Non-woven fabric defect detection method

Abstract

The invention provides a non-woven fabric defect detection method, and relates to the technical field of defect detection. Comprising the following steps: acquiring characterization characteristics of the non-woven fabric; based on a preset non-woven fabric pre-generated image module, pre-generated non-woven fabric images of the current non-woven fabric under different backlight intensities are generated by utilizing the characterization characteristics of the non-woven fabric, and the pre-generated non-woven fabric images comprise a pre-generated non-woven fabric defect image and a pre-generated non-woven fabric defect-free image; inputting the pre-generated non-woven fabric defect image under each backlight intensity into a preset defect identification model for identification, and selecting a pre-optimized backlight intensity; in actual detection, non-woven fabric defect-free images under a plurality of preset backlight intensities are obtained; calculating a deviation value between the actually obtained non-woven fabric defect-free image and the pre-generated non-woven fabric defect-free image under the corresponding backlight intensity; and obtaining the actually detected backlight intensity according to the deviation value. The non-woven fabric defect detection device can detect the defects of the non-woven fabric more accurately.

Description

Technical Field

[0001] The present application relates to the technical field of defect detection, and in particular to a method for detecting defects in nonwoven fabrics. Background Art

[0002] As an important industrial material, non-woven fabrics are widely used in medical and health care, filter materials, geotextiles, clothing and household products. Their quality directly affects the performance and safety of the product. Therefore, it is of great significance to conduct defect detection on non-woven fabrics during the production process.

[0003] With the development of computer vision and artificial intelligence technology, image recognition methods based on machine learning and deep learning are gradually applied to the field of non-woven fabric defect detection. These methods can improve the accuracy and efficiency of detection to a certain extent by training models.

[0004] However, the backlight intensity has an important influence on the appearance of non-woven fabric defects, and the optical properties of non-woven fabrics are affected by factors such as material type, fiber size, arrangement and arrangement density. Some non-woven fabrics can also produce fluorescence.

[0005] Non-woven fabrics with strong reflectivity will weaken the backlight effect, while non-woven fabrics that can stimulate fluorescence will brighten the backlight effect. This will affect the backlight projection. Using a uniform backlight intensity will cause some defects to not get proper backlight projection, reducing the possibility of recognition. Summary of the invention

[0006] In order to solve the above technical problems or at least partially solve the above technical problems, the present application provides a non-woven fabric defect detection method, which can more accurately detect the defects of the non-woven fabric.

[0007] The present application provides a nonwoven fabric defect detection method, the nonwoven fabric defect detection method comprising the following steps: Obtain the surface fiber design structure and material properties of the current non-woven fabric to obtain the characterization characteristics of the non-woven fabric; Based on a preset non-woven fabric pre-generated image module, using the characterization features of the non-woven fabric, a pre-generated non-woven fabric image of the current non-woven fabric under different backlight intensities is generated, wherein the pre-generated non-woven fabric image includes a pre-generated non-woven fabric defect image and a pre-generated non-woven fabric defect-free image; Input the pre-generated non-woven fabric defect image under each backlight intensity into a preset defect recognition model for recognition, and select the backlight intensity corresponding to the pre-generated non-woven fabric defect image with the highest recognition rate as the pre-optimal backlight intensity; In actual inspection, defect-free images of nonwoven fabrics under multiple preset backlight intensities are obtained; Calculating a deviation value between an actually acquired non-woven fabric defect-free image and a pre-generated non-woven fabric defect-free image under a corresponding backlight intensity; According to the deviation value, the pre-optimized backlight intensity is adjusted by a preset backlight intensity adjustment function to obtain the actual detected backlight intensity; The nonwoven fabric is illuminated by utilizing the actual detected backlight intensity, and the defects of the nonwoven fabric are detected by using a camera and a preset defect recognition model.

