A PE carpet texture anomaly detection method and system based on image recognition

Abstract

The present application belongs to the technical field of PE carpet, and particularly relates to a PE carpet texture abnormality detection method and system based on image recognition. The method comprises extracting corresponding detection texture features from a detection area on the surface of the PE carpet, comparing the detection texture features with an abnormal texture template, and identifying a texture abnormality area according to an abnormality detection value. The present application classifies standard images to distinguish normal texture images from abnormal defect images after obtaining the standard images. A plurality of abnormal defect images are summarized, image features thereof are extracted and edge processing is performed, and finally a corrected abnormal texture template is constructed. The above-mentioned comprehensive template is constructed from a plurality of real defect samples and is corrected, thereby overcoming the problems of insufficient representation and feature ambiguity that may occur when a single or unprocessed sample is used as a template.

Description

Technical Field

[0001] This invention belongs to the field of PE carpet technology, specifically relating to a method and system for detecting texture anomalies in PE carpets based on image recognition. Background Technology

[0002] In the polymer material processing and decorative material production industry, the inspection of product surface quality plays an important role in ensuring product quality and enhancing brand value. Polyethylene carpets, as a new type of flooring material widely used in commercial and home environments, have surface texture uniformity, integrity, and aesthetics as core indicators for measuring their quality.

[0003] Currently, the quality inspection of surface textures of products such as PE carpets is carried out by manual visual inspection. However, this method is labor-intensive and slow, making it difficult to match the pace of modern high-speed production lines and easily creating a bottleneck in the quality control process. Furthermore, manual judgment is highly subjective, and the test results are easily affected by the differences in the inspector's personal experience, making it difficult to unify the test standards and leading to missed and false detections.

[0004] To address the above problems, this invention proposes a method and system for detecting texture anomalies in PE carpets based on image recognition. Summary of the Invention

[0005] The purpose of this invention is to provide a method for detecting texture anomalies in PE carpets based on image recognition. This method can divide the surface of the PE carpet into multiple detection areas, and the divided areas are relatively small in size, which can improve the detection accuracy and accurately output which area has an anomaly.

[0006] The present invention adopts the following technical solution: A method for detecting texture anomalies in PE carpets based on image recognition, comprising: In response to extracting the corresponding detection texture features from the detection area on the PE carpet surface, the detection texture features are compared with a pre-constructed abnormal texture template to identify texture abnormal areas in the PE carpet. The construction process of the abnormal texture template includes: A standard image is generated based on the original sampled image obtained from the PE carpet surface. The standard image is then classified to obtain anomaly defect images. Anomaly texture templates are then constructed based on the image features of the anomaly defect images.

[0007] Preferably, generating a standard image based on the original sampled image obtained from the PE carpet surface includes: Acquire the original sampled image and noise images of multiple regions adjacent to the region of the original sampled image; extract the brightness values ​​of each pixel in the original sampled image and the noise images, and perform weighted calculation of the brightness values ​​in combination with preset standard parameters to obtain the brightness difference evaluation value; and correct the original sampled image according to the offset direction of the brightness difference evaluation value to output the corrected image as the standard image.

[0008] Preferably, acquiring the original sampled image includes: acquiring a set of adjacent sampled images, and setting the first image in the set of adjacent sampled images as the reference image; The system sequentially verifies other images in a group of adjacent sampled images to obtain verification results; compares the verification results with a set of preset judgment parameters to obtain a reliable judgment result for identifying reliable images in the group of adjacent sampled images; and selects and determines the image with the largest brightness value as the original sampled image from all reliable images.

[0009] Preferably, acquiring a set of adjacent sampled images includes: Multiple images from different directions are acquired, and the image features of each image are extracted to form a vector value. The vector values ​​are weighted by a preset directional weight to obtain multiple indicators to be sorted. Based on the indicators to be sorted in descending order of value, the order of multiple images is rearranged to generate a reordered image sequence as a group of adjacent sampled images.

[0010] Preferably, the process of constructing the abnormal texture template further includes: The abnormal defect images are classified and summarized to obtain classification results; using the classification results as a reference, the image information of the target texture region in the abnormal defect images is extracted as the content to be verified; and the edge processing of the content to be verified is performed to obtain the corrected image information, and an abnormal texture template is generated based on the corrected image information.

