Battery pack injection mold production defect detection and identification method

Abstract

The invention relates to the technical field of image detection, in particular to a battery pack injection mold production defect detection and identification method. Obtaining a first suspected defect degree according to gray level distribution characteristics in a preset neighborhood window of edge pixel points in the appearance image and change characteristics of the edge lines; obtaining a second suspected defect degree according to a texture distribution feature and a texture strength feature in a preset neighborhood window of the edge pixel point, and a gray difference feature and a gradient difference feature between the preset neighborhood window and a preset neighborhood large window; and obtaining an enhancement coefficient of the edge pixel point according to the first suspected defect degree and the second suspected defect degree. According to the invention, the preset sharpening intensity is adjusted according to the enhancement coefficient and the gradient features of the edge pixel points, the appearance image is sharpened and enhanced according to the adaptive sharpening intensity, and the injection mold defect is analyzed according to the sharpened and enhanced appearance image, so that the accuracy of defect identification is improved.

Description

Technical Field

[0001] The present invention relates to the field of image detection technology, and in particular to a method for detecting and identifying production defects of a battery pack injection mold. Background Art

[0002] In the battery pack production process, injection molds are key components in manufacturing the outer shell and other parts, making quality control and inspection crucial. Because defects in battery pack injection molds typically occur internally, defect identification can be performed through trial production of molded parts. Internal mold defects can lead to defects such as flash, flow marks, and shrinkage cavities. Because the presence of defects in molded parts is not only influenced by mold quality but also by process parameters, small batches of molded parts are necessary to improve the accuracy of defect analysis. Defects can be identified based on the appearance of a specific number of molded parts.

[0003] For multiple injection molded parts, machine vision technology can be used for batch defect detection. To accurately identify subtle defects, the appearance image needs to be enhanced to highlight the defect features. The existing USM unsharp masking algorithm is an image enhancement technology that can effectively highlight the edge features of injection molded part defects, especially for low-contrast defects such as tiny textures and burrs. In traditional USM algorithms, the sharpening strength is a fixed parameter. Excessive sharpening strength will lead to excessive enhancement of strong edges and amplification of noise; while too low sharpening strength will lead to insufficient enhancement of defect details and blurred edges. Therefore, a fixed sharpening strength makes it difficult to balance noise suppression and defect enhancement, ultimately reducing the accuracy of injection molded part and mold defect detection. Summary of the Invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a method for detecting and identifying production defects of battery pack injection molds. The technical solutions adopted are as follows: Obtain appearance images of battery pack injection molded parts; A first suspected defect degree is obtained based on the grayscale distribution characteristics and edge line change characteristics within a preset neighborhood window of the edge pixel points in the appearance image; a second suspected defect degree is obtained based on the texture distribution characteristics and texture intensity characteristics within the preset neighborhood window of the edge pixel points, and the grayscale difference characteristics and gradient difference characteristics between the preset neighborhood window and the preset neighborhood large window; Obtaining an enhancement coefficient for the edge pixel point according to the first suspected defect degree and the second suspected defect degree; adjusting a preset sharpening intensity according to the enhancement coefficient and a gradient feature of the edge pixel point to obtain an adaptive sharpening intensity; According to the adaptive sharpening intensity, the appearance image is sharpened and enhanced, defect recognition is performed according to the appearance image after sharpening and enhancement, and injection mold defects are analyzed according to a defect recognition result.

[0005] Further, the step of obtaining the first suspected defect degree of the edge pixel point according to the gray scale distribution feature in the preset neighborhood window of the edge pixel point in the appearance image and the change feature of the edge line comprises: According to an absolute value of a difference between the number of two gray scales with the most pixel points in the gray scale histogram of the preset neighborhood window, a gray scale uniformity degree is obtained; a reciprocal of a variance of curvatures of all edge pixel points on the edge line in the preset neighborhood window is calculated to obtain an edge regularity degree; and a product of the gray scale uniformity degree and the edge regularity degree is calculated and negatively correlated to obtain the first suspected defect degree of the edge pixel point.