[0008] Optionally, the surface fiber design structure includes fiber size, arrangement orientation and arrangement density, and the material properties include non-woven fabric material type and fluorescence efficiency; The preset non-woven fabric pre-generated image module is constructed by the following steps: The basic network structure generation module uses 3D modeling software to establish the basic network structure of non-woven fabrics based on the surface fiber design structure; Defect structure generation module, which uses 3D modeling software to select whether to insert a defect structure into the basic network structure; According to the type of non-woven fabric material and the fluorescence efficiency, the application material and fluorescence efficiency parameters of the basic network structure are set in the three-dimensional modeling software; Add backlight source to the basic network structure; Changing the intensity of the backlight source, and using an optical rendering engine to render the basic network structure, and setting a virtual camera to obtain non-woven fabric rendering images under different backlight intensities; Constructing an optical mask processing model, wherein the network framework of the optical mask processing model is a generative adversarial network, collecting a plurality of actual non-woven fabric images and corresponding non-woven fabric rendering images, taking the actual non-woven fabric surface images as labels and the corresponding non-woven fabric rendering images as inputs, training the generative adversarial network, thereby obtaining the optical mask processing model; The non-woven fabric rendering images under different backlight intensities are input into the optical mask processing model for processing to obtain the pre-generated non-woven fabric images under different backlight intensities.

[0009] Optionally, the preset defect recognition model is trained by the following steps: The network framework of the defect recognition model is a convolutional neural network; A plurality of non-woven fabric defect images with different defect types are collected, and the plurality of non-woven fabric defect images are used as inputs and the defect types are used as labels, so as to train a defect recognition model.

[0010] Optionally, calculating the deviation between the non-woven fabric defect-free image actually obtained and the pre-generated non-woven fabric defect-free image under the corresponding backlight intensity, and adjusting the pre-optimized backlight intensity according to the deviation value by a preset backlight intensity adjustment function to obtain the actual detected backlight intensity, comprising the following steps: The deviation value between the actually acquired non-woven fabric defect-free image and the pre-generated non-woven fabric defect-free image under the corresponding backlight intensity is calculated by a structural similarity algorithm; According to the size of the deviation value under each preset backlight intensity, determine the backlight intensity corresponding to the minimum deviation value; The preset backlight intensity adjustment function is: in, To actually detect the backlight intensity, To pre-optimize the backlight intensity, is the backlight intensity corresponding to the minimum deviation value, and c is the first empirical parameter.

[0011] Compared with the prior art, the technical solution provided by this application has the following advantages: One of its beneficial technical effects is that since the fiber thickness, arrangement and material properties of non-woven fabrics will affect the reflection and fluorescence effects of light, the projection of backlight is affected, which in turn affects the detectability of defects. The traditional method of fixing the backlight intensity cannot adapt to the detection needs of different non-woven fabrics.

[0012] In order to solve this problem, the present application first obtains the surface fiber design structure and material properties of the current non-woven fabric to obtain the characterization features of the non-woven fabric, which can be used to reflect the response characteristics of the non-woven fabric to light.

[0013] Based on this, the pre-generated non-woven fabric image module is used to generate pre-generated non-woven fabric images of the current non-woven fabric at different backlight intensities, including defective images and non-defective images. The pre-generated defective images at each backlight intensity are input into the preset defect recognition model, the defect recognition rate corresponding to each backlight intensity is evaluated, and the backlight intensity with the highest recognition rate is selected as the pre-optimal backlight intensity to obtain the pre-optimal backlight intensity.

[0014] In addition, considering the influence of the actual ambient light, during the actual detection process, multiple defect-free images of non-woven fabrics under preset backlight intensities are obtained, and the deviation values ​​between these actual images and the pre-generated defect-free images are calculated. According to the deviation value, the pre-optimized backlight intensity is accurately adjusted using the preset backlight intensity adjustment function to obtain the actual detection backlight intensity.