[0011] Preferably, comparing the detected texture features with a pre-constructed anomalous texture template to identify texture anomalous regions in the PE carpet includes: For each detected texture feature, its similarity to the abnormal texture template is calculated, and the similarity is determined as the anomaly detection value; the similarity is compared with the preset standard similarity parameter, and when the similarity is greater than the standard similarity parameter, the detection area corresponding to the detected texture feature is identified as a texture abnormal area.

[0012] Preferably, comparing the detected texture features with a pre-constructed anomalous texture template to identify texture anomalous regions in the PE carpet further includes: After comparing the similarity with the preset standard similarity parameters, the deviation between the two is calculated, and an offset signal is generated based on the deviation. The offset signal is then output as the recognition result of the texture abnormal region.

[0013] This invention also discloses an image recognition-based PE carpet texture anomaly detection system, comprising: An abnormal texture template construction module is used to construct an abnormal texture template based on an abnormal defect image generated from the original sampled image obtained from the PE carpet surface; The detection texture feature extraction module is used to extract the corresponding detection texture features from the detection area on the PE carpet surface; And a texture anomaly recognition module, which, in response to the texture features extracted by the texture feature extraction module, compares the texture features with the anomaly texture template constructed by the anomaly texture template construction module to identify texture anomaly areas in the PE carpet.

[0014] Preferably, the abnormal texture template building module is configured as follows: Acquire a set of adjacent sampled images, and set the first image in the set of adjacent sampled images as the reference image; The system sequentially verifies other images in a group of adjacent sampled images to obtain verification results; compares the verification results with a set of preset judgment parameters to obtain a reliable judgment result for identifying reliable images in the group of adjacent sampled images; and selects and determines the image with the largest brightness value as the original sampled image from all reliable images.

[0015] Preferably, acquiring a set of adjacent sampled images includes: Multiple images from different directions are acquired, and the image features of each image are extracted to form a vector value. The vector values ​​are weighted by a preset directional weight to obtain multiple indicators to be sorted. Based on the indicators to be sorted in descending order of value, the order of multiple images is rearranged to generate a reordered image sequence as a group of adjacent sampled images.

[0016] Beneficial effects 1. This invention acquires images from different directions, reorders these images, and selects the reliable image with the highest brightness value as the original sampled image. At the same time, it combines the original sampled image with adjacent noisy images to calculate the brightness difference evaluation value. The original sampled image is then corrected to generate a standard image. Through the above image preprocessing and standardization process, the interference of unstable factors caused by shooting angle, uneven lighting, and environmental noise on image quality is eliminated, ensuring that the image data used for subsequent analysis is obtained under a unified benchmark.

[0017] 2. After obtaining a standard image, the present invention classifies it to distinguish between normal texture images and abnormal defect images; multiple abnormal defect images are summarized, their image features are extracted and edge processing is performed, and finally a corrected abnormal texture template is constructed. The above method of constructing a comprehensive template derived from multiple real defect samples and after correction processing overcomes the problems of insufficient representativeness or blurred features that may occur when using a single or unprocessed sample as a template.

[0018] 3. This invention marks multiple detection areas on the surface of a PE carpet and extracts the detection texture features of each area one by one; compares the detection texture features with an abnormal texture template to identify texture abnormal areas; calculates the deviation between the similarity and the standard similarity parameter, and generates an offset signal as the identification result, thereby realizing the location of texture abnormality, providing direction for subsequent repair or removal work, and providing richer quality assessment information than traditional binary judgment by outputting a quantified offset signal. Attached Figure Description

[0019] Figure 1 This is one of the method flowcharts of the present invention; Figure 2 This is the second flowchart of the method of the present invention; Figure 3 This is a system module diagram of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely for explaining the invention and are not intended to limit the scope of protection of the invention.

