US-12620214-B2 - Method and apparatus for employing deep learning neural network to predict cropland data layer

Abstract

A computer-implemented method for predicting a cropland data layer (CDL) for a current year includes: retrieving a first set of records from a historical CDL database, where the first set corresponds to sampled areas of a region taken over a period for a number of years; retrieving a second set of records from a historical imagery database, where the second set corresponds to the sampled areas of the region, the period, and the number of years; employing the second set as inputs to train a deep learning network to generate the first set; retrieving a third set of records from a current imagery database, where the third set corresponds to a prescribed region, and where the third set corresponds to the time period and the current year; and using the third set as inputs and executing the trained deep learning network to generate a predicted CDL for the current year.

Inventors

• Ernesto Brau
• R. Shane Bussmann
• Ethan Sargent

Assignees

• CIBO TECHNOLOGIES, INC.

Dates

Publication Date

20260505

Application Date

20210531

Claims (20)

1. 1 . A computer-implemented method for predicting a cropland data layer for use within a current growing year, the computer-implemented method comprising: retrieving a first set of records from a historical cropland data layer database, wherein the first set of records corresponds to randomly sampled areas of a geographic region taken over a prescribed time period for a prescribed number of years; retrieving a second set of records from a historical imagery database, wherein the second set of records corresponds to the randomly sampled areas of the geographic region, the prescribed time period, and the prescribed number of years; employing the second set of records as inputs to train a deep learning convolutional neural network to generate the first set of records and using parameters generated during training to configure a trained deep learning convolutional neural network for execution, the employing comprising: cleansing the second set of records from the historical imagery database by removing duplicate information, inferring missing values, substituting for unconventional characters and symbols, or removing outlier values; retrieving a third set of records from a current imagery database, wherein the third set of records corresponds to a prescribed geographic region, and wherein the third set of records corresponds to the prescribed time period and the current growing year; and using the third set of records as inputs and executing the trained deep learning convolutional neural network to generate a predicted cropland data layer for the current growing year, the using comprising: stitching adjacent records from the current imagery database using coordinates of corresponding farms.
2. 2 . The computer-implemented method as recited in claim 1 , wherein the deep learning convolutional neural network and the trained deep learning convolutional neural network each comprise 5 layers.
3. 3 . The computer-implemented method as recited in claim 1 , wherein each of the second and third sets of records each comprise 128×128 pixel images.
4. 4 . The computer-implemented method as recited in claim 3 , wherein the each of the 128×128 pixel images comprise Sentinel satellite red channel, blue channel, green channel, near infrared channel, and cloud mask channel.
5. 5 . The computer-implemented method as recited in claim 4 , wherein the prescribed time period comprises May through October, and wherein the number of 128×128 pixel images for each of the prescribed number of years comprises 12 images.
6. 6 . The computer-implemented method as recited in claim 1 , wherein the prescribed number of years comprises three 3 years previous to the current growing year.
7. 7 . The computer-implemented method as recited in claim 1 , wherein the predicted cropland data layer comprises a raster of color-coded pixels, and wherein each pixel comprises a 30 meter×30 meter area within the prescribed geographic region, and wherein each pixel's color is indicative of a particular type of land covering for the 30 meter×30 meter area.
8. 8 . A non-transitory computer-readable storage medium storing instructions that, when executed by a computer, cause the computer to perform a method for predicting a cropland data layer for use within a current growing year, the method comprising: retrieving a first set of records from a historical cropland data layer database, wherein the first set of records corresponds to randomly sampled areas of a geographic region taken over a prescribed time period for a prescribed number of years; retrieving a second set of records from a historical imagery database, wherein the second set of records corresponds to the randomly sampled areas of the geographic region, the prescribed time period, and the prescribed number of years; employing the second set of records as inputs to train a deep learning convolutional neural network to generate the first set of records and using parameters generated during training to configure a trained deep learning convolutional