CN113516619B - Product surface flaw identification method based on image processing technology - Google Patents

Abstract

The invention discloses a method for identifying product surface defects based on image processing technology. First, a product detection point is set on the product assembly line, and a high-definition camera is used to shoot the product at the monitoring point to obtain an image of the product; secondly, based on the obtained product Image, the grayscale operation is performed on the image to obtain the grayscale image; the defect point recognition algorithm is used to calculate and output the area of the defect point and the coordinates of the center point on the image for the grayscale image again; finally, the laser marking machine is triggered according to the output value, and the product Laser labeling is carried out to mark the unqualified words. The defect identification method based on computer processing proposed by the invention can well identify product surface defect points, and does not generate misjudgment phenomenon for products meeting product quality requirements with smooth surfaces. It can solve the problem of checking whether auto parts are qualified in industrial production, improve industrial production efficiency, save costs, and is suitable for being extended to factory assembly line product testing.

Description

A product surface defect recognition method based on image processing technology

technical field

The invention relates to the technical field of image processing, in particular to a product surface defect recognition method based on image processing technology.

Background technique

In the production process of auto parts, human eyes are often used to identify surface defects of auto parts. However, relying on manual methods has a high false detection rate, and the accuracy of detection is affected by the subjective judgment and fatigue of workers. In order to improve the accuracy of detection, it has become a development trend to use machine vision to replace traditional human vision detection methods.

In actual work, the defect points of the product belong to the local features of the image, and the local feature values of the image do not change with the rotation, translation, affine, etc. of the picture. At present, the main algorithms in local feature detection include AKAZE, KAZE, BRISK or SIFT and other algorithms. Both are binary descriptor algorithms, each with its own advantages and disadvantages. AKAZE and KAZE are non-linear algorithms that take a long time to process images and are not suitable for use on industrial assembly lines. The BRISK or SIFT algorithm is a linear algorithm with a fast processing speed, but from the analysis of the pictures of the experimental results, it is easy to identify the non-defective part as a feature, resulting in an increased probability of misjudgment.

In addition, the texture of the product surface is a unique feature of the surface of the object. The texture can be used to study the spatial dependence of the image and analyze the surface characteristics of the object. At present, texture detection algorithms include Tamura texture analysis method, Gabor wavelet texture feature extraction method, LBP texture statistical feature extraction and other methods. In the course of the experiment, by constructing the GABOR filter, multiple experiments were carried out on the surface image of the defective product to verify. The left image of the original image was processed by the GABOR filter to obtain the gabor image. From the image, the texture similarity of the product is extremely high , product imperfections cannot be identified through texture analysis of images.

Contents of the invention

In view of the above existing problems, the present invention provides a product surface defect recognition method based on image processing technology, which can quickly identify product surface defects without causing misjudgment of products that meet product quality requirements and have a smooth surface.

The technical solution that realizes the object of the present invention is:

A product surface defect recognition method based on image processing technology, characterized in that it comprises the following steps:

Step 1: Set product inspection points on the product assembly line, use high-definition cameras to shoot products at the monitoring points, and obtain product images;

Step 2: Based on the obtained product image, perform a grayscale operation on the image to obtain a grayscale image;

Step 3: Calculate and output the area of the defect point and the coordinates of the center point on the image using the defect point recognition algorithm for the grayscale image;

Step 4: Trigger the laser marking machine according to the output value, carry out laser marking on the product, and mark the unqualified words.

Further, the specific operation steps of the blemish point identification algorithm described in step 3 include:

Step 31: Perform normalized quantization processing on the input grayscale image;

Step 32: Set the scale of the Gaussian filter, use the Gaussian filter to perform a Gaussian convolution operation, and perform noise reduction processing on the normalized quantized image;

Step 33: Perform binarization operations on the images of different scales after noise reduction, and when the pixel is greater than the set threshold minThreshold for obtaining image pixel corners, obtain the image pixel corners, and filter out the pixel values of non-defective points in the image;

Step 34: For the multiple processed binarized pictures obtained, use the library function findContours () of OpenCV to calculate multiple connected domains, and calculate the center coordinates and radius of each connected domain;

Step 35: According to the center coordinates and radius obtained in step 34, if the pixels are superimposed or connected, all of them are put together to form a large connected domain, and finally multiple connected regions are divided at the pixel level;

Step 36: Determine the distance between the edge points of two connected regions, and judge whether the two connected regions are classified into a group, and if so, they are regarded as a block, otherwise they are regarded as separate connected regions;

Step 37: Calculate the size of each block, and if the value is greater than the set threshold, the feature of the block is regarded as a product defect point.