[0021] Example 1 Please see Figure 1 , Figure 2 As shown, this embodiment provides a method for detecting texture anomalies in PE carpets based on image recognition, including the following steps: S1. Image acquisition and processing; By acquiring and optimizing images from multiple angles, we ensure that the acquired image data has the highest stability and can represent the true condition of the carpet, avoiding accidental noise introduced by sudden changes in lighting or minor vibrations in a single shot. Specifically: Industrial cameras deployed at different locations on the production line are used to acquire multiple images of the same carpet area from multiple preset directions, and the multiple images are initially arranged. The texture quantization features of each image are extracted. For example, by analyzing the gray-level distribution pattern and gradient change direction in the pixel neighborhood, complex texture information such as fiber direction, density, and roughness is transformed into a set of numerical values ​​or vectors that can be mathematically calculated and compared to form feature vector values ​​for objectively evaluating the texture state.

[0022] To select the best shooting angle that clearly reflects texture details, the following method was used: By combining preset directional weights to perform weighted calculations on the feature vector values, multiple indicators to be sorted are obtained. These directional weights can be preset based on production experience data. For example, a higher weight can be given to the shooting direction that forms a specific angle with the main direction of the carpet fibers, as this angle can usually better expose texture defects.

[0023] The indicators to be sorted are sorted in descending order of value. Based on this sorting result, the order of multiple images is rearranged according to the direction corresponding to the indicators, generating a reordered image sequence. This reordered image sequence serves as a set of high-quality adjacent sampled images, providing reliable input for subsequent selection of the best single-frame image. To accurately select the most suitable single-frame image for analysis from this image sequence and perform reliability verification, the specific operation is as follows: The first image in a set of adjacent sampled images, i.e., the best image after sorting, is designated as the reference image. Other images in the set of adjacent sampled images are then verified sequentially, for example, by comparing the differences between them and the reference image in brightness, contrast, and structure region by region. The verification result of the quantitative consistency degree is calculated, and the verification result and the reference image are used together to form the test data. The test data is compared with a set of preset judgment parameters to obtain a reliable judgment result. For example, the consistency degree must be higher than a predetermined threshold, which is preferably 0.98. The reliable judgment result is used to indicate which images in a set of adjacent sampled images are reliable images, i.e., images that are highly consistent with the reference image in core structure and effectively exclude motion blur or instantaneous illumination interference.

[0024] To ensure that the final image used for analysis has the best dynamic range to facilitate subsequent feature extraction, the image with the highest overall brightness value is selected from all the images deemed trustworthy, and this image with the highest brightness value is determined as the original sampled image.

[0025] After determining the original sampled image, in order to eliminate systematic deviations caused by uneven ambient lighting or differences in camera sensor response, a standard image needs to be generated. Dynamic correction is then used to bring the brightness distribution of the image to a uniform and ideal state.

[0026] This involves acquiring the original sampled image and noise images of multiple regions adjacent to the original sampled image. The noise images refer to images of a pre-defined reference area on the production line with uniform texture and standard reflectivity, reflecting the baseline state of the current lighting environment. Both the original sampled image and the noise images are used as the analysis object. The brightness values ​​of each pixel in both are extracted, and weighted calculations are performed using pre-defined standard parameters to obtain a brightness difference evaluation value. This evaluation value quantifies the degree of deviation between the current lighting conditions and ideal conditions. The pre-defined standard parameters are preferably the mean and variance of grayscale values ​​that an image should have under ideal lighting conditions.

[0027] Based on the offset direction of the brightness difference evaluation value, for example, if the evaluation value is positive, indicating that the image is generally too bright, the original sampled image is subjected to corresponding correction processing, such as applying grayscale mapping adjustment based on lookup table or adjusting its contrast curve, and the corrected image is output. The corrected image is defined as the standard image, which has uniform grayscale distribution characteristics and eliminates local illumination interference.

[0028] S2, Image Classification and Anomaly Template Construction; Based on preset image segmentation rules, the standard image is divided into multiple visual blocks of uniform size. According to the texture quantization standard, each visual block is analyzed: areas with obvious texture defects such as fiber breakage, fracture, protrusion, and color spots are identified as abnormal defect images, while areas without the above phenomena are classified as normal texture images.

[0029] Texture defects are visually manifested as interruptions in continuous textures, abnormal local bulges, or color variations.

