Textile defect detection device and textile defect detection method

Abstract

The present application relates to the technical field of textile detection. The present application discloses a kind of textile defect detection device and textile defect detection method, it includes gradient light absorption component below textile, is made of light absorption bottom layer and light reflection porous adjustment surface layer, surface layer porosity gradient distribution along detection direction to amplify the local change of penetration light.The upper side has pulsed polarized light source group, emits polarized light to form reflected light and penetration light. High-spectral-light-intensity fusion detection device is arranged on the side, is separated from light source, and spectrum and light intensity distribution are collected. One side of textile has zoned dynamic tension mechanism, and the distance between it and gradient light absorption component is fixed and controlled. Multi-modal data processing unit is arranged on the side of detection device, is connected with detection device, pre-stores dynamic reference library, and is used to identify defect and quantize edge gradient. The present application improves the accuracy and adaptability of textile defect detection through multi-dimensional innovation. Gradient light absorption component amplifies the light intensity change of defect, and improves the identification accuracy of small defect to 0.3mm.

Description

TECHNICAL FIELD

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

[0002] In the current field of textile defect detection, traditional manual detection methods still account for a certain proportion, but they rely on the experience and visual acuity of the detection personnel, and have significant limitations. Long-time work is easy to cause personnel fatigue, resulting in an increase in missed detection and misdiagnosis rates, especially for small defects with a diameter of less than 1 mm or deep fiber defects. At the same time, manual detection is low in efficiency, and a single production line needs to be equipped with multiple detection personnel, and the detection speed is difficult to match the high-speed production rhythm of the modern textile industry, which seriously restricts the improvement of production efficiency and the stability of quality control.

[0003] Although the existing optical detection technology has replaced manual detection to some extent, there are still many technical bottlenecks. Most devices use a single wavelength light source or a fixed polarization direction, which is difficult to simultaneously meet the detection needs of different types of defects such as surface stains and deep yarn unevenness of textiles, resulting in limited recognition ability for complex defects. In addition, the amplification ability of traditional optical devices for defect signals is insufficient. When the textile has low-contrast defects, the intensity change of the penetrating light or the reflected light is weak, and the detection device is difficult to capture effective signals, and the missed detection problem is prominent. At the same time, the material differences and surface flatness fluctuations of textiles easily lead to light signal interference, further reducing the detection accuracy.

[0004] The existing detection methods generally lack a dynamic adaptation mechanism, and it is difficult to cope with diversified detection scenes. Some devices do not consider the stability of the distance between the textile and the detection component, or do not establish a dynamic reference model for different materials, resulting in that the detection results are greatly affected by environmental factors and fabric characteristics. In addition, most technologies only rely on a single spectrum or light intensity signal for analysis, lack multi-dimensional data fusion, and are difficult to realize accurate classification and grading of defect types, cannot provide quantitative basis for production quality control, and restrict the development of fine quality management in the textile industry. SUMMARY

[0005] In order to solve the problems in the prior art, the present application discloses a textile defect detection device capable of detecting small defects of textiles, improving detection accuracy and improving detection speed.

[0006] The present application discloses a textile defect detection device, which comprises:

[0007] The gradient light absorption assembly is arranged below the textile and is composed of a bottom layer with light absorption effect and a porous adjusting surface layer with light reflection effect. The porous adjusting surface layer has a plurality of pores, and the porosity of the porous adjusting surface layer is gradiently distributed along the detection direction to amplify the local change of the penetration light intensity through the porosity difference.

[0008] The pulse polarized light source group is arranged above the textile and can emit polarized light. Part of the polarized light is reflected by the textile to form reflected light, and the other part of the polarized light passes through the textile to form penetration light.

[0009] The hyperspectral-light intensity fusion detection device is arranged above one side of the textile and is disposed on both sides of the pulse polarized light source group. The hyperspectral-light intensity fusion detection device is used to collect the spectrum and light intensity distribution of the reflected light of the penetration light after passing through the porous adjusting surface layer and the surface of the textile.

[0010] The partitioned dynamic tension mechanism is arranged on one side of the textile and is used to fix the textile and control the distance between the textile and the gradient light absorption assembly.

[0011] The multi-modal data processing unit is arranged on one side of the textile close to the hyperspectral-light intensity fusion detection device and is connected to the hyperspectral-light intensity fusion detection device. The multi-modal data processing unit pre-stores a dynamic reference library constructed based on a flawless textile sample, and is used to compare and analyze and identify the defect area of the textile and quantify the edge gradient.

[0012] When the textile has defects, the absorption amount of the penetration light in the bottom layer is different due to the porosity difference of different regions of the porous adjusting surface layer, thereby causing different changes in the reflection intensity of the porous adjusting surface layer.

[0013] Further, the bottom layer of the gradient light absorption assembly is a high light absorption material layer with a light absorption rate of ≥95%.

[0014] The pore size of the porous adjusting surface layer is microns, and the gradient distribution range of the porosity along the detection direction is 10%-50%.

[0015] The bottom layer adopts a nano carbon black coating or a black ceramic material, the porous adjusting surface layer adopts an electrochromic material or an optical grade polymer, and the thickness of the porous adjusting surface layer is 50-200 μm.

[0016] Further, the pulse polarized light source group can emit polarized light of at least two wavelengths, has pulse modulation function and polarization direction switching function, and the light source intensity can be dynamically adjusted.

[0017] The polarized light emitted by the pulse polarized light source group includes green light of 500-550 nm and red light of 600-650 nm, the pulse frequency is 10-50 Hz, and the polarization direction can be alternately switched between 0° and 90°.

