Snapshot hyperspectral imaging using wide dilation networks

Abstract

Hyperspectral (HS) cameras record the spectrum at multiple wavelengths for each pixel in an image, and are used, e.g., for quality control and agricultural remote sensing. We introduce a fast, cost-efficient and mobile method of taking HS images using a regular digital camera equipped with a passive diffraction grating filter, using machine learning for constructing the HS image. The grating distorts the image by effectively mapping the spectral information into spatial dislocations, which we convert into a HS image by a convolutional neural network utilizing novel wide dilation convolutions that accurately model optical properties of diffraction. We demonstrate high-quality HS reconstruction using a model trained on only 271 pairs of diffraction grating and ground truth HS images.

Detection of cutaneous oxygen saturation using a novel snapshot hyperspectral camera: a feasibility study

“…To assess the imaging characteristics of the hyperspectral c a m e r a , i t s o u t p u t w a s c o m p a r e d w i t h c o n t a c t measurements obtained using a fiber optic probe and multi-diameter reflectance spectroscopy (17,18). During each measurement, the reflectance spectra are collected at effective diameters of 200, 600, and 1,000 μm; StO 2 , and blood volume fraction, was determined using the largest effective single fiber diameter with an interrogation depth of approximately 500 μm (19). The pathlength of light in the interrogated volume for the point spectroscopy is fixed by the fiber (equal to approximately ½ the fiber diameter).…”

“…Furthermore, a correction for the second-order light effect due to light diffraction ( 21), provided by the manufacturer, was performed after the normalization of images. Lastly, in order to calculate the blood StO 2 at each pixel in the hyperspectral images, all spectral bands (N=41) were selected in the full range of 470-950 nm to enable fitting using a method described previously, although implemented for hyperspectral imaging (19,22,23).…”

“…As previously described, reflectance spectra were analyzed using an analytical model developed to characterize the wavelength-dependent optical properties in order to extract physiological and morphological characteristics from the tissue of interest (19). Briefly, attenuation caused by tissue absorption, is modeled using a modified Beer-Lambert law and is a function of both the tissue absorption coefficient (μa) and the single fiber photon path length.…”

“…Minimizing the chi-squared metric between the measured reflectance data and predictions of our model was achieved by using a Levenberg-Marquardt algorithm in order to estimate the parameter values for the microvascular hemoglobin StO 2 , the blood volume fraction and average blood vessel diameter (19). To estimate the confidence intervals of the parameters, the square root of the diagonal covariance matrix was used (9).…”

Visible Light Spectrum Extraction from Diffraction Images by Deconvolution and the Cepstrum

“…Even though some methods attempt to handle this by using global [17] or local [30] image structure, we prefer using field stops, similar to [19], to eliminate the overlaps. By limiting the scene to what is visible through the field stop, we can enforce all higher-order diffraction components to be outside of the area covered by the zeroth-order components; see Figure 2 for an illustration of the images used in our setup.…”