Minor discrepancies in the measurement data may lead to significant variations in the reconstructed spectrum when measuring and reconstructing the spectrum by the absorption method. In some cases, the reconstructed spectrum may contain negative values that do not conform to the physical law. To address the issue that noise in the measurement data has a significant effect on the reconstructed spectrum, the segment-smoothed Bayesian iterative method is proposed in this paper. The energy spectrum is reconstructed using the Bayesian iteration method under different levels of noise, and the accuracy and noise sensitivity of the reconstructed spectrum are evaluated. An optimization method, which adds smoothing constraints to the Bayesian iterative method, is proposed to reduce noise interference in the reconstructed spectrum. According to the spectrum characteristics, a two-coefficient segment-smoothing method is proposed with the peak value as the dividing line. The spectrum is reconstructed by segmented smoothing and global smoothing Bayesian iterative methods, respectively. The noise sensitivity of the reconstructed spectrum, unfolded by the segment-smoothed Bayesian iterative method, has been significantly reduced. An energy spectrometer based on the absorption method is developed. The spectrum is reconstructed based on the experimental data using the segment-smoothed Bayesian iterative method and the Bayesian iteration method. The spectrum reconstructed using the segment-smoothed Bayesian iterative method is more consistent with the theoretical spectrum, indicating that this method exhibits superior performance.