As spectrum-based detection methods have advantages of fastness and nondestructiveness, they have been widely applied in the component analysis. The spectral analysis model is a key factor affecting the detecting accuracy. The property difference of spectrometers leads to an accuracy decrease of the analysis model for different spectrometer data, which severely restricts the development of spectral analysis technology. The model transfer is one of the effective methods to solve this problem. However, the spectral data have characteristics of high-dimensionality and nonlinearity. Besides, there are nonlinear differences between the spectra of different spectrometers, which brings challenges to model transfer. To tackle this issue, a new model transfer strategy based on support vector regression (SVR) parameter calibrating has been proposed in this study. First, a shallow 1-D convolutional neural network (CNN)-SVR model is established to provide an optimal calibration model of the master spectrometer for the model transfer strategy. Then, the CNN-SVR model is transferred to the slave spectrometer through model-based transfer learning. Finally, to improve the prediction accuracy when there are nonlinear differences between the spectra of the master and slave spectrometers, the sample spectra of the slave spectrometers are used to calibrate the SVR parameters. According to the experimental results, our proposed model transfer strategy is superior to other model transfer strategies when encountering nonlinear differences between the spectra of the master and slave spectrometers. Moreover, it provides a new idea for calibration models sharing between different spectrometers.