Generating synthetic ground-level data based on generator conditioned for particular agricultural area
Implementations are described herein for conditioning a generator machine learning model to generate synthetic ground-level data that is biased towards a given agricultural area based on high-elevation images. In various implementations, a plurality of ground-level images may be accessed that depict crops within a specific agricultural area. A first set of high-elevation image(s) may also be accessed that depict the specific agricultural area. The ground-level images and the first set of high-elevation image(s) may be used to condition an air-to-ground generator machine learning model to generate synthetic ground-level data from high-elevation imagery depicting the specific agricultural area. A second set of high-elevation image(s) that depict a specific sub-region of the specific agricultural area may then be accessed and processed using the air-to-ground generator machine learning model to generate synthetic ground-level data that infers one or more conditions of the specific sub-region of the specific agricultural area.
1. A method implemented using one or more processors, comprising:
accessing a plurality of ground-level images that depict crops within a first area;
accessing a first set of one or more high-elevation images that depict a second area, wherein the second area includes the first area;
training an air-to-ground generator machine learning model to generate first synthetic ground-level data based on the first set of the one or more high-elevation images of the second area and the plurality of ground-level images of the first area;
accessing a second set of one or more high-elevation images that depict a third area, wherein the second area includes the first area and the third area, and wherein the second set of one or more high-elevation images include one or more portions of one or more of the high-elevation images of the first set, or one or more additional high-elevation images;
inferring a condition of the third area using the trained air-to-ground generator machine learning model, wherein the condition relates to a property of the crops within the first area; and
processing the second set of one or more high-elevation images using the trained air-to-ground generator machine learning model to generate second synthetic ground-level data depicting the third area that includes the condition.
2. The method of claim 1 , wherein the second synthetic ground-level data that infers the condition of the third area includes synthetic ground-level images.
3. The method of claim 2 , wherein the conditioning training includes:
processing the plurality of ground-level images based on a ground-to-air generator machine learning model to generate one or more synthetic high-elevation images;
processing the one or more synthetic high-elevation images based on a high-elevation discriminator machine learning model to generate high-elevation discriminator output; and
based on the high-elevation discriminator output, training the ground-to-air generator machine learning model.
4. The method of claim 3 , wherein the conditioning training further includes:
processing the one or more synthetic high-elevation images based on the air-to-ground generator machine learning model to generate ground-level images;
processing the ground-level images based on a ground-level discriminator machine learning model to generate ground-level discriminator output; and
based on the ground-level discriminator output, training the air-to-ground generator machine learning model.
5. The method of claim 1 , wherein at least one of the high-elevation images of the first set of one or more high-elevation images comprises includes a synthetic high-elevation image generated based on processing one or more of the ground-level images using a ground-to-air generator machine learning model.
6. The method of claim 2 , further including processing the synthetic ground-level images based on a phenotyping machine learning model to generate one or more agricultural inferences about crops in the third area.
7. The method of claim 6 , further including:
providing, to a remotely-operated client application, one or more of the synthetic ground-level images for display on a user interface of the remotely-operated client application;
receiving, from the remotely-operated client application, user selection data indicating a user selection of at least a portion of at least one of the one or more synthetic ground-level images displayed on the user interface of the remotely-operated client application; and
providing, to the remotely-operated client application and responsive to receiving the user selection data, at least one of the one or more agricultural inferences for display on the user interface.
8. The method of claim 1 , wherein the first set of one or more high-elevation images were captured by one or more satellites.
9. The method of claim 1 , wherein the first set of one or more high elevation images were captured by one or more unmanned aerial vehicles (UAVs).
10. The method of claim 1 , wherein the plurality of ground-level images were captured by a ground-based robot travelling through the first area.
11. A system comprising one or more processors and memory storing instructions that, in response to execution of the instructions by the one or more processors, cause the one or more processors to:
access a plurality of ground-level images that depict crops within a first area;
access a first set of one or more high-elevation images that depict a second area, wherein the second area includes the first area;
train an air-to-ground generator machine learning model to generate first synthetic ground-level data based on the first set of the one or more high-elevation images of the second area and the plurality of ground-level images of the first area;
access a second set of one or more high-elevation images that depict a third area, wherein the second area includes the first area and the third area, and wherein the second set of one or more high-elevation images include one or more portions of one or more of the high-elevation images of the first set, or one or more additional high-elevation images;
infer a condition of the third area using the trained air-to-ground generator machine learning model, wherein the condition relates to a property of the crops within the first area; and
process the second set of one or more high-elevation images using the trained air-to-ground generator machine learning model to generate second synthetic ground-level data depicting the third area that includes of the condition.
12. The system of claim 11 , wherein the second synthetic ground-level data that infers the condition of the third area includes synthetic ground-level images.
13. The system of claim 12 , wherein to train the air-to-ground generator machine learning model includes:
processing the plurality of ground-level images based on a ground-to-air generator machine learning model to generate one or more synthetic high-elevation images;
processing the one or more synthetic high-elevation images based on a high-elevation discriminator machine learning model to generate high-elevation discriminator output; and
based on the high-elevation discriminator output, training the ground-to-air generator machine learning model.
14. The system of claim 13 , wherein to train the air-to-ground generator machine learning model further includes:
processing the one or more synthetic high-elevation images based on the air-to-ground generator machine learning model to generate ground-level images;
processing the ground-level images based on a ground-level discriminator machine learning model to generate ground-level discriminator output; and
based on the ground-level discriminator output, training the air-to-ground generator machine learning model.
