With the growing use of computational spectral imaging in fields such as biomedical imaging, environmental monitoring, and remote sensing, there is an increasing demand for efficient, compact, and tunable snapshot spectral imaging systems. However, traditional diffraction optical elements (DOEs) are limited by manufacturing precision and diffraction efficiency, leading to significant errors in practical applications that affect both imaging quality and spectral reconstruction accuracy. In contrast, using liquid crystal on silicon spatial light modulators (LCoS-SLMs) to dynamically simulate the phase modulation characteristics of DOEs offers a flexible and efficient solution for snapshot spectral imaging. LCoS-SLMs not only provide higher quantization accuracy, overcoming the phase precision limitations of traditional DOEs, but also improve diffraction efficiency and spectral reconstruction by loading wavelength-specific phase modulation patterns. This approach effectively mitigates errors from DOE manufacturing and assembly, enhancing the adaptability and stability of the spectral imaging system and paving the way for the development of high-performance snapshot spectral imaging.

Although LCoS-SLMs have found widespread use in areas such as super-resolution imaging, light-field imaging, and computational holography, their potential in snapshot spectral imaging remains underexplored. To address the challenges of manufacturing precision and reduced diffraction efficiency in traditional DOEs, we propose a dynamic phase modulation scheme using LCoS-SLMs for efficient snapshot spectral imaging. This approach takes advantage of the high quantization accuracy (up to 256 levels) and real-time programmability of LCoS-SLMs to replicate the phase modulation function of DOEs. By applying optimized wavelength-specific phase patterns, we enhance spectral separation and improve reconstruction accuracy. With its high diffraction efficiency and dynamic adjustability, the developed imaging system achieves high-precision spectral reconstruction across the entire visible spectrum (400-700 nm). In comparison to traditional DOE systems, the peak signal-to-noise ratio (PSNR) of the reconstruction improves by 4.3 dB, and the structural similarity index (SSIM) rises to 0.98. Furthermore, the system supports flexible adjustments of focal length and field of view without the need for additional optical components, effectively reducing system complexity and cost. Relevant research results were recently published in Photonics Research, Volume 13, Issue 2, 2025. [Chong Zhang, Xianglei Liu, Lizhi Wang, Shining Ma, Yuanjin Zheng, Yue Liu, Hua Huang, Yongtian Wang, Weitao Song, "Lensless efficient snapshot hyperspectral imaging using dynamic phase modulation," Photonics Res. 13, 511 (2025)]

The schematic of the LESHI system is illustrated in Fig. 1. A light source illuminates the object, and the reflected light passes through a polarizer, is reflected by a beam splitter, and then interacts with the LCoS-SLM, which is loaded with optimized DOE patterns. The liquid crystal layer of the LCoS-SLM, with varying refractive indices for different wavelengths, introduces wavelength-dependent phase delays, functioning similarly to a DOE, thereby enabling the splitting of the continuous hyperspectral data cube. As the light wave passes through the liquid crystal layer, the modulation of each pixel alters the phase of the light wave. Finally, the phase-modulated light is reflected from the LCoS-SLM, passes through the beam splitter, and is recorded by a color CMOS camera.

Fig. 1. Schematic of the lensless efficient snapshot hyperspectral imaging (LESHI) system.

The working principle of LESHI is illustrated in Fig. 2a. In the forward propagation of the model, LESHI first compresses the spectral dataset into a 3-channel RGB snapshot, then reconstructs the 31-channel spectral cube from the snapshot, and finally computes the loss function between the reconstructed image and the ground truth. During backward propagation, the model optimizes its variables (e.g., the phase modulation pattern for each pixel and the neural network parameters) by minimizing the loss function using gradient descent methods.

Figure 2b illustrates the process of acquiring the point spread function (PSF) for diffraction optical imaging based on DOE patterns using LCoS-SLM. The Distributed Diffractive Optics (DDO) model employs spatiotemporal multiplexing techniques by sequentially loading multiple DOE patterns, enabling distributed imaging of the same scene across different spectral bands. In this system, multiple simulated DOE patterns are loaded onto the LCoS-SLM to achieve the DDO model in batches. As shown in Fig. 2c, three simulated DOE patterns capture the scene in the 400-500 nm, 500-600 nm, and 600-700 nm spectral ranges. Figure 2d presents the structure of the reconstruction network, ResU-Net, used for image reconstruction.

Fig. 2. Working principle of LESHI. (a) Pipeline of LESHI. (b) Schematic of PSF acquisition process in diffractive optical imaging based on LCoS-SLM with DOE patterns. (c) DDO model design based on LCoS-SLM. DDO fuses the PSFs of individual DOEs of the different bands and adds the model of the diffraction efficiency to form a degenerate PSF model. (d) Structure of the ResU-net reconstruction algorithm.

In the future, the system will undergo comprehensive optimization by incorporating the required scene object information into the model training process to enhance the model's generalization capability. Additionally, deep unfolding networks and a plug-and-play mechanism will be explored to increase the network's flexibility in handling spectral cubes of varying sizes. Ultimately, the entire network model will be miniaturized through the optimization of network parameters, enabling the trained model to be loaded onto FPGA hardware instead of a GPU, which will significantly improve spectral reconstruction speed.