Further, the Gaussian filter scale value is set to 5.

Compared with the prior art, this method has the following beneficial effects:

The product surface defect recognition method based on image processing technology proposed by the present invention, through the defect point recognition algorithm proposed in the present invention, can identify the number of defect points well, achieve accuracy, and solve the problem of checking whether auto parts are qualified in industrial production The problem is to avoid the long time spent on manual inspection and the high detection error rate, thereby improving industrial production efficiency and saving costs, and is suitable for promotion to factory assembly line product inspection.

Description of drawings

Fig. 1 is a schematic diagram of the processing flow of the defect point identification algorithm in the present invention;

Fig. 2 is product surface texture treatment effect among the present invention;

Fig. 3 is a comparative figure of eigenvalue detection algorithm experimental results among the present invention;

Fig. 4 (a)-(b) is the effect figure after sample 1 processing in the embodiment of the present invention;

Figure 5(a)-(b) are the effect diagrams of sample 2 in the embodiment of the present invention after treatment.

Detailed ways

In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

A product surface flaw recognition method based on image processing technology proposed by the present invention, its operation steps include:

A product surface defect recognition method based on image processing technology, comprising the following steps:

Step 1: Set product inspection points on the product assembly line, set the camera sampling interval time according to the assembly line speed, and use high-definition cameras to shoot products at the monitoring points to obtain product images;

Step 2: Based on the obtained product image, store it in the computer, and perform a grayscale operation on the image to obtain a grayscale image;

Step 3: Calculate and output the area of the defect point and the coordinates of the center point on the image using the defect point recognition algorithm for the grayscale image;

Step 4: Trigger the laser marking machine according to the output value, carry out laser marking on the product, and mark the unqualified words.

Preferably, referring to accompanying drawing 1, it can be seen that the operation steps of the blemish point identification algorithm include:

Step 31: Perform normalization and quantization processing on the input grayscale image, and the features of the processed grayscale image are invariant to affine transformations such as translation, rotation, and scaling;

Step 32: Set the scale of the Gaussian filter, use the Gaussian filter to perform a Gaussian convolution operation, and perform noise reduction processing on the normalized quantized image;

Step 33: Perform multiple binarization operations on the images of different scales after noise reduction, and when the pixel is greater than the set threshold minThreshold of the image pixel corner (eigenvalue), obtain the image pixel corner (eigenvalue);

Preferably, the value of minThreshold is set to 0.0001;

Step 34: For the processed binarized image, use the findContours function method to generate multiple connected domains, and calculate the center of each connected domain;

Described findContours function is the function method that is used to identify the outline of target, and it is the library function of OpenCV, by setting the parameter of this library function, obtains a plurality of connected domains; The prototype of this findContours function is findContours(image, contours, hierarchy ,mode,method,offset=Point()), the input of each parameter in this application is:

image: The input is the processed binarized image;

Contours: It is a vector of a set of points composed of Point points. Each set of Point points is a contour. As many contours as there are, there are as many elements in the vector contours;

hierarchy: There is a one-to-one correspondence between the elements of the vector and the elements in the contour vector contours, the default value is -1;

mode: is the retrieval mode of the outline, and its optional values are: CV_RETR_EXTERNAL, CV_RETR_LIST, CV_RETR_CCOMP, CV_RETR_TREE, and this parameter is set to CV_RETR_EXTERNAL in this application;

method: the approximation method used to define the profile, which can be selected as: CV_CHAIN_APPROX_NONE, CV_CHAIN_APPROX_SIMPLE, CV_CHAIN_APPROX_TC89_L1, CV_CHAIN_APPROX_TC89_KCOS, this parameter is set to CV_CHAIN_APPROX_NONE in this application;

point: The offset of all contour information relative to the corresponding point of the original image, which is equivalent to adding the offset to each detected contour point, and Point can also be a negative value;

Step 35: Among the binarized images obtained after multiple binarization operations in step 34, those with values are considered to be connected pixels, which can form a connected region of pixels, each region has a center coordinate and radius, according to the Center coordinates and radius, if the pixels are superimposed or connected, they will all be put together to form a large connected area, and finally multiple connected areas will be divided at the pixel level;

After being processed in step 35, the defect point is composed of a plurality of connected pixels at the pixel level;

Step 36: On the image coordinates, judge the distance between the edge points of two connected regions. If the shortest distance is less than minDist (referring to the shortest distance on the contour of the identified defect point), it is considered continuous, and thus judge the two Whether the connected area is classified as a group, if so, it is regarded as a block, otherwise it is regarded as a separate connected area;

Step 37: Calculate the size of each block. If the value is greater than the set threshold threshold, the block feature is regarded as a product defect point. In this application, the value of the threshold threshold is 1280.

