Anomaly detection device for unknown defects

Abstract

An anomaly detection device for unknown defects includes: an anomaly detection module performing image inspection on a sample, based on a defect threshold, to obtain a first detection result determining that the sample is a defective product or a second result determining that the sample is a quasi-good product; an image processing module detecting a dimensional defects on the second detection result based on a size threshold; an image segmentation processing module performing image segmentation on the second detection result to recognize a defective object on the sample; a post-processing module performing a post-processing to the recognized defective object; and a judgement module performing a physical quantity determination on the defective object which has been processed by the post-processing module.

Claims

1. An anomaly detection device for unknown defects, comprising: an anomaly detection module for performing image detection on a sample to be tested, and obtaining a first detection result indicating that the sample to be tested is a defective product or a second detection result indicating that it is a near-good product based on a defect threshold; an image processing module for detecting dimensional defects in the second detection result based on a size threshold; an image segmentation processing module for performing image segmentation on the second detection result to identify defective objects in the sample; a post-processing module for performing post-processing on the identified defective objects; and a judgment module for performing a physical quantity judgment on the defective objects processed by the post-processing module.

2. The anomaly detection device for the unknown defect as described in claim 1, wherein, The anomaly detection module is a deep learning model or machine learning model trained using pre-collected OK images as training images, and the OK images are images of multiple samples that have been confirmed as good products.

3. The anomaly detection device for unknown defects as described in claim 2, wherein, During the training process, the anomaly detection module further randomly generates defects and noise on the OK image to produce a simulated NG image, and simultaneously uses the OK image and the simulated NG image as training images; and after training, the anomaly detection module has the defect threshold, thereby gaining the ability to identify the test sample as a near-good product.

4. An anomaly detection device for an unknown defect as described in claim 2, wherein, The training of the anomaly detection module can be carried out using the OK image and multiple pre-collected NG images, i.e. defective images, thereby accelerating the convergence of the training process.

5. An anomaly detection device for an unknown defect as described in claim 3 or 4, wherein, The OK image and the simulated NG image or the NG image were not pre-labeled.

6. An anomaly detection device for an unknown defect as described in claim 1, wherein, The sample was judged as defective, meaning that the sample was found to have medium to large defects that exceed the defect threshold.

7. An anomaly detection device for an unknown defect as described in claim 1, wherein, The image processing module is an automatic optical inspection device that checks the second detection result for size defects based on the size threshold and decides whether to adjust the control conditions and parameters.

8. An anomaly detection device for an unknown defect as described in claim 1, wherein, The image segmentation processing module is a deep learning model or machine learning model trained using multiple pre-collected defective images of samples as training data, and the defective images are pre-labeled.

9. An anomaly detection device for an unknown defect as described in claim 1, wherein, The post-processing module is an automatic optical inspection device that inspects the identified defective area and determines whether to further adjust the control conditions and parameters.

10. An anomaly detection device for an unknown defect as described in claim 1, wherein, The judgment module further performs the physical quantity judgment on the defect inspected by the image processing module.

Images