Rapid detection system and detection method for textile defects

Abstract

The invention discloses a rapid detection system and detection method for textile defects. The method comprises the following steps: acquiring an image of a textile detection area in real time; the texture period of the textile detection area image is analyzed, and periodic features are extracted to serve as a preliminary template; a convolutional neural network model is called to carry out defect detection on the preliminary template, and the positions of defects are marked; defects are extracted from the preliminary template, and a periodic texture template is formed; processing the defects, setting a threshold value of error defects, marking the error defects in the defects, and uploading the marked defects to a model database for learning; comparing the periodic texture template with the texture period of the textile, marking a potential defect area, comparing with an error defect, and cancelling the defect mark if the similarity is higher than a set threshold value; according to the method, the periodic texture template is compared with the textile texture period, the potential defect area is quickly locked, the potential defect area is detected again, and whether the defect is an error defect or not is judged.

Description

Technical Field

[0001] The invention relates to the technical field of fabric defect detection, and in particular to a rapid detection system and method for textile defects. Background Art

[0002] With the development of image processing technology, more and more applications involve the detection, identification, and tracking of a certain target. By extracting and tracking the target in the scene, the shape of the target can be recorded, and through computer processing, the analysis results of the target can be further obtained. In the use of digital image acquisition in automated defect detection, industrial cameras or integrated single-chip sensor micro lenses are mainly used for photoelectric conversion. In defect detection, there are some wrinkles or stains on the surface of the fabric, as well as feature deviations, but within the qualified quality range of the textile, such defects are easily judged as defects by the model, such defects are easily covered or misjudged, and defects are difficult to detect accurately, requiring re-detection, and low detection efficiency. Summary of the invention

[0003] The present invention overcomes the deficiencies of the prior art and provides a rapid detection system and method for textile defects.

[0004] To achieve the above object, the technical solution adopted by the present invention is: a method for rapid detection of textile defects, comprising the following steps:

[0005] Step S1, acquiring an image of a textile detection area in real time;

[0006] Step S2, analyzing the texture period of the textile inspection area image, and extracting periodic features as a preliminary template of the texture period;

[0007] Step S3, calling the convolutional neural network model to perform defect detection on the preliminary template and mark the location of the defect;

[0008] Step S4, extracting defects from the preliminary template to form a periodic texture template;

[0009] Step S5, process the defects, set the threshold of the error defects, mark the error defects among the defects, and upload the defects marked in step S3 to the model database for learning;

[0010] Step S6: compare the periodic texture template with the textile texture period, mark the potential defect area, and compare it with the error defect. If the similarity is higher than the set threshold, cancel the defect mark.

[0011] In a preferred embodiment of the present invention, in the training step S1, a high-resolution camera is used to photograph the textile, and the shooting angle is perpendicular to the surface of the textile.

[0012] In a preferred embodiment of the present invention, the training step S2 includes the following sub-steps:

[0013] Step S21, preprocessing the image and performing grayscale processing;

[0014] Step S22, extracting periodic features of period length, direction, and repeatability;

[0015] Step S23, combining the obtained period length, direction and repeatability, segmenting the image to obtain several groups of repeated templates, and selecting one group of templates as the preliminary template of the texture period.

[0016] In a preferred embodiment of the present invention, in the training step S22, the image is converted from the spatial domain to the frequency domain by Fourier transform, the peak position in the spectrum is extracted, and the length and direction of the period are obtained; the repetition period of the texture is extracted using the autocorrelation function.

[0017] In a preferred embodiment of the present invention, the training step S3 includes the following sub-steps:

[0018] Step S31, providing a number of defect samples, marking the defect samples, and classifying the defect samples;

[0019] Step S32, making the classified defect samples into a data set;

[0020] Step S33: Use the data set to train the convolutional neural network model.

[0021] In a preferred embodiment of the present invention, in the training step S5, the defect sample is compared with the defect-free periodic texture template in terms of gray value distribution and texture direction consistency, a similarity threshold is set, and areas below the threshold are regarded as error defects.