[0018] Further, the hyperspectral-light intensity fusion detection device comprises a hyperspectral camera and a light intensity sensor array, which are arranged at an angle of 30-60 degrees with the surface of the textile and synchronously collect spectral and light intensity distribution data of reflected light and penetrating light.

[0019] The spectral range of the hyperspectral camera is 400-1000 nm, the detection accuracy of the light intensity sensor array is less than or equal to 0.1 lux, and the sampling frequency of both is greater than or equal to 100 fps.

[0020] Further, the partitioned dynamic tension mechanism is composed of clamping units independently driven by multiple servo motors, which can adjust the tension of the textile in different partitions, so that the distance between the textile and the gradient light absorption assembly is controlled to be 1-5 mm, and the flatness deviation is less than 100 microns.

[0021] The multi-modal data processing unit fuses spectral features and light intensity deviations through a convolutional neural network, and the edge gradient quantization accuracy of the defect is less than or equal to 0.3 mm.

[0022] The present application discloses a textile defect detection method for any one of the above-mentioned textile defect detection devices, S1: constructing a dynamic reference library, fixing a sample of a non-defective textile by a partitioned dynamic tension mechanism, irradiating and collecting reflected spectral and light intensity distribution data by using a pulsed polarized light source group, and establishing a reference feature model;

[0023] S2: Real-time spectral and light intensity acquisition, repeating the irradiation and acquisition operation of S1 for the textile to be tested to obtain real-time detection data;

[0024] S3: Defect feature analysis, comparing real-time data with the dynamic reference library through a multi-modal data processing unit, combining the amplified light intensity changes of the gradient light absorption assembly, and identifying suspected defect areas;

[0025] S4: Multi-dimensional verification, repeating the detection of the suspected areas by switching detection parameters to verify the consistency of the defect features;

[0026] S5: Outputting defect information, quantifying the edge gradient of the defect area and grading the output.

[0027] Further, when constructing the dynamic reference library in S1, at least 30 groups of sample data of non-defective samples under different material qualities and different environmental conditions need to be collected, and common reference features are extracted through principal component analysis;

[0028] In S2, when collecting real-time spectral and light intensity, the pulsed polarized light source group alternately emits polarized light of different wavelengths, and the hyperspectral-light intensity fusion detection device synchronously records the multi-band spectral and light intensity distribution matrix;

[0029] The polarized light of different wavelengths in S2 includes green light and red light, the green light is used to enhance the spectral contrast of surface defects, and the red light is used to amplify the light intensity difference of deep fiber defects.

[0030] Further, in S3, the flaw feature analysis adopts wavelet transform to denoise the light intensity data, extracts the contour features of the flaw area through the edge detection algorithm, and distinguishes the flaw types by combining the spectral angle matching.

[0031] Further, in S4, the multi-dimensional verification includes switching the light wavelength and polarization direction of the pulsed polarized light source group, adjusting the surface reflectivity of the gradient light absorption assembly, and if the flaw feature deviation of multiple detections is less than or equal to 5%, it is determined as a real flaw.

[0032] In S5, the flaw grading is based on light intensity deviation amplitude, spectral shift and edge gradient, and is divided into three levels of slight, moderate and severe, each level corresponding to a clear parameter threshold range.

[0033] Further, the textile is pretreated before detection, including surface cleaning and humidity adjustment, so that the textile moisture regain is maintained at 8%-15%, reducing environmental interference.

[0034] The multi-modal data processing unit adopts a convolutional neural network to perform feature fusion on real-time data, and the model training sample size is greater than or equal to 1000 groups.

[0035] The beneficial effects of the present application are:

[0036] The textile flaw detection device and method of the present application significantly improves the accuracy and adaptability of textile flaw detection through multi-dimensional technical innovation. The porosity gradient design of the gradient light absorption assembly breaks through the limitation of weak flaw signal in traditional detection, and the high light absorption material at the bottom and the porous adjusting surface layer work together to amplify the light intensity change difference of the flaw area, so that the recognition accuracy of small flaws is improved to within 0.3mm, solving the problem of missed detection of low contrast flaws in traditional methods. The multi-wavelength and polarization direction switching design of the pulsed polarized light source group, combined with the hyperspectral-light intensity fusion detection device, realizes the synchronous capture of surface and deep flaws, and the green light enhances the spectral contrast of surface stains, and the red light amplifies the light intensity difference of deep uneven yarns, improving the accuracy rate compared with single wavelength detection. The partition dynamic tension mechanism controls the distance and flatness through independent servo motors, adapts to various fabrics from light silk to heavy denim, avoids detection deviation caused by material differences, and widens the application range of the device.

[0037] The high efficiency and reliability of the application further optimize the textile detection process. The multi-modal data processing unit fuses spectrum and light intensity data, combines a dynamic reference library constructed based on more than 30 groups of flawless samples, realizes feature matching through a convolutional neural network, and the flaw recognition accuracy is greater than or equal to 95%, and the missed detection rate is reduced by 80% compared with manual detection. The multi-dimensional verification and grading output in the detection method provide a quantitative basis for quality control. The components work cooperatively, the sampling frequency is greater than or equal to 100fps, which meets the real-time detection demand of high-speed production line, and the efficiency is improved by more than 10 times compared with manual cloth inspection, and the pretreatment reduces environmental interference, ensuring the stability of long-term operation. In summary, the device not only breaks through the precision bottleneck of traditional detection technology, but also takes into account the efficiency and compatibility of industrial production, providing an integrated solution for textile quality detection. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 Figure 1 is a flowchart of a textile flaw detection device according to an embodiment of the present application.

