Traditional methods for improving the quality of low-illumination imaging include external supplementary light and expanding the aperture, and complete source information enhancement by increasing the physical light input. However, these methods have been proposed to cause problems, such as light source pollution and shortening of the depth of field. Therefore, this paper proposes an end-to-end multispectral fusion scheme to achieve high-quality computational imaging, thereby effectively restoring the color and details of objects in low-illumination scenes while making up for the shortcomings of traditional methods. First, the multi-channel spectral information was integrated through a custom-designed deep learning network, after which scene noise was effectively eliminated. As a result, we discovered that the proposed method had a high degree of freedom, could adjust number of channels and network parameters according to the needs of specific conditions, replaced the traditional camera module, and optimized the image signal processing process. Next, we conducted detailed ablation experiments. The results show that after spectral fusion, the mean square error (MSE) and perceptual loss of image quality obtained by the proposed method reduce by 54.43% and 35.12%, respectively, compared with that obtained by traditional methods based on RGB data. Thus, the successful proposal and verification of this study's method should serve as a background for new high-quality imaging solutions within complex application scenarios, such as augmented reality/virtual reality (AR/VR), medical imaging, and autonomous driving.