neural network for execution, the employing comprising cleansing the second set of records from the historical imagery database by removing duplicate information, inferring missing values, substituting for unconventional characters and symbols, or removing outlier values; retrieving a third set of records from a current imagery database, wherein the third set of records corresponds to a prescribed geographic region, and wherein the third set of records corresponds to the prescribed time period and the current growing year; and using the third set of records as inputs and executing the trained deep learning convolutional neural network to generate a predicted cropland data layer for the current growing year, the using comprising: stitching adjacent records from the current imagery database using coordinates of corresponding farms.
9. 9 . The non-transitory computer-readable storage medium as recited in claim 8 , wherein the deep learning convolutional neural network and the trained deep learning convolutional neural network each comprise 5 layers.
10. 10 . The non-transitory computer-readable storage medium as recited in claim 8 , wherein each of the second and third sets of records each comprise 128×128 pixel images.
11. 11 . The non-transitory computer-readable storage medium as recited in claim 10 , wherein the each of the 128×128 pixel images comprise Sentinel satellite red channel, blue channel, green channel, near infrared channel, and cloud mask channel.
12. 12 . The non-transitory computer-readable storage medium as recited in claim 11 , wherein the prescribed time period comprises May through October, and wherein the number of 128×128 pixel images for each of the prescribed number of years comprises 12 images.
13. 13 . The non-transitory computer-readable storage medium as recited in claim 8 , wherein the prescribed number of years comprises three 3 years previous to the current growing year.
14. 14 . The non-transitory computer-readable storage medium as recited in claim 8 , wherein the predicted cropland data layer comprises a raster of color-coded pixels, and wherein each pixel comprises a 30 meter×30 meter area within the prescribed geographic region, and wherein each pixel's color is indicative of a particular type of land covering for the 30 meter×30 meter area.
15. 15 . A computer program product for predicting a cropland data layer for use within a current growing year, the computer program product comprising: a computer readable non-transitory medium having computer readable program code stored thereon, the computer readable program code comprising: program instructions to retrieve a first set of records from a historical cropland data layer database, wherein the first set of records corresponds to randomly sampled areas of a geographic region taken over a prescribed time period for a prescribed number of years; program instructions to retrieve a second set of records from a historical imagery database, wherein the second set of records corresponds to the randomly sampled areas of the geographic region, the prescribed time period, and the prescribed number of years; program instructions to employ the second set of records as inputs to train a deep learning convolutional neural network to generate the first set of records and to use parameters generated during training to configure a trained deep learning convolutional neural network for execution, wherein the second set of records from the historical imagery database are cleansed by removing duplicate information, inferring missing values, substituting for unconventional characters and symbols, or removing outlier values; program instructions to retrieve a third set of records from a current imagery database, wherein the third set of records corresponds to a prescribed geographic region, and wherein the third set of records corresponds to the prescribed time period and the current growing year; and program instructions to use the third set of records as inputs and to execute the trained deep learning convolutional neural network to generate a predicted cropland data layer for the current growing year, wherein adjacent records from the current imagery database are stitched together using coordinates of corresponding farms.
16. 16 . The computer program product as recited in claim 15 , wherein the deep learning convolutional neural network and the trained deep learning convolutional neural network each comprise 5 layers.
17. 17 . The computer program product as recited in claim 15 , wherein each of the second and third sets of records each comprise 128×128 pixel images.
18. 18 . The computer program product as recited in claim 17 wherein the prescribed time period comprises May through October, and wherein the number of 128×128 pixel images for each of the prescribed number of years comprises 12 images.
19. 19 . The computer program product as recited in claim 15 , wherein the prescribed number of years comprises three 3 years previous to the current growing year.