[0006] Further, the step of obtaining the second suspected defect degree of the edge pixel point according to the texture distribution feature and the texture intensity feature in the preset neighborhood window of the edge pixel point, the gray scale difference feature and the gradient difference feature between the preset neighborhood window and the preset neighborhood large window comprises: A product of an energy of a gray scale co-occurrence matrix of the preset neighborhood window and an inverse difference matrix is calculated to obtain a texture uniformity feature value; a product of an information entropy of a gray scale value in the preset neighborhood window and an information entropy of a gradient direction is calculated to obtain a window feature value; a reciprocal of an absolute value of a difference between the window feature values of the preset neighborhood window and the preset neighborhood large window is calculated to obtain a region feature similarity degree; an average value of absolute values of gray scale differences between all edge points in the preset neighborhood window and adjacent pixel points in the gradient direction is calculated to obtain a local gray scale difference value; and a product of the texture uniformity feature value, the region feature similarity degree and the local gray scale difference value is calculated and negatively correlated to obtain the second suspected defect degree of the edge pixel point.

[0007] Further, the step of obtaining the enhancement coefficient of the edge pixel point according to the first suspected defect degree and the second suspected defect degree comprises: An average value of the first suspected defect degree and the second suspected defect degree is calculated to obtain the enhancement coefficient of the edge pixel point.

[0008] Further, the step of adjusting the preset sharpening intensity according to the enhancement coefficient and the gradient feature of the edge pixel point to obtain the adaptive sharpening intensity comprises: A product of a reciprocal of the gradient value of the edge pixel point and the enhancement coefficient is calculated and normalized to obtain an adjustment degree; and a product of the adjustment degree and a maximum value of the preset sharpening intensity is calculated to obtain the adaptive sharpening intensity of the edge pixel point.

[0009] Furthermore, the step of performing sharpening enhancement on the appearance image according to the adaptive sharpening strength includes: The appearance image is sharpened and enhanced according to the adaptive sharpening strength and the USM algorithm.

[0010] The present invention has the following beneficial effects: In the present invention, since the grayscale difference characteristics on both sides of the defect edge and the normal boundary edge are different, and there are differences in the change characteristics of the two edge lines, obtaining the first suspected defect degree can reflect the possibility that the edge pixel point is on the defect edge. Since there are differences in characteristics between the normal repetitive texture and the defect texture on the surface of the battery pack injection molded part, obtaining the second suspected defect degree can reflect the possibility that the edge pixel point is on the defect edge based on the texture characteristics. The enhancement coefficient of the edge pixel point is obtained according to the first suspected defect degree and the second suspected defect degree, and the enhancement degree of the edge pixel point can be accurately obtained, thereby achieving the emphasis on enhancing the defect edge while suppressing the occurrence of artifact noise. Obtaining the adaptive sharpening strength can accurately characterize the sharpening enhancement degree at each edge pixel point, thereby improving the enhancement effect and defect recognition accuracy of the defect area. Finally, the appearance image is sharpened and enhanced according to the adaptive sharpening strength, the defect is recognized based on the appearance image after sharpening and enhancement, and the injection mold defect is analyzed based on the defect recognition result; the image enhancement effect and the accuracy of defect area recognition are improved, thereby making the defect detection of the injection mold more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0012] Figure 1 A flow chart of a method for detecting and identifying production defects in a battery pack injection mold provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0013] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail a method for detecting and identifying defects in the production of battery pack injection molds proposed by the present invention, including its specific implementation, structure, features, and effects. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable form.

[0014] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0015] The following describes in detail a method for detecting and identifying production defects of a battery pack injection mold provided by the present invention with reference to the accompanying drawings.

[0016] See also Figure 1 , which shows a flow chart of a method for detecting and identifying production defects of a battery pack injection mold provided by one embodiment of the present invention, the method comprising the following steps: Step S1, obtaining an appearance image of a battery pack injection molded part.

