High-resolution infrared speckle wavemeter under weak illumination via sum-frequency generation in a thin-film...

The computational infrared speckle wavemeter features a compact structure and high resolution but typically relies on InGaAs cameras for speckle pattern recording. Compared to silicon-based cameras, InGaAs cameras are significantly more expensive and less sensitive, limiting their practical applications. While second-harmonic generation (SHG) can shift infrared light into a wavelength range detectable by silicon-based cameras, its conversion efficiency drops sharply at low input power, resulting in weak signals and reduced detection performance. To overcome these limitations, we propose a wavemeter scheme based on sum-frequency generation (SFG) in a thin-film lithium niobate waveguide. This approach enables efficient infrared-to-visible conversion, enhancing detection sensitivity under weak-light conditions. At an input power of 1.0 mW, spectral line reconstruction with a 5 pm resolution is achieved using the transmission matrix method. Furthermore, by incorporating a ResNet-50 neural network, the system achieves a recognition accuracy exceeding 96.5% even at an input power as low as 0.5 mW, with 1 pm wavelength spacing. The proposed method offers a low-cost, high-sensitivity, and compact solution for near-infrared weak-light detection, expanding the applicability of speckle-based computational wavemeters in practical scenarios.