In order to solve the problems of peak shift, weak feature spectrum misjudgment, and artificial spectral line recognition in the wavelength calibration process of visible light spectroradiometers, the adaptive threshold method and the differential method were used to detect and recognize the feature spectrum. Meanwhile, the relationship between wavelength and pixel was determined using the method of least squares and Lagrange interpolation. Theoretical analysis and experimental verification were carried out and the optimum wavelength calibration equation was obtained. The results show that the feature spectrum is detected and recognized sequentially based on the intensity of the spectral lines. The optimal equation was obtained by the Lagrange interpolation method based on 404.6565 nm, 435.8335 nm, 546.0750 nm, and 579.0670 nm, with a residual standard deviation of only 3.2747×10-13 nm, the wavelength calibration error distribution is within the range of ±0.2 nm. Therefore, this method improves the efficiency of wavelength calibration. It also avoids noise misjudgment and spectral line recognition error and provides reference for wavelength calibration of other spectral instruments.