To address the issue of baseline interference in the subsequent processing of spectral signals, a spectral signal region baseline fitting algorithm was proposed in this study. A certain number of smaller data points were extracted from the original spectrum, and an initial baseline was obtained through linear interpolation. In cases where there were overlapping peaks in the spectrum, the selected smaller data points may contain non-baseline outliers. Smoothing iterations were introduced, and a method for baseline determination based on the rate of change of the slope was proposed. This algorithm was compared with different baseline fitting methods on simulated and actual spectra. The results show that, in the process of fitting simulated baselines, a relative standard deviation of 8.25% is obtained with this algorithm, which is the lowest compared to that obtained with other methods. The correlation of the calibration curve based on this algorithm is the highest at 99.85%, and the smallest root mean square error in prediction is 0.5912. The baseline fitting algorithm proposed in this study demonstrates high accuracy and stability in various types of atomic emission spectra and can effectively estimate the continuous background of atomic emission spectra.