[0039] Figure 2 Figure 2 is a structural diagram of a textile flaw detection device according to an embodiment of the present application.

[0040] Figure 3 Figure 3 is another structural diagram of a textile flaw detection device according to an embodiment of the present application.

[0041] Figure 4 Figure 4 is a structural diagram of a gradient light absorption assembly according to an embodiment of the present application.

[0042] In the figure: gradient light absorption assembly 11, bottom layer 111, porous adjustment surface layer 112, pulsed polarized light source group 12, hyperspectral-light intensity fusion detection device 13, hyperspectral camera 131, light intensity sensor array 132, partitioned dynamic tension mechanism 14, multi-modal data processing unit 15. DETAILED DESCRIPTION

[0043] In order to enable personnel in the art to better understand the application scheme, the technical solutions in the specific embodiments of the application will be described clearly and completely below.

[0044] The application discloses a textile flaw detection device, which comprises a gradient light absorption assembly arranged below the textile, wherein the gradient light absorption assembly is composed of a bottom layer with light absorption effect and a porous adjustment surface layer with light reflection effect, the porous adjustment surface layer has a plurality of pores, the porosity of the porous adjustment surface layer is gradiently distributed along a detection direction, and the porosity difference is used to amplify local changes in penetrating light intensity.

[0045] The pulse polarized light source group placed above the textile can emit polarized light, part of which is reflected by the textile to form reflected light, and the other part passes through the textile to form penetrating light;

[0046] The hyperspectral-light intensity fusion detection device is disposed on one side of the textile above and is disposed on both sides of the pulse polarized light source group, and is used for collecting the spectrum and light intensity distribution of the reflected light of the penetrating light after passing through the multi-porous adjusting surface layer and the surface of the textile;

[0047] The partitioned dynamic tension mechanism is disposed on one side of the textile and is used for fixing the textile and controlling the distance between the textile and the gradient light absorption assembly;

[0048] The multi-modal data processing unit is disposed on one side of the textile close to the hyperspectral-light intensity fusion detection device and is connected to the hyperspectral-light intensity fusion detection device, the multi-modal data processing unit pre-stores a dynamic reference library constructed based on a flawless textile sample, is used for comparative analysis and identification of the defect area of the textile and quantification of the edge gradient; the edge gradient is the spatial transition width of the defect edge from the normal area to the defect area, the unit is mm, and the quantification accuracy is ≤0.3mm; the light intensity gradient is the light intensity change rate per unit distance at the defect edge, the unit is lux / mm.

[0049] When the textile has defects, the porosity difference of different areas of the multi-porous adjusting surface layer causes the different absorption amounts of the penetrating light in the bottom layer, and then causes the different change amounts of the reflected light intensity of the multi-porous adjusting surface layer.

[0050] As the detection basis, the gradient light absorption assembly uses high light absorption material as the bottom layer, which can efficiently absorb the light penetrating through the textile. The porous adjusting surface layer has a porous structure, and the porosity is gradiently distributed along the detection direction. When the textile is flawless, the amount of light penetrating is uniform, and the difference in reflected light intensity of different porosity areas of the porous adjusting surface layer is small. When the textile has defects, the defect area will transmit more light, and these additional light is more likely to pass through the pores of the high porosity area of the porous adjusting surface layer and be absorbed by the bottom layer, resulting in a significant change in the reflected light intensity of this area, which is significantly greater than that of the low porosity area, thereby amplifying the defect signal. The polarized light source group emits polarized light obliquely to the textile, part of which is reflected by the surface of the textile to form reflected light, and the other part penetrates through the textile to form penetrating light. The hyperspectral-light intensity fusion detection device and the light source are placed on both sides to synchronously collect the spectral and light intensity data of the reflected light and the penetrating light reflected by the porous adjusting surface layer at a specific angle, ensuring that the changes in the light signal caused by the surface features and internal defects of the textile are captured simultaneously. The partitioned dynamic tension mechanism fixes the textile through independently driven clamping units, controls the distance between the textile and the gradient light absorption assembly, and maintains the flatness of the textile. This avoids the light signal interference caused by wrinkles or distance fluctuations of the textile, providing a reference condition for stable detection. The multi-modal data processing unit compares the real-time collected light signal with the pre-stored dynamic reference library of flawless samples, identifies the defect area by analyzing the light intensity change amplitude and spectral feature difference. At the same time, the unit calculates the light signal change rate of the defect edge from the normal area to the defect area, quantifies the edge gradient, and realizes the accurate description of the defect morphology.

[0051] The porosity gradient design of the gradient light absorption assembly breaks through the limitation of weak defect signal in traditional detection, amplifies the light intensity change difference of different areas, improves the recognition accuracy of small defects, and solves the problem of missed detection of low contrast defects in traditional methods. The multi-modal data processing unit fuses spectral and light intensity data, realizes multi-dimensional comparison combined with the dynamic reference library, can not only distinguish the defect type, but also realize defect grading through edge gradient quantification, and the accuracy of the detection result is improved compared with single signal analysis. The partitioned dynamic tension mechanism can adjust the tension and distance according to the material of the textile, adapt to various fabrics from light silk to thick denim, and avoid detection deviation caused by differences in fabric characteristics. The textile defect detection device works cooperatively, from light source irradiation to data output, and is fully automated. The detection speed can reach more than 30 frames per second, which meets the real-time detection needs of high-speed production lines, improves the efficiency of manual fabric inspection by more than 10 times, and reduces the missed detection rate by 80%.

