Post-weld-shift compensation strategy with improved U-net edge detection for reducing coupling loss of 25Gbps TOSA

Post-weld-shift can lead to power attenuation in the coupling and packaging of opto-electronic devices. In this paper, we present a compensation method based on U-net deep learning architecture for post-weld-shift based on the image recognition of weld point features. The MPFG-net architecture, equipped with CG-MLP and MIF modules, markedly improves the image recognition performance for local weld point features. The experimental results indicate that the proposed net achieves an exceptional weld-point edge recognition accuracy of 78%, outperforming the leading mainstream networks. The compensation strategy was established based on the proposed MPFG-net, combined precise weld-point alignment with laser energy modulation, achieving a 15% improvement in optical power compensation.