Disease recognition from images having a large field of view

1. A computer-implemented method of detecting infected objects from large field-of-view images, comprising:

receiving, by a processor, a digital image capturing multiple objects;

generating, by the processor, a plurality of scaled images from the digital image respectfully corresponding to a plurality of scales;

computing a group of feature matrices for the digital image;

for each of the plurality of scaled images:

selecting a list of candidate regions from the scaled image each likely to capture a single object; and

for each of the list of candidate regions:

mapping the candidate region back to the digital image to obtain a mapped region;

identifying a corresponding portion from each of the group of feature matrices based on the mapping; and

determining whether the candidate region is likely to capture the single object infected with a disease based on the group of corresponding portions;

choosing a group of final regions from the lists of mapped regions based on the determining; and

causing display of information regarding the group of final regions.

2. The computer-implemented method of claim 1 , the multiple objects being multiple leaves in a crop field, and the single object being one of the multiple leaves.

3. The computer-implemented method of claim 1 ,

the determining comprising computing a probability of infection associated with the disease,

the choosing being based on the probability of infection larger than a certain threshold.

4. The computer-implemented method of claim 1 ,

the determining comprises computing a probability of infection associated with the disease,

the choosing comprising:

identifying one of the lists of mapped regions having a largest probability of infection;

eliminating any mapped region that overlaps with the one mapped region for an amount exceeding a certain threshold;

repeating the previous two steps until a stopping criterion is satisfied.

5. The computer-implemented method of claim 1 ,

the computing comprising executing one or more convolutional layers of a convolutional neural network (CNN) on the digital image, the set of convolutional layers configured to extract feature values of the single object at the plurality of scales,

the determining comprising:

executing a pooling layer of the CNN on each of the group of corresponding portions of the feature matrices;

executing a fully-connected layer of the CNN on output data of the pooling layer, the fully-connected layer configured to classify a combination of feature values with respect to different classes corresponding to the single object respectively infected with different diseases.

6. The computer-implemented method of claim 1 , the selecting comprising:

generating a plurality of regions from the scaled image using a sliding window of a specific size;

for each of the plurality of regions:

calculating a feature vector for the region;

determining whether the region is likely to capture the single object based on the feature vector.

7. The computer-implemented method of claim 6 , the calculating comprising constructing a histogram of oriented gradients (HOG) for the region.

8. The computer-implemented method of claim 6 , determining whether the region is likely to capture the single object comprising executing a support vector machine (SVM) on the feature vector, the SVM configured to classify a set of feature values with respect to different classes corresponding to the presence or absence of the single object.

9. The computer-implemented method of claim 1 , the single object is a corn leaf, and the disease is gray leaf spot (GLS), Goss's Wilt (GW), or Northern Leaf Blight (NLB).

10. The computer-implemented method of claim 1 , the information indicating, for one of the group of final regions, a position of the final region within the digital image and the corresponding disease.

11. A non-transitory computer-readable storage medium storing one or more instructions which, when executed by one or more processors, cause the one or more processors to perform a method of detecting infected objects from large field-of-view images, the method comprising:

receiving a digital image capturing multiple objects;

generating a plurality of scaled images from the digital image respectfully corresponding to a plurality of scales;

computing a group of feature matrices for the digital image;

for each of the plurality of scaled images:

selecting a list of candidate regions from the scaled image each likely to capture a single object; and

for each of the list of candidate regions:

mapping the candidate region back to the digital image to obtain a mapped region;

identifying a corresponding portion from each of the group of feature matrices based on the mapping; and

determining whether the candidate region is likely to capture the single object infected with a disease based on the group of corresponding portions;

choosing a group of final regions from the lists of mapped regions based on the determining; and

causing display of information regarding the group of final regions.

12. The non-transitory computer-readable storage medium of claim 11 , the multiple objects being multiple leaves in a crop field, and the single object being one of the multiple leaves.

13. The non-transitory computer-readable storage medium of claim 11 ,

the determining comprising computing a probability of infection associated with the disease,

the choosing being based on the probability of infection larger than a certain threshold.

14. The non-transitory computer-readable storage medium of claim 11 ,

the determining comprises computing a probability of infection associated with the disease,

the choosing comprising:

identifying one of the lists of mapped regions having a largest probability of infection;

eliminating any mapped region that overlaps with the one mapped region for an amount exceeding a certain threshold;

repeating the previous two steps until a stopping criterion is satisfied.

15. The non-transitory computer-readable storage medium of claim 11 ,

the computing comprising executing one or more convolutional layers of a convolutional neural network (CNN) on the digital image, the set of convolutional layers configured to extract feature values of the single object at the plurality of scales,

the determining comprising:

executing a pooling layer of the CNN on each of the group of corresponding portions of the feature matrices;

executing a fully-connected layer of the CNN on output data of the pooling layer, the fully-connected layer configured to classify a combination of feature values with respect to different classes corresponding to the single object respectively infected with different diseases.

16. The non-transitory computer-readable storage medium of claim 11 , the selecting comprising:

generating a plurality of regions from the scaled image using a sliding window of a specific size;

for each of the plurality of regions:

calculating a feature vector for the region;

determining whether the region is likely to capture the single object based on the feature vector.

17. The non-transitory computer-readable storage medium of claim 16 , the calculating comprising constructing a histogram of oriented gradients (HOG) for the region.

18. The non-transitory computer-readable storage medium of claim 16 , determining whether the region is likely to capture the single object comprising executing a support vector machine (SVM) on the feature vector, the SVM configured to classify a set of feature values with respect to different classes corresponding to the presence or absence of the single object.

19. The non-transitory computer-readable storage medium of claim 11 , the single object is a corn leaf, and the disease is gray leaf spot (GLS), Goss's Wilt (GW), or Northern Leaf Blight (NLB).

20. The non-transitory computer-readable storage medium of claim 11 , the information indicating, for one of the group of final regions, a position of the final region within the digital image and the corresponding disease.