CN112070710A

CN112070710A - Method for detecting industrial product defect image

Info

Publication number: CN112070710A
Application number: CN201910530470.1A
Authority: CN
Inventors: 不公告发明人
Original assignee: Beijing Pingheng Intelligent Technology Co ltd
Current assignee: Beijing Pingheng Intelligent Technology Co ltd
Priority date: 2019-06-10
Filing date: 2019-06-10
Publication date: 2020-12-11

Abstract

The industrial product defect image detection method is divided into two parts which are sequentially executed: (1) firstly, acquiring a certain number of industrial product images without defects in real time, and carrying out statistical analysis on the industrial product images to obtain an acceptance domain; (2) then analyzing the known sample image (i.e. knowing whether the sample has defects) to obtain the number of pixel points with the gray value of 255, and when the number value is within the range of the acceptance domain, judging the sample as a normal sample without defects by the method; when the magnitude value is outside the range of the acceptance domain, the method judges the sample as a defective sample. The method uses the accuracy as an index to measure the quality of the forecast. The method inhibits the interference of imaging noise caused by various errors in the industrial image acquisition process to industrial detection to a certain extent, and has good detection effect on the film defects with small background and target gray difference.

Description

Method for detecting industrial product defect image

Technical Field

The invention relates to the field of industrial product defect detection, in particular to a method for detecting industrial product defect images.

Background

For industrial product defect image detection, selecting a proper image segmentation threshold value is important. If the threshold value is selected too large, the target defect is misjudged as the background, and the condition of missing detection occurs; on the contrary, if the threshold is too small, the background is erroneously determined as the target defect, and the normal product is erroneously determined as the defect. The film defects (generally including scratches, creases, sharp concave-convex points, bubbles and the like) in industrial products are detected, because the gray difference between a target defect and a background is very small, an image segmentation threshold value is not easy to select, and in addition, the influence of random errors, system errors and large errors caused by human factors in the process of acquiring images in real time in the industry causes difficulty in detecting the defects of the industrial film to a certain extent.

At present, a common image segmentation algorithm is that people like wang jin jiang determine a segmentation threshold value by a bimodal method through smooth filtering and histogram analysis, and the film defect image is simply and effectively segmented. However, this method is only effective when the gray values of the target defect are very different from the background and the histogram shows significant double peaks. The background difference algorithm described by harlefa (a.khalifa) et al has a narrow application range, and the conventional background difference algorithm is effective only when the background is constant, but is not suitable for the situation where the background image changes slowly or violently. The maximum entropy threshold segmentation method is proposed by Kapur and the like, and the entropy relates to logarithm operation, so that the time overhead is large, and the requirement on real-time performance is difficult to meet.

Disclosure of Invention

In order to solve the problems, the invention provides an industrial product defect image detection method based on statistical analysis, which eliminates the influence of various errors in the process of acquiring images in real time in industry to imaging to a certain extent, can provide a better image segmentation threshold value for the defects of the industrial thin film with small difference between the target defects and the background gray level, and realizes the judgment on whether the industrial thin film has defects or not by adjusting parameters and selecting a proper acceptance domain.

The solution of the invention is divided into two parts which are executed in sequence: (1) firstly, acquiring a certain number of industrial product images without defects in real time, and carrying out statistical analysis on the industrial product images to obtain an acceptance domain; (2) then analyzing the known sample image (i.e. whether the known sample has defects) to obtain the number of pixel points with the gray value of 255, and when the numerical value is within the range of the acceptance domain, judging the sample as a normal sample without defects; when the magnitude value is outside the range of the acceptance domain, the sample is determined to be a defective sample.