[0052] Light intensity can only reflect the intensity change of light, suitable for detecting light signal fluctuations caused by poor light transmittance, but cannot distinguish the type and material characteristics of defects. In addition, the background color and dye composition of textiles will affect the baseline value of light intensity signal, and only relying on light intensity is easy to misjudge; while spectral features can strip the interference of background color, accurately locate defects by comparing the spectral patterns of normal and defective areas. The essence of the fusion of hyperspectral and light intensity data is to combine intensity changes and wavelength characteristics, both capturing the spatial location of defects through light intensity and distinguishing defect types through spectrum, ultimately improving the accuracy of detection.

[0053] The detection process is the core of light signal modulation and analysis. After the polarized light emitted by the pulsed polarized light source irradiates the textile, it is divided into two signals: reflected light and transmitted light. The reflected light carries the surface features of the textile, and the transmitted light is modulated by the gradient light absorption component. When the textile has defects, the amount of transmitted light increases in the defect area. These additional light is more easily absorbed by the bottom layer through the pores in the high-porosity area of the multi-pore adjustment surface layer, resulting in a significantly larger change in the intensity of the reflected light in this area than in other areas, forming an amplified defect signal. After the high-spectral-light-intensity fusion detection device captures the two light signals, the multi-modal data processing unit compares them with the dynamic baseline library, and through the comprehensive analysis of light intensity deviation, spectral shift, and edge gradient, it accurately identifies the defect area and quantifies its characteristics. This process realizes efficient and accurate detection of textile defects through gradient amplification, multi-signal fusion, and dynamic calibration.

[0054] The flowchart of the textile defect detection device is shown in Figure 1 .

[0055] In the process of detecting textile defects, the average light intensity of the same material and environment without defects in the dynamic baseline library is used as the baseline value to set a clear light intensity fluctuation judgment range: when using a pulsed polarized light source group to emit 500-550 nm green light to detect surface defects such as stains and surface fiber damage, if the real-time light intensity of the area to be detected fluctuates more than ±8% compared to the baseline value; or when using 600-650 nm red light to detect deep defects such as uneven yarn and deep fiber breakage, if the real-time light intensity fluctuates more than ±12% compared to the baseline value, the area is marked as a suspected defect area. Subsequently, a multi-dimensional verification process is required, switching the polarization direction of the light source to 0° and 90° alternately, adjusting the surface reflectivity of the gradient light absorption component, if the deviation of the light intensity fluctuation amplitude of the area is ≤5% in multiple detections, and combined with the spectral shift features captured by the hyperspectral camera, it can be finally determined that the area has real defects; if the light intensity fluctuation amplitude is always within ±8% or ±12% of the corresponding wavelength, it is determined that the textile has no defects and meets the quality standards.

[0056] The structure diagram of the textile defect detection device is shown inFigure 2 The textile defect detection device includes a gradient light absorption assembly 11, a pulsed polarized light source group 12, a hyperspectral-light intensity fusion detection device 13, a partitioned dynamic tension mechanism 14, and a multi-modal data processing unit 15. The pulsed polarized light source group 12 and the hyperspectral-light intensity fusion detection device 13 can adjust the position and angle as needed.

[0057] Example 1: Cotton fabric defect detection. In actual detection, there are small fluctuations in the light source, sensor accuracy, and slight unevenness on the surface of the textile. The measured value has an error of within ±1.5% from the theoretical value.

[0058] Detection object: A piece of cotton fabric to be detected, which may have small holes, local stains, and other defects.

[0059] The gradient light absorption assembly is placed directly below the fabric. The bottom layer is a nano-carbon black coating with an absorption rate of 96% and almost no reflection. The porous adjustment surface layer is an optical-grade polyethylene porous plate with a thickness of 100 pm. The plate is uniformly distributed with small holes with a diameter of 5 pm. The porosity is gradiently distributed from left to right along the detection direction, with 10% on the left side and 50% on the right side. The solid part of the porous adjustment surface layer can reflect 40% of the incident light.

[0060] The pulsed polarized light source group is installed on the left side above the fabric. It can emit 530 nm green polarized light with a polarization direction of 0° and a pulse frequency of 30 Hz. The light is obliquely incident on the surface of the fabric.

[0061] The hyperspectral-light intensity fusion detection device is installed on the right side above the fabric, separated from the light source on both sides, and forms a 45° angle with the surface of the fabric. It contains a hyperspectral camera with a detection spectral range of 400-800 nm and a light intensity sensor array with an accuracy of 0.05 lux, which is used to synchronously capture the color characteristics and brightness changes of the reflected light.

[0062] The partitioned dynamic tension mechanism clamps the edges of the fabric on both sides. Four servo motors are used to adjust the tension, so that the fabric remains flat and the distance between the gradient light absorption assembly and the fabric is stable at 2 mm.

[0063] The multi-modal data processing unit has a dynamic reference library built based on 100 samples of flawless cotton cloth. It contains the spectral curve and light intensity distribution model of the normal area, and is connected to the detection device for real-time data analysis.

[0064] Step 1: Detection process of flawless cotton cloth

[0065] The 530 nm green light emitted by the pulsed polarized light source is incident on the flawless cotton cloth. Most of the light is reflected by the cotton fibers, forming surface reflected light. Only 15% of the light penetrates the cotton cloth and reaches the gradient light absorption assembly below.