[0022] In a preferred embodiment of the present invention, in the training step S5, multi-dimensional features of error defect color, texture direction, and brightness are extracted and uploaded to the database for training.

[0023] In a preferred embodiment of the present invention, the training step S6 includes the following sub-steps:

[0024] Step S61, aligning the periodic texture template with other templates in terms of direction and period;

[0025] Step S62, evaluating the periodic texture template and other templates with similarity values, and setting a similarity threshold, and marking positions below the set threshold as potential defect areas;

[0026] Step S63, matching the potential defect area with the features in the error defect database, calculating the similarity of the multi-dimensional feature vectors of the potential defect and the error defect, if the similarity is higher than the set threshold, it is an error defect, and the defect mark is cancelled.

[0027] A rapid detection system for textile defects, comprising: an image acquisition module, a texture periodic analysis module, a defect detection module, a periodic texture template generation module, a defect processing module, and a template comparison module.

[0028] The training image acquisition module is used to collect high-resolution images of the textile inspection area as the basic data source for subsequent texture analysis and defect detection;

[0029] The training texture periodicity analysis module is used to extract the periodicity characteristics of the texture period length, direction, and repeatability from the collected textile images, so as to segment the images to obtain several groups of repeated templates, and select one group of templates as the preliminary templates of the texture periodicity;

[0030] The defect detection module is trained to detect the preliminary template using a convolutional neural network to identify and mark the location of defects;

[0031] The training periodic texture template generation module is used to extract the defect area in the preliminary template, remove the defects, and generate a defect-free periodic texture template;

[0032] The training defect processing module is used to detect defects and perform feature analysis. According to the set similarity threshold, the defect area below the threshold is marked as an error defect and uploaded to the error defect database;

[0033] The training template comparison module is used to compare the periodic texture template with other texture periods, mark potential defect areas, compare the features of potential defects with those in the error defect database, and unmark areas with similarities higher than a set threshold.

[0034] In a preferred embodiment of the present invention, the training defect processing module includes: a feature extraction unit, a threshold determination unit and a database update unit;

[0035] The training feature extraction unit extracts multi-dimensional features such as gray value distribution, texture direction consistency, color, brightness, etc. of defects;

[0036] The training threshold determination unit is used to set the similarity threshold, and the area below the threshold is marked as an error defect;

[0037] The training database updating unit is used to upload the error defect features to the database.

[0038] The present invention solves the defects existing in the background technology and has the following beneficial effects:

[0039] (1) The present invention provides a method for rapid detection of textile defects. By extracting a texture cycle template, the amount of area that needs to be directly detected is reduced. The defect-free texture cycle template is used as a reference to compare with other texture cycles, and defects that may exist in other cycles are quickly located. The complex defect detection problem is decomposed into a consistency analysis of the texture cycle, which greatly reduces the amount of calculation, thereby improving the detection efficiency. At the same time, potential defects are screened out by comparison, and error defects can be quickly separated, avoiding unnecessary operations caused by error defects, thereby improving the accuracy of defect detection.

[0040] (2) The present invention identifies error defects among defects by setting a threshold value of error defects, which can effectively eliminate error defects in textile defect detection, reduce false alarms of defects, and improve the efficiency of defect detection. The characteristics of error defects are uploaded to the database and integrated into the learning model, thereby optimizing the judgment criteria of the model and improving the model's ability to recognize defects.

[0041] (3) The present invention quickly locks in potential defect areas by periodically comparing the periodic texture template with the textile texture, and re-detects the potential defect areas to determine whether they are error defects. By confirming the defects twice, it is ensured that real defects will not be missed due to the interference of error defects, thereby enhancing the ability to identify defects. By eliminating error defects, the subsequent analysis and processing time of non-defect areas is reduced, the resource allocation of the detection system is optimized, and the detection efficiency is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work.