Drawings

FIG. 1 is a schematic flow chart of a method of industrial defect image detection of the present invention;

FIG. 2 is a captured image of an industrial film without a defect site;

FIG. 3 is an image of an industrial film with scratch defects;

FIG. 4 is a view of the flowchart of FIG. 3, which is compiled after being written in vs2013 by c + +, and the defect portion is white;

FIG. 5 is an image of an industrial film with a crease defect, the crease being located within a red circle;

FIG. 6 is a diagram of an image compiled after being written in vs2013 by c + + according to the flow shown in FIG. 1 and generated by FIG. 5, wherein a defect part is white and is located in a red circle;

FIG. 7 is an image of an industrial film with sharp asperity defects;

FIG. 8 is a view of the flowchart of FIG. 7, which shows the defect portion is white, in a view of the image compiled after being written in vs2013 in c + +, according to the flow shown in FIG. 1;

Detailed Description

The method is further illustrated below with reference to the examples and the description of the figures.

1) Firstly, counting the industrial thin film images of positive samples (samples without defects) with the same size collected in real time, and calculating the average value of the gray value of each pixel point of the images. In the present embodiment, the number of the collected positive sample images is 5, one of which is shown in fig. 2;

2) further calculating the standard deviation of the gray value of each pixel point;

3) multiplying an adjustable parameter by the standard deviation of the gray value of each pixel point to obtain the threshold value of each pixel point;

4) and traversing each pixel point of each image in the positive sample image set, and comparing the separation value (the absolute value of the difference between the gray value and the average value) of the pixel points at the same position with the threshold value. If the dispersion value is larger than the threshold value, setting the gray value of the pixel point to be 255; if the dispersion value is less than or equal to the threshold value, setting the gray value of the pixel point to be 0;

5) counting the number of pixel points with the gray value of 255 of each image in the positive sample image set;

6) in order to improve the prediction accuracy, in this embodiment, corresponding adjustment parameters are finally determined by continuously debugging different types of defects, and of course, the mean value of the adjustment parameters corresponding to different types of defects may also be taken as a uniform adjustment parameter, so as to perform uniform processing on the different types of defects, and the mean value of the number of pixels whose image gray values in the positive sample image set are equal to 255 is obtained on the basis of 5), and the mean value smaller than or equal to the mean value is selected as an acceptance domain, and the mean value larger than the mean value is selected as a rejection domain;

7) finally, in the present embodiment, three industrial thin film defect sample images as shown in fig. 3, 5, and 7 collected under the same condition are tested by the method, the number of pixels with the calculated image gray value equal to 255 falls within the reject region, and the samples determined to have defects are determined to have defects, and the defects are marked with white marks as shown in fig. 4, 6, and 8, respectively.

The final result shows that the method has obvious segmentation effect on the industrial thin film defect image with small difference between the target defect and the background gray value. With the increase of the number of the positive sample images acquired in real time, the segmentation effect is more obvious, and the prediction accuracy is further improved.

Claims

1. The method for detecting the industrial product defect image is characterized by being divided into two parts which are sequentially executed:

1) firstly, acquiring a certain number of industrial product images without defects in real time, and carrying out statistical analysis on the industrial product images to obtain an acceptance domain;

2) analyzing a known sample image (namely whether a known sample has defects) to obtain the number of pixel points with the gray value of 255, and judging the sample as a normal sample without defects when the numerical value is within the range of an acceptance domain; when the magnitude value is outside the range of the acceptance domain, the method judges the sample as a defective sample.

2. The industrial product defect image detection method according to claim 1, characterized in that:

1) counting positive sample (sample without defects) images which are collected in real time and have the same size, and calculating the average value of the gray value of each pixel point of the images;

4) traversing each pixel point of each image in the positive sample image set, comparing the deviation value (the absolute value of the difference between the gray value and the average value) of the pixel point at the same position with the threshold value, and if the deviation value is greater than the threshold value, setting the gray value of the pixel point to be 255; if the dispersion value is less than or equal to the threshold value, setting the gray value of the pixel point to be 0;

5) and counting the number of pixels with the gray value of 255 in each image in the positive sample image set so as to determine a proper receiving domain.

3. The method as claimed in claim 1, wherein the accuracy index is used to measure how well the method predicts, and in order to improve the prediction accuracy, different adjustment parameters and acceptance domains need to be tested until the result is satisfactory.