Opt Lett. 2023 Mar 1;48(5):1156-1159. doi: 10.1364/OL.479622.
Computational hyperspectral cameras with broadband encoded filter arrays enable high precision spectrum reconstruction with only a few filters. However, these types of hyperspectral cameras have limited application, because it is difficult for conventional encoded filter arrays to balance among the spectrum regulation capacity, angle insensitivity, and processibility. This Letter presents a new, to the best of our knowledge, encoded filter composed of superposition Fabry-Perot resonance cavity (SFP) that can simultaneously take all three aspects into consideration. By learning the parameters of an SFP encoder and a neural network decoder in an end-to-end manner, a computational hyperspectral camera based on an SFP filter array presents up to 2.24 times higher spectral reconstruction accuracy, 10 times wider working angle, and can be produced with a low-cost manufacturing process.
带宽带编码滤光片阵列的计算光谱相机仅用几个滤光片就能实现高精度的光谱重建。然而,这些类型的光谱相机的应用有限,因为传统的编码滤光片阵列很难在光谱调节能力、角度不敏感性和可加工性之间取得平衡。本研究提出了一种新的编码滤光片,由超构法布里-珀罗共振腔(SFP)组成,可同时兼顾这三个方面。通过端到端的方式学习 SFP 编码器和神经网络解码器的参数,基于 SFP 滤光片阵列的计算光谱相机的光谱重建精度提高了 2.24 倍,工作角度扩大了 10 倍,并且可以采用低成本的制造工艺进行生产。