A method and system for detecting defects on a textile surface

Abstract

The present application relates to the technical field of textile detection, and particularly relates to a textile surface defect detection method and system, which is used for online detection of a textile with a heat-sensitive coating on the surface, and comprises the following steps: applying a preset periodic intensity modulation to a detection light source used for online detection, so as to generate a reference sequence as a synchronism analysis benchmark; collecting a series of brightness values of a suspected defect point preliminarily identified on the surface of the textile within a preset time period under the periodic intensity modulation, so as to constitute a brightness response sequence representing optical response characteristics of the suspected defect point; calculating a time synchronization degree between the brightness response sequence and the reference sequence; and distinguishing between temporary thermal effect artifacts and real physical defects, so as to effectively distinguish between temporary optical artifacts caused by thermal effects and real physical defects, reduce a false positive rate, and improve detection accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of textile detection, and particularly relates to a textile surface defect detection method and system. BACKGROUND

[0002] In modern textile industry, especially in the field with extremely strict product quality requirements, it has become a key link to guarantee product qualification rate and production efficiency to use an automatic optical detection system to detect the surface defects of textiles on line. Such a system usually relies on high-intensity illumination and high-resolution cameras to capture the subtle images of the surface of the fabric moving at high speed on the production line, and to identify defects such as broken yarns, stains and holes through image analysis. However, with the development of functional textiles, the special physical properties of some new materials have brought new challenges to the traditional optical detection method. In particular, for those textiles with functional coating on the surface, the coating material may show sensitivity to the physical quantities (such as temperature) in the detection environment, thereby generating interference signals that are difficult to distinguish from real defects in the detection process, resulting in a large number of false positives.

[0003] In order to be able to identify the subtle structural defects such as broken yarns, thin or thick roads that may exist on the fabric substrate and have an impact on product performance, as well as the defects of the coating itself, a plurality of high-brightness linear LED light sources are installed in the detection dark box. These light sources continuously and intensively irradiate the surface of the fabric from a specific angle, ensuring that the line array camera can capture clear enough surface images with high signal-to-noise ratio. However, as the continuous production time is prolonged, the system begins to frequently issue alarms, reporting that a large number of small and irregularly shaped “dark spots” or “bright spots” defects have been found on the surface of the fabric. When the operator suspends the production line for manual re-inspection, no physical abnormalities can be found in the static state, and these so-called defects disappear without a trace.

[0004] The root of the problem lies in the fact that prolonged, continuous irradiation by high-intensity LED light sources causes the fabric surface to absorb considerable heat. Due to subtle differences in density and heat capacity at the microscopic level (such as at yarn interlacing points), numerous "micro-hot spots" with slightly higher temperatures than their surroundings form on the fabric surface. The functional coating on this fabric surface is temperature-sensitive, and the optical reflectivity of the coating above these "micro-hot spots" undergoes temporary, slight changes, resulting in blurred "dark spots" or "bright spots" in the image from the linear array camera. These temporary optical artifacts caused by thermal effects are almost indistinguishable in size, shape, and contrast from real, permanent defects caused by unstable coating processes (such as a small drop of incompletely spread paint or a tiny missed spot due to an air bubble). This leads to a dilemma where the detection system cannot distinguish from a static image frame whether the underlying physical cause is a temporary thermal effect or a permanent material defect, resulting in a high false alarm rate and a high risk of missed detections.

[0005] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method and system for detecting surface defects in textiles.

[0007] In a first aspect, the present invention provides a method for detecting surface defects in textiles, used for online detection of textiles with a heat-sensitive coating on their surface, the method comprising the following steps:

[0008] A preset periodic intensity modulation is applied to the detection light source used in online detection to generate a reference sequence as a benchmark for synchronization analysis.

[0009] For suspected defects initially identified on the surface of the textile, under the periodic intensity modulation, a series of brightness values ​​of the suspected defects are collected within a preset time period to form a brightness response sequence characterizing its optical response properties.

[0010] Calculate the degree of time synchronization between the brightness response sequence and the reference sequence;

[0011] Based on the degree of time synchronization, the suspected defect point is determined to be either a real physical defect or a temporary optical artifact caused by thermal effects.

[0012] The core innovation of the present application is that by applying a preset periodic intensity modulation to the detection light source and analyzing the time synchronization degree between the suspected defect point brightness response sequence and the modulation synchronization analysis reference sequence, the dynamic optical response difference between the real physical defects and the temporary optical artifacts caused by thermal effects under periodic light is utilized to solve the problem of distinguishing between the two types of defects in the prior art, and to filter the false positives caused by thermal artifacts.

[0013] In a second aspect, a textile surface defect detection system is provided for online detection of a textile with a heat-sensitive coating on the surface, the system comprising:

[0014] A modulation module for applying a preset periodic intensity modulation to a detection light source used for online detection to generate a reference sequence as a synchronization analysis reference;

[0015] An acquisition module for acquiring a series of brightness values of a suspected defect point preliminarily identified on the surface of the textile within a preset time period under the periodic intensity modulation to form a brightness response sequence representing the optical response characteristics of the suspected defect point;

[0016] A calculation module for calculating the time synchronization degree between the brightness response sequence and the reference sequence;

[0017] A determination module for determining whether the suspected defect point is a real physical defect or a temporary optical artifact caused by thermal effects based on the time synchronization degree.