[0017] After the battery pack injection mold has produced a batch of battery pack injection-molded parts in a small-batch trial, pictures of the battery pack injection-molded parts are taken with a camera. The pictures are converted to grayscale and denoised using the existing non-local mean denoising method to obtain the appearance images of the battery pack injection-molded parts.

[0018] Step S2, obtaining a first suspected defect degree based on the grayscale distribution characteristics and edge line change characteristics within the preset neighborhood window of the edge pixel points in the appearance image; obtaining a second suspected defect degree based on the texture distribution characteristics and texture intensity characteristics within the preset neighborhood window of the edge pixel points, and the grayscale difference characteristics and gradient difference characteristics between the preset neighborhood window and the preset neighborhood large window.

[0019] When enhancing the appearance image, in order to make the defect area more obvious, the possible defect area should be enhanced. First, the existing Canny edge detection algorithm is used to obtain the edge lines and edge pixels in the appearance image. It should be noted that when performing edge detection, the algorithm parameter settings should meet the requirements of obtaining all edge lines in the image as much as possible to prevent the edges of tiny defects from being lost due to the inability to obtain them, thereby affecting the accuracy of defect detection.

[0020] In the appearance image, not only the defective area has edges, but also the boundaries of the normal area and the surface texture have obvious or subtle edges. Therefore, when enhancing the image, the edges of the defective area should be enhanced, while the normal edges should be weakened and enhanced to prevent edge artifacts. Since the boundary edge is the pattern design boundary of the injection molded part, its edge is regular and neat, while the flash edge is the parting surface overflow, and its edge is irregularly curved. In addition, the boundary edge is only a region dividing line, the color of the injection molded part is relatively uniform, and the grayscale change on both sides of the edge is relatively small; while the flash is thin and has strong light transmittance, and there is a clear light and dark boundary on both sides of its edge; therefore, the first suspected defect level can be obtained based on the grayscale distribution characteristics within the preset neighborhood window of the edge pixel points in the appearance image and the change characteristics of the edge line.

[0021] Preferably, in an embodiment of the present invention, the step of obtaining the first suspected defect degree includes: obtaining grayscale uniformity based on the absolute value of the difference in the number of two grayscale levels with the largest number of pixels in the grayscale histogram of the preset neighborhood window; when the grayscale uniformity is larger, it means that the number of pixels corresponding to the grayscale level with the largest number of pixels is larger, and the grayscale value within the preset neighborhood window is more uniform, then the edge line where the edge pixel point is located is more likely to be a normal boundary edge line; when the grayscale uniformity is smaller, it means that the number of pixels of the two grayscale levels is closer, then the edge pixel point is more likely to be an edge at a flash, with light and dark differences on both sides of the edge. In an embodiment of the present invention, the preset neighborhood window is a window with a side length of 6 centered on the edge pixel point, which can be determined by the implementer according to the implementation scenario. Calculate the inverse of the variance of the curvature of all edge pixels on the edge line within the preset neighborhood window to obtain edge regularity; in the process of calculating the inverse in the embodiment of the present invention, if the denominator is 0, a preset minimum positive number is introduced instead. In this embodiment, the preset minimum positive number is 0.01. The curvature of an edge pixel is obtained by the coordinates of the two other edge pixels and the edge pixel in its eight-neighborhood. The greater the difference in curvature of the edge pixel in the preset neighborhood window, the greater the variance, which means that the edge line changes more irregularly, and is more likely to be a flash edge, and the smaller the edge regularity. The smaller the difference in curvature of the edge pixel, the greater the edge regularity, which means that the edge line changes more regularly and is more likely to be a boundary edge. The product of grayscale uniformity and edge regularity is calculated and negatively correlated to obtain the first suspected defect degree of the edge pixel. The greater the first suspected defect degree, the more likely the edge pixel is a pixel on a flash edge. The formula for obtaining the first suspected defect degree includes: Where R represents the first suspected defect degree, S represents the variance of the curvature of all edge pixels in the preset neighborhood window, represents the edge regularity, D represents the grayscale uniformity, Indicates normalization.