[0015] Finally, the non-woven fabric is illuminated with the adjusted actual detection backlight intensity, and the defects of the non-woven fabric are detected by combining the camera and the preset defect recognition model. In this way, the method of the present application fully considers the optical properties of the non-woven fabric and the actual detection environment, and improves the accuracy and reliability of defect detection by automatically optimizing the backlight intensity, thereby solving the detection error caused by the incompatibility of the backlight intensity in the traditional method.

[0016] Therefore, the non-woven fabric defect detection method provided in the present application can detect the defects of the non-woven fabric more accurately.

[0017] The second beneficial technical effect is that, in practical applications, due to the influence of changes in ambient lighting, the pre-determined pre-preferred backlight intensity may not maintain the best effect under all detection conditions, resulting in a deviation between the pre-generated defect-free image and the actual acquired defect-free image. When the backlight intensity point with the smallest deviation value deviates far from the pre-preferred backlight intensity, it means that the current pre-preferred backlight intensity fails to fully adapt to the actual detection environment. If it is not adjusted, it may lead to a decrease in the accuracy of defect detection. To solve this problem, the present application obtains an actual defect-free image under multiple preset backlight intensities and compares it with the corresponding pre-generated defect-free image to calculate the deviation value. Based on these deviation values, the pre-preferred backlight intensity is dynamically adjusted using a preset backlight intensity adjustment function to make it closer to the direction of the actual minimum deviation point, and the amplitude of the adjustment amount is adjusted according to the size of the deviation value. Specifically, when the backlight intensity point with the smallest deviation value is far away from the pre-preferred backlight intensity, the adjustment function will adjust the pre-preferred backlight intensity to a greater extent to minimize the deviation as much as possible; and when the point with the smallest deviation value is close to the pre-preferred backlight intensity, the adjustment amplitude is correspondingly reduced to reduce the possibility of deviation caused by the adjustment. This dynamic adjustment mechanism can effectively compensate for the influence of external factors such as ambient light, ensure that the backlight intensity setting is more accurate, and reduce the occurrence of false detection and missed detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Flow chart of the non-woven fabric detection method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The technical solution in this application will be described below in conjunction with the accompanying drawings.

[0020] In the following description, many specific details are set forth to facilitate a full understanding of the present application, but the present application can also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only part of the embodiments of the present application, not all of the embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0021] like Figure 1As shown, the present application provides a non-woven fabric defect detection method, and the non-woven fabric defect detection method comprises the following steps: S101: Obtain the surface fiber design structure and material properties of the current non-woven fabric to obtain the characterization characteristics of the non-woven fabric; Specifically, the surface fiber design structure includes fiber size, arrangement orientation and arrangement density, and the material properties include non-woven fabric material type and fluorescence efficiency.

[0022] These parameters need to be manually input and set according to the specific parameters of the current non-woven fabric production.

[0023] S102: Based on a preset non-woven fabric pre-generated image module, using the characterization features of the non-woven fabric, generating pre-generated non-woven fabric images of the current non-woven fabric under different backlight intensities, wherein the pre-generated non-woven fabric images include pre-generated non-woven fabric defect images and pre-generated non-woven fabric non-defect images; The preset non-woven fabric pre-generated image module is constructed by the following steps: The basic network structure generation module uses 3D modeling software to establish the basic network structure of non-woven fabrics based on the surface fiber design structure.

[0024] Specifically, the 3D modeling software is 3DEXPERIENCE.

[0025] Among them, the basic network structure of the non-woven fabric can be imported into the structural model designed by the user before production, or the established non-woven fabric basic network structure template can be transformed according to the actual fiber size, arrangement orientation and arrangement density to obtain the basic network structure of this non-woven fabric.

[0026] Defect structure generation module, which uses 3D modeling software to select whether to insert a defect structure into the basic network structure; Specifically, it realizes collecting various types of defect structures, and randomly inserting random types and random numbers of defect structures into the basic network structure according to whether to generate a pre-generated non-woven fabric defect image or a pre-generated non-woven fabric defect-free image; After the defect structure is inserted, the insertion position of each defect structure and the type of defect structure corresponding to the insertion position are recorded for verification during defect identification when the pre-optimized backlight intensity is subsequently selected.

