	US 12,277,754 B2
	Generating synthetic training data and training a remote sensing machine learning model
Yueqi Li, San Jose, CA (US)
Assigned to Deere & Company, Moline, IL (US)
Filed by Deere & Company, Moline, IL (US)
Filed on May 26, 2022, as Appl. No. 17/825,615.
Prior Publication US 2023/0386183 A1, Nov. 30, 2023
Int. Cl. G06V 10/774 (2022.01); G06V 20/10 (2022.01); G06V 20/13 (2022.01)

CPC G06V 10/7747 (2022.01) [G06V 20/13 (2022.01); G06V 20/188 (2022.01)]

20 Claims

1. A method for training a remote sensing machine learning model based on at least one synthetic satellite training image, the method implemented using one or more processors and comprising:

accessing a plurality of ground truth low-elevation images that depict one or more particular crops growing in one or more agricultural areas, wherein the ground truth low-elevation images are captured within a first elevation range that is below a second elevation range that corresponds to low earth orbit;

identifying, for the plurality of ground truth low-elevation images, terrain conditions observed in the one or more agricultural areas;

generating data of a plurality of low-elevation training images based on the ground truth low-elevation images, wherein the data of the plurality of low-elevation training images includes:

a first subset comprising the plurality of ground truth low-elevation images and the corresponding observed terrain conditions, and

a second subset comprising synthetic terrain conditions and synthetic low-elevation images generated based on the synthetic terrain conditions, wherein the synthetic terrain conditions comprise variations of the observed terrain conditions;

processing the plurality of low-elevation training images using a synthetic satellite image machine learning model to generate the at least one synthetic satellite training image, wherein the plurality of low-elevation training images respectively correspond to different geographic portions of a particular agricultural region and the at least one synthetic satellite training image corresponds to the particular agricultural region;

processing the at least one synthetic satellite training image using the remote sensing machine learning model to generate inferred terrain conditions for the particular agricultural region; and

training the remote sensing machine learning model based on a comparison of the inferred terrain conditions and the corresponding observed terrain conditions and the synthetic terrain conditions.