Method and apparatus for reconstructing hyperspectral image using artificial intelligence
View Patent ↗A method and an apparatus for reconstructing a hyperspectral image using artificial intelligence are provided. The method includes receiving an encoded sensor input for an image and reconstructing a hyperspectral image of the image for the encoded sensor input based on a previously generated nonlinear learning model.
1. A method for reconstructing a hyperspectral image, the method comprising:
generating a nonlinear learning model through learning using a predetermined hyperspectral image dataset;
receiving coded data for an image; and
reconstructing a hyperspectral image of the image for the coded data based on the generated nonlinear learning model and a nonlinear optimization technique,
wherein the nonlinear optimization technique jointly regularizes fidelity of nonlinear spectral representations for the generated nonlinear learning model and sparsity of gradients in a spatial domain.
2. The method of claim 1 , wherein the generating of the nonlinear learning module comprises:
generating the nonlinear learning model by training a convolutional autoencoder through the learning using the hyperspectral image dataset.
3. The method of claim 2 , wherein the generating of the nonlinear learning module comprises:
training an encoder network in a nonlinear space using the convolutional autoencoder; and
generating the nonlinear learning model using a decoder network.
4. The method of claim 3 , wherein the convolutional autoencoder comprises:
an encoder network for transforming the hyperspectral image dataset into nonlinear representations; and
a decoder network for generating an original dataset from the nonlinear representations.
5. The method of claim 1 , wherein the nonlinear learning model is a learning model for outputting nonlinear reconstruction of the hyperspectral image.
6. The method of claim 1 , wherein the
nonlinear optimization technique is iteratively performed using an alternating direction method of multipliers (ADMM).
7. The method of claim 1 , wherein the receiving comprises:
receiving the coded data for the image using compressive hyperspectral imaging.
8. A method for reconstructing a hyperspectral image, the method comprising:
training a spectral prior using a predetermined hyperspectral image dataset;
receiving coded data for an image; and
reconstructing a hyperspectral image of the image for the coded data based on the trained spectral prior and a nonlinear optimization technique,
wherein the nonlinear optimization technique jointly regularizes fidelity of nonlinear spectral representations for the trained spectral prior and sparsity of gradients in a spatial domain.
9. The method of claim 8 , wherein
the nonlinear optimization technique is iteratively performed using an alternating direction method of multipliers (ADMM).
10. An apparatus for reconstructing a hyperspectral image, the apparatus comprising:
a generation unit configured to generate a nonlinear learning model through learning using a predetermined hyperspectral image dataset;
a reception unit configured to receive coded data for an image; and
a reconstruction unit configured to reconstruct a hyperspectral image of the image for the coded data based on the generated learning model and a nonlinear optimization technique,
wherein the nonlinear optimization technique jointly regularizes fidelity of nonlinear spectral representations for the generated nonlinear learning model and sparsity of gradients in a spatial domain.
11. The apparatus of claim 10 , wherein the generation unit is configured to:
generate the nonlinear learning model by training a convolutional autoencoder through the learning using the hyperspectral image dataset.
12. The apparatus of claim 11 , wherein the generation unit is configured to:
train an encoder network in a nonlinear space using the convolutional autoencoder; and
generate the nonlinear learning model using a decoder network.
13. The apparatus of claim 12 , wherein the convolutional autoencoder comprises:
an encoder network for transforming the hyperspectral image dataset into nonlinear representations; and
a decoder network for generating an original dataset from the nonlinear representations.
14. The apparatus of claim 10 , wherein the nonlinear learning model is a learning model for outputting nonlinear reconstruction of the hyperspectral image.