[0043] Figure 1 is a flow chart of a method for rapid detection of textile defects according to a preferred embodiment of the present invention;

[0044] Figure 2 It is a schematic diagram of the framework of a rapid detection system for textile defects according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0047] Application Overview:

[0048] As mentioned above, in the field of fabric defect detection technology, the effectiveness of using images for detection will affect the correctness and effectiveness of making decisions based on images.

[0049] With the development of computer vision technology, automatic detection methods based on image processing have gradually been applied to textile defect detection. In textile defect detection, machine vision is used to generate images to detect defects in textile fabrics. This type of method usually obtains the surface image of the textile through an image acquisition device, and combines the algorithm to analyze the texture features in the image to identify abnormal areas. However, there are some wrinkles or stains on the surface of the fabric, as well as feature deviations, but within the qualified quality range of the textile. These are easily judged as defects by the model. However, when using machine vision to generate images to detect defects in textile fabrics, such defects cannot be judged, which can easily lead to missed detection or misjudgment of defects.

[0050] Therefore, in response to the above technical problems, an embodiment of the present invention provides a rapid detection system and method for textile defects, which quickly locks potential defect areas by periodically comparing periodic texture templates with textile textures, and re-detects the potential defect areas to determine whether they are error defects.

[0051] like Figure 1 As shown, a method for rapid detection of textile defects comprises the following steps:

[0052] Step S1, acquiring an image of a textile detection area in real time.

[0053] Step S2, analyzing the texture period of the textile inspection area image, and extracting periodic features as a preliminary template of the texture period;

[0054] It includes the following sub-steps:

[0055] Step S21, perform denoising and contrast enhancement preprocessing on the image, optimize image quality, eliminate detection interference, and grayscale the optimized image to convert the image from color space to grayscale space, retain the brightness information of the textile texture, and reduce the computational complexity.

[0056] Step S22, extracting the periodic characteristics of period length, direction, and repeatability, converting the image from the spatial domain to the frequency domain through Fourier transform, extracting the peak position in the spectrum, and obtaining the length and direction of the period; and extracting the repetition period of the texture using the autocorrelation function;

[0057] Specifically, the grayscale image is converted from the spatial domain to the frequency domain to obtain a spectrum diagram; in the spectrum, the periodic characteristics of the texture will be manifested as obvious peak positions and directions. The period length is determined by calculating the distance from the peak to the origin; the period direction is determined by the angle of the peak.

[0058] The autocorrelation function of the image is calculated in the spatial domain to analyze the repeatability of the texture. The local maximum position of the autocorrelation function corresponds to the repetition period of the texture.

[0059] For example, for a textile with horizontal stripes, its spectrum may show obvious periodic peaks in the horizontal direction, and the peak spacing corresponds to the period length; the direction of the stripes is horizontal (0°).

[0060] Step S23, combining the obtained period length, direction and repeatability, segmenting the image to obtain several groups of repeated templates, and selecting one group of templates as the preliminary template of the texture period.

[0061] Step S3, calling the convolutional neural network model to perform defect detection on the preliminary template and mark the location of the defect; performing a sliding window comparison between the preliminary template and each local area of ​​the target image through the convolutional neural network model to identify the defect location of the preliminary template;

[0062] By deeply analyzing the preliminary template of the textile texture cycle, detecting defects and marking their specific locations, deep learning is used to accurately identify abnormal areas in complex textile textures, laying the foundation for subsequent template optimization and error analysis.

[0063] Step S4, extracting defects from the preliminary template to form a periodic texture template;

[0064] The periodic texture template represents the normal texture pattern of the fabric, that is, the regular structure within the texture period. When the texture of the image to be tested is compared with the standard template during the detection process, the area that does not match the standard pattern can be effectively identified, that is, the defect area, thereby improving the accuracy of defect detection. Moreover, by comparing the periodic texture template with the area to be tested of the textile, the defect area can be accurately located, reducing false alarms and missed alarms, and improving the detection effect.

[0065] By extracting the texture cycle template, the amount of area that needs to be directly inspected is reduced. The defect-free texture cycle template is used as a benchmark for comparison with other texture cycles to quickly locate possible defects in other cycles. The complex defect detection problem is decomposed into a consistency analysis of the texture cycle, which greatly reduces the amount of calculation and thus improves the detection efficiency.

