Singapore Institute of Manufacturing Technology, Agency for Science, Technology and Research, 2 Fusionopolis Way, Singapore 138634, Singapore.
Critical Analytics for Manufacturing Personalized-Medicine, Singapore-MIT Alliance for Research and Technology Centre, 1 Create Way, Singapore 138602, Singapore.
Sensors (Basel). 2021 Jul 24;21(15):5034. doi: 10.3390/s21155034.
Hyperspectral imaging (HSI) provides additional information compared to regular color imaging, making it valuable in areas such as biomedicine, materials inspection and food safety. However, HSI is challenging because of the large amount of data and long measurement times involved. Compressed sensing (CS) approaches to HSI address this, albeit subject to tradeoffs between image reconstruction accuracy, time and generalizability to different types of scenes. Here, we develop improved CS approaches for HSI, based on parallelized multitrack acquisition of multiple spectra per shot. The multitrack architecture can be paired up with either of the two compatible CS algorithms developed here: (1) a sparse recovery algorithm based on block compressed sensing and (2) an adaptive CS algorithm based on sampling in the wavelet domain. As a result, the measurement speed can be drastically increased while maintaining reconstruction speed and accuracy. The methods were validated computationally both in noiseless as well as noisy simulated measurements. Multitrack adaptive CS has a ∼10 times shorter measurement plus reconstruction time as compared to full sampling HSI without compromising reconstruction accuracy across the sample images tested. Multitrack non-adaptive CS (sparse recovery) is most robust against Poisson noise at the expense of longer reconstruction times.
高光谱成像 (HSI) 提供了比常规彩色成像更多的信息，因此在生物医学、材料检查和食品安全等领域具有重要价值。然而，由于涉及大量数据和长时间的测量，HSI 具有挑战性。压缩感知 (CS) 方法可以解决这个问题，但需要在图像重建准确性、时间和对不同类型场景的泛化能力之间进行权衡。在这里，我们基于每拍摄一次获取多个光谱的并行多轨迹采集，开发了用于 HSI 的改进 CS 方法。多轨迹架构可以与我们在这里开发的两种兼容 CS 算法中的任意一种相结合：(1)基于块压缩感知的稀疏恢复算法，以及 (2)基于小波域采样的自适应 CS 算法。结果，在保持重建速度和准确性的同时，可以大大提高测量速度。这些方法在无噪声和有噪声的模拟测量中进行了计算验证。与没有牺牲重建准确性的全采样 HSI 相比，多轨迹自适应 CS 的测量加重建时间约缩短了 10 倍。多轨迹非自适应 CS（稀疏恢复）在重建时间较长的情况下对泊松噪声最具鲁棒性。
