US 11,982,999 B2
Defect detection task processing method, device, apparatus and storage medium
Meijuan Zhang, Beijing (CN); Yaoping Wang, Beijing (CN); Zhaoyue Li, Beijing (CN); Yuanyuan Lu, Beijing (CN); Wangqiang He, Beijing (CN); Dong Chai, Beijing (CN); and Hong Wang, Beijing (CN)
Assigned to Beijing Zhongxiangying Technology Co., Ltd., Beijing (CN); and BOE TECHNOLOGY GROUP CO., LTD., Beijing (CN)
Appl. No. 17/429,013
Filed by Beijing Zhongxiangying Technology Co., Ltd., Beijing (CN); and BOE TECHNOLOGY GROUP CO., LTD., Beijing (CN)
PCT Filed Oct. 30, 2020, PCT No. PCT/CN2020/125439
§ 371(c)(1), (2) Date Aug. 6, 2021,
PCT Pub. No. WO2022/088082, PCT Pub. Date May 5, 2022.
Prior Publication US 2023/0030296 A1, Feb. 2, 2023
Int. Cl. G05B 23/02 (2006.01); G05B 13/02 (2006.01)
CPC G05B 23/0224 (2013.01) [G05B 13/028 (2013.01); G05B 2223/02 (2018.08)] 18 Claims
OG exemplary drawing
 
1. A defect detection task processing method, comprising:
receiving, by a server comprising at least one hardware processor, a detection task, and determining a task type of the detection task;
storing, by the server, the detection task in a task queue if the task type is a target task type; and
when a processor of the server is idle, executing, by the server, the detection task in a preset order and generating a feedback signal;
wherein the detection task of the target task type includes an inference task and a training task and executing the training task comprises:
modifying, by the server, configuration information according to a preset rule based on product information in the detection task, by: extracting, by the server, a product information field in the detection task to obtain the product information; and setting, by the server, a size of a training image, a number of training times, a number of test times, a defect threshold, and a learning rate decay strategy according to the preset rule based on the product information;
acquiring, by the server, training data comprising the training image and an initial model according to the product information;
modifying the size of the training image of the training data to a size corresponding to the size of the training image of the configuration information;
using, by the server, the modified training data to train the initial model according to the configuration information to obtain a target model; and
storing, by the server, the target model.