[0066] When the penetrating light reaches the porous adjusting surface layer, it falls on the sparse hole area with 10% porosity on the left side, 90% solid part and 10% small hole. 90% of the penetrating light is reflected by the solid part of the porous adjusting surface layer, and 10% of the penetrating light is absorbed by the bottom layer. At this time, the relative value of the reflected light intensity is = 15% x 90% = 13.5%;

[0067] Falls on the dense hole area with 50% porosity on the right side, wherein the dense hole area of the porous adjusting surface layer is 50% solid part without porosity and 50% small hole. 50% of the penetrating light is reflected by the solid part of the porous adjusting surface layer, and 50% of the penetrating light is absorbed by the bottom layer. At this time, the relative value of the reflected light intensity is = 15% x 50% = 7.5%.

[0068] The detection device shows that the light intensity sensor array measures that the left side reflected light intensity is stable at 15 lux, and the right side is stable at 8 lux; the spectral curve collected by the hyperspectral camera is completely matched with the green light spectrum of the flawless cotton cloth in the dynamic reference library.

[0069] The multi-modal data processing unit determines that there is no flaw, and updates the real-time fluctuation range of the dynamic reference library.

[0070] Second step: detection process of cotton cloth containing a small hole

[0071] There is a small hole with a diameter of 0.5mm in the middle of the fabric, which is difficult to detect with the naked eye.

[0072] When the light source irradiates the hole area, the proportion of penetrating light increases from 15% to 60% due to the absence of fiber blockage, and an additional 45% of penetrating light reaches the gradient light absorption component,

[0073] Falls on the sparse hole area with 10% porosity on the left side. At this time, the relative value of the reflected light intensity is = 60% x 90% = 54%, which is increased by 40.5% compared with 13.5% of the normal area;

[0074] Falls on the dense hole area with 50% porosity on the right side. At this time, the relative value of the reflected light intensity is = 60% x 50% = 30%, which is increased by 22.5% compared with 7.5% of the normal area.

[0075] At this time, the corresponding absolute light intensity is that the light intensity sensor array measures that the light intensity of the hole corresponding to the left side area increases to 75 lux, which is changed by 60 lux compared with the normal 15 lux, and the light intensity of the right side area increases to 40 lux, which is changed by 32 lux compared with the normal 8 lux. The hyperspectral camera finds that although the main peak of the spectral curve of the hole area is still 530nm, the light intensity peak value is significantly increased.

[0076] The multi-modal data processing unit compares the real-time data with the dynamic benchmark library, finds that the light intensity change amplitude far exceeds the ±2 lux threshold, and the left side change of 60 lux is much greater than the right side change of 32 lux. Through this gradient difference amplification, the hole position is directly located. At the same time, the light intensity change rate from the normal area to the hole area is calculated, and the light intensity increases by 5 lux every 0.1 mm distance. The light intensity gradient is quantified as 50 lux / mm, the edge gradient is quantified as 0.1 mm, and it is determined as a real defect.

[0077] The light intensity sensor array has high spatial resolution and can capture the light intensity distribution of a small area on the surface of the textile. When a hole is detected, the sensor records the continuous change of light intensity from the normal area to the hole area. At the edge of the hole, the light intensity increases by 5 lux every 0.1 mm. The multi-modal data processing unit calculates the edge gradient by calculating the light intensity change / distance: 5 lux ÷ 0.1 mm = 50 lux / mm, to quantify the light intensity mutation rate of the edge from the normal area to the defect area.

[0078] The porosity of the gradient light absorption assembly is gradiently distributed along the detection direction. This spatial distribution feature forms a fixed law for the reflection or absorption ratio of penetrating light at different positions. When a hole exists, the amount of penetrating light at the corresponding area increases significantly, and this increased penetrating light covers a certain spatial range of the gradient light absorption assembly, i.e. the position of the hole on the fabric corresponds to a certain continuous area of the gradient light absorption assembly below. The light intensity sensor array has an array structure and has spatial resolution, and can real-time collect the light intensity distribution at different positions on the surface of the textile. In the area of the gradient light absorption assembly corresponding to the hole area, the reflected light intensity change in the low porosity area on the left side is much greater than the change in the high porosity area on the right side. This gradient difference with more significant change on the left side forms a local abnormal area in the spatial data of the sensor array, i.e. the position with the most intense light intensity mutation and the most consistent porosity gradient law. At the same time, in the spectral data collected by the hyperspectral camera, the light intensity peak value of the hole area is increased, which completely corresponds to the abnormal area of the sensor array in the spatial position, further verifying that this area is the hole. The multi-modal data processing unit can accurately locate the specific position of the hole by matching the light intensity gradient difference distribution in space and the spectral abnormal position.

[0079] As an embodiment, the bottom layer of the gradient light absorption assembly is a high light absorption material layer with an absorption rate of ≥95%. The pore size of the porous adjustment surface layer is microns, and the gradient distribution range of the porosity along the detection direction is 10%-50%. The bottom layer uses a nano-carbon black coating or a black ceramic material, the porous adjustment surface layer uses an electrochromic material or an optical-grade polymer, and the thickness of the porous adjustment surface layer is 50-200 μm.