Example

1. Experimental environment

Operating system: WIN 10; development platform: Python3.8+Opencv+Jupyternotebook;

CPU: Quad-core Core(TM)_i7, GeForce RTX2060SUPER.

2. Performance analysis and comparative effect

Conduct experiments with reference to steps 1-4 in the technical solution, and compare the inventive algorithm with the local feature recognition algorithm and texture extraction algorithm through the results. In terms of extracting defect point features, the local feature recognition algorithm extracts too many feature values. As shown in Table 1, some non-product defect points will also be extracted as features. Combined with reference to Figure 3, the small colored circles or points on the figure are the identified eigenvalues. It can be seen from the figure that from the experimental results of some current mainstream local eigenvalue extraction algorithms, the characteristics of the extracted defect points , the local feature recognition algorithm extracts too many feature values, and cannot correctly identify defect points. Because the texture of the defect point on the product is very similar to the background texture, it is very difficult for the texture algorithm to extract the feature of the defect point, and it is impossible to identify the defect point of the product, as shown in Figure 2.

Table 1: Comparison of the inventive algorithm and the eigenvalue algorithm for extracting feature points

Sample 1 in Table 1 is a picture captured by a camera with a resolution of 665×1037, 24-bit true color, and a size of 92.9K, as shown in Figure 4(a). After processing, the effect of Figure 4(b) is obtained, marked by a computer program, and the red circle in the picture is the defect point of the product.

Sample 2 is a picture captured by a camera with a resolution of 4032×3024, 24-bit true color, and a size of 1.16M, as shown in Figure 5(a). After being processed by the inventive algorithm, the effect of Fig. 5(b) is obtained. Marked by computer program, the red circle in the picture is the defect point of the product.

The content not described in detail in this specification belongs to the prior art known to those skilled in the art. Although the patent of the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments, or perform equivalent replacements for some of the technical features. Within the spirit and principles of the present invention, any modifications, equivalent replacements, improvements, etc., shall be included within the protection scope of the present invention.

Claims (2)

1. A product surface flaw identification method based on an image processing technology is characterized by comprising the following steps:

step 1: setting product detection points on a product production line, and shooting the products at the monitoring points by using a high-definition camera to obtain images of the products;

and 2, step: performing gray scale operation on the picture based on the obtained product image to obtain a gray scale image;

and step 3: calculating and outputting the area of the flaw point and the coordinate of the central point on the image by using a flaw point identification algorithm aiming at the gray level image;

and 4, step 4: triggering a laser marking machine according to the output value, carrying out laser marking on the product, and marking an unqualified typeface;

the specific operation steps of the fault point identification algorithm in the step 3 comprise:

step 31: carrying out normalization quantization processing on an input gray level image;

step 32: setting the scale size of a Gaussian filter, performing Gaussian convolution operation by using the Gaussian filter, and performing noise reduction processing on the image subjected to normalization quantization processing;

step 33: carrying out binarization operation on the noise-reduced images with different scales, and when the pixel is larger than a set threshold minThreshold for obtaining a pixel corner point of the image, obtaining the pixel corner point of the image, and filtering out the pixel value of a non-defective point in the image;

step 34: calculating a plurality of connected domains by using a library function findContours () of OpenCV for the obtained plurality of processed binary pictures, and calculating the center coordinate and the radius of each connected domain;

step 35: if the pixels are overlapped or connected according to the central coordinates and the radius obtained in the step 34, all the pixels are put together to form a large connected domain, and finally a plurality of connected domains are divided on the pixel level;

step 36: judging the distance between the edge points of the two connected regions, judging whether the two connected regions belong to a group, if so, taking the two connected regions as a block, otherwise, taking the two connected regions as separated connected regions;

step 37: and calculating the size of each block, and regarding the block characteristic as a product flaw when the value is larger than a set threshold value.

2. The method of claim 1, wherein the gaussian filter scale value is set to 5.

Images