Defect inspection device

1. A defect inspection device comprising:

a robot arm including a hold unit for holding an object and a driving unit for moving the object;

a first camera unit photographing an exterior of the object;

an illumination unit irradiating light to the exterior of the object; and

a control unit determining whether there is a defect in the object based on an image of the object photographed by the first camera unit, wherein the control unit determines whether there is the object based on two or more images photographed under different illumination conditions, respectively, and

wherein the control unit

calculates a first normal image photographed under a first illumination condition and a second normal image photographed under a second illumination condition which are normal images which are a basis of anomaly determination by using both images as inputs of channels included in a first sub model included in a defect inspection model, respectively,

calculates a first determination object image photographed under the first illumination condition and a second determination object image photographed under the second illumination condition which are anomaly determination object images by using both images as inputs of channels included in a second sub model included in the defect inspection model, respectively, and

determines whether there is the defect in the object based on a calculation result of the first sub model and the second sub model.

2. The defect inspection device of claim 1 , further comprising:

a display unit displaying at least one image of the image of the object photographed by the first camera unit and an image in which whether there is the defect in the object is displayed.

3. The defect inspection device of claim 1 , further comprising:

a transport unit transporting the object for defect inspection; and

a classification transport unit classifying and transporting the object based on a result of the defect inspection of the control unit for the object.

4. The defect inspection device of claim 1 , wherein the driving unit includes at least one driving shaft for rotating the object.

5. The defect inspection device of claim 1 , wherein the driving unit adjusts an arrangement state of the object so that the first camera unit photographs the object in a different arrangement state to acquire images photographed at different timings.

6. The defect inspection device of claim 1 , further comprising:

a second camera unit for photographing the object before being held by the hold unit,

wherein the control unit allows the hold unit to be aligned to hold the object based on the arrangement state of the object based on the image of the object acquired by the second camera unit.

7. The defect inspection device of claim 1 , wherein the illumination unit includes two or more sub illumination units irradiating light, and the illumination unit has a space formed therein,

the first camera unit is located on one side of the space and photographs an object arranged on the other side of the space by the robot arm through the space, and

the control unit allows the robot arm to arrange the object on the other side of the space so that the first camera unit photographs the object.

8. The defect inspection device of claim 7 , wherein in the sub illumination unit, a light emitting unit configured in a corn or polypyramid shape, and irradiating light to an internal surface of a horn shape is located.

9. The defect inspection device of claim 7 , wherein the two or more sub illumination units are configured in the corn or polypyramid shape, and ratios of widths of lower surfaces and widths of upper surfaces of the horn shape of respective sub illumination units are different from each other.

10. The defect inspection device of claim 1 , wherein the illumination unit irradiates two or more different types of light to the exterior of the object.

11. The defect inspection device of claim 1 , wherein the control unit determines whether there is the defect in the object based on two or more images acquired by photographing an object in a first arrangement state under different illumination conditions and two or more images acquired by photographing an object in a second arrangement state under different illumination conditions.

12. The defect inspection device of claim 1 , wherein the control unit sets the two or more images as inputs of two or more channels included in the defect inspection model including one or more network functions, respectively to extract features of the two or more input images, respectively by using the defect inspection model, and determines whether there is anomaly for each of the two or more images based on the extracted features.

13. The defect inspection device of claim 12 , wherein when it is determined that there is the anomaly in at least one image of the two or more images, the control unit determines that there is the defect in the object.

14. The defect inspection device of claim 1 , wherein the defect inspection model further includes a comparator connected to at least one of the first sub model and the second sub model in series.

15. The defect inspection device of claim 1 , wherein a first channel of the first sub model into which the first normal image is input shares at least one link having the same weight as a first channel of the second sub model into which the first determination object image is input, and

a second channel of the first sub model into which the second normal image is input shares at least one link having the same weight as a second channel of the second sub model into which the second determination object image is input.

16. The defect inspection device of claim 1 , wherein the control unit

generates comparison information of at least one first layer based on at least one layer of the first channel of the first sub model and at least one layer of the first channel of the second sub model, and transfers the first layer comparison information to a corresponding layer of the first channel of the comparator to calculate anomaly information of the first determination object image, and

generates comparison information of at least one second layer based on at least one layer of the second channel of the first sub model and at least one layer of the second channel of the second sub model, and transfers the second layer comparison information to a corresponding layer of the second channel of the comparator to calculate the anomaly information of the second determination object image.