[0080] The bottom layer of the gradient light absorption assembly is made of high light absorption material, with a light absorption rate of 95% or above, which can efficiently absorb the light passing through the pores of the porous adjusting surface layer, providing a stable low background basis for light signal contrast. The bottom layer material can be a nano-carbon black coating or a black ceramic material, both of which have excellent light absorption performance and can be selected according to the actual application scenario. The pore size of the porous adjusting surface layer is micron level, which can ensure the effective penetration of light in the pores and form light signal modulation through the pore distribution. The porosity of the porous adjusting surface layer is gradiently distributed along the detection direction in the range of 10%-50%, making the absorption and reflection ratio of the penetrating light in different areas present regular differences, creating conditions for flaw signal amplification. The porous adjusting surface layer is made of electrochromic material or optical-grade polymer, both of which have their own advantages. Electrochromic material can change the reflection characteristics through external control, while optical-grade polymer has stable physical and chemical properties. The thickness of the porous adjusting surface layer is controlled within 50-200μm, which can ensure the stability of the structure and achieve the ideal balance between reflection and penetration of light in the porous adjusting surface layer. The bottom layer and the porous adjusting surface layer work together, the strong light absorption characteristics of the bottom layer ensure that the light signal at the pore is effectively absorbed, and the porosity gradient distribution of the porous adjusting surface layer makes the change of penetrating light in the flaw area form obvious differences at different positions, which together realize the accurate amplification of flaw signal.

[0081] The light absorption rate of the bottom layer high light absorption material is ≥95%, which can minimize the reflection interference of the penetrating light, make the light signal difference between normal area and flaw area more prominent, improve the contrast of flaw identification, and enable small flaws to be clearly captured. The micron-level pore size design of the porous adjusting surface layer is suitable for the size characteristics of textile fibers, avoiding light signal distortion caused by excessively large or small pore size, ensuring that the ratio of light penetration and reflection meets the detection requirements, and providing a basis for accurate detection. The gradient distribution of porosity 10%-50% can produce different amplitude signal responses for different degrees of flaws, making the textile flaw detection device have good detection sensitivity for various types and sizes of flaws, expanding the detection range. The selection of electrochromic material or optical-grade polymer and the thickness control of 50-200μm make the porous adjusting surface layer have good optical performance and structural stability, which can adapt to long-term industrial detection environment, reduce maintenance cost, and prolong the service life of the assembly.

[0082] The detection principle of the gradient light absorption assembly is based on the synergistic effect of the bottom layer and the porous adjusting surface layer. When light irradiates the textile and penetrates to the gradient light absorption assembly, the high light absorption material of the bottom layer absorbs the light passing through the pores of the porous adjusting surface layer, and the solid part of the porous adjusting surface layer reflects part of the light. Due to the 10%-50% gradient distribution of the porosity of the porous adjusting surface layer along the detection direction, when the textile has defects, the amount of penetrating light in the defect area increases. In the area with high porosity, more penetrating light will pass through the pores and be absorbed by the bottom layer, resulting in a larger change in the reflected light intensity in this area than in the area with low porosity, thereby amplifying the defect signal. The signal amplification achieved by the gradient distribution of porosity, combined with the high light absorption performance of the bottom layer, makes it easier for the detection device to capture the light signal changes caused by defects, providing clear and reliable basis for subsequent data analysis and defect identification.

[0083] As an embodiment, the pulse polarized light source group can emit polarized light of at least two wavelengths, has pulse modulation function and polarization direction switching function, and the light source intensity can be dynamically adjusted. The polarized light emitted by the pulse polarized light source group includes green light of 500-550 nm and red light of 600-650 nm, the pulse frequency is 10-50 Hz, and the polarization direction can be alternately switched between 0° and 90°.

[0084] The pulse polarized light source group has a multi-wavelength light emitting module built-in, which can emit polarized light of at least two wavelengths, including green light of 500-550 nm and red light of 600-650 nm, and meet the diversified detection needs through the characteristic differences of different wavelengths. The pulse modulation module can convert continuous light into pulse light, and the pulse frequency can be adjusted in the range of 10-50 Hz, so that the light signal output by the light source presents periodic changes, which is convenient for the detection device to synchronously collect signals. The polarization direction switching function is realized by a polarization controller, which can control the polarization light to be alternately switched between 0° and 90°, and by changing the vibration direction of the light, the recognition ability for different types of defects is enhanced. The light source intensity adjustment is completed through a power control unit, which can dynamically adjust the output light intensity according to the changes of the textile material, color and detection environment, to ensure the stability and applicability of the light signal in the detection process. Emitting polarized light of at least two wavelengths, especially green light of 500-550 nm and red light of 600-650 nm, the green light can enhance the light signal contrast of the surface defects of the textile, and the red light can highlight the deep fiber defects, improving the detection ability for different depths and different types of defects. The polarization direction is alternately switched between 0° and 90°, which can adapt to the different arrangement directions of the textile fibers, making the interaction between light and fibers present differentiation, so as to more clearly expose the defects of uneven fiber arrangement, increasing the comprehensiveness of detection. The light source intensity can be dynamically adjusted, which can provide appropriate light intensity for different textiles such as light and dark, thin and thick, etc., avoiding signal saturation caused by too strong light intensity or signal blur caused by too weak light intensity, ensuring the accuracy of detection.

[0085] As an embodiment, the hyperspectral-intensity fusion detection device includes a hyperspectral camera and an intensity sensor array, which are arranged at an angle of 30°-60° with the textile surface to synchronously collect spectral and intensity distribution data of reflected and transmitted light. The spectral range of the hyperspectral camera is 400-1000 nm, the detection accuracy of the intensity sensor array is ≤0.1 lux, and the sampling frequency of both is ≥100 fps.

