SYMBOL RECOGNITION AND APPLICATION USING MACHINE LEARNING

Abstract

In the ongoing wave of digitalization, with the advent of the intelligence era, the automation and intelligent detection of symbols have gained significant importance in traffic management and industrial production. This study focuses on symbol detection in two critical domains: traffic sign recognition and industrial symbol detection on iron ladle bodies. Manual detection methods are no longer sufficient to meet the demand for real-time, high-accuracy information processing, necessitating the use of intelligent technologies like machine learning to enable automation. In intelligent transportation systems, rapid and accurate identification of traffic signs improves driver assistance systems, especially under adverse weather or lighting conditions, enhancing road safety. In industrial environments, detecting symbols on iron ladle bodies ensures production safety and efficiency, reducing manual operation costs. This study proposes machine learning-based recognition methods tailored to both scenarios. A high-precision traffic sign detection algorithm based on Faster R-CNN is developed. The gamma transformation is applied to improve feature expression under uneven lighting. The model incorporates a Convolutional Block Attention Module to address network depth issues, enhances shallow feature acquisition, and reduces parameters. A Feature Pyramid Network improves the detection of signs of different sizes. The model achieves 99.79% mAP on GTSDB (↑10.35%) and 87.62% on CCTSDB2021 (31.04%). For industrial symbol detection, a lightweight YOLOv8-based model is proposed for resource-constrained environments. Using PP-HGNetV2 and Ghost-HGNet modules, it reduces parameter count while maintaining performance. BiFPN is introduced for efficient multi-scale feature fusion. On the Iron Ladle dataset, the model achieves 97.57% mAP with 1.29M parameters, 1.72M fewer than the original, with only a 0.27% performance drop. This work demonstrates the effectiveness of machine learning in enabling accurate, efficient symbol detection in complex, real-time traffic and industrial applications.