Granted Patent B2

US 12,430,697 · App. 16/699,292 · Granted Sep 30, 2025

Inventors: Jeffrey G. White (Raleigh, NC); Bradley A. Miller (Ames, IA); Julianne Bielski (Durham, NC)

Assignees: SOILMETRIX, INC.; IOWA STATE UNIVERSITY RESEARCH FOUNDATION, INC.; NORTH CAROLINA STATE UNIVERSITY

G06Q50/02G06Q10/0639G06V20/188

Abstract

Claims (29)

1. A system for optimizing agricultural zone attributes comprising:

a data processor in communication with a data server;

a user device, in communication with the data server, displaying optimized management zones from the following steps performed by the data processor;

constructing at least two data cubes, by the data processor, comprising a first input dataset of geospatial images, by the data processor, from remote sensing of field conditions subjected to digital analysis, collected multi-temporal satellite or aerial imagery bands and calculated soil indices from the collected multi-temporal satellite or aerial imagery bands, and a SSURGO Soil Parent Material Data;

labeling, by the data processor, the first input dataset based on past land management data and yields of a particular geographic areas;

training, by the data processor, a first soil property model with the first input data set and the labels using a supervised learning algorithm;

generating, with the first soil property model by the data processor, predictions for a first soil property;

iteratively changing, by the data processor, a neural network architecture performing the supervised learning for each soil property model and tuning hyperparameters to improve accuracy relative to the labelled training until accuracy is at a sufficient level; and, in parallel, applying, by the data processor, an unsupervised artificial intelligence algorithm to at least one data cube comprising the first input dataset, and

outputting, by the data processor, a plurality of potential sets of management zones using the unsupervised artificial intelligence algorithm, based on at least predictions for the first soil property;

geospatially joining, by the data processor, the predictions for the first soil property from the output of the first soil property model generated using the supervised learning algorithm with each of the plurality of potential sets of management zones generated from the unsupervised artificial intelligence algorithms;

comparing, by the data processor, each of the plurality of potential sets of management zones that have been geospatially joined with the predictions for the first soil property from the output of the first soil property model to optimize intra-zone differences and maximize inter-zone differences of the first soil property based on a data cube of a prediction of a soil property that forms the best performing zone in comparison to at least a second data cube; and

delivering, by the data processor, an optimized management zone for the first soil property to the user device from the plurality of potential sets of management zones.

2. The system of claim 1 where an evaluation of zone effectiveness incorporates ground truth or Deep Learning algorithm output.

3. The system of claim 1 , wherein the first data set comprises of a data cube with geospatial images with all data layers of such geospatial images sharing the same geographic coordinate system and projection;

wherein the first input data set of the data cube is labeled with ground truth data regarding past land management data and yields of a particular geographic area;

wherein the first soil property model is trained with the data cube using the supervised learning algorithm; and

wherein the data processor applies the data cube in parallel with the supervised learning algorithm and the unsupervised artificial intelligence algorithms and outputs optimized management zone for the first soil property to the user device from the plurality of potential sets of management zones.

4. A method for optimizing agricultural zone attributes comprising:

constructing, by a data processor, at least two data cubes comprising a first input dataset of geospatial images using remote sensing of field conditions subjected to digital analysis, collected multi-temporal satellite or aerial imagery bands and calculated soil indices from the collected multi-temporal satellite or aerial imagery bands, and SSURGO Soil Parent Material Data—

labeling, by the data processor, the first input dataset based on past land management data and yields of a particular geographic areas;

training, by the data processor, with the first input data set using a supervised learning algorithm a first soil property model with the first input data set and the labels;

generating, by the data processor, predictions for a first soil property with the first soil property model; and, in parallel,

iteratively changing, by the data processor, a neural network architecture performing the supervised learning for each soil property model and tuning hyperparameters to improve accuracy relative to the labelled training until accuracy is at a sufficient level;

applying, by the data processor, an unsupervised artificial intelligence algorithm to at least one data cube comprising the first input dataset, and outputting a plurality of potential sets of management zones using the unsupervised artificial intelligence algorithm, based on at least predictions for the first soil property;

geospatially, by the data processor, joining the predictions the predictions for the first soil property from the output of the first soil property model generated using the supervised learning algorithm with each of the plurality of potential sets of management zones generated from the unsupervised artificial intelligence algorithm;

comparing, by the data processor, each of the plurality of potential sets of management zones that have been geospatially joined with the predictions for the first soil property from the output of the first soil property model to optimize intra-zone differences and maximize inter-zone differences of the first soil property based on a data cube of a prediction of a soil property that forms the best performing zone in comparison to at least a second data cube; and

delivering, by the data processor, an optimized management zone for the first soil property to the a user device.

5. The method of claim 4 , where the supervised learning algorithm is a Deep Learning algorithm.

6. The method of claim 4 , where an evaluation of zone effectiveness incorporates ground truth or Deep Learning algorithm output.

Assignments (4)

Continuity (2)

References Cited (21)