MicroCharNet: An Ultra-Lightweight Model for License Plate Character Detection

Forum topic · 小凯 · 2026-07-15

Summary

Researchers Huy Che, Dinh-Duy Phan, and Duc-Lung Vu propose MicroCharNet, an ultra-lightweight neural network for license plate character detection in intelligent transportation systems. The architecture combines a compact backbone of C2f blocks with CoordAtt modules to enhance feature extraction while preserving spatial information, a lightweight C3k2-based neck that fuses multi-level features, and a single-stage anchor-free detection head for end-to-end prediction. Experiments on the UFPR-ALPR dataset show MicroCharNet achieves competitive detection accuracy with only 0.08M parameters and 0.096 GFLOPs, outperforming several recent YOLO baselines while being dramatically smaller. Hardware-level evaluations confirm efficient real-time deployment on edge devices, demonstrating that carefully designed ultra-lightweight architectures can effectively balance accuracy and computational efficiency for resource-constrained license plate recognition.

Paper Overview

Field: Computer Vision (CV) Authors: Huy Che, Dinh-Duy Phan, Duc-Lung Vu Published: 2026-07-13 arXiv: 2607.11830

Abstract

License plate character detection is a crucial component of intelligent transportation systems, where high accuracy and computational efficiency are required for real-time deployment. Although recent deep learning-based methods have substantially improved detection performance, many high-accuracy models rely on large-scale architectures that incur substantial computational overhead, limiting their applicability to resource-constrained devices.

This paper proposes MicroCharNet, an ultra-lightweight model specifically designed for license plate character detection. Key architectural components:

Compact backbone composed of C2f blocks, integrated with a CoordAtt module to enhance feature extraction while preserving spatial information
Lightweight C3k2-based neck that fuses multi-level features
Single-stage anchor-free detection head for end-to-end prediction

Results

On the UFPR-ALPR dataset, MicroCharNet achieves competitive detection accuracy with only 0.08M parameters and 0.096 GFLOPs
It outperforms multiple recent YOLO baselines despite its tiny size
Hardware-level evaluation confirms its efficiency for real-time deployment on edge devices

These results demonstrate that carefully designed ultra-lightweight architectures can effectively balance accuracy and efficiency in license plate character detection — truly a case of "less is more."