[0030] Texture quantization criteria are a set of predefined rules for automatically classifying visual patches in images. They define numerical criteria for distinguishing normal textures from abnormal defects, such as thresholds based on pixel grayscale gradients, color distribution, or structural continuity.

[0031] Furthermore, in order to construct an abnormal texture template that can represent typical abnormal features, normal texture images and abnormal defect images are classified separately to obtain classification results. Using the classification results as a reference, image information of the target texture region is extracted from the abnormal defect image, that is, only the pixel data of the defect itself is extracted, and this image information is used as the content to be verified.

[0032] Furthermore, to eliminate the interference of edge burrs and minor noise on template construction, edge processing is performed on the content to be verified. Specifically: Edge recognition processing is applied to define the initial contour of the defect by calculating the brightness gradient between pixels. Image morphology processing is then performed on this contour, specifically a combination of expanding and contracting the contour to smooth the edges and fill in the tiny internal breaks, resulting in the corrected image information.

[0033] Based on the corrected image information, a comprehensive corrected abnormal texture template is generated by pixel-level superposition and averaging of multiple corrected defective image information, or by extracting and integrating their common key texture values. This corrected abnormal texture template is then designated as the abnormal texture template. After being fixed, the abnormal texture template is used for comparison and identification of unknown images in subsequent production processes.

[0034] S3. Detection region segmentation and feature extraction; In a standard image, multiple detection areas are automatically defined at certain intervals, such as 10 mm by 10 mm, according to a preset grid division rule.

[0035] Subsequently, for each detection region, its corresponding detection texture features are extracted. These features are a set of quantified parameters used to describe the texture state of the region, such as directional indicators reflecting the texture direction, periodic distribution parameters describing the texture repetition pattern, and edge density values ​​that quantify the edge complexity.

[0036] S4. Anomaly comparison and region identification; The constructed anomaly texture template is invoked, and the detected texture features of all detection regions are compared one by one. Specifically: For each detected texture feature, the degree of similarity between it and the abnormal texture template in terms of numerical features is calculated. For example, by placing the feature vectors of the two in a multi-dimensional space and calculating the angle between them, or by comparing the differences between the two in brightness, contrast and structural information point by point, a numerical value of quantitative similarity is obtained, which is determined as the anomaly detection value. The abnormal texture regions are identified based on the abnormal detection values. This identification process includes comparing the abnormal detection values ​​with a preset standard similarity parameter, which is preferably a threshold of 0.85 determined based on a large amount of experimental data. When the abnormal detection value is greater than the standard similarity parameter, it means that the texture features of the current detection area are highly matched with the known defect features. The detection area corresponding to the detected texture features is then identified as a texture abnormal area, and the area can be highlighted on the user interface to issue a warning. When the abnormal detection value is less than or equal to the standard similarity parameter, the detection area is determined to be a non-abnormal area, and no alarm is required.

[0037] Furthermore, in order to provide richer detection information, after comparing the abnormal detection value with the preset standard similarity parameter, the deviation between the two is calculated; based on the deviation, an offset signal is generated; for example, the deviation can be normalized and mapped to a continuous value between -1 and 1, which is the offset signal. The offset signal is output as the identification result of texture abnormality area. This offset signal can not only be used to determine whether there is an abnormality, but also to quantify the severity of the abnormality or the degree of deviation from the standard defect template, providing data support for subsequent quality control grading and production process adjustment.

[0038] Example 2 Please see Figure 3 As shown, this embodiment provides a PE carpet texture anomaly detection system based on image recognition, including the following modules: Abnormal texture template building module: In offline or early production phases, based on acquired sample images, one or more high-precision anomalous texture templates are constructed for subsequent comparisons. The workflow covers the entire process from raw image acquisition and optimization to final template generation, specifically: During the image acquisition stage, the module controls the image acquisition device to acquire multiple images of the PE carpet sample from different directions. To ensure the best image quality, the module extracts the image features of each image to form a vector value, and performs weighted calculation on the vector value in combination with preset directional weights to obtain multiple unsorted indicators. Based on the unsorted indicators sorted from largest to smallest, the module rearranges the order of these multiple images to generate a reordered image sequence as a group of adjacent sampled images.