[0027] The application material and fluorescence efficiency of the basic network structure are set in the three-dimensional modeling software according to the type of non-woven fabric material and the fluorescence efficiency.

[0028] Add backlight source to the basic network structure; The intensity of the backlight source is changed, and the basic network structure is rendered using an optical rendering engine, and a virtual camera is set to obtain non-woven fabric rendering images under different backlight intensities.

[0029] Specifically, the optical rendering engine here is the optical rendering engine provided in 3DEXPERIENCE.

[0030] An optical mask processing model is constructed, wherein the network framework of the optical mask processing model is a generative adversarial network. A plurality of actual non-woven fabric images and corresponding non-woven fabric rendering images are collected, and the actual non-woven fabric surface images are used as labels and the corresponding non-woven fabric rendering images are used as inputs. The generative adversarial network is trained to obtain the optical mask processing model.

[0031] The non-woven fabric rendering images under different backlight intensities are input into the optical mask processing model for processing to obtain the pre-generated non-woven fabric images under different backlight intensities.

[0032] S103: inputting the pre-generated non-woven fabric defect image under each backlight intensity into a preset defect recognition model for recognition, and selecting the backlight intensity corresponding to the pre-generated non-woven fabric defect image with the highest recognition rate as the pre-optimal backlight intensity; Specifically, the preset defect recognition model is trained through the following steps: The network framework of the defect recognition model is a convolutional neural network; A plurality of non-woven fabric defect images with different defect types are collected, and the plurality of non-woven fabric defect images are used as inputs and the defect types are used as labels, so as to train a defect recognition model.

[0033] In actual detection, before applying the defect recognition model, the image is usually binarized and segmented to obtain the area where the abnormal binary points in the image are located as the region of interest.

[0034] Then the image in the region of interest is input into the defect recognition model to complete the defect recognition. However, this belongs to the prior art and will not be described in detail here.

[0035] S104: In actual detection, a camera is used to obtain a plurality of defect-free images of the non-woven fabric under a preset backlight intensity.

[0036] Specifically, the multiple preset backlight intensities here include at least a pre-preferred backlight intensity and a plurality of left and right backlight intensities. To pre-optimize the backlight intensity, is a manually set step length, n is a manually set number, then the set of multiple preset backlight intensities is ( , …, , …, ).

[0037] Calculating a deviation value between an actually acquired non-woven fabric defect-free image and a pre-generated non-woven fabric defect-free image under a corresponding backlight intensity; According to the deviation value, the pre-optimized backlight intensity is adjusted by a preset backlight intensity adjustment function to obtain the actual detected backlight intensity; Specifically, the following steps are included: The deviation value between the actually acquired non-woven fabric defect-free image and the pre-generated non-woven fabric defect-free image under the corresponding backlight intensity is calculated by a structural similarity algorithm.

[0038] The structural similarity algorithm is an existing technology for comparing the similarity between images, which will not be described in detail here.

[0039] According to the size of the deviation value under each preset backlight intensity, determine the backlight intensity corresponding to the minimum deviation value .

[0040] The preset backlight intensity adjustment function is: in, To actually detect the backlight intensity, To pre-optimize the backlight intensity, is the backlight intensity corresponding to the minimum deviation value, and c is the first empirical parameter set artificially.

[0041] S105: irradiating the non-woven fabric with the actual detected backlight intensity, and detecting defects of the non-woven fabric using a camera and a preset defect recognition model.

[0042] This step is actually consistent with S103, and the defect recognition model is used to identify the surface defects in the captured non-woven fabric image. This belongs to the prior art and will not be described in detail here.