[0066] In the preliminary template, the detected defect area is replaced with the adjacent normal texture area using the texture repair algorithm, so that the repaired area is consistent with the overall periodic texture of the template;

[0067] Fourier transform and autocorrelation analysis are used to extract the periodic features in the template, obtain the frequency distribution of the period, locate the primitive units of the preliminary template according to the periodic characteristics, extract the primitives of the periodic texture from the preliminary template, ensure that the primitives do not contain defects, and repeatedly tile and splice the primitives to generate a complete periodic texture template; this provides a basis for the subsequent rapid comparison of the periodic texture template directly with the target image, significantly reducing the computational complexity.

[0068] Step S5, process the defects, set the threshold of the error defects, mark the error defects among the defects, and upload the defects marked in step S3 to the model database for learning;

[0069] Process the defects and classify them into accurate defects and error defects;

[0070] Error defects specifically include some wrinkles or stains on the fabric surface, as well as feature deviations. However, they are within the qualified quality range of textiles and are easily judged as defects by the model. Feature deviations include: size deviation, color deviation, and texture deviation.

[0071] The threshold setting for error defects is set by the area of ​​the defect, that is, the ratio of the area of ​​the defect to the periodic texture template: threshold = defect area / area of ​​the periodic texture template; based on sample statistics, an area ratio is defined, the area ratio of point defects is ≤0.1%, and the length ratio of linear defects is ≤2%. Defects below this ratio are judged as error defects;

[0072] By setting the threshold, the position and number of error defects in the defects are determined, the obtained error defects are summarized, and the defects and the summarized error defects are uploaded to the model database for learning. By continuously uploading and learning the defect features, the training data set size of the convolutional neural network is increased, thereby enhancing the model's ability to recognize similar defects, and can more efficiently adapt to the inspection needs of different types of textiles.

[0073] Step S6, comparing the periodic texture template with the textile texture period, marking the potential defect area, and comparing it with the error defect. If the similarity is higher than the set threshold, the defect mark is cancelled;

[0074] It includes the following sub-steps:

[0075] Step S61, aligning the periodic texture template with other templates in terms of direction and period.

[0076] Step S62: Evaluate the periodic texture template and other templates based on similarity values, set a similarity threshold, and mark positions below the set threshold as potential defect areas.

[0077] Step S63, matching the potential defect area with the features in the error defect database, calculating the similarity of the multi-dimensional feature vectors of the potential defect and the error defect, if the similarity is higher than the set threshold, it is an error defect, and the defect mark is cancelled.

[0078] In step S62, by measuring the structural similarity between the periodic texture template and other templates, the similarity of the two images in dimensions such as brightness, contrast, and structure is measured, wherein the similarity value ranges from (0, 1), wherein 1 represents complete similarity, and a corresponding similarity threshold is established based on experience, such as a similarity threshold of 0.95, and defect locations below the similarity threshold are potential defect areas;

[0079] The similarity of the multi-dimensional feature vectors of potential defects and error defects is calculated. If the similarity is higher than the set threshold, it is an error defect and the defect mark is cancelled. By setting the threshold of the error defect, the error defect among the defects can be identified. In the detection of textile defects, the error defects can be effectively eliminated, the false alarm of defects can be reduced, and the efficiency of defect detection can be improved. The characteristics of the error defects are uploaded to the database and integrated into the learning model, the judgment criteria of the model are optimized, and the model's ability to recognize defects is improved.

[0080] The present invention quickly locks potential defect areas by periodically comparing the periodic texture template with the textile texture period, and re-detects the potential defect areas to determine whether they are error defects. By confirming the defects twice, it is ensured that real defects will not be missed due to the interference of error defects, thereby enhancing the ability to identify defects. By eliminating error defects, the subsequent analysis and processing time of non-defect areas is reduced, the resource allocation of the detection system is optimized, and the detection efficiency is improved.

