Artificial neural network-based method for selecting surface type of object

1. An artificial neural network-based method for selecting a surface type of an object, suitable for selecting a surface type of a plurality objects, the method comprising:

individually training a plurality of predictive models by performing surface type identification on a same plurality of object images to obtain a determination defect rate of each of the predictive models, wherein the plurality of object images are images of surfaces of some objects among the plurality of objects;

cascading the plurality of predictive models in an order determined according to the respective determination defect rates of the predictive models into an artificial neural network system; and

feeding object images of the remaining objects of the plurality of objects to the artificial neural network system, so as to select the surface type of the remaining objects.

2. The artificial neural network-based method for selecting a surface type of an object of claim 1 , further comprising:

transforming each of the object images to a matrix;

wherein, one of the plurality of predictive models performs the surface type identification by using the matrix.

3. The artificial neural network-based method for selecting a surface type of an object of claim 1 , further comprising:

normalizing the plurality of object images; and

transforming the plurality of normalized object images to the matrices;

wherein, one of the plurality of predictive models performs the surface type identification by using each of the matrices.

4. The artificial neural network-based method for selecting a surface type of an object of claim 1 , further comprising:

respectively superimposing a plurality of object images of the same object into a plurality of initial images;

wherein, one of the plurality of predictive models performs the surface type identification by using each of the initial images.

5. The artificial neural network-based method for selecting a surface type of an object of claim 1 , further comprising:

respectively superimposing a plurality of object images of the same object into a plurality of initial images; and

transforming each of the initial images into a matrix;

wherein, one of the plurality of predictive models performs the surface type identification by using each of the matrices.

6. The artificial neural network-based method for selecting a surface type of an object of claim 1 , wherein each of the predictive models is implemented by a convolutional neural network (CNN) algorithm.

7. The artificial neural network-based method for selecting a surface type of an object of claim 1 , wherein each of the object images is formed by combining a plurality of detection images.

8. The artificial neural network-based method for selecting a surface type of an object of claim 1 , wherein the plurality of predictive models have different quantities of neural network layers.

9. The artificial neural network-based method for selecting a surface type of an object of claim 1 , wherein the plurality of predictive models have different neuron configurations.

10. The artificial neural network-based method for selecting a surface type of an object of claim 1 , further comprising:

feeding the plurality of object images corresponding to the remaining objects to the artificial neural network system to perform the surface type identification.

11. The artificial neural network-based method for selecting a surface type of an object of claim 1 , further comprising:

selecting the remaining objects by prioritizing one among the predictive models that has a higher determination defect rate.

12. The artificial neural network-based method for selecting a surface type of an object of claim 1 , wherein the plurality of object images correspond to surface types of the plurality of objects with a known defect rate, and one of the respective determination defect rates of the predictive models is higher than the known defect rate.

13. The artificial neural network-based method for selecting a surface type of an object of claim 1 , further comprising:

performing a deep learning process by using different training conditions to build the plurality of predictive models.

14. The artificial neural network-based method for selecting a surface type of an object of claim 1 , further comprising:

performing a plurality of deep learning processes to respectively build the plurality of predictive models.

15. The artificial neural network-based method for selecting a surface type of an object of claim 1 , wherein the predictive models are cascaded in the order from highest determination defect rate to lowest determination defect rate.

16. The artificial neural network-based method for selecting a surface type of an object of claim 1 , wherein the plurality of predictive models are respectively applied in sub neural network systems of the artificial neural network configured in stages according to the determination defect rate of the respective predictive model, such that each of the object images of the remaining objects fed into the artificial neural network is sequentially categorized as normal or abnormal in stages, with only object images categorized as normal by a previous stage being inputted to a subsequent stage.

17. The artificial neural network-based method for selecting a surface type of an object of claim 16 , wherein the sub neural network systems are ordered from highest determination defect rate to lowest determination defect rate.