Iterative relabeling using spectral neighborhoods
View Patent ↗A first image is received. An initial label is assigned to at least some pixels in the first image, including by assigning a first label to a first pixel. A determination is made, based at least in part on labels of spectral neighbors of the first pixel, that the first pixel's label should be replaced with a different label. The first pixel's label is updated with the different label. The first pixel's label is iteratively refined until convergence.
1. A system, comprising:
a processor configured to:
receive a first image;
assign an initial label to at least some pixels in the first image, including by assigning a first label to a first pixel;
determine, based at least in part on labels of spectral neighbors of the first pixel that the first pixel's label should be replaced with a different label;
update the first pixel's label with the different label; and
iteratively refine the first pixel's label, using labels of at least some of the spectral neighbors of the first pixel, until at least one of: (1) a rate of change of pixel label values over successive repetitions falls below a threshold, and (2) a non-zero upper bound of number of iterations is reached; and
a memory coupled to the processor and configured to provide the processor with instructions.
2. The system of claim 1 wherein the determination is based at least in part on a majority vote of the spectral neighbors.
3. The system of claim 1 wherein the spectral neighbors comprise nearest neighbors.
4. The system of claim 1 wherein the initial label is assigned based at least in part by a classifier trained using a training set.
5. The system of claim 1 wherein the initial label is assigned based at least in part by a classifier that takes as input the received first image and a boundary map.
6. The system of claim 1 wherein the initial label is assigned using a historical classification.
7. The system of claim 6 wherein the historical classification is determined from crop data.
8. A method, comprising:
receiving a first image;
assigning an initial label to at least some pixels in the first image, including by assigning a first label to a first pixel;
determining, based at least in part on labels of spectral neighbors of the first pixel that the first pixel's label should be replaced with a different label;
updating the first pixel's label with the different label; and
iteratively refining the first pixel's label, using labels of at least some of the spectral neighbors of the first pixel, until at least one of: (1) a rate of change of pixel label values over successive repetitions falls below a threshold, and (2) a non-zero upper bound of number of iterations is reached.
9. The method of claim 8 wherein determining that the first pixel's label should be replaced with a different label is based at least in part on a majority vote of the spectral neighbors.
10. The method of claim 8 wherein the spectral neighbors comprise nearest neighbors.
11. The method of claim 8 wherein the initial label is assigned based at least in part by a classifier trained using a training set.
12. The method of claim 8 wherein the initial label is assigned based at least in part by a classifier that takes as input the received first image and a boundary map.
13. The method of claim 8 wherein the initial label is assigned using a historical classification.
14. The method of claim 13 wherein the historical classification is determined from crop data.
15. A computer program product embodied in a tangible computer readable storage medium and comprising computer instructions for:
receiving a first image;
assigning an initial label to at least some pixels in the first image, including by assigning a first label to a first pixel;
determining, based at least in part on labels of spectral neighbors of the first pixel that the first pixel's label should be replaced with a different label;
updating the first pixel's label with the different label; and
iteratively refining the first pixel's label, using labels of at least some of the spectral neighbors of the first pixel, until at least one of: (1) a rate of change of pixel label values over successive repetitions falls below a threshold, and (2) a non-zero upper bound of number of iterations is reached .
16. The system of claim 1 wherein the processor is further configured to provide as output a classification map which identifies, for a set comprising at least some pixels included in the first image, a numeric value corresponding to a label for each of the pixels included in the set.
17. The system of claim 16 wherein the numeric value corresponds to a land use classification.
18. The system of claim 16 wherein the processor is further configured to provide the output to a yield predictor.
19. The system of claim 16 wherein the processor is further configured to refine the output using a boundary map.
20. The system of claim 1 wherein the processor is further configured to receive a second image that also includes the first pixel, assign a label to the first pixel as it appears in the second image, and determine a final label for the first pixel by combining a first label result for the first pixel in the first image and a second label result for the first pixel in the second image.
21. The system of claim 20 wherein the processor is configured to determine the final label based at least in part on a speed at which convergence occurred, respectively, for processing of the first image and the second image.
22. The system of claim 20 wherein the first image and second image are included in a time series of observations.
23. The system of claim 20 wherein the first image and second image are observations made by two different sources.
24. The method of claim 8 further comprising providing as output a classification map which identifies, for a set comprising at least some pixels included in the first image, a numeric value corresponding to a label for each of the pixels included in the set.
25. The method of claim 24 wherein the numeric value corresponds to a land use classification.
26. The method of claim 24 further comprising providing the output to a yield predictor.
27. The method of claim 24 further comprising refining the output using a boundary map.
28. The method of claim 8 further comprising receiving a second image that also includes the first pixel, assigning a label to the first pixel as it appears in the second image, and determining a final label for the first pixel by combining a first label result for the first pixel in the first image and a second label result for the first pixel in the second image.
29. The method of claim 28 further comprising determining the final label based at least in part on a speed at which convergence occurred, respectively, for processing of the first image and the second image.
30. The method of claim 28 wherein the first image and second image are included in a time series of observations.
31. The method of claim 28 wherein the first image and second image are observations made by two different sources.