Method and device for testing product quality

Patent metadata

Field	Value
Publication number	US-12417528-B2
Application number	US-202017758788-A
Country	US
Kind code	B2
Filing date	Nov 18, 2020
Priority date	Jan 15, 2020
Publication date	Sep 16, 2025
Grant date	Sep 16, 2025

Abstract

A method and device for testing product quality are disclosed. The method for testing product quality comprises: acquiring an image of a product to be tested; testing the image by using a pre-trained neural network model to obtain a testing result output by the neural network model; when the testing result indicates that the product to be tested is a defective product, performing a secondary judgment on the testing result according to position information of defective feature pixels in the image in the testing result, and determining whether the product to be tested is qualified according to a secondary judgment result. The method has high test accuracy, ensures the quality of product and facilitates reducing the labor cost of test.

First claim

What is claimed is:

1. A method for testing product quality, comprising:
acquiring an image of a product to be tested;
testing the image by using a pre-trained neural network model to obtain a testing result output by the neural network model; and
when the testing result indicates that the product to be tested is a defective product, performing a secondary judgment on the testing result according to position information of defective feature pixels in the image in the testing result, and determining whether the product to be tested is qualified according to a secondary judgment result,
wherein performing a secondary judgment on the testing result according to position information of defective feature pixels in the image in the testing result comprises:
obtaining a contour of the defective feature pixels according to information of an effective region in the testing result, wherein the effective region is an image region including the defective feature pixels;
calculating an area value of the contour according to the pixels included in the figure of the contour, and maintaining the testing result of defective product when the area value is greater than a preset area threshold; and
when the area value is less than or equal to the preset area threshold, determining the product to be tested as a qualified product.

2. The method according to claim 1 , wherein performing the secondary judgment on the testing result according to position information of defective feature pixels in the image in the testing result comprises:
dividing the image into a target region and a non-target region according to a relationship between positions of defective features and product quality, wherein the target region refers to a region in the image where occurrence of defective feature affects the product quality, and the non-target region refers to a region in the image where occurrence of defective feature does not affect the product quality;
according to the position information of the defective feature pixels in the image in the testing result, matching positions of the defective feature pixels in the image with the target region and the non-target region respectively;
when the positions of the defective feature pixels in the image are successfully matched with the target region, maintaining the testing result of defective product; and
when the positions of the defective feature pixels in the image are successfully matched with the non-target region, judging the product to be tested as a qualified product.

3. The method according to claim 2 , wherein
the testing result further includes defect category information and confidence corresponding to a defect category when the product to be tested is a defective product;
performing the secondary judgment on the testing result according to position information of defective feature pixels in the image in the testing result comprises:
setting a defect threshold corresponding to the target region; and
comparing the confidence corresponding to the defect category with the defect threshold, and maintaining the testing result of defective product when the confidence of the defect category is greater than the preset defect threshold.

4. The method according to claim 1 , wherein
acquiring the image of the product to be tested comprises:
acquiring a first image and a second image of the product to be tested under different exposure duration of a camera respectively, wherein the exposure duration corresponding to the first image is less than the exposure duration corresponding to the second image;
testing the image by using a pre-trained neural network model to obtain a testing result output by the neural network model comprises:
testing the first image and the second image respectively by using the pre-trained neural network model; and
when at least one of the first image and the second image includes a defective feature, obtaining a testing result that the product to be tested is a defective product.

5. The method according to claim 1 , further comprising: training the neural network model before testing the image using the pre-trained neural network model, specifically comprising:
acquiring a first image and a second image of a defective product, and marking an effective region where a defective feature is located on the first image and the second image of the defective product respectively, wherein at least part of defective features contained in the first image and the second image are different, and the effective region includes defective feature pixels;
obtaining a first training sample according to the first image of the defective product, information of the effective region and defect category information corresponding to the defective feature, and training a first neural network model by using the first training sample to obtain a stable first neural network model; and
obtaining a second training sample according to the second image of the defective product, information of the effective region and defect category information corresponding to the defective feature, and training a second neural network model by using the second training sample to obtain a stable second neural network model.

6. The method according to claim 5 , wherein testing the first image and the second image respectively by using the pre-trained neural network model comprises:
using the first neural network model to test the first image of the product to be tested, when it is found that the first image of the product to be tested includes more than two defect categories, arranging the defect category in a preset order according to position information of each defect category appearing in the first image of the product to be tested, and determining a defect category as a representative defect category included in the first image according to an arrangement result;
using the second neural network model to test the second image of the product to be tested, when it is found that the second image of the product to be tested includes more than two defect categories, arranging the defect category in a preset order according to position information of each defect category appearing in the second image of the product to be tested, and determining a defect category as a representative defect category included in the second image according to an arrangement result; and
taking the representative defect category included in the first image or the representative defect category included in the second image as a defect category of the product to be tested, so as to classify the product to be tested according to the defect category of the product to be tested.

7. The method according to claim 1 , wherein
the product to be tested is a magnetic circuit product;
acquiring the image of the product to be tested comprises:
obtaining a black image and a white image of the magnetic circuit product;
testing the image by using a pre-trained neural network model to obtain a testing result output by the neural network model comprises:
testing the black image by using the pre-trained neural network model, and obtaining a testing result output by the neural network model that the black image includes at least one of the following defect categories: impurities, wool/fiber, crush and breakage; and
testing the white image by using the pre-trained neural network model, and obtaining a testing result output by the neural network model that the white image includes at least one of the following defect categories: impurities, wool/fiber, crush, breakage and glue overflow.

Assignees

• Goertek Inc

Inventors

• Yu Yangyang
• Li Shiyu

Classifications

• G06T2207/20081

  Training; Learning · CPC title

• G06T2207/10144

  Varying exposure · CPC title

• G06V10/44

  Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components · CPC title

• G06V10/751

  Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching · CPC title

• G06V10/82

  using neural networks · CPC title