[0086] The hyperspectral-intensity fusion detection device integrates a hyperspectral camera and an intensity sensor array, which work together to collect spectral and intensity distribution data, respectively. The hyperspectral-intensity fusion detection device is arranged at an angle of 30°-60° with the textile surface, which avoids the interference of direct light from the light source and efficiently captures the signals reflected by the gradient light-absorbing component after the reflected and transmitted light from the textile. This ensures that the collected light signals are representative. The spectral range of the hyperspectral camera covers 400-1000 nm, which can capture the characteristics of the reflected and transmitted light from the textile at different wavelengths, providing spectral basis for distinguishing different types of defects. The detection accuracy of the intensity sensor array reaches ≤0.1 lux, which can sense weak changes in light intensity. Meanwhile, the sampling frequency of both the hyperspectral camera and the intensity sensor array is ≥100 fps, ensuring real-time synchronous data collection and meeting the requirement of rapid detection.

[0087] The hyperspectral-intensity fusion detection device is aligned with the textile at a specific angle. After the pulsed polarized light source irradiates the textile, part of the light is reflected by the textile surface, and part of the light is transmitted through the textile and then reflected by the gradient light-absorbing component. The hyperspectral camera in the hyperspectral-intensity fusion detection device captures the spectral information of these two types of light. By analyzing the spectral characteristics at different wavelengths, the differences between the normal and defective areas of the textile can be distinguished. For example, stains can cause enhanced absorption of light at specific wavelengths, and fiber defects can cause changes in spectral reflectance. The intensity sensor array synchronously collects the intensity distribution data of these two types of light. Since the defective area causes changes in light intensity different from the normal area, and these changes are more obvious after being amplified by the gradient light-absorbing component, the sensor can accurately capture these light intensity differences. The spectral and intensity data are synchronously transmitted to the multi-modal data processing unit, which compares and analyzes them in combination with the dynamic reference library, ultimately realizing the identification and positioning of the defective area of the textile.

[0088] As an embodiment, the partitioned dynamic tension mechanism is composed of multiple clamping units driven by servo motors, which can adjust the tension of the textile in different zones, control the distance between the textile and the gradient light-absorbing component to be 1-5 mm, and the flatness deviation to be <100 μm. The multi-modal data processing unit fuses the spectral characteristics and intensity deviations through a convolutional neural network, and the gradient quantization accuracy of the defect edge is ≤0.3 mm.

[0089] The partition dynamic tension mechanism is composed of multiple clamping units, each of which is driven by an independent servo motor. Through the precise control of the servo motor, each clamping unit can apply different tension to different areas of the textile, realizing partition adjustment. This adjustment can stably control the distance between the textile and the gradient light absorption assembly to be 1-5mm, while ensuring that the flatness deviation of the textile is less than 100μm, providing reliable spatial conditions for the stable propagation and collection of light signals.

[0090] When the partition dynamic tension mechanism is working, each servo motor drives the corresponding clamping unit to apply appropriate tension to different areas according to the characteristics of the textile such as material and thickness. Through this partition adjustment, the textile is evenly spread out and maintains a stable distance of 1-5mm with the gradient light absorption assembly, avoiding problems such as light signal refraction and scattering caused by uneven textile or improper distance, ensuring that the light signal maintains a stable path and intensity during propagation. After the multi-modal data processing unit receives the spectral features and light intensity deviation transmitted by the hyperspectral-light intensity fusion detection device, the convolutional neural network processes these data layer by layer. Through learning the feature patterns of flawless samples, the network can quickly identify abnormal information that differs from normal features, and these abnormal information corresponds to possible defect areas.

[0091] The present application discloses a kind of textile defect detection method, comprising:

[0092] S1: construct dynamic reference library, through partition dynamic tension mechanism fixed flawless textile sample, utilize pulse polarized light source group to irradiate and collect reflected spectrum and light intensity distribution data, establish reference feature model;

[0093] S2: real-time spectrum and light intensity acquisition, repeat the irradiation and acquisition operation of S1 to the textile to be measured, obtain real-time detection data;

[0094] S3: defect feature analysis, compare real-time data with dynamic reference library by multi-modal data processing unit, combined with the light intensity change amplified by gradient light absorption assembly, identify suspected defect area;

[0095] S4: multi-dimensional verification, switch detection parameters to repeat detection to suspected area, verify the consistency of defect features;

[0096] S5: output defect information, quantize the edge gradient of defect area and output in stages.

[0097] S1, when constructing the dynamic benchmark library, the flawless textile sample is fixed by the partition dynamic tension mechanism to keep it flat and stable distance from the gradient light absorption assembly. The pulse polarized light source group irradiates the sample according to the set parameters, the hyperspectral-intensity fusion detection device synchronously collects the reflectance spectrum and intensity distribution data, and the multi-modal data processing unit analyzes these data, extracts common features and establishes a benchmark feature model as a reference standard for subsequent detection.

[0098] S2, in the real-time spectrum and intensity collection stage, the same fixing method, light source parameters and collection method as S1 are used for the textile to be tested. The partition dynamic tension mechanism ensures that the textile to be tested is in the same detection state as the sample. After irradiation by the pulse polarized light source group, the hyperspectral-intensity fusion detection device collects real-time spectral and intensity data, providing a basis for subsequent comparison.

[0099] S3, in the defect feature analysis, the multi-modal data processing unit compares the real-time detection data with the benchmark feature model in the dynamic benchmark library, calculates the differences in spectral features, intensity distribution, etc. Meanwhile, combined with the amplified intensity changes of the gradient light absorption assembly, regions with large deviations from normal features are selected and marked as suspected defect regions.

