Three-dimensional spectrum situation completion method and device based on generative adversarial network

Patent metadata

Field	Value
Publication number	US-12430817-B2
Application number	US-202218032573-A
Country	US
Kind code	B2
Filing date	Jan 25, 2022
Priority date	Mar 30, 2021
Publication date	Sep 30, 2025
Grant date	Sep 30, 2025

Abstract

A three-dimensional (3D) spectrum situation completion method and device based on a generative adversarial network includes performing graying and coloring preprocessing based on incomplete 3D spectrum situations from historical or empirical spectrum data obtained by a UAV through sampling a target region, obtaining three-channel incomplete 3D spectrum situation maps displayed in colors, forming a training set based on the incomplete 3D spectrum situation maps; training the generative adversarial network based on the training set and obtaining a trained generator network in the generative adversarial network, performing graying and coloring preprocessing based on a measured incomplete 3D spectrum situation obtained by the UAV through sampling a specified measurement region, obtaining a three-channel measured incomplete 3D spectrum situation map displayed in colors, and using the measured incomplete 3D spectrum situation map as input data to the generator network to obtain a three-channel measured complete 3D spectrum situation map displayed in colors.

First claim

What is claimed is:

1. A three-dimensional (3D) spectrum situation completion method based on a generative adversarial network performed by an unmanned aerial vehicle (UAV) connected to a computer, comprising:
performing graying and coloring preprocessing based on incomplete 3D spectrum situations from historical or empirical spectrum data obtained by an unmanned aerial vehicle (UAV) through sampling a target region, obtaining three-channel incomplete 3D spectrum situation maps displayed in colors, and forming a training set based on the obtained three-channel incomplete 3D spectrum situation maps displayed in colors;
training the generative adversarial network based on the training set and obtaining a trained generator network in the generative adversarial network, wherein the generative adversarial network is configured with a sampling processing function, and the sampling processing function is used to perform, based on sampled and unsampled position points in the three-channel incomplete 3D spectrum situation maps displayed in colors that are inputted into the generative adversarial network, sampling processing on a three-channel complete 3D spectrum situation maps displayed in colors that are outputted by the generator network in the generative adversarial network; and
performing graying and coloring preprocessing based on a measured incomplete 3D spectrum situation obtained by the UAV through sampling a specified measurement region, obtaining a three-channel measured incomplete 3D spectrum situation map displayed in colors, and using the three-channel measured incomplete 3D spectrum situation map as input data to the generator network to obtain a three-channel measured complete 3D spectrum situation map displayed in colors,
wherein the performing, based on the sampled and unsampled position points in the three-channel incomplete 3D spectrum situation map displayed in colors that is inputted into the generative adversarial network, sampling processing on the three-channel complete 3D spectrum situation map displayed in colors that is outputted by the generator network in the generative adversarial network comprises:
determining a numerical matrix, wherein the numerical matrix comprises a first specified numerical column of the sampled position points and a second specified numerical column of the unsampled position points, and the first specified numerical column is different from the second specified numerical column; and
calculating a Hadamard product of the numerical matrix and the three-channel complete 3D spectrum situation map displayed in colors.

2. The 3D spectrum situation completion method based on the generative adversarial network according to claim 1 , wherein the graying and coloring preprocessing comprises:
calculating a normalized value for the incomplete 3D spectrum situation based on incomplete historical or empirical spectrum data; and
determining, based on the normalized value, a single-channel incomplete 3D spectrum situation map displayed in grayscale.

3. The 3D spectrum situation completion method based on the generative adversarial network according to claim 2 , wherein the graying and coloring preprocessing further comprises:
copying twice the single-channel incomplete 3D spectrum situation map;
obtaining a two-channel incomplete 3D spectrum situation map;
expanding the single-channel incomplete 3D spectrum situation map by the two-channel incomplete 3D spectrum situation map, obtaining a three-channel incomplete 3D spectrum situation map displayed in grayscale; and
coloring with a specified color unsampled position points in the three-channel incomplete 3D spectrum situation map displayed in grayscale to obtain the three-channel incomplete 3D spectrum situation map displayed in colors, wherein the colors comprise the grayscale and the specified color.

4. The 3D spectrum situation completion method based on the generative adversarial network according to claim 3 , wherein in the generative adversarial network,
the input data of the generator network is configured as the three-channel incomplete 3D spectrum situation map displayed in colors; and
output data of the generator network is used as the three-channel complete 3D spectrum situation map displayed in colors.

5. The 3D spectrum situation completion method based on the generative adversarial network according to claim 4 , wherein in the generative adversarial network,
the generator network comprises an input-side 3D convolutional layer, two down-sampling module, six residual modules, two residual modules, and an output-side 3D convolutional layer;
the input data of the generator network is processed sequentially by the input-side 3D convolutional layer, the down-sampling module, the residual module, the up-sampling module, and the output-side 3D convolutional layer of the generator network; and
the residual module of the generator network has a dilated convolutional layer.

6. The 3D spectrum situation completion method based on the generative adversarial network according to claim 4 , wherein in the generative adversarial network,
input data of a discriminator network is configured as the three-channel incomplete 3D spectrum situation map displayed in colors, or a three-channel complete 3D spectrum situation map after sampling processing.

7. The 3D spectrum situation completion method based on the generative adversarial network according to claim 6 , wherein in the generative adversarial network,
the discriminator network comprises an input-side 3D convolutional layer, two down-sampling modules, three residual modules, and an output-side 3D convolutional layer;
the input data of the discriminator network is processed sequentially by the input-side 3D convolutional layer, the down-sampling modules, the residual modules, and the output-side 3D convolutional layer of the discriminator network; and
the residual modules of the discriminator network have a dilated convolutional layer.

8. The 3D spectrum situation completion method based on the generative adversarial network according to claim 7 , wherein in the generative adversarial network,
a hyperparameter set of the residual modules of the discriminator network is used to input an output data of a specified number of channels to the output-side 3D convolutional layer of the discriminator network; and
the output-side 3D convolutional layer of the discriminator network has a controlled hyperparameter set.

9. A three-dimensional (3D) spectrum situation completion method based on a generative adversarial network performed by an unmanned aerial vehicle (UAV) connected to a computer, comprising:
performing graying and coloring preprocessing based on incomplete 3D spectrum situations from historical or empirical spectrum data obtained by an unmanned aerial vehicle (UAV) through sampling a target region, obtaining three-channel incomplete 3D spectrum situation maps displayed in colors, and forming a training set based on the obtained three-channel incomplete 3D spectrum situation maps displayed in colors;
training the generative adversarial network based on the training set and obtaining a trained generator network in the generative adversarial network, wherein the generative adversarial network is configured with a sampling processing function, and the sampling processing function is used to perform, based on sampled and unsampled position points in the three-channel incomplete 3D spectrum situation maps displayed in colors that are inputted into the generative adversarial network, sampling processing on a three-channel complete 3D spectrum situation maps displayed in colors that are outputted by the generator network in the generative adversarial network; and
performing graying and colo

Assignees

• Univ Nanjing Aeronautics & Astronautics

Inventors

• Huang Yang
• Zhu Qiuming
• Hu Tianyu
• Wu Qihui
• Gong Zhiren
• Wu Xuan
• Zhong Weizhi
• Mao Kai
• Zhang Xiaofei
• Lu Yiwei

Classifications

• G06T11/10 Primary

  Texturing; Colouring; Generation of textures or colours (retouching, inpainting or scratch removal G06T5/77) · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T7/90

  Determination of colour characteristics · CPC title

• G06N3/045

  Combinations of networks · CPC title