Beam spot quality detection of 650 nm laser diode using lightweight transformer model

Abstract

The 650 nm laser diode (LD) with high beam quality has superhigh application demand in fields such as information storage, medicine and display. The accurate detection of 650 nm LD beam spots is crucial for its production and application. In this paper, to detect and classify different 650 nm LD beam spots quickly and accurately, a beam spot detection network based on the vision transformer architecture is proposed for the first time. The presented model allows better extraction of beam spot features and is extremely hardware-friendly. Meanwhile, we designed a new Transformer Block, which can optimize local information acquisition, and enhance memory access efficiency. Moreover, the first 650 nm LD beam spot image dataset, RED-LaserSpot-4K, was created. The experimental results show that the model is both lightweight and powerful, with TOP-1 Accuracy of 93.2 %, only 1.0G FLOPs and 12.3 M parameters. It also can provide a practice foundation for enhancing the performance of 650 nm LD.