[0043] In summary, the non-woven fabric defect detection method provided in the embodiment of the present application has the following beneficial effects: One of its beneficial technical effects is that since the fiber thickness, arrangement and material properties of non-woven fabrics will affect the reflection and fluorescence effects of light, the projection of backlight is affected, which in turn affects the detectability of defects. The traditional method of fixing the backlight intensity cannot adapt to the detection needs of different non-woven fabrics.

[0044] In order to solve this problem, the present application first obtains the surface fiber design structure and material properties of the current non-woven fabric to obtain the characterization features of the non-woven fabric, which can be used to reflect the response characteristics of the non-woven fabric to light.

[0045] Based on this, the pre-generated non-woven fabric image module is used to generate pre-generated non-woven fabric images of the current non-woven fabric at different backlight intensities, including defective images and non-defective images. The pre-generated defective images at each backlight intensity are input into the preset defect recognition model, the defect recognition rate corresponding to each backlight intensity is evaluated, and the backlight intensity with the highest recognition rate is selected as the pre-optimal backlight intensity to obtain the pre-optimal backlight intensity.

[0046] In addition, considering the influence of the actual ambient light, during the actual detection process, multiple defect-free images of non-woven fabrics under preset backlight intensities are obtained, and the deviation values ​​between these actual images and the pre-generated defect-free images are calculated. According to the deviation value, the pre-optimized backlight intensity is accurately adjusted using the preset backlight intensity adjustment function to obtain the actual detection backlight intensity.

[0047] Finally, the non-woven fabric is illuminated with the adjusted actual detection backlight intensity, and the defects of the non-woven fabric are detected by combining the camera and the preset defect recognition model. In this way, the method of the present application fully considers the optical properties of the non-woven fabric and the actual detection environment, and improves the accuracy and reliability of defect detection by automatically optimizing the backlight intensity, thereby solving the detection error caused by the incompatibility of the backlight intensity in the traditional method.

[0048] Therefore, the non-woven fabric defect detection method provided in the present application can detect the defects of the non-woven fabric more accurately.

[0049] The second beneficial technical effect is that, in practical applications, due to the influence of changes in ambient illumination, the pre-determined pre-preferred backlight intensity may not maintain the best effect under all detection conditions, resulting in a deviation between the pre-generated defect-free image and the actual acquired defect-free image. When the backlight intensity point with the smallest deviation value deviates far from the pre-preferred backlight intensity, it means that the current pre-preferred backlight intensity fails to fully adapt to the actual detection environment. If it is not adjusted, it may lead to a decrease in the accuracy of defect detection. To solve this problem, the present application obtains an actual defect-free image under multiple preset backlight intensities and compares it with the corresponding pre-generated defect-free image to calculate the deviation value. Based on these deviation values, the pre-preferred backlight intensity is dynamically adjusted using a preset backlight intensity adjustment function to make it closer to the direction of the actual minimum deviation point, and the amplitude of the adjustment amount is adjusted according to the size of the deviation value. Specifically, when the backlight intensity point with the smallest deviation value is far away from the pre-preferred backlight intensity, the adjustment function will adjust the pre-preferred backlight intensity to a greater extent to minimize the deviation as much as possible; and when the point with the smallest deviation value is close to the pre-preferred backlight intensity, the adjustment amplitude is correspondingly reduced to reduce the possibility of deviation caused by the adjustment. This dynamic adjustment mechanism can effectively compensate for the influence of external factors such as ambient light, ensure that the backlight intensity setting is more accurate, and reduce the occurrence of false detection and missed detection.

[0050] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. In addition, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements. Moreover, in the description of the embodiments of the present application, unless otherwise specified, " / " means or, for example, A / B can represent A or B; "and / or" in this article is only a kind of association relationship describing the associated objects, indicating that there can be three relationships, for example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. Furthermore, in the description of the embodiments of the present application, “plurality” refers to two or more than two.

