Defect AOI Detection
1.Ultrasonic image inspection: Use the result of ultrasonic to detect whether the internal structure meets the standard, and mark the color on the map according to the inspection result.
2.Dust detection: detect whether there is dust on the surface of the object, and mark the location of the dust. In the figure below, the identified dust is marked with a red circle.
3.Scratch detection: Detect whether there are scratches on the surface of the object, and mark the gray scratches. In the figure below, the identified scratches are marked with a green frame or the scratches are colored.
4.Appearance contour detection: Take out the contour of the object and check whether the contour is consistent with the sample picture.
5.Skew detection: Detect the angle of the object in the screen, once the object tilt angle is too large, a warning will appear.
6.Width detection: Detect the width of a specific part of the object. Once the width exceeds the standard, a warning will appear.
7.Anomaly detection Arch.: The input image will first go through the Teacher-Student network structure, the Student will focus on the Global information and the Local information, and the Teacher will focus on the Global information, and because the Teacher’s information is Global, the degree of confidence usually higher than Student, so the Teacher will refer to a small amount of Student information (those Local information) when the model is judged.
The model selection uses GAN (Generative Adversarial Network) to generate an adversarial network. Downsampling is used in the first half of the model to obtain image feature information, and then Upsampling is used in the second half to restore the inverted image on the Spatial domain and remove blemishes. The generator of the model will be responsible for generating a flawless image after restoration, the Discriminator will be responsible for judging whether the generated image is really flawless, and finally through GradCAM++ to restore the Artifact Location (defect location) that the model focuses on.