Snapshot spectral imaging technology can obtain the target's two-dimensional (2D) spatial information and one-dimensional (1D) spectral information within a single exposure. Compared with the scanning spectral imaging technology, it reduces the requirement on the stability of the platform and can capture the temporal-spatial-spectral datacube of dynamic targets. Therefore, it has broad application prospects in biological imaging, medical diagnosis, food production monitoring, dangerous substance leakage warning, and dynamic target monitoring. This paper summarizes the data acquisition schemes of typical snapshot spectral imaging techniques, which are categorized into direct and indirect measurement. The direct measurement transfers the multidimensional radiation to the grayscale response directly and builds the one-to-one correspondence between the datacube voxels and detector pixels. Therefore, the number of datacube voxels should be less than the number of detector pixels. The indirect measurement modulates the multidimensional radiation and measures the coupled spatial-spectral information. The datacube voxels have to be calculated based on the indirect measurements. The number of datacube voxels is no longer limited by the number of detector pixels. However, the reconstruction algorithms have high computation costs and render limited performance in real scenarios. There are two main development trends of snapshot spectral imaging technology. On the one hand, extending the dimension of detection can provide a more comprehensive analysis of the target. For example, the snapshot spectral volumetric imaging technology can detect the volumetric target in real time and obtain four-dimensional (4D) data, including three-dimensional (3D) spatial and 1D spectral information. There are two main methodologies to realize snapshot spectral volumetric imaging. One methodology integrates a spectral imaging system with a volumetric imaging system. The two imaging systems work independently and their detection data are merged by post-processing. Common volumetric imaging modalities can be divided into active systems and passive systems. The active systems use auxiliary light sources and are suitable for indoor scenarios, such as structured light imager, laser scanning imager, and Time-of-Flight (ToF) imager. The passive systems, such as stereo imager and light field imager, utilize the parallax to estimate depth, which are less sensitive to ambient light. The other method to perform snapshot spectral volumetric imaging is to modulate the spectral-volumetric data and capture the coupled spatial-spectral measurement in a single exposure. Compared with the independent measurement scheme, the coupled measurement scheme has an advantage in system compactness and robustness. However, with the detection dimension increasing, the resolution of the reconstructed datacube acquired by the direct measurement is limited by the resolution of the detector. Therefore, the indirect measurement of spectral volumetric information is becoming a research hotspot. On the other hand, the snapshot spectral imaging technologies based on direct measurement sacrifice the spatial resolution for the spectral resolution. To improve target detection ability , a lot of algorithms are proposed to enrich the spatial details, which are called spectral image Super-Resolution ( SR). There are two solutions to perform spectral image SR. One solution is to capture an additional RGB/ panchromatic image with a higher spatial resolution of the same target at the same time, such as using a beam splitter. Then the RGB/panchromatic measurements are fused with the Low-Resolution (LR) spectral image to generate the High-Resolution ( HR) spectral image. The fusion method requires additional hardware implementation and the result is sensitive to the spatial alignment error. On the contrary, the single spectral image SR uses the LR spectral image as the input, and the HR spectral image is obtained with no need of auxiliary HR images. Benefiting from the development of deep learning in single image SR, the methods of distributing channel attention in both spatial and spectral dimensions are exploited to increase spatial resolution and keep spectral fidelity at the same time. The spectral image SR networks are expected to directly learn the end-to-end mapping relationship between LR and HR spectral images and have great prospects in the future. In summary, for snapshot spectral imaging technology, the development of technical principles has cutting-edge research significance, and the development of data processing technologies has practical significance for promoting its application in real scenarios.