[0018] Compared with the prior art, the present application has the following advantages:

[0019] By applying periodic intensity modulation to the detection light source and analyzing the time synchronization of the brightness response of the suspected defect point and the modulation signal, and if necessary, combining frequency analysis based on geometric features, the temporary thermal effect artifact and the real physical defect are distinguished, which has the advantages of effectively distinguishing the temporary optical artifact caused by thermal effects and the real physical defect, reducing the false positive rate and improving the detection accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 The method flowchart of the present application.

[0021] Figure 2 The system structure diagram of the present application.

[0022] In the figure: 201, modulation module; 202, acquisition module; 203, calculation module; 204, determination module. DETAILED DESCRIPTION

[0023] Embodiments of the present application are described in detail below with reference to the attached drawing figures, wherein the same or like component have the same or similar designations. The embodiments described below are presented by way of example only and are not intended to limit the present application as defined by the claims.

[0024] The terms "first", "second", etc. are used only for descriptive purposes and do not connote or imply relative importance or an ordering between or among the indicated technical features. Thus, a feature defined with "first", "second", etc. can include one or more of the features implicitly or explicitly. In the description of the present application, the meaning of "a plurality" is two or more, unless otherwise expressly specified.

[0025] The conventional existing automated optical inspection system has the problem of being difficult to distinguish between real physical defects and temporary optical artifacts induced by the thermal effect of the detection light source when detecting the textile with a surface covered with a thermosensitive coating. High-intensity illumination causes local temperature rise on the microstructure of the fabric, which in turn causes temporary changes in the optical properties of the thermosensitive coating, forming artifacts in the online array camera image that are similar to the visual features of real defects. This similarity makes it difficult to accurately determine the physical cause of the signal by relying solely on static image features, resulting in a high false alarm rate of the detection system.

[0026] For example, assume that a production line is detecting a composite textile with a surface covered with a thermosensitive coating. To ensure clear images of the high-speed moving fabric, the system uses a high-intensity LED light source for continuous illumination. This illumination causes local temperature rise on the microstructure of the fabric surface, and the thermosensitive coating is sensitive to temperature, so its optical properties change temporarily, forming temporary optical artifacts in the image that are similar to the visual features of real defects. These artifacts are difficult to distinguish from permanent defects in static images, resulting in frequent false alarms of the detection system.

[0027] If the above problem is not solved, the detection system will produce many false alarms, requiring frequent manual review, reducing production efficiency. A high false alarm rate can cause operators to become fatigued by the alarms, increasing the risk of missing real defects. The inability to reliably distinguish between temporary artifacts and permanent defects will compromise the system's quality control capabilities, resulting in unreliable detection results.

[0028] To this end, the present application provides a textile surface defect detection method as shown in Figure 1 The method comprises the following steps:

[0029] S101, applying a predetermined periodic intensity modulation to the detection light source used for online detection to generate a reference sequence as a synchronization analysis reference;

[0030] S102, for the suspected defect point on the textile surface which is preliminarily identified, a series of brightness values of the suspected defect point in a preset time period are collected under periodic intensity modulation to constitute a brightness response sequence representing optical response characteristics thereof;

[0031] S103, a time synchronization degree between the brightness response sequence and the reference sequence is calculated;

[0032] S104, based on the time synchronization degree, it is determined whether the suspected defect point is a real physical defect or a temporary optical artifact caused by thermal effect.

[0033] The detection light source applies a preset periodic intensity modulation, which means that the light intensity of the light source used to irradiate the textile surface changes regularly according to a predetermined time period, for example, the intensity change can be in the form of a sine wave, square wave, pulse sequence, etc. It can be achieved by controlling the driving current or voltage of the light source, for example, by pulse width modulation (PWM) to control the brightness of the LED light source. The main purpose is to make the change of the optical response of the textile surface caused by the thermal effect of the light source show the characteristics related to the periodic modulation, so as to provide time dimension information for subsequent differentiation between real defects and thermal false alarms. The reference sequence generated as a synchronization analysis reference is a sequence that is synchronized with the modulation law in time according to the preset periodic intensity modulation law applied to the detection light source. The sequence can be ideal modulation waveform data or actual monitored light source intensity change data. It can use the control signal output by the signal generator as a reference. The main purpose is to use it as a standard to measure whether the brightness response sequence of the suspected defect point is synchronized with the light source modulation. The collection of a series of brightness values of the suspected defect point within a preset time period is to use an image sensor or a photoelectric detector to continuously record the light signal intensity of the position of the suspected defect point identified initially within a period of time while the detection light source is periodically modulated in intensity, and convert it into a digital brightness value sequence. It can be achieved by using a high-speed linear array camera or a point sensor to continuously sample specific pixels or regions. The main purpose is to obtain the dynamic optical response data of the suspected defect point under periodic light change. The brightness response sequence representing the optical response characteristics is a sequence that reflects the change of the brightness of the suspected defect point with time within a preset time period, which is arranged in time sequence according to the brightness values collected. The shape, amplitude, phase and other characteristics of the sequence are related to the response characteristics of the suspected defect point to the periodic light modulation. The main purpose is to use it as input data for subsequent time synchronization degree calculation. The calculation of the time synchronization degree between the brightness response sequence and the reference sequence is to quantify the correlation or consistency of the brightness response sequence and the reference sequence in the time dimension by using signal processing or statistical methods, for example, it can calculate the cross-correlation coefficient, phase difference, frequency component matching degree, etc. It can be achieved by using a digital signal processor (DSP) or a general-purpose processor to execute related algorithms. The main purpose is to determine whether the brightness change of the suspected defect point is synchronized with the periodic modulation of the detection light source.The determining whether the suspected defect point is a real physical defect or a temporary optical artifact caused by thermal effect based on the time synchronization degree refers to comparing the calculated time synchronization degree value with a preset discrimination threshold, thereby making a conclusion about the nature of the suspected defect point. If the synchronization degree is high, it is inclined to be determined as a thermal artifact; if the synchronization degree is low, it is inclined to be determined as a real physical defect. The main purpose is to distinguish and filter real defects and thermal artifacts by using the dynamic response difference of the suspected defect point to periodic light modulation.

