCR engines. Additional evaluation across noisy schematics and detailed error analysis will be conducted to improve system robustness and post-processing reliability.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Deep learning
Subjects:	Q Science > Q Science (General) > Q300-390 Cybernetics
Divisions:	Faculty of Artificial Intelligence & Engineering (FAIE)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	18 Mar 2026 08:17
Last Modified:	19 Mar 2026 02:00
URII:	http://shdl.mmu.edu.my/id/eprint/15585

Downloads

Edit (login required)