Billet Number Recognition Based on ESRGAN and Improved YOLOv5

Abstract

Aiming at solving the problem of billet number recognition and further improve the rate of recognition, this paper proposes a billet number recognition algorithm based on ESRGAN and improved YOLOv5, which is referred to as YOLOv5-Billet in this paper. According to the actual situation, we propose a target recognition algorithm that uses the lightweight network MobileNetv3 to replace the backbone feature extraction network of YOLOv5. We also introduce ESRGAN to improve the quality of the input image and further improve the recognition effect. A series of rigorous experiments show that YOLOv5-Billet has a certain improvement in billet recognition speed and precision. Specifically, the average precision of YOLOv5-Billet is improved by 10.8%, and the detection speed is also improved by 5.12fps, reaching 40.35fps. Through the experimental verification of real-time collection of data sets, compared with a variety of classical target detection methods, the recognition precision and detection speed are improved to varying degrees. This model maintains the lightweight characteristics of YOLOv5 and meets the requirements of real-time monitoring of the number of blanks.