[0039] In the original sampled image screening stage, the first image in a set of adjacent sampled images is set as the reference image, and the other images in the sequence are verified in turn. For example, the structural similarity or feature point matching degree with the reference image is calculated to obtain the verification result. The verification result is compared with a set of preset judgment parameters to obtain the credible judgment result used to indicate the credible image.

[0040] Among all images deemed reliable, the image with the highest brightness value is selected and used as the original sampled image for subsequent processing, thereby ensuring the integrity and clarity of the image information to the greatest extent.

[0041] In the standard image generation stage, in order to eliminate the influence of uneven ambient lighting or sensor noise, the original sampled image and noise images of multiple adjacent areas are acquired. The brightness values ​​of each pixel in the original sampled image and the noise images are extracted and weighted by combining preset standard parameters to obtain a brightness difference evaluation value. This evaluation value reflects the systematic brightness shift of the image. Based on the shift direction of the brightness difference evaluation value, the module performs pixel-level brightness correction on the original sampled image and outputs a corrected image as the standard image.

[0042] During the template generation stage, the standard image is classified to identify and separate regions containing texture abnormalities, thereby obtaining abnormal defect images.

[0043] To improve the representativeness of the template, the module classifies and summarizes multiple abnormal defect images to obtain classification results. Using these classification results as a reference, the module extracts the image information of the target texture region from the corresponding abnormal defect image as the content to be verified.

[0044] Furthermore, edge processing is performed on the content to be verified to obtain corrected image information, ensuring the clarity and accuracy of the template edges. The preferred processing method is edge smoothing or sharpening. Finally, an abnormal texture template with a clear structure and well-defined features is generated based on this corrected image information.

[0045] The detection texture feature extraction module is used to continuously extract texture features for comparison from the surface of PE carpet on the production line. Specifically, it processes video streams or continuous image frames from image acquisition devices in real time. For each detection area on the PE carpet surface, it uses a preset image feature extraction algorithm to extract detection texture features that can characterize the texture state of that area.

[0046] Texture anomaly detection module: The main responsibility is to compare the real-time extracted texture features with the pre-built templates and make a final judgment. This module receives the detected texture features output by the texture feature extraction module and the abnormal texture templates generated by the abnormal texture template construction module. When performing recognition, for each detected texture feature, this module calculates the similarity between it and the abnormal texture template and determines the calculated similarity as the abnormal detection value. The similarity calculation can use various techniques, such as template matching and feature vector distance calculation.

[0047] The abnormal detection value is compared with the preset standard similarity parameter. When the abnormal detection value is greater than the standard similarity parameter, it indicates that the texture of the current detection area is highly similar to the known abnormal texture. The module will identify the detection area corresponding to the detected texture feature as a texture abnormal area.

[0048] To provide richer diagnostic information, after the comparison is completed, the deviation between the abnormal detection value and the standard similarity parameter is further calculated, and a corresponding offset signal is generated based on the magnitude of the deviation. This offset signal can quantitatively represent the severity of the abnormality and is output as the final identification result of the texture abnormal area for use by downstream systems such as alarm systems and sorting robots.

[0049] The above description is merely a preferred embodiment of this application and is not intended to limit this application. For those skilled in the art, this application can have various modifications and variations. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for detecting texture anomalies in PE carpets based on image recognition, characterized in that, include: In response to extracting the corresponding detection texture features from the detection area on the PE carpet surface, the detection texture features are compared with a pre-constructed abnormal texture template to identify texture abnormal areas in the PE carpet. The construction process of the abnormal texture template includes: A standard image is generated based on the original sampled image obtained from the PE carpet surface. The standard image is then classified to obtain anomaly defect images. Anomaly texture templates are then constructed based on the image features of the anomaly defect images.