[0081] In the present invention, in the training step S1, a high-resolution camera is used to photograph the textile, and the shooting angle is perpendicular to the surface of the textile.

[0082] In a preferred embodiment of the present invention, the training step S3 includes the following sub-steps:

[0083] Step S31, provide several defect samples, mark the defect samples, and classify the defect samples; specifically collect different types of defect samples from the textile production process, such as color difference, broken thread, hole, etc., and these defect samples cover common textile textures and possible defect types to ensure data diversity and representativeness; through sample labeling and classification, build a high-quality training data set so that the model can accurately identify and distinguish different types of defects.

[0084] Step S32, making the classified defect samples into a data set; specifically, standardizing the labeled defect samples, including: image size adjustment and data enhancement;

[0085] Several defects are randomly combined with several non-defect samples to obtain a large number of new defect samples, and the positions of the defects on the non-defect samples are random to simulate the actual defect distribution. This step is repeated until a sufficient number of new defect samples are generated. The acquired defect samples and new defect samples are integrated to obtain a data set.

[0086] By extracting defect features from a small number of defect samples and randomly combining them with non-defect samples, a large number of different defect samples can be automatically generated, which can increase the number of samples in the original data set, enrich the diversity of samples, improve the generalization ability of the model, and reduce the cost of data set production.

[0087] Step S33: Use the data set to train the convolutional neural network model.

[0088] In the present invention, in the training step S5, the defect sample is compared with the defect-free periodic texture template in terms of gray value distribution and texture direction consistency, a similarity threshold is set, and areas below the threshold are regarded as error defects.

[0089] In the present invention, in the training step S5, multi-dimensional features of error defect color, texture direction, and brightness are extracted and uploaded to the database for training.

[0090] like Figure 2 As shown, a rapid detection system for textile defects includes: an image acquisition module, a texture periodic analysis module, a defect detection module, a periodic texture template generation module, a defect processing module, and a template comparison module.

[0091] The training image acquisition module is used to collect high-resolution images of the textile inspection area as the basic data source for subsequent texture analysis and defect detection;

[0092] The training texture periodicity analysis module is used to extract the periodicity characteristics of the texture period length, direction, and repeatability from the collected textile images, so as to segment the images to obtain several groups of repeated templates, and select one group of templates as the preliminary templates of the texture periodicity;

[0093] The defect detection module is trained to detect the preliminary template using a convolutional neural network to identify and mark the location of defects;

[0094] The training periodic texture template generation module is used to extract the defect area in the preliminary template, remove the defects, and generate a defect-free periodic texture template;

[0095] The training defect processing module is used to detect defects and perform feature analysis. According to the set similarity threshold, the defect area below the threshold is marked as an error defect and uploaded to the error defect database;

[0096] The training template comparison module is used to compare the periodic texture template with other texture periods, mark potential defect areas, compare the features of potential defects with those in the error defect database, and unmark areas with similarities higher than a set threshold.

[0097] In the present invention, the training defect processing module includes: a feature extraction unit, a threshold determination unit and a database update unit;

[0098] The training feature extraction unit extracts multi-dimensional features such as gray value distribution, texture direction consistency, color, brightness, etc. of defects;

[0099] The training threshold determination unit is used to set the similarity threshold, and the area below the threshold is marked as an error defect;

[0100] The training database updating unit is used to upload the error defect features to the database.

[0101] The above is based on the ideal embodiment of the present invention. Through the above description, relevant personnel can make various changes and modifications without departing from the technical concept of the present invention. The technical scope of the present invention is not limited to the content in the specification, and the technical scope must be determined according to the scope of the claims.

Claims

1. A method for rapid detection of textile defects, characterized in that: The following steps are involved: Step S1, acquiring an image of a textile detection area in real time; Step S2, analyzing the texture period of the textile inspection area image, and extracting periodic features as a preliminary template of the texture period; Step S3, calling the convolutional neural network model to perform defect detection on the preliminary template and mark the location of the defect; Step S4, extracting defects from the preliminary template to form a periodic texture template; Step S5, process the defects, set the threshold of the error defects, mark the error defects among the defects, and upload the defects marked in step S3 to the model database for learning; Step S6: compare the periodic texture template with the textile texture period, mark the potential defect area, and compare it with the error defect. If the similarity is higher than the set threshold, cancel the defect mark.