[0022] Furthermore, because battery pack injection molded parts often have repetitive textures on their surfaces, these repetitive textures are caused by internal indentations within the injection mold. These textures are characterized by high texture uniformity and clarity, and are distributed over a large area. Flow mark defects, on the other hand, are wavy surface defects near the gate, with relatively complex local variations, poor texture uniformity, and a fuzzy appearance, confining them to a localized area. Therefore, the second level of suspected defects can be determined based on the texture distribution and intensity characteristics within a preset neighborhood window of edge pixels, as well as the grayscale and gradient difference characteristics between the preset neighborhood window and the preset large neighborhood window.

[0023] Preferably, in an embodiment of the present invention, the step of obtaining the second suspected defect degree includes: calculating the product of the energy and the inverse moment of the gray level co-occurrence matrix of the preset neighborhood window to obtain the texture uniformity eigenvalue; it should be noted that the calculation of the energy and the inverse moment of the gray level co-occurrence matrix belongs to the existing technology, and the specific steps are not repeated; when the energy and the inverse moment are larger, the texture uniformity eigenvalue is larger, which means that the texture distribution of the preset neighborhood window is more uniform, the smaller the change is, and the more likely it is a normal surface texture. Calculate the product of the information entropy of the gray value in the preset neighborhood window and the information entropy of the gradient direction to obtain the window eigenvalue; the window eigenvalue reflects the gray distribution characteristics and gradient direction distribution characteristics in the preset neighborhood window; it should be noted that the information entropy belongs to the existing technology, and the specific calculation steps are not repeated. Calculate the inverse of the absolute value of the difference between the window feature values ​​of the preset neighborhood window and the preset neighborhood large window to obtain the regional feature similarity; in an embodiment of the present invention, the preset neighborhood large window is 3 times the preset neighborhood window, and the window center position is the same; when the regional feature similarity is greater, it means that the texture feature in the preset neighborhood window is similar to the texture feature of other adjacent positions, and the preset neighborhood window is more likely to be a normal repetitive texture; when the regional feature similarity is smaller, the preset neighborhood window is more likely to be a local flow mark defect area. Calculate the average of the absolute value of the grayscale difference between all edge points in the preset neighborhood window and the adjacent pixel points in the gradient direction to obtain the local grayscale difference value; when the grayscale difference between the edge point and the adjacent pixel points in the gradient direction is greater, the local grayscale difference value is greater, which means that the edge gradient feature is more obvious, the edge is clearer, and the more likely it is a normal repetitive texture; when the local grayscale difference value is smaller, it means that the edge gradient feature is weaker, the edge is more blurred, and the more likely it is a flow mark defect area. Calculate the product of texture uniformity eigenvalue, regional feature similarity, and local grayscale difference value and negatively correlate them to obtain the second suspected defect degree of the edge pixel point; the greater the second suspected defect degree, the more likely the edge pixel point is at the edge of the defect. The formula for obtaining the second suspected defect degree includes: Where W represents the texture uniformity eigenvalue, E represents the energy of the gray-level co-occurrence matrix, and H represents the inverse moment. represents the texture uniformity feature value, F represents the local grayscale difference value, Represents the window characteristic value of the preset neighborhood window, Represents the window characteristic value of the preset neighborhood large window, represents the similarity of regional features, Indicates normalization.

[0024] Step S3, obtaining an enhancement coefficient of edge pixels according to the first suspected defect degree and the second suspected defect degree; adjusting the preset sharpening intensity according to the enhancement coefficient and the gradient characteristics of the edge pixels to obtain an adaptive sharpening intensity.