[0034] The scheme of the present application overcomes the limitation of relying only on static image features for determination by introducing time dimension analysis means. Specifically, first, a preset periodic intensity modulation is applied to the detection light source used for online detection, which is equivalent to applying an external excitation with a specific time rule to the textile surface. At the same time, a reference sequence is generated according to the modulation rule, which represents the ideal excitation signal timing. Then, for the suspected defect point identified initially, a series of brightness values of the suspected defect point in a period of time are continuously collected under the periodic modulation illumination of the light source, forming a brightness response sequence representing the dynamic optical response of the point. This brightness response sequence contains the actual response information of the suspected defect point to the periodic light excitation. Since the textile with a heat-sensitive coating is sensitive to the thermal effect caused by light, the brightness change of the temporary optical artifact caused by thermal effect tends to follow the periodic modulation of the light source, showing a high time synchronization with the reference sequence. While the real physical defect is structural, its optical properties are relatively stable, and even under periodic light, its brightness change (if any) usually does not show a high degree of synchronization with the light source modulation. Therefore, by calculating the time synchronization degree between the brightness response sequence and the reference sequence, the correlation between the brightness change of the suspected defect point and the light source modulation can be quantified. Finally, based on the calculated time synchronization degree, a discrimination standard is set, for example, a synchronization threshold, and the synchronization degree higher than the threshold is determined as a temporary optical artifact caused by thermal effect, and the synchronization degree lower than the threshold is determined as a real physical defect. The whole process forms a complete dynamic response analysis chain, enabling the system to distinguish the nature of the defect based on the dynamic behavior difference of the suspected defect point to the periodic light.

[0035] In one embodiment, the detection light source used for online detection can be an array of LEDs. The array of LEDs is subjected to a predetermined periodic intensity modulation, for example by controlling the driving current of the LEDs through a pulse width modulation (PWM) technique, so that the brightness of the LEDs is periodically varied at a fixed frequency (e.g. 100 Hz) and waveform (e.g. square wave or sine wave). In synchronization with the PWM control signal, a digitized reference sequence is generated, which reflects the expected variation pattern of the light source intensity. As the textile passes through the detection region, the surface of the fabric is continuously imaged by a line array camera. When a suspected defect point is preliminarily identified by an image processing algorithm, the system locks the corresponding pixel position of the suspected defect point in the consecutive image frames. Within a predetermined time period (e.g. 100 ms), a series of brightness values of the pixel position are extracted from the consecutive image frames, thus forming a brightness response sequence of the suspected defect point. The brightness response sequence reflects the variation of the brightness of the point over time under the periodic modulation of the light source. To calculate the degree of time synchronization, cross-correlation analysis can be performed on the brightness response sequence and the reference sequence, and a cross-correlation function between them is calculated. The peak value size and position of the cross-correlation function can represent the degree of time synchronization of the two sequences. If a peak value appears at zero time lag or a fixed time lag, it indicates that the brightness response sequence and the reference sequence have a high degree of time synchronization. Based on the comparison of the size of the cross-correlation peak value with a predetermined threshold value, if the peak value is higher than the threshold value, it is determined that the suspected defect point is a temporary optical artifact caused by thermal effect; if the peak value is lower than the threshold value, it is determined to be a real physical defect.

[0036] As an embodiment of the present application, in the case where the degree of time synchronization is lower than a predetermined synchronization threshold, the method further comprises:

[0037] obtaining a geometric feature frequency determined by the physical size of the three-dimensional periodic structure of the textile and the running speed of the textile;

[0038] analyzing the frequency composition of the brightness response sequence to determine its dominant frequency;

[0039] based on the matching relationship between the dominant frequency and the geometric feature frequency, re-determining whether the suspected defect point is a real physical defect or a temporary optical artifact.

[0040] The geometric feature frequency refers to a frequency determined by the physical size of the three-dimensional periodic structure of the textile and the running speed of the textile, is an inherent frequency of the periodic structure of the textile in the image sequence, and reflects the periodic change of the structure of the textile in a moving state. The frequency composition of the brightness response sequence refers to the energy distribution or intensity distribution of the sequence at different frequencies obtained by frequency analysis of the brightness response sequence. The dominant frequency refers to the frequency component with the most prominent energy or intensity in the frequency composition of the brightness response sequence, which represents the most significant periodic pattern in the brightness change. The matching relationship refers to whether there is a preset correlation or consistency between the dominant frequency and the geometric feature frequency, for example, whether the frequency values of the two are close or equal within the allowed error range.

