An Enhanced Super-Resolution Generative Adversarial Network (ESRGAN) for Improving Image Quality on Construction Vehicle License Plates

In recent years, with the continuous advancement of artificial intelligence technology in the field of construction engineering, the use of computer vision methods to address issues in engineering management scenarios has become a major research focus. However, due to the complex environmental factors present in construction sites, the application of computer vision technology in this context is often affected to varying degrees. In this paper, we focus on common images in construction scenes that are affected by dust and exposure, which often have low target resolution, blur, and abnormal lighting distribution. Taking the task of license plate recognition for construction vehicles as an example, we propose a method based on the ESRGAN image super-resolution algorithm to improve the quality of license plate images and ultimately enhance license plate text recognition accuracy. We constructed a mixed dataset through on-site shooting and code synthesis for training the super-resolution model, and tested the model on a self-built license plate image test set. The accuracy of license plate text recognition in the verification experiment using the super-resolved license plate images reached 74.5%, which was a significant improvement compared to the 65% accuracy achieved with the original images. The test results show that the model can effectively improve the resolution of license plate images while addressing issues of blurriness and abnormal lighting distribution to some extent. The proposed method in this paper has a positive effect on downstream research tasks in engineering management based on computer vision and has significant research implications for enhancing the quality and efficiency of engineering management.