20. 20 . The computer program product as recited in claim 15 , wherein the predicted cropland data layer comprises a raster of color-coded pixels, and wherein each pixel comprises a 30 meter×30 meter area within the prescribed geographic region, and wherein each pixel's color is indicative of a particular type of land covering for the 30 meter×30 meter area.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS This application is related to the following co-pending U.S. Patent Applications, each of which has a common assignee and common inventors, the entireties of which are herein incorporated by reference. FILINGSER. NO.DATETITLE17/334,970May 31, 2021METHOD AND APPARATUS FOR EMPLOYING DEEP LEARNING NEURAL NETWORK TO PREDICTMANAGEMENT ZONES17/334,974May 31, 2021METHOD AND APPARATUS FOR EMPLOYING DEEP LEARNING NEURAL NETWORK TO INFERREGENERATIVE COVER CROP PRACTICES17/334,978May 31, 2021METHOD AND APPARATUS FOR EMPLOYING DEEP LEARNING TO INFER IMPLEMENTATION OFREGENERATIVE IRRIGATION PRACTICES17/334,983May 31, 2021METHOD AND APPARATUS FOR EMPLOYING DEEP LEARNING TO INFER IMPLEMENTATION OFREGENERATIVE TILLAGE PRACTICES BACKGROUND OF THE INVENTION Field of the Invention This invention relates in general to the field of regenerative agricultural management practices, and more specifically to methods and systems for processing imagery data to predict crop-specific land cover, to determine agricultural management zones within fields, and to detect use of regenerative practices within agricultural management zones. Description of the Related Art Climate change is one of the most studied and discussed topics on the planet, and this level of global concern has sparked numerous initiatives to reduce Earth's carbon footprint. Initiatives include zero waste recycling and reuse programs, clean energy programs, conservation measures, sustainable transportation programs, and carbon offset and trading programs. This application focuses on carbon offsets from an agricultural perspective, how they are determined, and how programs to generate those offsets are monitored and verified. As one skilled in the art will appreciate, billions of dollars are spent each year by countries, corporations, small businesses, and individuals to reduce greenhouse gas emissions. But more often than not, the impact of carbon footprint reduction programs is difficult to quantify, mainly because such an effort is labor intensive and relies heavily on self-reporting. Every year the United States Department of Agriculture (USDA) generates a rasterized, geo-referenced, crop-specific land cover map for the continental United States that is known as the Cropland Data Layer (CDL). The CDL is generated from moderate resolution satellite imagery and extensive agricultural ground truth, and is used by all manner of agricultural-related entities such as universities and private research firms, commercial producers, growers, equipment manufacturers, underwriters, real-estate concerns, bankers, conservationists, carbon brokers, and political entities. The CDL consists of a raster of color-coded pixels, where each pixel comprises a 30 meter by 30 meter geographic area (0.09 hectare pixel resolution), and where each pixel's color is indicative of a particular type of “crop.” The crops indicated range from conventional cash crops (e.g., corn, cotton, rice) and also include colors that indicate fallow/idle cropland, wetlands, ice/snow, developed land, forests, pastures, etc., thus mapping land in the continental United States to its use from an agricultural perspective. Albeit extremely useful, generation of the CDL is not timely, for the CDL for a given year is not released to the public until the first quarter of the following year, which is quite limiting to those agricultural entities that require more timely data. Accordingly, what is needed are methods and systems for predicting a cropland data layer at the end of a current year growing season. What is also needed are methods and apparatus for predicting a cropland data layer at the end of a current year growing season based solely on satellite imagery data. What is further needed are deep learning methods and systems that are trained on historical CDL data and corresponding satellite imagery data to predict CDL at the end of a current growing season. What is additionally needed are methods and apparatus that employ transfer learning techniques to detect agricultural management practices zones within parcels based solely on satellite imagery. What is finally need are methods and apparatus for inferring implementation and maintenance of agricultural management practices within determined management practices zones. SUMMARY OF THE INVENTION The present invention, among other applications, is directed to solving the above-noted problems and addresses other problems, disadvantages, and limitations of the prior art. In one embodiment, a computer-implemented method for predicting a cropland data layer for use within a current growing year is provided, the computer-implemented method including: retrieving a first set of records from a historical cropland data layer database, where the first set of records corresponds to randomly sampled areas of a geographic region taken over a prescribed time period for a prescribed number of years; retrieving a second set of records from a his