The multi-channel Raman imaging system is often affected by the nonlinear factors such as fluorescence background and noise, which reduces the Raman spectral reconstruction accuracy. Therefore, a reconstruction algorithm based on Gaussian kernel principal component analysis was proposed, in which the calibration samples are optimized by similarity factor; Then the calibration samples were mapped to high-dimensional space in a nonlinear form by using kernel function; The basis function was extracted from the mapped data set, and the basis function coefficients were obtained by pseudo-inverse method. Polymethyl methacrylate was used in the experiment and the Raman spectral reconstruction accuracy was evaluated in terms of relative root mean square error. The experimental results show that the proposed algorithm has higher reconstruction accuracy and anti-noise property than the traditional pseudo-inverse and wiener estimation methods. And the proposed algorithm can effectively reduce the impact of bad data and nonlinear factors in the calibration samples and imaging system. Therefore, the proposed algorithm can provide an effective Raman spectral reconstruction algorithm for multi-channel Raman imaging.