Understanding the distribution of the spectral radiation field of laser-induced plasma at multiple wavelengths will be helpful to acquire the distribution of the temperature field, electron density and other physical quantities of the plasma, which is significant for improving the laser welding process. Making clear the physical process of how spectral radiation transfers to detector target is the premise of reconstructing spectral radiation field of plasma. In this paper, a spectral radiation transmission model of plasma in laser welding based on optical imaging is proposed: By studying the process that plasma spectral radiation emitting to the lens of the detector based on radiometry, the coefficient ω1 that the spectral radiation received by the lens taking up that emitting to the surrounding space of a point in the plasma is deduced so that the spectral radiation transfer model is established; by studying the imaging process that spectral radiation emits from plasma through the lens to the detector target based on the principle of optical imaging, the coefficient ω2 that the spectral radiation received by a pixel taking up that received by the lens is deduced so that an optical imaging model is established. Then the spectral radiation ratio coefficient ω is obtained with ω1 and ω2, thus the spectral radiation transmission model is established. Through simulation experiments, the effect of spectral radiation field reconstruction of laser welding plasma with symmetrical and asymmetrical temperature distribution between 1 000 and 3 000 ℃ was investigated. The results showed that the reconstruction accuracy with the model presented is 5.39%, 6.5%, 7.25% and 6.11% for the plasma with symmetrical temperature distribution at 400, 500, 600 and 700 nm respectively, and 9.34%, 10.07%, 10.68% and 9.72% for that with asymmetrical temperature distribution, which means the proposed model can be used for reconstructing the spectral radiation field of laser welding plasma at 400～700 nm well. Compared with other models, this model has better reconstruction accuracy and can meet the needs of industrial applications.