US 12,141,987 B2
Method for extracting roof edge image for installing solar panel by using machine learning
Yungcheol Byun, Jeju-si (KR); Jihyeok Yang, Jeju-si (KR); and Debaprya Hazra, Jeju-si (KR)
Assigned to NANOOMENERGY CO., LTD., Jeju-si (KR)
Appl. No. 17/789,034
Filed by NANOOMENERGY CO., LTD., Jeju-si (KR)
PCT Filed Jul. 31, 2020, PCT No. PCT/KR2020/010178
§ 371(c)(1), (2) Date Jun. 24, 2022,
PCT Pub. No. WO2021/132829, PCT Pub. Date Jul. 1, 2021.
Claims priority of application No. 10-2019-0175767 (KR), filed on Dec. 26, 2019.
Prior Publication US 2023/0057612 A1, Feb. 23, 2023
Int. Cl. G06K 9/00 (2022.01); G06T 7/13 (2017.01); G06T 7/181 (2017.01); G06V 10/26 (2022.01); G06V 10/44 (2022.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01)
CPC G06T 7/13 (2017.01) [G06T 7/181 (2017.01); G06V 10/267 (2022.01); G06V 10/44 (2022.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30184 (2013.01)] 6 Claims
OG exemplary drawing
 
1. A method of extracting a roof edge image for solar panel installation by using machine learning for a generative adversarial neural network (GAN) based image conversion model, the method comprising:
inputting original rooftop image data (A) and an image data set (B) in which the rooftop edge is detected into the roof edge image extraction system;
the original rooftop image data (A) being inputted to a second generator unit of the system and passing through the second generator unit of the system and the second generator unit outputting an image (O) similar to a target image (T);
image data (B) in which the rooftop edge is detected being inputted to a first generator unit of the system and passing the first generator unit of the system;
the first generator unit mapping noise samples and inserting obstructions into the input image data (B);
a first discriminator unit of the system having access to the original input image data set and receiving the output from the first generator unit as an input, training to discriminate each region between the original image and the received input;
parameters for segmentation being altered and rectified according to the adversarial training result between the first generator unit and the first discriminator unit, obstruction hiding the edge of the roof being segmented, and a second discriminator unit receiving the image in which desirable roof edges are detected;
the second discriminator unit receiving a first pair (A, T) of the original image data and the output image output from the first discriminator unit and a second pair (A, O) of the original image data and the image output from the second generator unit, respectively, predicting the correct output image from each pair, and weights of the parameters assigned to the second generator and the second discriminator being optimized according to the prediction accuracy, the second generator unit and the second discriminator unit being trained again; and
the roof edge image extraction system removing the obstructions hiding the edge portion of the roof image, extracting the edge of roof image, connecting the edge parts automatically and generating the complete roof edge image.