[0041] The scheme of the present application introduces a secondary judgment mechanism based on frequency analysis when the time synchronization degree is lower than the preset synchronization threshold. First, the geometric feature frequency determined by the physical size of the three-dimensional periodic structure of the textile and the running speed of the textile is obtained, which represents the inherent frequency response of the periodic structure of the textile in the movement process. Then, the frequency composition of the suspected defect point brightness response sequence is analyzed to determine the dominant frequency, which reflects the main periodic component of the brightness change of the suspected defect point. Finally, based on the matching relationship between the dominant frequency and the geometric feature frequency, the type of the suspected defect point is judged again. If the dominant frequency matches the geometric feature frequency, it means that the brightness change period of the suspected defect point is consistent with the structure period of the textile, and it is very likely to be a temporary optical artifact caused by the periodic structure of the textile. On the contrary, if they do not match, it means that the brightness change of the suspected defect point is not related to the structure of the textile, and it is more likely to be a real physical defect. This secondary judgment based on frequency characteristics, combined with the preliminary judgment based on time synchronization, forms a multi-dimensional defect discrimination method, effectively making up for the shortcomings of a single discrimination dimension, especially in the case of unclear time synchronization, it can more accurately distinguish between artifacts and real physical defects caused by the structure of the textile, thereby improving the accuracy of defect detection.

[0042] As an embodiment of the present application, the step of analyzing the frequency composition of the brightness response sequence to determine the dominant frequency and judging again whether the suspected defect point is a real physical defect or a temporary optical artifact based on the matching relationship between the dominant frequency and the geometric feature frequency comprises:

[0043] obtaining the frequency composition of the brightness response sequence;

[0044] centering on the geometric feature frequency, a frequency interval is demarcated;

[0045] based on the frequency composition of the brightness response sequence, the energy accumulation value within the frequency interval is calculated;

[0046] According to the prominence of the energy accumulation value, the suspected defect point is determined to be a real physical defect.

[0047] wherein the frequency composition of the sequence of luminance responses refers to the energy or amplitude distribution of the sequence of luminance responses at different frequencies by converting it from time domain to frequency domain through frequency transformation such as Fourier transform, which is the basis for subsequent frequency analysis, and the purpose is to convert the characteristics of the sequence of luminance responses from time dimension to frequency dimension for observation and analysis. The frequency interval centered on the geometric feature frequency refers to the geometric feature frequency determined according to the physical size of the three-dimensional periodic structure of the textile and the running speed of the textile, and a specific frequency range is determined around the frequency, and the frequency interval limits the frequency range of interest, because real physical defects often produce more significant frequency response near the geometric feature frequency, and the purpose is to focus the analysis on this key area, exclude the interference of other frequency components, and improve the accuracy of the analysis. Based on the frequency composition of the sequence of luminance responses, the energy accumulation value located in the frequency interval is calculated, which refers to the summation or integration of the energy of the sequence of luminance responses in the frequency domain within the delimited frequency interval, and the energy accumulation value reflects the energy concentration degree of the sequence of luminance responses near the geometric feature frequency, and the purpose is to quantify the response strength of the suspected defect point in this key frequency range. According to the prominence of the energy accumulation value, the suspected defect point is determined to be a real physical defect, which refers to comparing the calculated energy accumulation value with a certain reference or threshold to evaluate its significance, and if the energy accumulation value is significantly higher than the background or the preset threshold, it is considered that the suspected defect point is more likely to be a real physical defect, and the purpose is to make a judgment about the physical properties of the suspected defect point according to the quantified energy concentration degree.

[0048] The scheme of the present application is based on the preliminary distinction between thermal artifacts and real defects based on time synchronization, and the secondary determination based on the matching relationship between the dominant frequency and the geometric feature frequency in the case of low time synchronization. The scheme further refines the method of frequency analysis. The scheme recognizes that simple dominant frequency matching may not be sufficient to deal with all cases, especially when structural noise produces energy concentration near the geometric feature frequency. Therefore, the scheme locks the focus of analysis on the key frequency range related to the periodic structure of the textile by obtaining the frequency composition of the brightness response sequence and defining a frequency interval centered on the geometric feature frequency. Then, by calculating the energy accumulation value in the frequency interval, the scheme can more comprehensively evaluate the total energy intensity of the suspected defect point in the key frequency range, rather than just focusing on a single dominant frequency. Finally, according to the prominence of the energy accumulation value, the scheme can effectively distinguish those artifacts or noises whose dominant frequency is close to the geometric feature frequency but the energy concentration is not high, because real physical defects usually show stronger and more concentrated energy response near the geometric feature frequency. This method, combined with the aforementioned time synchronization analysis and dominant frequency matching, forms a multi-stage, multi-dimensional judgment process, significantly enhancing the system's ability to distinguish real defects from temporary optical artifacts, thereby improving the accuracy and reliability of defect detection.

