Fusion-based classifier, classification method, and classification system

1. A fusion-based classifier, the classifier comprising:

a storage medium, stores a plurality of modules; and

a processor, coupled to the storage medium, wherein the processor accesses and executes the plurality of modules, wherein the plurality of modules comprise:

a sub-classifier, generating a plurality of probability vectors according to input data, wherein each of the plurality of probability vectors comprises a plurality of elements respectively corresponding to a plurality of classes;

a fusion layer, selecting a first probability vector having a first class element corresponding to a first class-of-interest from the plurality of probability vectors according to the first class-of-interest, wherein the first class element has an extremum value between a plurality of first class elements corresponding to the plurality of probability vectors respectively, wherein the plurality of first class elements corresponding to the first class-of-interest; and

an output layer, determining a class of the input data according to the first probability vector.

2. The classifier according to claim 1 , wherein the fusion layer, further selects from the plurality of probability vectors, a second probability vector having a second extremum value corresponding to a second class-of-interest according to the second class-of-interest, and the plurality of modules further comprising:

a second fusion layer performs probability fusion on the first probability vector and the second probability vector to generate a third probability vector.

3. The classifier according to claim 2 , wherein determining the class of the input data according to the first probability vector, comprising:

determining the class of the input data according to the third probability vector.

4. The classifier according to claim 1 , wherein the first class-of-interest corresponds to one of the plurality of elements, and the extremum value represents one of a maximum value and a minimum value.

5. The classifier according to claim 2 , wherein the first class-of-interest corresponds to one of the plurality of elements, and the second class-of-interest corresponds to another one of the plurality of elements, and the extremum value represents one of a maximum value and a minimum value.

6. The classifier according to claim 1 , wherein determining the class of the input data according to the first probability vector, comprising:

performing a likelihood ratio test on the first probability vector according to the first class-of-interest, and determining the class of the input data according to a result of the likelihood ratio test.

7. The classifier according to claim 1 , wherein determining the class of the input data according to the first probability vector, comprising:

determining the class of the input data according to a threshold corresponding to the first class-of-interest.

8. The classifier according to claim 3 , wherein determining the class of the input data according to the third probability vector, comprising:

performing a likelihood ratio test on the third probability vector according to the first class-of-interest and the second class-of-interest, and determining the class of the input data according to a result of the likelihood ratio test.

9. The classifier according to claim 3 , wherein determining the class of the input data according to the third probability vector, comprising:

determining the class of the input data according to a first threshold corresponding to the first class-of-interest and a second threshold corresponding to the second class-of-interest.

10. The classifier according to claim 1 , wherein the input data comprises a plurality of cropped parts of a feature map output from a convolution neural network.

11. A classification method based on probability fusion, the classification method comprising:

generating a plurality of probability vectors according to input data, wherein each of the plurality of probability vectors comprises a plurality of elements respectively corresponding to a plurality of classes;

selecting a first probability vector having a first class element corresponding to a first class-of-interest from the plurality of probability vectors according to the first class-of-interest, wherein the first class element has an extremum value between a plurality of first class elements corresponding to the plurality of probability vectors respectively, wherein the plurality of first class elements corresponding to the first class-of-interest;

determining a class of the input data according to the first probability vector.

12. The classification method according to claim 11 , comprising:

selecting a second probability vector having a second extremum value corresponding to a second class-of-interest from the plurality of probability vectors according to the second class-of-interest, and performing the probability fusion on the first probability vector and the second probability vector to generate a third probability vector.

13. The classification method according to claim 12 , comprising:

determining the class of the input data according to the third probability vector.

14. The classification method according to claim 11 , wherein the first class-of-interest corresponds to one of the plurality of elements, and the extremum value represents one of a maximum value and a minimum value.

15. The classification method according to claim 12 , wherein the first class-of-interest corresponds to one of the plurality of elements, and the second class-of-interest corresponds to another one of the plurality of elements, and the extremum value represents one of a maximum value and a minimum value.

16. The classification method according to claim 11 , wherein determining the class of the input data according to the first probability vector, comprising:

performing a likelihood ratio test on the first probability vector according to the first class-of-interest, and determining the class of the input data according to a result of the likelihood ratio test.

17. The classification method according to claim 11 , wherein determining the class of the input data according to the first probability vector, comprising:

determining the class of the input data according to a threshold corresponding to the first class-of-interest.

18. The classification method according to claim 13 , wherein determining the class of the input data according to the third probability vector, comprising:

performing a likelihood ratio test on the third probability vector according to the first class-of-interest and the second class-of-interest, and determining the class of the input data according to a result of the likelihood ratio test.

19. The classification method according to claim 13 , wherein determining the class of the input data according to the third probability vector, comprising:

determining the class of the input data according to a first threshold corresponding to the first class-of-interest and a second threshold corresponding to the second class-of-interest.

20. The classification method according to claim 11 , wherein the input data comprises a plurality of cropped parts of a feature map output from a convolution neural network.

21. A classification system based on probability fusion, the classification system comprising:

an automatic optical inspection equipment, obtaining image data of an article; and

a processor, configured to control:

a classifier, comprising:

a sub-classifier, generating a plurality of probability vectors according to the image data, wherein each of the plurality of probability vectors comprises a plurality of elements respectively corresponding to a plurality of classes;

a fusion layer, selecting a first probability vector having a first class element corresponding to a first class-of-interest from the plurality of probability vectors according to the first class-of-interest, wherein the first class element has an extremum value between a plurality of first class elements corresponding to the plurality of probability vectors respectively, wherein the plurality of first class elements corresponding to the first class-of-interest;

an output layer, determining a class of an appearance defect according to the first probability vector.