Method and system for crop recognition and boundary delineation

Patent metadata

Field	Value
Publication number	US-11527062-B2
Application number	US-202016832769-A
Country	US
Kind code	B2
Filing date	Mar 27, 2020
Priority date	Dec 30, 2016
Publication date	Dec 13, 2022
Grant date	Dec 13, 2022

Abstract

A computer-implemented method for determining field boundaries and crop forecasts in each field is provided. The method includes deriving vegetation indices for each geo-spatial pixel of each image of multi-spectral imagery at a plurality of points in time, constructing minimum bounding boxes for each image according to the vegetation indices, and generating, based on a neural network analysis of each image and the minimum bounding boxes, a geo-spatial plot of crops including a predicted plot of future crop usage for an area including each field in the multi-spectral imagery.

First claim

What is claimed is:

1. A computer-implemented method for determining field boundaries and crop forecasts in each field, comprising:
deriving vegetation indices for each geo-spatial pixel of each image of multi-spectral imagery at a plurality of points in time;
constructing minimum bounding boxes for each image according to the vegetation indices, the constructing comprising deriving pixel statistics corresponding to temporal differences of the vegetation indices, determining threshold bands according to local maxima and local minima of the pixel statistics, and binarizing the temporal differences of the vegetation indices to determine areas within each image having vegetation indices within the threshold bands;
generating, based on a neural network analysis of each image and the minimum bounding boxes, a geo-spatial plot of crops including a predicted plot of future crop usage for an area including each field in the multi-spectral imagery; and
backpropagating a loss between a semantic segmentation of the multi-spectral imagery and historical labelled multi-spectral imagery.

2. The method of claim 1 , further comprising cropping each image according to the minimum bounding boxes.

3. The method of claim 1 , wherein constructing minimum bounding boxes includes:
constructing minimum bounding boxes using contours determined from the binarized temporal differences of the vegetation indices.

4. The method of claim 3 , wherein the pixel statistics are determined according to a Fourier transform of the vegetation indices as a function of time.

5. The method of claim 1 , further comprising training a crop prediction model of the neural network, including:
labeling the multi-spectral imagery according to historical ground survey data by geo-spatially fusing each image of the multi-spectral imagery with corresponding ground survey data; and
evaluating semantic segmentation of the multi-spectral imagery.

6. The method of claim 1 , wherein the neural network includes a convolutional neural network that semantically segments each image of the multi-spectral imagery according to a crop prediction model.

7. The method of claim 1 , further comprising updating crop yield predictions according to the geo-spatial plot of crops.

8. The method of claim 1 , further comprising evaluating soil quality in each field in the geo-spatial plot of crops according to homogeneity of the crops.

9. The method of claim 8 , further comprising determining a next crop to plant in each field according to the soil quality in each field.

10. A system for determining field boundaries and crop forecasts in each field, comprising:
a big data platform for curating and indexing multi-spectral imagery and ground survey data for an agricultural area across time;
a boundary detection engine that derives vegetation indices for each geo-spatial pixel of each image of multi-spectral imagery at a plurality of points in time and constructs minimum bounding boxes for each image according to the vegetation indices, and includes deriving pixel statistics corresponding to temporal differences of the vegetation indices, determining threshold bands according to local maxima and local minima of the pixel statistics, and binarizing the temporal differences of the vegetation indices to determine areas within each image having vegetation indices within the threshold bands; and
a neural network that generates a geo-spatial plot of crops including a predicted plot of future crop usage for the area including each field in the multi-spectral imagery based on the minimum bounding boxes, and backpropagates a loss between a semantic segmentation of the multi-spectral imagery and historical labelled multi-spectral imagery.

11. The system of claim 10 , further comprising an image cropping and normalization engine that crops each image according to the minimum bounding boxes.

12. The system of claim 10 , wherein the boundary detection engine constructions minimum bounding boxes including:
constructing minimum bounding boxes using contours determined from the binarized temporal differences of the vegetation indices.

13. The system of claim 10 , further comprising a neural network training engine that trains training a model of the neural network, including:
labeling the multi-spectral imagery according to historical ground survey data by geo-spatially fusing each image of the multi-spectral imagery with corresponding ground survey data; and
evaluating semantic segmentation of the multi-spectral imagery.

14. The system of claim 10 , wherein the neural network includes a convolutional neural network that semantically segments each image of the multi-spectral imagery according to a crop prediction model.

15. A computer program product for determining field boundaries and crop forecasts in each field, the computer program product comprising a non-transitory computer readable storage medium having program instructions embodied therewith, the program instructions being executable by a computer to cause the computer to perform a method comprising:
deriving vegetation indices for each geo-spatial pixel of each image of multi-spectral imagery at a plurality of points in time;
constructing minimum bounding boxes for each image according to the vegetation indices, the constructing minimum bounding boxes comprising deriving pixel statistics corresponding to temporal differences of the vegetation indices, determining threshold bands according to local maxima and local minima of the pixel statistics, and binarizing the temporal differences of the vegetation indices to determine areas within each image having vegetation indices within the threshold bands;
generating, based on a neural network analysis of each image and the minimum bounding boxes, a geo-spatial plot of crops including a predicted plot of future crop usage for the area including each field in the multi-spectral imagery; and
backpropagating a loss between a semantic segmentation of the multi-spectral imagery and historical labelled multi-spectral imagery.

16. The computer program product of claim 15 , further comprising cropping each image according to the minimum bounding boxes.

17. The computer program product of claim 15 , wherein constructing minimum bounding boxes includes:
constructing minimum bounding boxes using contours determined from the binarized temporal differences of the vegetation indices.

18. The computer program product of claim 15 , further comprising training a crop prediction model of the neural network, including:
labeling the multi-spectral imagery according to historical ground survey data by geo-spatially fusing each image of the multi-spectral imagery with corresponding ground survey data; and
evaluating semantic segmentation of the multi-spectral imagery.

19. The computer program product of claim 15 , wherein the neural network includes a convolutional neural network that semantically segments each image of the multi-spectral imagery according to a crop prediction model.

20. The computer program product of claim 15 , further comprising determining a next crop to plant in each field according to a soil quality in each field.

Assignees

• IBM

Inventors

• Albrecht Conrad M
• Lu Siyuan
• Marianno Fernando J
• Hamann Hendrik F
• Freitag Marcus O
• Klein Levente I

Classifications

• G06V10/82

  using neural networks · CPC title

• G06T2207/30188

  Vegetation; Agriculture · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06V20/188 Primary

  Vegetation · CPC title

• G06T7/11 Primary

  Region-based segmentation · CPC title