Time series generator trained using satellite data
Implementations are described herein for utilizing the spectral-, spatial-, and temporal-information of a satellite image time series to facilitate crop control or monitoring. In various implementations, a plurality of training examples may be assembled for a generative model. Each training example of the plurality of training examples may include a respective high-resolution image capturing a respective region and a corresponding low-resolution satellite image time series capturing the respective region. The plurality of training examples can be used to train the generative model, to acquire a trained generative model. A high-resolution image depicting one or more agricultural conditions for a given region, can be received and processed using the trained generative model, to generate a synthetic low-resolution satellite image time series, where the synthetic low-resolution satellite image time series represent the one or more agricultural conditions.
1. A method implemented using one or more processors, the method comprising:
receiving an image having a first resolution, wherein the image having the first resolution depicts one or more agricultural conditions in a given region; and
processing the image having the first resolution using a trained generative model to generate a synthetic satellite image time series having a second resolution as an output of the trained generative model, the first resolution greater than the second resolution, and wherein the synthetic low resolution satellite image time series having the second resolution represent the one or more agricultural conditions, wherein the trained generative model is trained using a plurality of training examples, and wherein the plurality of training examples includes a first training example including (i) a respective satellite image having a third resolution capturing a respective region at a particular time and (ii) a corresponding satellite image time series having a fourth resolution capturing the respective region, the third resolution greater than the fourth resolution.
2. The method of claim 1 , wherein the corresponding low-resolution satellite image time series having the fourth resolution capturing the respective region are captured during a time interval that includes the particular time.
3. The method of claim 1 , wherein:
the respective satellite image having the third resolution is captured by a sensor having a fifth resolution of a first satellite, and
the corresponding satellite image time series having the fourth resolution are captured by a sensor having a sixth resolution of a second satellite, wherein the fifth resolution is greater than the sixth resolution.
4. The method of claim 1 , wherein the image having the first resolution is captured from a satellite, an unmanned aerial vehicle (UAV), or a ground-based vehicle.
5. The method of claim 1 , further including processing the generated synthetic satellite image time series having the second resolution using a trained discriminative model to generate a prediction result or a classification result as an output of the trained discriminative model.
6. The method of claim 1 , further including:
based on the one or more agricultural conditions in the given region, generating a ground truth label for the generated synthetic satellite image time series having the second resolution;
generating a training instance, wherein the generated training instance includes the generated synthetic satellite image time series having the second resolution and the generated ground truth label; and
training a discriminative model using the generated training instance.
7. The method of claim 6 , wherein the training the discriminative model using the generated training instance includes:
processing the generated synthetic satellite image time series having the second resolution using the discriminative model to generate an output of the discriminative model;
comparing the generated output with the ground truth label to determine a difference between the generated output and the ground truth label; and
updating one or more weights of the discriminative model based on the determined difference.
8. The method of claim 1 , wherein the first training example further includes metadata associated with the respective satellite image having the third resolution.
9. The method of claim 8 , wherein the metadata associated with the respective satellite image having the third resolution is a soil condition, a terrain condition, or a weather condition at the particular time.
10. The method of claim 1 , wherein the synthetic satellite image time series having the second resolution has a substantially same image resolution as the corresponding satellite image time series having the fourth resolution capturing the given region.
11. A system comprising one or more processors and a memory storing a plurality of instructions, wherein the plurality of instructions are executed to cause the one or more processors to:
receive an image having a first resolution, wherein the high-resolution image having the first resolution depicts one or more agricultural conditions in a given region; and
process the image having the first resolution using a trained generative model to generate a synthetic satellite image time series having a second resolution as an output of the trained generative model, the first resolution greater than the second resolution, and the synthetic low-resolution satellite image time series having the second resolution represent the one or more agricultural conditions, wherein the trained generative model is trained using a plurality of training examples, and wherein the plurality of training examples includes a first training example including (i) a respective image having a third resolution capturing a respective region at a particular time and (ii) a corresponding satellite image time series having a fourth resolution capturing the respective region, the third resolution greater than the fourth resolution.
12. The system of claim 11 , wherein the corresponding satellite image time series having the fourth resolution capturing the respective region are captured during a time interval that includes the particular time.
13. The system of claim 11 , wherein:
the respective image having the third resolution is captured by a sensor having a fifth resolution of a first satellite, and
the corresponding satellite image time series having the fourth resolution are captured by a sensor having a sixth resolution of a second satellite, wherein the fifth resolution is greater than the sixth resolution.
14. The system of claim 11 , wherein the respective image having the third resolution is captured from an unmanned aerial vehicle (UAV) or a ground-based vehicle.
15. The system of claim 11 , wherein the plurality of instructions are executed to further cause the one or more processors to process the generated synthetic satellite image time series having the second resolution using a trained discriminative model to generate a prediction result or a classification result as an output of the trained discriminative model.
16. The system of claim 11 , wherein the plurality of instructions are executed to further cause the one or more processors to:
based on the one or more agricultural conditions in the given region, generate a ground truth label for the generated synthetic satellite image time series having the second resolution;
generate a training instance, wherein the generated training instance includes the generated synthetic satellite image time series having the second resolution and the generated ground truth label; and
train a discriminative model using the generated training instance.
17. The system of claim 16 , wherein the plurality of instructions are executed to further cause the one or more processors to train the discriminative model, wherein to train the discriminative model causes the one or more processors to:
process the generated synthetic satellite image time series having the second resolution using the discriminative model to generate an output of the discriminative model;
compare the generated output with the ground truth label to determine a difference between the generated output and the ground truth label; and
update one or more weights of the discriminative model based on the determined difference.
18. The system of claim 11 , wherein the first training example further includes metadata associated with the respective image having the third resolution.
19. The system of claim 18 , wherein the metadata associated with the respective image having the third resolution is a soil condition or a weather condition at the particular time.
20. A method implemented using one or more processors, the method comprising:
receiving an image having a first resolution, wherein the image having the first resolution depicts one or more agricultural conditions in a given region; and
in response to receiving the image having the first resolution:
generating a label associated with the given region in the image having the first resolution;
processing the image having the first resolution using a trained generative model to generate a synthetic satellite image time series having a second resolution as an output of the trained generative model, the first resolution greater than the second resolution, and the synthetic satellite image time series having the second resolution represent the one or more agricultural conditions, wherein the trained generative model is trained using a plurality of training examples, and the plurality of training examples includes a first training example including (i) a respective satellite image having a third resolution capturing a respective region and (ii) a corresponding satellite image time series having a fourth resolution capturing the respective region, the third resolution greater than the fourth resolution; and
generating a training instance for an untrained discriminative model to include (1) the generated label, and (2) the generated synthetic satellite image time series having the second resolution.