15. The system of claim 11 , wherein at least one of the high-elevation images of the first set of one or more high-elevation images includes a synthetic high-elevation image generated based on processing one or more of the ground-level images using a ground-to-air generator machine learning model.
16. The system of claim 12 , further including processing the synthetic ground-level images based on a phenotyping machine learning model to generate one or more agricultural inferences about crops in the third area.
17. The system of claim 16 , wherein execution of the instructions causes the one or more processors to:
provide, to a remotely-operated client application, one or more of the synthetic ground-level images for display on a user interface of the remotely-operated client application;
receive, from the remotely-operated client application, user selection data indicating a user selection of at least a portion of at least one of the one or more synthetic ground-level images displayed on the user interface of the remotely-operated client application; and
provide, to the remotely-operated client application and responsive to receiving the user selection data, at least one of the one or more agricultural inferences for display on the user interface.
18. The system of claim 11 , wherein the plurality of ground-level images were captured by a ground-based robot travelling through the first area.
19. At least one non-transitory computer-readable medium comprising instructions that, in response to execution of the instructions by one or more processors, cause the one or more processors to:
access a plurality of ground-level images that depict crops within a first area;
access a first set of one or more high-elevation images that depict a second area, wherein the second area includes the first area;
train an air-to-ground generator machine learning model to generate first synthetic ground-level data from based on the first set of one or more high-elevation imagery images of the second area and the plurality of ground-level images of the first area;
access a second set of one or more high-elevation images that depict a third area, wherein the second area includes the first area and the third area, and wherein the second set of one or more high-elevation images include one or more portions of one or more of the high-elevation images of the first set, or one or more additional high-elevation images;
infer a condition of the third area using the trained air-to-ground generator machine learning model, wherein the condition relates to a property of the crops within the first area; and
process the second set of one or more high-elevation images using the trained air-to-ground generator machine learning model to generate second synthetic ground-level data depicting the third area that includes the condition.
20. The at least one non-transitory computer-readable medium of claim 19 , wherein at least one of the high-elevation images of the second set of one or more high-elevation images that depict the third area includes a synthetic high-elevation image generated based on processing one or more additional ground-level images, each capturing at least a portion of the third area, using a ground-to-air generator machine learning model.
21. A method implemented using one or more processors, comprising:
accessing a plurality of ground-level images that depict crops within a specific agricultural area;
accessing a first set of one or more high-elevation images that depict the specific agricultural area;
using the plurality of ground-level images and the first set of one or more high-elevation images, conditioning an air-to-ground generator machine learning model to generate synthetic ground-level data from high-elevation imagery depicting the specific agricultural area;
accessing a second set of one or more high-elevation images that depict a specific sub-region of the specific agricultural area, wherein the second set of one or more high-elevation images include: one or more portions of one or more of the high-elevation images of the first set, or one or more additional high-elevation images;
processing the second set of one or more high-elevation images using the air-to-ground generator machine learning model to generate synthetic ground-level data that infers one or more conditions of the specific sub-region of the specific agricultural area, wherein the synthetic ground-level data that infers the one or more conditions of the specific sub-region includes synthetic ground-level images; and
processing the synthetic ground-level images based on a phenotyping machine learning model to generate one or more agricultural inferences about crops in the specific sub-region of the specific agricultural area.
22. The method of claim 21 , further including:
providing, to a remotely-operated client application, one or more of the synthetic ground-level images for display on a user interface of the remotely-operated client application;
receiving, from the remotely-operated client application, user selection data indicating a user selection of at least a portion of at least one of the one or more synthetic ground-level images displayed on the user interface of the remotely-operated client application; and
providing, to the remotely-operated client application and responsive to receiving the user selection data, at least one of the one or more agricultural inferences for display on the user interface.
23. A system comprising one or more processors and memory storing instructions that, in response to execution of the instructions by the one or more processors, cause the one or more processors to:
access a plurality of ground-level images that depict crops within a specific agricultural area;
access a first set of one or more high-elevation images that depict the specific agricultural area;
use the plurality of ground-level images and the first set of one or more high-elevation images, conditioning an air-to-ground generator machine learning model to generate synthetic ground-level data from high-elevation imagery depicting the specific agricultural area;
access a second set of one or more high-elevation images that depict a specific sub-region of the specific agricultural area, wherein the second set of one or more high-elevation images include: one or more portions of one or more of the high-elevation images of the first set, or one or more additional high-elevation images;
process the second set of one or more high-elevation images using the air-to-ground generator machine learning model to generate synthetic ground-level data that infers one or more conditions of the specific sub-region of the specific agricultural area, wherein the synthetic ground-level data that infers one or more conditions of the specific sub-region includes synthetic ground-level images; and
process the synthetic ground-level images based on a phenotyping machine learning model to generate one or more agricultural inferences about crops in the specific sub-region of the specific agricultural area.
24. The system of claim 23 , wherein execution of the instructions causes the one or more processors to:
provide, to a remotely-operated client application, one or more of the synthetic ground-level images for display on a user interface of the remotely-operated client application;
receive, from the remotely-operated client application, user selection data indicating a user selection of at least a portion of at least one of the one or more synthetic ground-level images displayed on the user interface of the remotely-operated client application; and
provide, to the remotely-operated client application and responsive to receiving the user selection data, at least one of the one or more agricultural inferences for display on the user interface.