AI Training database for COF Film defects and application of AI Training database

Abstract

The invention discloses an AI Training database generation method, an AI Training database, a COF Film defect detection method and a dynamic AI Training database. The AI Training database generation method comprises the following steps: S11, manually distinguishing an image set; S12, taking the image set as a Training material, and reading the Training material into an AI Training Server for learning; S13, forming a Training database V0; S14, taking an image as an inspection image; and S15, judging the inspection image by using the AI Training Server having performed learning for judgment, andmanually re-judging the judgment result of the AI Training Server. According to the method, artificial intelligence is used for self-learning to recognize COF Film defects, so naked-eye judgment of personnel can be replaced, labor cost is saved, and meanwhile, and the productivity of a detection station is improved.

Description

technical field [0001] The invention relates to a COF Film production process, in particular to an AI Training database generation method, an AITraining database, a COF Film defect detection method and a dynamic AI Training database. Background technique [0002] In the COF Film production process, the product testing process is as follows: figure 1 After the tin plating process, there will be an optical inspection station AOI (Automated Optical Inspection), and there will be an optical inspection station AVI (Automated Visual Inspection) after the product process is completed. After the inspection is completed, photos of suspected NG will be taken for file , the next site is to manually re-judgment the photos left in the optical inspection file. [0003] The detection process has the following deficiencies: [0004] ① High demand for manpower; [0005] ② The level of manpower detection ability is uneven, which is easy to cause Overkill or Underkill; [0006] ③ For those...

AI Technical Summary

Problems solved by technology

[0004]①There is a lot of manpower demand;

[0005]②The level of human detection ability is uneven, which may easily lead to Overkill or Underkill;

[0006]③ For those with poor rejudgment ability, the manpower training cycle is longer and the cost is higher;

[0007]④The Tact Time of the detection site is longer;

[0008]⑤ For the feedback and analysis of the main bad modes, it takes more time for experienced personnel to sort out the results, and the real-time performance is poor, which is not conducive to timely improvement of the process

Images