[0049] As an embodiment of the present application, the step of determining the suspected defect point as a real physical defect according to the prominence of the energy accumulation value comprises:

[0050] Obtaining the frequency composition of one or more adjacent regions of the suspected defect point on the textile;

[0051] Based on the frequency composition of one or more adjacent regions, determining a local background energy reference in the frequency interval;

[0052] Comparing the energy accumulation value with the local background energy reference, and determining the suspected defect point as a real physical defect based on the comparison result.

[0053] The one or more adjacent regions of the suspected defect on the textile refer to one or more regions selected around the suspected defect for analyzing the background texture characteristics thereof, which can be achieved by selecting a fixed-size rectangular region, a circular region, or other shape region selected according to a preset rule; the frequency composition refers to frequency spectrum information obtained by frequency analysis of the image region, representing the energy distribution of the region at different spatial frequencies, which can be achieved by using a two-dimensional Fourier transform method or the like; the frequency interval refers to a frequency range centered on the geometric feature frequency; the local background energy benchmark refers to a typical energy level representing the background texture of the textile in the adjacent region of the suspected defect within the frequency interval, which can be achieved by calculating the average energy value, the median, or a value determined by using other statistical methods; the energy accumulation value refers to the total energy or a representative value of the suspected defect within the frequency interval; and the comparison result refers to the relationship between the energy accumulation value and the local background energy benchmark, which can be determined by whether the energy accumulation value exceeds the benchmark plus a threshold.

[0054] The scheme of the present application obtains the frequency composition of the adjacent region of the suspected defect, thereby understanding the frequency characteristics of the textile background around the suspected defect. The texture of the textile, as a periodic structure, will produce energy concentration at specific frequencies, and these frequencies constitute background noise. By analyzing the frequency composition of the adjacent region, the intensity and distribution of the background noise can be estimated. Based on the frequency composition of the one or more adjacent regions, a local background energy benchmark within the frequency interval is determined. The local background energy benchmark represents the typical energy level of the textile background within a specific frequency range. The energy accumulation value is compared with the local background energy benchmark, and the suspected defect is determined to be a real physical defect based on the comparison result. If the energy accumulation value of the suspected defect is significantly higher than the local background energy benchmark, it can be considered that the energy accumulation is caused by the defect rather than the texture of the textile itself. Conversely, if the energy accumulation value is close to the local background energy benchmark, it can be considered that the energy accumulation is caused by the texture of the textile itself, thereby excluding false positives. This comparison method by introducing a local background energy benchmark is an improvement over the method of determining only based on the energy accumulation value of the suspected defect itself, which takes into account the fluctuation of the local background texture, making the determination process more adaptable to complex textile texture backgrounds. In combination with the preliminary determination based on time synchronization and geometric feature frequency matching in the previous scheme, the present scheme further refines the determination logic in the frequency domain, effectively improves the ability to distinguish real defects from texture noise in complex backgrounds, and forms a more robust and accurate defect recognition system.

[0055] As an embodiment of the present application, the step of obtaining the frequency composition of the one or more adjacent regions of the suspected defect on the textile comprises:

[0056] determining a directional characteristic exhibited by the textile background around the suspected defect point;

[0057] selecting one or more neighboring regions on the textile based on the directional characteristic;

[0058] obtaining a frequency composition corresponding to the selected one or more neighboring regions.

[0059] wherein the directional characteristic refers to the dominant direction or angular distribution exhibited by the texture of the textile in space, which can be extracted using image processing techniques such as Fourier transform or Gabor filter bank analysis to obtain the texture direction information of local regions of the image, wherein the neighboring regions refer to one or more image regions adjacent or close to the suspected defect point in space, whose shape and size can be set according to actual application and texture characteristics, and wherein the frequency composition refers to the energy distribution of the image region in the frequency domain, which can be obtained by performing frequency transform on the image region.

[0060] The scheme of the present application determines the directional characteristic exhibited by the textile background around the suspected defect point, and selects neighboring regions based on the directional characteristic, thereby obtaining a frequency composition that is more representative of the real local background texture. This is because the texture of the textile often has directionality, such as the warp and weft direction. If the selected neighboring regions do not match the directional characteristic, their frequency composition may not accurately reflect the background texture around the suspected defect point. By considering the directional characteristic to select the neighboring regions, the texture of the selected regions can be ensured to be more consistent with the background texture around the suspected defect point. Based on the frequency composition of these more representative neighboring regions, a local background energy benchmark within the frequency interval can be more accurately determined. Comparing the energy accumulation value of the suspected defect point with this more accurate local background energy benchmark can more effectively eliminate the interference of the texture of the textile itself on defect judgment. It is precisely because the neighboring regions that match the directional characteristic of the background texture are selected that the obtained background energy benchmark is more reliable, which enables the subsequent comparison and determination process to more accurately distinguish between real defects and artifacts caused by texture, thereby improving the accuracy of defect detection.

[0061] As an embodiment of the present application, the step of determining a directional characteristic exhibited by the textile background around the suspected defect point comprises:

[0062] performing two-dimensional frequency transform on the background image around the suspected defect point to obtain a corresponding frequency spectrum;

[0063] determining the directional characteristic based on the distribution orientation of the energy in the frequency spectrum.

