Anomaly detection
Disclosed herein are methods and systems for anomaly detection within a field using images or sensor data. An anomaly detection process determines whether a pixel or a cluster of pixels fit within calculated or modeled expectations to identify an anomaly. The process can further analyze the anomaly to determine a likely cause. Example anomaly detection techniques account for intra-field inhomogeneity and for non-crop regions. Example anomaly detection techniques also can identify anomalous regions.
1. A method for detecting anomalies in agricultural land, the method comprising:
computing first pixel statistics of pixels of a first hyperspectral image of agricultural land, the pixels within boundaries specified by field boundary data;
generating a pixel mask using the first pixel statistics or historic anomalies;
applying the pixel mask to the first hyperspectral image to generate a second hyperspectral image;
forming groups of pixels of the second hyperspectral image using zone strata of the agricultural land;
computing second pixel statistics using the groups of pixels of the second hyperspectral image;
identifying a pixel clump within the second hyperspectral image using the second pixel statistics, the pixel clump representing an anomalous area within the agricultural land;
predicting a source of anomaly of the pixel clump; and
generating an anomaly map using the pixel clump and the predicted source of anomaly of the pixel clump.
2. The method of claim 1 , further comprising:
determining a mean topographical position of the pixel clump using a digital elevation model of the agricultural land,
wherein predicting a source of anomaly of the pixel clump comprises determining a likely source of anomaly of the pixel clump using the mean topographical position of the pixel clump.
3. The method of claim 1 further comprising:
calculating a mean spectral signature for the pixel clump,
wherein predicting a source of anomaly of the pixel clump comprises determining a likely source of anomaly of the pixel clump using the mean spectral signature of the pixel clump.
4. The method of claim 1 , further comprising:
determining a mean topographical position of the pixel clump using a digital elevation model of the agricultural land; and
calculating a mean spectral signature of the pixel clump,
wherein predicting a source of anomaly of the pixel clump comprises determining a likely source of anomaly of the pixel clump using the mean topographical position of the pixel clump and the mean spectral signature of the pixel clump.
5. The method of claim 4 , wherein determining the likely source of anomaly of the pixel clump comprises:
determining that the likely source of anomaly of the pixel clump is excessive moisture responsive to determining that the mean spectral signature of the pixel clump has reflectance below a first threshold at about 960 nanometers, or at about 1450 nanometers, or has reflectance above a second threshold at about 550 nanometers, and the pixel clump has a topographical wetness index that exceeds a third threshold.
6. The method of claim 4 , wherein determining the likely source of anomaly of the pixel clump comprises:
determining that the likely source of anomaly of the pixel clump is insufficient moisture responsive to determining that the mean spectral signature of the pixel clump has a reflectance above a first threshold at about 960 nanometers, or above a second threshold at about 1450 nanometers, or has reflectance above a third threshold at about 550 nanometers, and the pixel clump has a topographical wetness index below a fourth threshold.
7. The method of claim 4 , wherein determining the likely source of anomaly of the pixel clump comprises determining that the likely source of anomaly is weed infestation responsive to determining the pixel clump has a greenness value that exceeds a threshold.
8. The method of claim 4 , wherein determining the likely source of anomaly of the pixel clump comprises:
determining that the likely source of anomaly of the pixel clump is insect damage responsive to determining that the mean spectral signature of the pixel clump has a ratio of red to near-infrared on the spectral index below a threshold.
9. The method of claim 4 , wherein determining the likely source of anomaly of the pixel clump comprises:
determining that the likely source of anomaly of the pixel clump is a disease responsive to determining that the mean spectral signature of the pixel clump has a ratio of red to near-infrared below a first threshold, the mean spectral signature of the pixel clump has ratio associated with the disease, and the mean topographical position of the pixel clump is within preferences associated with the disease.
10. The method of claim 1 , wherein the pixel statistics comprise mean statistics and standard deviation statistics.
11. The method of claim 10 , wherein generating a pixel mask using the first pixel statistics comprises adding a pixel of the first hyperspectral image to the pixel mask responsive to determining that the pixel is more than three standard deviations from a mean pixel value.
12. The method of claim 10 , wherein identifying a pixel clump comprises adding a pixel of the second hyperspectral image to a clump responsive to determining that the pixel is more than one standard deviation from a mean pixel value.
13. The method of claim 10 comprising stratifying the second hyperspectral image into the zone strata using crop variation within the agricultural land due to intra-field inhomogeneity.
14. The method of claim 1 , wherein identifying a pixel clump comprises applying cleanup operations.
15. The method of claim 14 , wherein performing cleanup operations comprises applying a majority filter to the second hyperspectral image.
16. A method of detecting agricultural anomalies in an agricultural field, the method comprising:
stratifying a hyperspectral image of an agricultural field into strata, the strata representing zones of crop variation due to non-anomalous intra-field inhomogeneity;
applying an extreme value filter to the hyperspectral image that masks individual pixels or clusters of pixels of the hyperspectral image that are more than three standard deviations from a mean pixel value;
calculating pixel statistics of the filtered hyperspectral image according to the strata;
using the pixel statistics to isolate one or more clumps of pixels of the filtered hyperspectral image, each clump representing an anomaly within the agricultural field; and
providing a location of at least one of the one or more anomalies.
17. The method of claim 16 , further comprising:
determining a mean topographical position of each pixel clump using a digital elevation model of the agricultural land; and
classifying the anomaly of each pixel clump using the mean topographical position and mean spectral signature of a respective clump.
18. The method of claim 16 , further comprising:
calculating a mean spectral signature for each pixel clump; and
classifying the anomaly of each pixel clump using a mean topographical position and mean spectral signature of a respective clump.
19. The method of claim 16 , further comprising classifying the anomaly of each pixel clump using the mean topographical position and mean spectral signature of the respective pixel clump.
20. The method of claim 16 , wherein using the statistics to isolate one or more clumps of pixels of the filtered hyperspectral image comprises:
identifying clumps of pixels that are more than one standard deviation from a mean pixel value of a respective strata.