[0051] The above description is only a specific implementation of the present application, so that those skilled in the art can understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments described herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting defects in nonwoven fabrics, characterized in that: The nonwoven fabric defect detection method comprises the following steps: Obtain the surface fiber design structure and material properties of the current non-woven fabric to obtain the characterization characteristics of the non-woven fabric; Based on a preset non-woven fabric pre-generated image module, using the characterization features of the non-woven fabric, a pre-generated non-woven fabric image of the current non-woven fabric under different backlight intensities is generated, wherein the pre-generated non-woven fabric image includes a pre-generated non-woven fabric defect image and a pre-generated non-woven fabric defect-free image; Input the pre-generated non-woven fabric defect image under each backlight intensity into a preset defect recognition model for recognition, and select the backlight intensity corresponding to the pre-generated non-woven fabric defect image with the highest recognition rate as the pre-optimal backlight intensity; In actual inspection, defect-free images of nonwoven fabrics under multiple preset backlight intensities are obtained; Calculating a deviation value between an actually acquired non-woven fabric defect-free image and a pre-generated non-woven fabric defect-free image under a corresponding backlight intensity; According to the deviation value, the pre-optimized backlight intensity is adjusted by a preset backlight intensity adjustment function to obtain the actual detected backlight intensity; The nonwoven fabric is illuminated by utilizing the actual detected backlight intensity, and the defects of the nonwoven fabric are detected by using a camera and a preset defect recognition model.

2. The nonwoven fabric defect detection method according to claim 1, characterized in that: The surface fiber design structure includes fiber size, arrangement orientation and arrangement density, and the material properties include non-woven fabric material type and fluorescence efficiency; The preset non-woven fabric pre-generated image module is constructed by the following steps: The basic network structure generation module uses 3D modeling software to establish the basic network structure of non-woven fabrics based on the surface fiber design structure; Defect structure generation module, which uses 3D modeling software to select whether to insert a defect structure into the basic network structure; According to the type of non-woven fabric material and the fluorescence efficiency, the application material and fluorescence efficiency parameters of the basic network structure are set in the three-dimensional modeling software; Add backlight source to the basic network structure; Changing the intensity of the backlight source, and using an optical rendering engine to render the basic network structure, and setting a virtual camera to obtain non-woven fabric rendering images under different backlight intensities; Constructing an optical mask processing model, wherein the network framework of the optical mask processing model is a generative adversarial network, collecting a plurality of actual non-woven fabric images and corresponding non-woven fabric rendering images, taking the actual non-woven fabric surface images as labels and the corresponding non-woven fabric rendering images as inputs, training the generative adversarial network, thereby obtaining the optical mask processing model; The non-woven fabric rendering images under different backlight intensities are input into the optical mask processing model for processing to obtain the pre-generated non-woven fabric images under different backlight intensities.

3. The nonwoven fabric defect detection method according to claim 1, characterized in that: The preset defect recognition model is trained by the following steps: The network framework of the defect recognition model is a convolutional neural network; A plurality of non-woven fabric defect images with different defect types are collected, and the plurality of non-woven fabric defect images are used as inputs and the defect types are used as labels, so as to train a defect recognition model.

4. The nonwoven fabric defect detection method according to claim 1, characterized in that: Calculating the deviation between the non-woven fabric defect-free image actually obtained and the pre-generated non-woven fabric defect-free image under the corresponding backlight intensity, and adjusting the pre-optimized backlight intensity according to the deviation value by a preset backlight intensity adjustment function to obtain the actual detected backlight intensity, including the following steps: The deviation value between the actually acquired non-woven fabric defect-free image and the pre-generated non-woven fabric defect-free image under the corresponding backlight intensity is calculated by a structural similarity algorithm; According to the size of the deviation value under each preset backlight intensity, determine the backlight intensity corresponding to the minimum deviation value; The preset backlight intensity adjustment function is: in, To actually detect the backlight intensity, To pre-optimize the backlight intensity, is the backlight intensity corresponding to the minimum deviation value, and c is the first empirical parameter.

Images