[0064] The two-dimensional frequency transform refers to a mathematical method for converting an image from a spatial domain to a frequency domain, and can be implemented by a two-dimensional Fourier transform, a two-dimensional discrete cosine transform, etc. The purpose is to reveal the components of the image in different frequencies and directions. The frequency spectrum refers to a frequency domain representation obtained after the two-dimensional frequency transform, which is usually a complex number matrix. The amplitude spectrum reflects the intensity of different frequency components in the image, and the phase spectrum reflects the position information of the components. The purpose is to visualize the frequency and direction information of the image. The distribution orientation of the energy in the frequency spectrum refers to the main directionality of the region with greater amplitude (energy) in the frequency domain plane. For an image with periodic texture, the energy will be concentrated in a specific direction perpendicular to the texture direction. The purpose is to infer the texture direction of the original image by analyzing the direction of energy concentration. The directional feature refers to the main trend or arrangement direction of the texture of the textile background in space, such as the direction of warp or weft. The purpose is to represent the structural characteristics of the textile background and provide a basis for subsequent selection of adjacent regions.

[0065] The scheme of the present application converts the image from a spatial domain to a frequency domain by performing a two-dimensional frequency transform on the background image around the suspected defect point, and obtains the corresponding frequency spectrum. This process converts complex spatial texture information into frequency and direction information that is easy to analyze. Subsequently, the directional feature is determined based on the distribution orientation of the energy in the frequency spectrum. Since the textile usually has a periodic texture structure, the energy of its frequency spectrum will be concentrated in a specific direction perpendicular to the texture direction. By analyzing the direction of energy concentration in the frequency spectrum, the texture direction of the original image can be directly and accurately extracted. For example, if the texture of the textile is horizontal, the energy of its frequency spectrum will be concentrated in the vertical direction. This frequency domain-based analysis method can effectively filter out the effects of noise and illumination changes, and more robustly extract the direction information of the periodic texture, compared to directly analyzing the image texture in the spatial domain. After determining the directional feature, one or more adjacent regions are selected on the textile according to the directional feature, and then the frequency composition corresponding to the selected one or more adjacent regions is obtained. The accurate directional feature enables the subsequently selected adjacent regions to better represent the background texture in that direction, thereby more accurately obtaining the background frequency composition and more accurately determining the local background energy reference, ultimately improving the accuracy of defect judgment. It is because the frequency domain analysis is used to determine the directional feature that the subsequent selection of adjacent regions and the determination of the background energy reference are more reliable, thereby effectively solving the problem of deviation in the selection of adjacent regions caused by inaccurate determination of the directional feature.

[0066] As an embodiment of the present application, the step of determining the directional feature based on the distribution orientation of the energy in the frequency spectrum includes:

[0067] performing energy integration in the angular direction on the frequency spectrum;

[0068] determine an angle direction with maximum energy according to the energy integration result;

[0069] take the angle direction with maximum energy as the directionality feature.

[0070] wherein the spectrum refers to two-dimensional data obtained by performing two-dimensional frequency transform on the image, representing energy distribution of the image at different frequencies and directions, which can be realized by using fast Fourier transform (FFT) algorithm or the like. Wherein the energy integration of the spectrum in the angle direction refers to the calculation of the energy accumulation of the spectrum in the angle direction in the spectrum space, which can be realized by dividing the spectrum into multiple angle sectors and calculating the total energy in each sector. Wherein the energy integration result refers to the distribution data of the energy accumulation value in different angle directions obtained by the energy integration in the angle direction, which can be represented by a one-dimensional array or curve, wherein each element or point corresponds to an angle direction and its corresponding energy integration value. Wherein the angle direction with maximum energy refers to the angle corresponding to the maximum energy value in the energy integration result, which can be determined by searching for the maximum value and its index (corresponding to the angle) in the energy integration result. Wherein the directionality feature refers to a parameter for describing the main trend of the texture or structure of the image, which can be represented by an angle value or a direction vector.

[0071] The scheme of the present application performs energy integration of the obtained spectrum in the angle direction, accumulates the energy information scattered in the two-dimensional spectrum along different angle directions, and obtains a one-dimensional curve reflecting the energy distribution with angle. This integration process effectively smooths the local fluctuations and noises in the spectrum, making the energy distribution trend in different directions clearer. Then, by analyzing the energy integration result, the angle direction with maximum energy is directly determined. This angle direction with maximum energy represents the dominant direction of the energy concentration in the spectrum, corresponding to the main trend of the texture or structure in the image. Finally, this angle direction with maximum energy is directly taken as the directionality feature of the image. The whole process converts the complex two-dimensional spectrum analysis into simple one-dimensional curve analysis, providing a robust and efficient way to accurately extract the directionality information of the image. The accurately extracted directionality feature can guide the subsequent steps to more accurately select the adjacent area with similar background texture to the suspected defect point, so as to more accurately estimate the local background energy reference, and thus improve the accuracy of the final defect determination, effectively distinguishing the real physical defects from the temporary optical artifacts.

[0072] As an embodiment of the present application, the step of determining a local background energy reference in the frequency interval includes:

[0073] sorting the energy values of the one or more adjacent areas in the frequency interval;

[0074] According to the sorting result, a specific quantile value in the sorting result is selected as the local background energy reference.

