In recent years, multispectral filter array sensors have received increasing attention as a method capable of simultaneously capturing high-quality and alignment-free images in multiple bands during a single acquisition. An important development in this direction is the sensor that simultaneously captures short-wave infrared (NIR) and color (RGB) images, called RGB-NIR sensor. Cameras using RGB-NIR sensors can easily obtain aligned RGB and NIR images at the same time and can be used for a variety of optical applications. RGB-NIR sensors face more challenges in areas such as demosaicing than regular RGB sensor array (Bayer array) sensors. There are no mature commercial image signal processing solutions on the market specifically for RGB-NIR sensors. The number of channels and the arrangement law of RGB-NIR sensor filter arrays are different from traditional RGB sensor arrays, so how to demosaic RGB-NIR sensors is a key issue. Among the existing RGB-NIR array demosaicing algorithms, the residual interpolation-based algorithm is versatile and perform well, and the computational cost is low. However, the regularization parameter of the standard guide filtering architecture is fixed, which leads to easy artifacts at the edges or textures of the image. And the residual interpolation-based algorithm use the standard guide filtering architecture, so the above problems inevitably occur. To address the above situation, this paper proposes a weighted guide filtering demosaicing algorithm that introduces weighting coefficients. The proposed demosaicing algorithm first obtains the G-channel image using gradient based threshold free algorithm combined with residual interpolation algorithm. Then the images of R, B, and IR channels are interpolated by weighted guide filtering. The linear coefficients are no longer simply determined by using the average value, but the weighting coefficients are introduced and calculated by the weighted average method. Finally, residual interpolation is applied to obtain the final image. To validate the algorithm, it was tested on a dataset as well as on real devices, respectively. For the dataset experiment, the TokyoTech dataset, which is currently the most commonly used dataset for RGB-NIR multispectral imaging, was selected for testing; for the real device experiment, a specific RGB-NIR sensor (OV2744 sensor) was used to build the image acquisition system. Finally, the RGB-NIR sensor camera was also compared with the common RGB sensor camera to explore the application scenarios of the RGB-NIR sensor camera. For the test experiment on the dataset, PSNR (Peak Signal To Noise Ratio) is selected as the evaluation metrics. The PSNR and the average PSNR of the four algorithms for the dataset of 16 images shown that the algorithm proposed in this paper has the best performance. The proposed algorithm reduces the artifacts at the edges. For experiments on real devices, the NIQE (Natural Image Quality Evaluator) metrics was taken to evaluate the results of the algorithm. The experimental results are shown when both white light source and infrared light source are turned on. The interpolation effect of the algorithm proposed in this paper is the best, and the average NIQE values of 5.54 and 3.77 are the smallest for both color images and infrared images, respectively. The same scene was taken again with the IR light source turned off. Without infrared light, the algorithm of this paper also performs optimally on both color and infrared images, with the average NIQE values of 5.33 and 4.56, respectively. Moreover, the proposed algorithm also has better robustness and the quality of the obtained images is more stable. Finally, it can be seen in the experiment comparing with the ordinary RGB sensor that the images obtained by the RGB-NIR sensor can clearly see the contents of the shadow region and effectively remove the shadows. Aiming at the problems of existing demosaicing algorithms for RGB-NIR sensors, this paper proposes a residual interpolation demosaicing algorithm based on weighted guide filtering, and experiments are carried out on both publicly available dataset as well as practical device application. From the experimental results, it can be seen that the algorithm proposed in this paper has the best performance compared with other algorithms. Both in the dataset and in the real device application, there is a significant improvement in the metrics, and the artifacts at the edges can be better reduced in the subjective vision. Meanwhile, the proposed algorithm has better robustness and more stable image quality. Finally, the application of RGB-NIR sensors on the application scenario of shadow removal is also explored. From the result, it can be seen that the RGB-NIR sensor image acquisition system can provide a significant shadow removal effect.