2. A rapid detection method for textile defects according to claim 1, characterized in that: In step S1, the textile is photographed using a high-resolution camera, with the photographing angle being perpendicular to the surface of the textile.

3. A rapid detection method for textile defects according to claim 1, characterized in that: In the step S2, the following sub-steps are included: Step S21, preprocessing the image and performing grayscale processing; Step S22, extracting periodic features of period length, direction, and repeatability; Step S23, combining the obtained period length, direction and repeatability, segmenting the image to obtain several groups of repeated templates, and selecting one group of templates as the preliminary template of the texture period.

4. A rapid detection method for textile defects according to claim 3, characterized in that: In step S22, the image is converted from the spatial domain to the frequency domain by Fourier transform, the peak position in the spectrum is extracted, and the length and direction of the period are obtained; the repetition period of the texture is extracted using the autocorrelation function.

5. A rapid detection method for textile defects according to claim 1, characterized in that: In the step S3, the following sub-steps are included: Step S31, providing a number of defect samples, marking the defect samples, and classifying the defect samples; Step S32, making the classified defect samples into a data set; Step S33: Use the data set to train the convolutional neural network model.

6. A rapid detection method for textile defects according to claim 1, characterized in that: In step S5, the defect sample is compared with the defect-free periodic texture template in terms of gray value distribution and texture direction consistency, a similarity threshold is set, and areas below the threshold are regarded as error defects.

7. A rapid detection method for textile defects according to claim 1, characterized in that: In step S5, multi-dimensional features of error defect color, texture direction, and brightness are extracted and uploaded to a database for training.

8. A rapid detection method for textile defects according to claim 1, characterized in that: In step S6, the following sub-steps are included: Step S61, aligning the periodic texture template with other templates in terms of direction and period; Step S62, evaluating the periodic texture template and other templates with similarity values, and setting a similarity threshold, and marking positions below the set threshold as potential defect areas; Step S63, matching the potential defect area with the features in the error defect database, calculating the similarity of the multi-dimensional feature vectors of the potential defect and the error defect, if the similarity is higher than the set threshold, it is an error defect, and the defect mark is cancelled.

9. A rapid detection system for textile defects, comprising: The image acquisition module, the texture periodicity analysis module, the defect detection module, the periodic texture template generation module, the defect processing module and the template comparison module are characterized in that: The image acquisition module is used to acquire high-resolution images of the textile inspection area as a basic data source for subsequent texture analysis and defect detection; The texture periodicity analysis module is used to extract the periodicity characteristics of the texture period length, direction, and repeatability from the collected textile image, thereby segmenting the image to obtain a plurality of groups of repeated templates, and selecting a group of templates as the preliminary templates of the texture period; The defect detection module uses a convolutional neural network to detect the preliminary template, identify and mark the location of the defect; The periodic texture template generation module is used to extract the defect area in the preliminary template, remove the defects, and generate a defect-free periodic texture template; The defect processing module is used to detect defects and perform feature analysis. According to a set similarity threshold, defect areas below the threshold are marked as error defects and uploaded to the error defect database; The template comparison module is used to compare the periodic texture template with other texture periods, mark potential defect areas, compare the features of potential defects with those in the error defect database, and unmark areas with similarities higher than a set threshold.

10. A rapid detection system for textile defects according to claim 9, characterized in that: The defect processing module includes: a feature extraction unit, a threshold determination unit and a database update unit; The feature extraction unit extracts multi-dimensional features such as gray value distribution, texture direction consistency, color, brightness, etc. of the defect; The threshold determination unit is used to set a similarity threshold, and areas below the threshold are marked as error defects; The database updating unit is used to upload the error defect features to the database.

Images