[0025] When the first suspected defect degree and the second suspected defect degree of an edge pixel point are larger, it means that the edge pixel point is more likely to be a defect edge and needs to be enhanced more. Therefore, an enhancement coefficient of the edge pixel point is obtained based on the first suspected defect degree and the second suspected defect degree. Preferably, in an embodiment of the present invention, the step of obtaining the enhancement coefficient includes: calculating the average of the first suspected defect degree and the second suspected defect degree to obtain the enhancement coefficient of the edge pixel point; when the enhancement coefficient is larger, it means that the edge pixel point needs to be enhanced more. Conversely, when the enhancement coefficient is smaller, the enhancement degree needs to be weakened to avoid the defect area features being covered and the normal edge being over-enhanced.

[0026] Furthermore, the preset sharpening intensity can be adjusted according to the enhancement coefficient and the gradient characteristics of the edge pixel points to obtain an adaptive sharpening intensity; preferably, in an embodiment of the present invention, the step of obtaining the adaptive sharpening intensity includes: calculating the product of the inverse of the gradient value of the edge pixel point and the enhancement coefficient and normalizing it to obtain the adjustment degree; since the pixel points with clear edges do not need to be over-enhanced, the larger the gradient value, the smaller the enhancement degree. The larger the adjustment degree, the more the edge pixel point needs to be enhanced. Calculate the product of the adjustment degree and the maximum value of the preset sharpening intensity to obtain the adaptive sharpening intensity of the edge pixel point; the greater the sharpening intensity, the more obvious the enhancement effect, but excessive sharpening intensity will cause artifacts. In an embodiment of the present invention, the maximum value of the preset sharpening intensity is 0.7, which can be determined by the implementer according to the implementation scenario. For non-edge pixels, the preset sharpening intensity in the embodiment of the present invention is 0.3, which prevents over-sharpening while improving the overall image quality. The implementer can determine it by the implementer according to the implementation scenario.

[0027] Step S4: sharpen and enhance the appearance image according to the adaptive sharpening strength, identify defects based on the sharpened and enhanced appearance image, and analyze the defects of the injection mold based on the defect identification results.

[0028] After obtaining the adaptive sharpening strength of edge pixels, the appearance image can be sharpened and enhanced based on the adaptive sharpening strength. This includes sharpening and enhancing the appearance image based on the adaptive sharpening strength and the Unsaturated Mask (USM) algorithm. It should be noted that the USM algorithm is a state-of-the-art technique, and the specific steps are not detailed here. This sharpening and enhancement significantly improves the display quality of defects while suppressing artifacts in other areas, thereby increasing the accuracy of defect recognition in injection molded parts.

[0029] Furthermore, defect identification can be performed based on the sharpened and enhanced appearance image, and injection mold defects can be analyzed based on the defect identification results. In an embodiment of the present invention, a large number of manually annotated defect images are first obtained, and the defects include flash, flow marks, and shrinkage holes. Then, a convolutional neural network is selected for training and learning, and the trained defect detection model is used for defect identification in the current appearance image. Implementers can use other defect identification methods, which are not described here. After defect identification is performed on all enhanced appearance images, the defect location and defect type are annotated to analyze whether the injection mold is defective. Implementers can set their own rules for judging injection mold defects based on the implementation scenario.

[0030] In summary, an embodiment of the present invention provides a method for detecting and identifying production defects of battery pack injection molds; a first suspected defect degree is obtained based on the grayscale distribution characteristics and edge line change characteristics within a preset neighborhood window of edge pixels in an appearance image; a second suspected defect degree is obtained based on the texture distribution characteristics and texture intensity characteristics within a preset neighborhood window of edge pixels, and the grayscale difference characteristics and gradient difference characteristics between the preset neighborhood window and the preset neighborhood large window; an enhancement coefficient of the edge pixel is obtained based on the first suspected defect degree and the second suspected defect degree. The present invention adjusts the preset sharpening intensity based on the enhancement coefficient and the gradient characteristics of the edge pixel, sharpens and enhances the appearance image based on the adaptive sharpening intensity, and analyzes injection mold defects based on the sharpened and enhanced appearance image, thereby improving the accuracy of defect identification.

