IP Library Granted Patent US 12,632,923
Granted Patent B2
US 12,632,923 · App. 17/925,737 · Granted May 19, 2026

Learning method, high resolution processing method, learning apparatus and computer program

Inventors: Yoko Sogabe (Musashino, JP); Shiori Sugimoto (Musashino, JP); Takayuki Kurozumi (Musashino, JP); Hideaki Kimata (Musashino, JP)
Assignee: NTT, Inc.
G06T3/4046G06T5/60G06T2207/10036G06T2207/20016G06T2207/20081G06T2207/20084
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,632,923
App. No.
17/925,737
Granted
May 19, 2026
Kind
B2
Abstract

An aspect of the present invention is a learning method for learning a model which reconstructs data and acquires reconstructed data, the learning method includes a plurality of stages, and the learning method includes first sub-processing which is based on an iterative calculation algorithm in which a Lagrange multiplier is not used, second sub-processing which is based on an iterative calculation algorithm in which the Lagrange multiplier is used, and weighting sub-processing which determines processing in each of the plurality of stages based on a weight for determining usage of the first sub-processing and the second sub-processing in each stage.

Claims (21)

1 . A learning method for learning a model which reconstructs data, the learning method comprising a plurality of stages, the learning method comprising:

acquiring, by an observation device, image data, where the image data is obtained by imaging an imaging target with compressed sensing;

first sub-processing of the image data, where the first sub-processing is based on a half quadratic splitting method;

second sub-processing of the image data, where the second sub-processing is based on an alternating direction method of multipliers;

weighting sub-processing of the image data, where the weighting sub-processing determines processing in each of the plurality of stages based on a weight for determining usage of the first sub-processing and the second sub-processing in each stage;

executing each stage of the plurality of stages according to the weighting sub-processing; and

generating a learned model based on parameters obtained from executing each stage in the plurality of stages, where the learned model outputs a hyperspectral image.

2 . The learning method according to claim 1 , further comprises performing processing at each layer of a deep neural network using training data, thereby obtaining the parameters and generating the learned model.

3 . A non-transitory computer-readable medium having computer-executable instructions that, upon execution of the instructions by a processor of a computer, cause a computer to execute the learning method according to claim 1 .

4 . The learning method according to claim 1 wherein the learned model uses an observation matrix as input data.

5 . The learning method according to claim 1 wherein the observation device includes an optical system configured to disperse light.

6 . The learning method according to claim 1 further comprises initially training the learned model using the first sub-processing of the image data and subsequently training the learned model using the second sub-processing of the image data.

7 . A learning apparatus for learning a model which reconstructs data, the learning apparatus having a plurality of stages and comprising:

a processor; and

a storage medium having computer program instructions stored thereon, when executed by the processor, perform, in each of the plurality of stages, to:

acquire image data, where the image data is obtained by imaging an imaging target with compressed sensing;

execute a first sub-processing of the image data, where the first sub-processing is based on a half quadratic splitting method;

execute a second sub-processing of the image data, where the second sub-processing is based on an alternating direction method of multipliers;

execute a weighting sub-processing which determines processing in each of the plurality of stages based on a weight for determining usage of the first sub-processing and the second sub-processing in each stage;

executing each stage of the plurality of stages according to the weighting sub-processing; and

generating a learned model based on parameters obtained from executing each stage in the plurality of stages, where the learned model outputs a hyperspectral image.

Assignments (2)
CHANGE OF NAME Recorded Oct 3, 2025
From: NIPPON TELEGRAPH AND TELEPHONE CORPORATION
To: NTT, INC.
Reel/Frame 072996/0032 →
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Nov 16, 2022
From: SOGABE, YOKO; SUGIMOTO, SHIORI; KUROZUMI, TAKAYUKI; KIMATA, HIDEAKI
To: NIPPON TELEGRAPH AND TELEPHONE CORPORATION
Reel/Frame 061795/0486 →
Continuity (1)
Related Publication 20230196512A1 · Jun 22, 2023
References Cited (10)
US 20210127101A1 · Roh · 2021 [cited by examiner]
US 20210350590A1 · Kim · 2021 [cited by examiner]
Diamond, Steven, et al. “Unrolled optimization with deep priors.” arXiv preprint arXiv:1705.08041v2 (2017). (Year: 2018). [cited by examiner]
Guo, Hongbo, Shi Liu, and Hongqi Guo. “Hybrid iterative reconstruction method for imaging problems in ECT.” IEEE Transactions on Instrumentation and Measurement 69.10 (2020): 8238-8249. (Year: 2020). [cited by examiner]
Kellman, Michael, et al. “Memory-efficient Learning for Large-scale Computational Imaging.” arXiv preprint arXiv:2003.05551v1 (2020). (Year: 2020). [cited by examiner]
Liu, Risheng, et al. “Proximal alternating direction network: A globally converged deep unrolling framework.” Proceedings of the AAAI Conference on Artificial Intelligence. vol. 32. No. 1. 2018. (Year: 2018). [cited by examiner]
Sun, Shiliang, et al. “A survey of optimization methods from a machine learning perspective.” IEEE transactions on cybernetics 50.8 (2019): 3668-3681. (Year: 2019). [cited by examiner]
Yang, Yan, et al. “ADMM-CSNet: A deep learning approach for image compressive sensing.” IEEE transactions on pattern analysis and machine intelligence 42.3 (2018): 521-538. (Year: 2018). [cited by examiner]
Lizhi Wang et al., “Hyperspectral Image Reconstruction Using a Deep Spatial-Spectral Prior”, In the IEEE conference on computer vision and pattern recognition (CVPR), pp. 8032-8041, IEEE, Jun. 2019. [cited by applicant]
Jian Zhang et al., “ISTA-Net: Interpretable Optimization-Inspired Deep Network for Image Compressive Sensing”, Computer vision foundation, pp. 1828-1837, 2018. [cited by applicant]