Normalizing counts of plant-parts-of-interest
Implementations are described herein for normalizing counts of plant-parts-of-interest detected in digital imagery to account for differences in spatial dimensions of plants, particularly plant heights. In various implementations, one or more digital images depicting a top of a first plant may be processed. The one or more digital images may have been acquired by a vision sensor carried over top of the first plant by a ground-based vehicle. Based on the processing: a distance of the vision sensor to the first plant may be estimated, and a count of visible plant-parts-of-interest that were captured within a field of view of the vision sensor may be determined. Based on the estimated distance, the count of visible plant-parts-of-interest may be normalized with another count of visible plant-parts-of-interest determined from one or more digital images capturing a second plant.
1. A method implemented using one or more processors, comprising:
obtaining a plurality of digital images acquired by a range-capable vision sensor carried by a ground-based vehicle past at least first and second plants, wherein differing spatial dimensions of the first and second plants cause parts of the first and second plants to be first and second distances, respectively, from the range-capable vision sensor as it passes the at least first and second plants;
processing one or more of the digital images depicting the first plant;
based on the processing:
estimating the first distance of the range-capable vision sensor to the first plant,
estimating a spatial dimension of the first plant, and
determining a count of visible plant-parts-of-interest of the first plant that were captured within a field of view of the range-capable vision sensor;
based on the estimated first distance, normalizing the count of visible plant-parts-of-interest of the first plant with another count of visible plant-parts-of-interest determined from one or more of the digital images depicting the second plant; and
predicting a crop yield using the normalized count of visible plant-parts-of-interest and the first spatial dimension of the first plant.
2. The method of claim 1 , wherein the estimated first distance of the range-capable vision sensor to the first plant is estimated based on a distribution of pixel-wise ranges between the range-capable vision sensor and the first plant, wherein the pixel-wise ranges are detected by the range-capable vision sensor and included in one or more of the digital images capturing the first plant.
3. The method of claim 2 , wherein the estimated first distance of the range-capable vision sensor to the first plant is calculated as a quantile of the distribution of pixel-wise ranges.
4. The method of claim 1 , wherein the count of visible plant-parts-of-interest is determined using a convolutional neural network.
5. The method of claim 1 , wherein the visible plant-parts-of-interest comprise bean pods.
6. The method of claim 1 , wherein the first plant and second plants are different varietals of soybean plants.
7. The method of claim 1 , wherein the spatial dimension of the first plant comprises a height of the first plant.
8. A system comprising one or more processors and memory storing instructions that, in response to execution by the one or more processors, cause the one or more processors to:
obtain a plurality of digital images acquired by a range-capable vision sensor carried by a ground-based vehicle past at least first and second plants, wherein differing spatial dimensions of the first and second plants cause parts of the first and second plants to be first and second distances, respectively, from the range-capable vision sensor as it passes the at least first and second plants;
process one or more of the digital images depicting the first plant;
based on the processing:
estimate the first distance of the range-capable vision sensor to the first plant,
estimate a spatial dimension of the first plant, and
determine a count of visible plant-parts-of-interest of the first plant that were captured within a field of view of the range-capable vision sensor;
based on the estimated first distance, normalize the count of visible plant-parts-of-interest of the first plant with another count of visible plant-parts-of-interest determined from one or more of the digital images depicting the second plant; and
predict a crop yield using the normalized count of visible plant-parts-of-interest and the first spatial dimension of the first plant.
9. The system of claim 8 , wherein the estimated first distance of the range-capable vision sensor to the first plant is estimated based on a distribution of pixel-wise ranges between the range-capable vision sensor and the first plant, wherein the pixel-wise ranges are detected by the range-capable vision sensor and included in one or more of the digital images capturing the first plant.
10. The system of claim 9 , wherein the estimated first distance of the range-capable vision sensor to the first plant is calculated as a quantile of the distribution of pixel-wise ranges.
11. The system of claim 8 , wherein the count of visible plant-parts-of-interest is determined using a convolutional neural network.
12. The system of claim 8 , wherein the visible plant-parts-of-interest comprise bean pods.
13. The system of claim 8 , wherein the first plant and second plants are different varietals of soybean plants.
14. The system of claim 8 , wherein the spatial dimension of the first plant comprises a height of the first plant.
15. At least one non-transitory computer-readable medium comprising instructions that, when executed by one or more processors, cause the one or more processors to:
obtain a plurality of digital images acquired by a range-capable vision sensor carried by a ground-based vehicle past at least first and second plants, wherein differing spatial dimensions of the first and second plants cause parts of the first and second plants to be first and second distances, respectively, from the range-capable vision sensor as it passes the at least first and second plants;
process one or more of the digital images depicting the first plant;
based on the processing:
estimate the first distance of the range-capable vision sensor to the first plant,
estimate a spatial dimension of the first plant, and
determine a count of visible plant-parts-of-interest of the first plant that were captured within a field of view of the range-capable vision sensor;
based on the estimated first distance, normalize the count of visible plant-parts-of-interest of the first plant with another count of visible plant-parts-of-interest determined from one or more of the digital images depicting the second plant; and
predict a crop yield using the normalized count of visible plant-parts-of-interest and the first spatial dimension of the first plant.
16. The at least one non-transitory computer-readable medium of claim 15 , wherein the spatial dimension of the first plant comprises a height of the first plant.
17. The at least one non-transitory computer-readable medium of claim 15 , wherein the estimated first distance of the range-capable vision sensor to the first plant is estimated based on a distribution of pixel-wise ranges between the range-capable vision sensor and the first plant, wherein the pixel-wise ranges are detected by the range-capable vision sensor and included in one or more of the digital images capturing the first plant.
18. The at least one non-transitory computer-readable medium of claim 17 , wherein the estimated first distance of the range-capable vision sensor to the first plant is calculated as a quantile of the distribution of pixel-wise ranges.
19. The at least one non-transitory computer-readable medium of claim 15 , wherein the count of visible plant-parts-of-interest is determined using a convolutional neural network.
20. The at least one non-transitory computer-readable medium of claim 15 , wherein the visible plant-parts-of-interest comprise bean pods.