[0031] It should be noted that the order in which the embodiments of the present invention are described above is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0032] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. A method for detecting and identifying production defects of battery pack injection molds, characterized in that: The method comprises the following steps: Obtain appearance images of battery pack injection molded parts; A first suspected defect degree is obtained based on the grayscale distribution characteristics and edge line change characteristics within a preset neighborhood window of the edge pixel points in the appearance image; a second suspected defect degree is obtained based on the texture distribution characteristics and texture intensity characteristics within the preset neighborhood window of the edge pixel points, and the grayscale difference characteristics and gradient difference characteristics between the preset neighborhood window and the preset neighborhood large window; Obtaining an enhancement coefficient for the edge pixel point according to the first suspected defect degree and the second suspected defect degree; adjusting a preset sharpening intensity according to the enhancement coefficient and a gradient feature of the edge pixel point to obtain an adaptive sharpening intensity; The appearance image is sharpened and enhanced according to the adaptive sharpening strength, defects are identified based on the sharpened and enhanced appearance image, and defects of the injection mold are analyzed based on the defect identification results.

2. A battery pack injection mold production defect detection and identification method according to claim 1, characterized in that: The step of obtaining a first suspected defect degree according to the grayscale distribution characteristics and edge line change characteristics in a preset neighborhood window of edge pixels in the appearance image comprises: The grayscale uniformity is obtained based on the absolute value of the difference in the number of two grayscale levels with the most pixels in the grayscale histogram of the preset neighborhood window; the edge regularity is obtained by calculating the inverse of the variance of the curvature of all edge pixels on the edge line within the preset neighborhood window; the product of the grayscale uniformity and the edge regularity is calculated and negatively correlated to obtain the first suspected defect degree of the edge pixel point.

3. The method for detecting and identifying production defects of a battery pack injection mold according to claim 1, characterized in that: The step of obtaining a second suspected defect degree according to the texture distribution characteristics and texture intensity characteristics within the preset neighborhood window of the edge pixel point, and the grayscale difference characteristics and gradient difference characteristics between the preset neighborhood window and the preset neighborhood large window comprises: Calculate the product of the energy and inverse moment of the grayscale co-occurrence matrix of the preset neighborhood window to obtain the texture uniform eigenvalue; calculate the product of the information entropy of the grayscale value in the preset neighborhood window and the information entropy in the gradient direction to obtain the window eigenvalue; calculate the inverse of the absolute value of the difference between the window eigenvalues ​​of the preset neighborhood window and the preset neighborhood large window to obtain the regional feature similarity; calculate the average value of the absolute value of the grayscale difference between all edge points in the preset neighborhood window and the adjacent pixel points in the gradient direction to obtain the local grayscale difference value; calculate the product of the texture uniform eigenvalue, the regional feature similarity, and the local grayscale difference value and negatively correlate them to obtain the second suspected defect degree of the edge pixel point.

4. A battery pack injection mold production defect detection and identification method according to claim 1, characterized in that: The step of obtaining the enhancement coefficient of the edge pixel point according to the first suspected defect degree and the second suspected defect degree includes: An average value of the first suspected defect degree and the second suspected defect degree is calculated to obtain an enhancement coefficient of the edge pixel point.

5. The method for detecting and identifying production defects of a battery pack injection mold according to claim 1, characterized in that: The step of adjusting the preset sharpening strength according to the enhancement coefficient and the gradient characteristics of the edge pixel points to obtain the adaptive sharpening strength includes: The product of the inverse of the gradient value of the edge pixel point and the enhancement coefficient is calculated and normalized to obtain the adjustment degree; the product of the adjustment degree and the preset sharpening intensity maximum is calculated to obtain the adaptive sharpening intensity of the edge pixel point.

6. A battery pack injection mold production defect detection and identification method according to claim 1, characterized in that: The step of sharpening and enhancing the appearance image according to the adaptive sharpening strength comprises: The appearance image is sharpened and enhanced according to the adaptive sharpening strength and the USM algorithm.

Images