Comprehensive remote sensing monitoring requires high sensitivity and dynamic range of the sensor, so a spectral imaging method based on a pixel by pixel gain switching is proposed. Unlike from the frame gain or row column gain switching imaging methods, this method can optimize the gain at pixel level by using the nondestructive readout characteristics of the 4T-APS CMOS detector. This method can take into account the imaging needs of water, vegetation, cloud and other factors and greatly improve the development efficiency of the payload. The basic principle is that the detector firstly carries out global exposure and then according to the saturation judgment of the multi-stage integral capacitance, selects the unsaturated highest gain signal to be transmitted down in the form of gain code plus signal. The real radiation value of the pixel is derived from the calibration coefficient of the gain code. Due to the multi spectral segment and piecewise response, it is important to establish the multi-gain spectral imaging model and analyze the noise to ensure the system's quantitative application. Based on the analysis of noise types, the Poisson Gaussian noise model with multi-gain is established. Based on the model, the probability of low gain readout is calculated. The results show that although the noise will affect the change of the readout gain, the influence range is very small. When the radiance is within 5 mW·cm^-2·μm^-1·sr^-1, the signal is less 0.05 mW·cm^-2·μm^-1·sr^-1 than the value of the gain gear, so the probability of normal readout is greater than 99.6%. With the enhancement of the signal, the photon noise increases, the gain decreases, and the influence range expand. According to the multi-gain SNR model, the changes of SNR in spectral mode and combined channel mode are analyzed. Finally, the push broom spectral imaging simulation of four gains is carried out using the wideband imaging spectrometer (WIS) data as the entrance pupil radiance, and the inherent characteristics of the multi-gain spectral image are analyzed. Based on the noise model, add 1~3σ random noise to the spectral image with center wavelength of 0.443 μm, the influence of the noise on the gain of ground objects is analyzed. The results show that on the premise of meeting the SNR index, the single gain dynamic range of the system is 74 dB, and the total dynamic range is 114 dB. This method improves the SNR of weak signals such as water and ensures the unsaturation of bright targets such as buildings and clouds. Imaging simulation and noise analysis are conducive to the subsequent development of the sensors and provide a reference for the design of similar spectral instruments.