[0075] The specific quantile value refers to a value at a specific position after a group of data is sorted from small to large, for example, a percentile, which can be implemented by calculating a value at a certain percentile position in the sorted data set, and the purpose is to represent a certain position of the data distribution and effectively exclude the influence of extreme values. The local background energy reference refers to a reference value for measuring the energy level of the background region around the suspected defect point in a specific frequency range, which can be determined by statistical analysis (such as calculating the average, median, quantile, etc.) on the energy values of the adjacent regions, and the purpose is to provide a reliable background reference for determining whether the suspected defect point is a real physical defect.

[0076] The scheme of the present application can reveal the overall distribution of the energy values in the frequency interval by sorting the energy values of one or more adjacent regions, rather than simply taking the average. After sorting, the arrangement of the energy values from low to high can be clearly seen, so that the concentrated area of the background energy and the abnormal value can be better understood. On this basis, according to the sorting result, a specific quantile value in the sorting result is selected as the local background energy reference. The selection of the quantile value can effectively exclude the influence of abnormal values. For example, if the average value is directly taken, a few extremely high energy values may significantly increase the background energy reference, resulting in some real defects being misjudged as background. By selecting a lower quantile value, the level of background energy can be more robustly estimated, reducing the possibility of misjudgment. At the same time, selecting an appropriate quantile value can also reflect the concentration degree of the background energy. For example, if the background energy distribution is relatively uniform, the difference between different quantile values may not be large. If the background energy is concentrated in a few regions, a lower quantile value may be more representative of the true level of the background. It is because of the sorting of the energy values of the adjacent regions and the selection of the appropriate quantile value that the estimation of the local background energy reference is more accurate, thereby improving the accuracy and robustness of defect detection. This more accurate reference determination method combined with the framework of judging based on the local background energy reference makes the entire judgment process more robust and reliable.

[0077] As an embodiment of the present application, the step of selecting a specific quantile value in the sorting result as the local background energy reference includes:

[0078] analyzing the distribution trend of the energy values in the sorting result;

[0079] determining a quantile value that can represent the concentrated area of the background energy according to the distribution trend;

[0080] taking the quantile value as the local background energy reference.

[0081] The analysis of the energy value distribution trend in the ranking results refers to examining the overall characteristics of the ranked energy value sequence through statistical or data analysis methods, such as the concentration, dispersion, and presence of multiple peaks. This can be achieved by calculating energy value histograms, density estimation, or calculating statistical moments (such as mean, standard deviation, skewness, and kurtosis). Determining a quantile value that represents the concentrated area of ​​background energy refers to adaptively selecting a quantile point based on the analyzed energy value distribution trend. The energy value corresponding to this quantile point can effectively characterize the main concentration level of background energy. The selection of the quantile value can be dynamically adjusted based on the peak position of the distribution, the width of the distribution, or preset rules (e.g., if the distribution is biased towards low energy, a lower quantile value is selected; if the distribution is wide, a quantile value that covers the main background energy range is selected).

[0082] This application's solution first analyzes the distribution trend of energy value ranking results in neighboring regions. This allows the system to understand the actual distribution pattern of background energy, such as whether it is concentrated or dispersed, and whether there are abnormally high or low values. Based on the analysis of the distribution trend, the system can adaptively determine a more representative quantile value. This quantile value is no longer a fixed percentage but is dynamically adjusted according to the actual data distribution, thereby more accurately capturing the concentration level of background energy. This adaptively determined quantile value is used as a local background energy benchmark for comparison with the energy accumulation value of suspected defect points. This method overcomes the benchmark bias problem that may occur when fixed quantile values ​​are used when the background energy distribution is uneven or there is interference. Within the defect detection framework based on time and frequency analysis, especially based on the method of selecting quantile values ​​as background benchmarks through neighboring region energy ranking, this solution optimizes the quantile value selection process, making the estimation of the background energy benchmark more accurate. This more precise benchmark enables a more effective distinction between the cumulative energy of real physical defects and the energy of background noise or artifacts during subsequent comparisons, thereby improving the accuracy and robustness of the entire detection method and effectively solving the problem of misjudgment caused by temporary optical artifacts due to thermal effects.

[0083] like Figure 2 The illustrated textile surface defect detection system is used for online inspection of textiles with a heat-sensitive coating. The system includes:

[0084] The modulation module 201 is used to apply a preset periodic intensity modulation to the detection light source used in online detection to generate a reference sequence as a benchmark for synchronization analysis.

[0085] The acquisition module 202 is configured to acquire a series of luminance values of the suspected defect point in a preset time period under periodic intensity modulation, so as to constitute a luminance response sequence representing the optical response characteristic of the suspected defect point.

[0086] The calculation module 203 is configured to calculate the time synchronization degree between the luminance response sequence and the reference sequence.

[0087] The determination module 204 is configured to determine whether the suspected defect point is a real physical defect or a temporary optical artifact caused by thermal effect based on the time synchronization degree.

[0088] The scheme of the present application realizes systematic detection of surface defects of the heat-sensitive coating textile by decomposing the detection process into four functional modules, i.e., modulation, acquisition, calculation and determination. The modulation module 201 applies periodic intensity variation to the detection light source, thereby providing a time reference for subsequent analysis. The acquisition module 202 synchronously acquires the luminance variation data of the suspected defect point under light source modulation. The calculation module 203 analyzes the time synchronization relationship between the acquired luminance variation data and the light source modulation reference. The determination module 204 distinguishes the temporary optical artifact caused by the thermal effect of the light source from the real physical defect according to the degree of synchronization. This modularized cooperative working mode enables the system to identify defects from dynamic response characteristics rather than static image features, effectively deals with the detection challenge brought by the heat-sensitive coating, and improves the accuracy and robustness of the detection.

