US 11,054,310 B2
Spectral sensor system employing a deep learning model for sensing light from arbitrary angles of incidence, and related hyperspectral imaging sensor
Xingze Wang, Durham, NC (US); Yibo Zhu, Redwood City, CA (US); and Xin Lei, San Carlos, CA (US)
Assigned to Coherent AI LLC, Redwood City, CA (US)
Filed by Coherent AI LLC, Redwood City, CA (US)
Filed on Sep. 5, 2019, as Appl. No. 16/562,324.
Prior Publication US 2021/0072081 A1, Mar. 11, 2021
Int. Cl. H04N 5/225 (2006.01); G01J 3/28 (2006.01); G06N 20/00 (2019.01)
CPC G01J 3/2803 (2013.01) [G06N 20/00 (2019.01); H04N 5/2254 (2013.01); H04N 5/2258 (2013.01); G01J 2003/283 (2013.01)] 22 Claims
OG exemplary drawing
1. A method implemented in a spectral sensing system, the spectral sensing system comprising an array of sampling optical elements, an array of optical sensors, and a machine learning model implemented by computing processors and memories, wherein each sampling optical element is configured to have optical properties that alter a reflectance, or a transmittance, or a spatial distribution of an incident light falling on the sampling optical element as a function of wavelength to produce an altered light, wherein the optical properties of different sampling optical elements are different, wherein each optical sensor is disposed to receive altered light from one of the sampling optical elements and to convert received light intensity to an electrical signal, and wherein the machine learning model is configured to perform computations on received model input to generate model output, the method comprising:
a process of using the machine learning model to measure an unknown spectrum, wherein the machine learning model is a trained machine learning model, the process including:
shining a target light wave having the unknown spectrum on the array of sampling optical elements at one or more unspecified angles of incidence;
recording the electrical signals output by the array of optical sensors in response to the target light wave being shone on the sampling optical elements, to obtain target sensor output;
inputting the target sensor output to the trained machine learning model; and
the trained machine learning model performing computation on the received model input to generate target model output, wherein the target model output represents the unknown spectrum.