2. The method for detecting texture anomalies in PE carpets based on image recognition according to claim 1, characterized in that, The standard image is generated based on the original sampled image obtained from the PE carpet surface, including: Acquire the original sampled image and noise images of multiple regions adjacent to the region of the original sampled image; extract the brightness values ​​of each pixel in the original sampled image and the noise images, and perform weighted calculation of the brightness values ​​in combination with preset standard parameters to obtain the brightness difference evaluation value; and correct the original sampled image according to the offset direction of the brightness difference evaluation value to output the corrected image as the standard image.

3. The method for detecting texture anomalies in PE carpets based on image recognition according to claim 2, characterized in that, Obtaining the original sampled image includes: obtaining a set of adjacent sampled images, and setting the first image in the set of adjacent sampled images as the reference image; The system sequentially verifies other images in a group of adjacent sampled images to obtain verification results; compares the verification results with a set of preset judgment parameters to obtain a reliable judgment result for identifying reliable images in the group of adjacent sampled images; and selects and determines the image with the largest brightness value as the original sampled image from all reliable images.

4. The method for detecting texture anomalies in PE carpets based on image recognition according to claim 3, characterized in that, Obtaining a set of adjacent sampled images includes: Multiple images from different directions are acquired, and the image features of each image are extracted to form a vector value. The vector values ​​are weighted by a preset directional weight to obtain multiple indicators to be sorted. Based on the indicators to be sorted in descending order of value, the order of multiple images is rearranged to generate a reordered image sequence as a group of adjacent sampled images.

5. The method for detecting texture anomalies in PE carpets based on image recognition according to claim 1, characterized in that, The process of constructing anomaly texture templates also includes: The abnormal defect images are classified and summarized to obtain classification results; using the classification results as a reference, the image information of the target texture region in the abnormal defect images is extracted as the content to be verified; and the edge processing of the content to be verified is performed to obtain the corrected image information, and an abnormal texture template is generated based on the corrected image information.

6. The method for detecting texture anomalies in PE carpets based on image recognition according to claim 1, characterized in that, The detected texture features are compared with a pre-constructed anomalous texture template to identify texture anomalous regions in the PE carpet, including: For each detected texture feature, its similarity to the abnormal texture template is calculated, and the similarity is determined as the anomaly detection value; the similarity is compared with the preset standard similarity parameter, and when the similarity is greater than the standard similarity parameter, the detection area corresponding to the detected texture feature is identified as a texture abnormal area.

7. The method for detecting texture anomalies in PE carpets based on image recognition according to claim 6, characterized in that, The process of comparing detected texture features with pre-built anomalous texture templates to identify texture anomalous regions in PE carpets also includes: After comparing the similarity with the preset standard similarity parameters, the deviation between the two is calculated, and an offset signal is generated based on the deviation. The offset signal is then output as the recognition result of the texture abnormal region.

8. A PE carpet texture anomaly detection system based on image recognition, characterized in that, include: An abnormal texture template construction module is used to construct an abnormal texture template based on an abnormal defect image generated from the original sampled image obtained from the PE carpet surface; The detection texture feature extraction module is used to extract the corresponding detection texture features from the detection area on the PE carpet surface; And a texture anomaly recognition module, which, in response to the texture features extracted by the texture feature extraction module, compares the texture features with the anomaly texture template constructed by the anomaly texture template construction module to identify texture anomaly areas in the PE carpet.

9. A PE carpet texture anomaly detection system based on image recognition according to claim 8, characterized in that, The abnormal texture template building module is configured as follows: Acquire a set of adjacent sampled images, and set the first image in the set of adjacent sampled images as the reference image; The verification results are obtained by sequentially verifying other images in a group of adjacent sampled images; The verification result is compared with a set of preset judgment parameters to obtain a reliable judgment result for indicating a reliable image in a group of adjacent sampled images; And among all the reliable images, the image with the highest brightness value is selected as the original sampled image.

10. A PE carpet texture anomaly detection system based on image recognition according to claim 8, characterized in that, Obtaining a set of adjacent sampled images includes: Multiple images from different directions are acquired, and the image features of each image are extracted to form a vector value. The vector values ​​are weighted by a preset directional weight to obtain multiple indicators to be sorted. Based on the indicators to be sorted in descending order of value, the order of multiple images is rearranged to generate a reordered image sequence as a group of adjacent sampled images.

Images