Speckle patterns have been widely employed as a method for precisely determining the wavelength of monochromatic light. In order to achieve higher wavelength precision, a variety of optical diffusing waveguides have been investigated with a focus on their wavelength sensitivity. However, it has been a challenge to find a balance among the cost, compactness, precision, and stability of the waveguide. In this work, we designed a compact cylindrical random scattering waveguide (CRSW) as the light diffuser by mixing TiO2 particles and ultra-violate adhesive. In the CRSW, speckle patterns are generated by input light scattering off TiO2 particles multiple times. Additionally, a thin layer of upconversion nanoparticles (UCNPs) was sprayed on the end face of CRSW to allow near-infrared (NIR) light to be converted to visible light, breaking the imaging limitations of visible cameras in the NIR range. We, then, further designed a convolution neural network (CNN) to recognize the wavelength of the speckle patterns with excellent robustness and ability to transfer learning. This resulted in the achievement of a high wavelength precision of 20 kHz (∼0.16 fm) at around 1550 nm with a temperature resistance of ±2 °C. Our results demonstrate a low-cost, compact, and simple NIR wavemeter, which is capable of ultra-high wavelength precision and good temperature stability. It has significant value for applications in high-speed and high-precision laser wavelength measurements.