[0089] The basic principles, main features and advantages of the present application are shown and described above. It should be understood by those skilled in the art that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only the principles of the present application. Various changes and improvements can be made without departing from the spirit and scope of the present application, and these changes and improvements all fall within the scope of the claimed present application.

Claims

1. A method for detecting surface defects in textiles, used for online detection of textiles with a heat-sensitive coating on their surface, characterized in that, The method includes the following steps: A preset periodic intensity modulation is applied to the detection light source used in online detection to generate a reference sequence as a benchmark for synchronization analysis. For suspected defects initially identified on the surface of the textile, under the periodic intensity modulation, a series of brightness values ​​of the suspected defects are collected within a preset time period to form a brightness response sequence characterizing its optical response properties. Calculate the degree of time synchronization between the brightness response sequence and the reference sequence; Based on the degree of time synchronization, the suspected defect point is determined to be either a real physical defect or a temporary optical artifact caused by thermal effects.

2. The method for detecting surface defects in textiles according to claim 1, characterized in that, If the time synchronization level is lower than a preset synchronization threshold, the method further includes: Obtain the geometric characteristic frequency determined by the physical dimensions of the three-dimensional periodic structure of the textile and the running speed of the textile; Analyze the frequency composition of the brightness response sequence to determine its dominant frequency; Based on the matching relationship between the dominant frequency and the geometric feature frequency, the suspected defect point is again determined to be a real physical defect or a temporary optical artifact.

3. The method for detecting surface defects in textiles according to claim 2, characterized in that, The steps of analyzing the frequency composition of the luminance response sequence to determine its dominant frequency, and then, based on the matching relationship between the dominant frequency and the geometric feature frequency, determining whether the suspected defect point is a real physical defect or a temporary optical artifact, include: The frequency composition of the brightness response sequence is obtained; A frequency range is defined with the geometric characteristic frequency as the center; Based on the frequency composition of the brightness response sequence, calculate the cumulative energy value within the frequency range; Based on the prominence of the accumulated energy value, the suspected defect point is determined to be a real physical defect.

4. The method for detecting surface defects in textiles according to claim 3, characterized in that, The step of determining whether a suspected defect is a real physical defect based on the prominence of the accumulated energy value includes: The frequency of the suspected defect point in one or more adjacent areas on the textile is obtained; Based on the frequency composition of one or more neighboring regions, a local background energy reference is determined within the frequency range; The accumulated energy value is compared with the local background energy benchmark, and the suspected defect point is determined to be a real physical defect based on the comparison result.

5. The method for detecting surface defects in textiles according to claim 4, characterized in that, The step of obtaining the frequency composition of one or more adjacent regions of the suspected defect point on the textile includes: Determine the directional characteristics of the textile background surrounding the suspected defect point; Based on the directional characteristics, one or more adjacent areas are selected on the textile. The frequencies are obtained from one or more selected neighboring regions.

6. The method for detecting surface defects in textiles according to claim 5, characterized in that, The step of determining the directional characteristics of the textile background surrounding the suspected defect includes: A two-dimensional frequency transformation is performed on the background image around the suspected defect point to obtain the corresponding spectrum; The directional characteristics are determined based on the energy distribution orientation in the spectrum.

7. The method for detecting surface defects in textiles according to claim 6, characterized in that, The step of determining the directional feature based on the energy distribution orientation in the spectrum includes: Energy integration is performed on the spectrum in the angular direction; Based on the energy integration result, determine the angular direction of maximum energy; The angular direction of maximum energy is taken as the directional feature.

8. The method for detecting surface defects in textiles according to claim 4, characterized in that, The step of determining a local background energy reference within the frequency range includes: The energy values ​​of the one or more neighboring regions within the frequency range are sorted. Based on the sorting results, a specific quantile value in the sorted results is selected as the local background energy benchmark.

9. A method for detecting surface defects in textiles according to claim 8, characterized in that, The step of selecting a specific quantile value from the sorting results as the local background energy reference includes: Analyze the distribution trend of energy values ​​in the sorting results; Based on the distribution trend, a quantile value that can represent the region of concentrated background energy is determined; The quantile value is used as the local background energy reference.

10. A textile surface defect detection system for online detection of textiles with a heat-sensitive coating, characterized in that, The system includes: The modulation module is used to apply a preset periodic intensity modulation to the detection light source used in online detection to generate a reference sequence as a benchmark for synchronization analysis. The acquisition module is used to acquire a series of brightness values ​​of suspected defects that have been initially identified on the surface of the textile under the periodic intensity modulation, within a preset time period, so as to form a brightness response sequence characterizing its optical response characteristics. The calculation module is used to calculate the degree of time synchronization between the brightness response sequence and the reference sequence; The determination module is used to determine, based on the degree of time synchronization, whether the suspected defect is a real physical defect or a temporary optical artifact caused by thermal effects.

Images