US 11,810,310 B2
Satellite image processing method, network training method, related devices and electronic device
Dongliang He, Beijing (CN); Henan Zhang, Beijing (CN); and Hao Sun, Beijing (CN)
Assigned to Beijing Baidu Netcom Science Technology Co., Ltd., Beijing (CN)
Filed by Beijing Baidu Netcom Science and Technology Co., Ltd., Beijing (CN)
Filed on Jun. 1, 2021, as Appl. No. 17/335,647.
Claims priority of application No. 202011478534.7 (CN), filed on Dec. 15, 2020.
Prior Publication US 2021/0295546 A1, Sep. 23, 2021
Int. Cl. G06T 7/55 (2017.01); G06N 3/08 (2023.01); G06N 3/045 (2023.01)
CPC G06T 7/55 (2017.01) [G06N 3/045 (2023.01); G06N 3/08 (2013.01); G06T 2207/10032 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30181 (2013.01)] 12 Claims
OG exemplary drawing
 
1. A network training method, comprising:
acquiring a training image set, the training image set comprising a training input satellite image and a training output satellite image corresponding to the training input satellite image, an image parameter of the training input satellite image being different from an image parameter of the training output satellite image, the image parameter comprising image transparency or an image quality parameter;
performing feature extraction on the training input satellite image through a target neural network to acquire a first feature, fusing the first feature with a second feature to acquire a target feature, and performing image reconstruction in accordance with the target feature and the training input satellite image to acquire a second target satellite image, the second feature being a feature acquired through reconstructing a feature extracted from the first feature, the target neural network comprising a first neural network or a second neural network;
determining difference information between the second target satellite image and the training output satellite image; and
updating a parameter of the target neural network in accordance with the difference information.