Wavelength calibration is carried out before a spectrometer is used normally. The usual calibration process measures a standard light source of known wavelength, such as that of a mercury argon lamp. The spectral lines formed by the standard light source are divided to find the peak value; then, the peak value of the light source is matched with the wavelength of the standard light source. At present, the calibration of the spectrometer needs to be carried out with human intervention, which requires a lot of work, and the accuracy is not high. In our previous work, we proposed an efficient and accurate peak-finding method and a matching method. This paper will expand on this basis to illustrate a new intelligent and automated wavelength calibration method. In terms of peak searching, we discussed the implementation and defects of the maximum discrete cosine transform, Gaussian mixed model, and polynomial fitting methods; we also compared the maximum linear matching method with the maximum matching method and proposed the advantages of our method. Our experiments show that spectrum segmentation and peak-seeking methods based on spectral energy can effectively solve the problem of segmentation and peak seeking in wavelength calibration and is superior to other fitting methods. The matching method, combining the Hough transform and random sample consensus, can effectively solve the matching problem in wavelength calibration without manual intervention. In addition, the average accuracy and average recall rate exceed the traditional manual matching method and maximum linear matching method.