[0100] S4, in the multi-dimensional verification, for suspected defect regions, the parameters of the pulse polarized light source group such as wavelength, polarization direction, pulse frequency, or the reflectivity of the gradient light absorption assembly are adjusted, and spectral and intensity collection is repeated. The multi-modal data processing unit analyzes the multiple detection results, and if the defect features remain consistent, the region is confirmed as a true defect region. S5, in the output defect information stage, the multi-modal data processing unit quantifies the edge gradient of the defect region and classifies the defect according to the pre-set grading standard, and finally outputs the detection results including defect location, type, grade, etc.

[0101] The textile defect detection method takes the dynamic benchmark library as a reference and realizes defect detection through comparative analysis. In S1, the spectral and intensity data of the flawless sample under standardized conditions are recorded and constructed into a benchmark model, reflecting the optical signal features of normal textiles. The real-time data collected in S2 contain the optical signal changes caused by possible defects. In S3, the multi-modal data processing unit compares the real-time data with the benchmark model. Due to the amplification of the intensity changes in the defect region by the gradient light absorption assembly, the difference between the optical signal of the defect region and that of the normal region is more significant, enabling accurate identification of suspected defect regions. In S4, repeated detection by switching detection parameters further verifies whether the features of the suspected regions are stable, eliminating false positives caused by accidental factors. In S5, the features such as edge gradient of the defect are quantified and classified, and the final output results comprehensively reflect the defect situation of the textile, realizing accurate detection and evaluation of textile defects.

[0102] As an embodiment, at least 30 groups of flawless sample data of different materials and different environmental conditions are collected when constructing the dynamic reference library in S1, and common reference features are extracted through principal component analysis. In S2, the pulsed polarized light source group alternately emits polarized light of different wavelengths, and the hyperspectral-light intensity fusion detection device synchronously records the multi-band spectrum and light intensity distribution matrix. The polarized light of different wavelengths in S2 includes green light and red light. The green light is used to enhance the spectral contrast of the multi-hole adjusting surface defect, and the red light is used to amplify the light intensity difference of the deep fiber defect.

[0103] When constructing the dynamic reference library in S1, at least 30 groups of flawless textile samples are selected, covering different materials such as cotton, hemp, silk, and chemical fiber. At the same time, different environmental conditions are simulated, such as adjusting temperature and humidity. Through the partition dynamic tension mechanism, these samples are fixed, and under the unified detection setting, the hyperspectral-light intensity fusion detection device collects the reflection spectrum and light intensity distribution data of each sample. The multi-modal data processing unit performs principal component analysis on these data to extract the common features of different samples in terms of spectrum and light intensity, and constructs a common reference feature model.

[0104] In the real-time spectrum and light intensity acquisition stage S2, the pulsed polarized light source group alternately emits polarized light of different wavelengths according to the set program, i.e. green light and red light. At the same time when the light source emits light, the hyperspectral-light intensity fusion detection device works synchronously, records the multi-band spectrum data and light intensity distribution matrix of the textile reflection, and forms a complete real-time detection data set. Among them, the emission of green light is aimed at the surface layer of the textile. When the green light irradiates the surface defect, it will interact with the defect, making the surface defect more obvious in spectrum, which is convenient for the hyperspectral camera to capture its features. Red light is more likely to penetrate the surface layer of the textile and reach the deep fiber. When there is a fiber defect in the deep layer, the light intensity of the red light will change accordingly, thereby amplifying the light intensity difference of the deep fiber defect, so that the light intensity sensor array can accurately perceive.

[0105] As an embodiment, in S3, wavelet transform is used to denoise the light intensity data, the edge detection algorithm is used to extract the contour features of the defect area, and the spectrum angle matching is combined to distinguish the defect type.

[0106] S3 carries out the flaw feature analysis, first applies wavelet transform to the collected light intensity data. Wavelet transform can decompose the light intensity data into different frequency components, by retaining the effective signal frequency related to the flaw, filtering out irrelevant high or low frequency noise such as environmental interference, equipment noise, etc., to obtain the denoised light intensity data. Subsequently, the edge detection algorithm is used to process the denoised light intensity data. The algorithm identifies the area of light intensity mutation, determines the boundary between the flaw area and the normal area, and then extracts the contour features of the flaw area, such as shape, size, etc. At the same time, the spectral angle matching method is used to compare the spectral features of the flaw area with the known flaw types. Spectral angle matching calculates the angle between different spectra, the smaller the angle, the more similar the spectral features, so as to distinguish different types of flaws such as stains, fiber breakage, uneven yarn, etc. In the whole process, wavelet transform, edge detection algorithm and spectral angle matching work together, from denoising, contour extraction to type distinction, gradually completing the comprehensive analysis of the flaw features.

[0107] The principle of wavelet transform denoising is based on the difference between noise and effective signal in the frequency domain. The light intensity change caused by the flaw has a specific frequency range, while the noise is usually distributed in other frequency intervals. By decomposing the light intensity data into different frequency channels through wavelet transform, the components in the noise channel are removed, and the components in the effective signal channel are retained, thereby achieving denoising and making the flaw signal more prominent. The core of the edge detection algorithm is to identify the points of light intensity mutation. In textiles, there is a significant difference in light intensity between the flaw area and the normal area, and this difference is manifested as a sharp change in light intensity at the boundary. The edge detection algorithm calculates the gradient change of light intensity to find points where the gradient value exceeds the threshold, and the curve formed by connecting these points is the contour of the flaw area, thereby achieving the extraction of the flaw contour. The principle of spectral angle matching to distinguish flaw types is that different types of flaws have unique spectral features. Due to the difference in composition and structure, each type of flaw has different reflection and absorption characteristics of light, resulting in a specific shape of the spectral curve. By calculating the angle between the spectrum of the flaw to be detected and the spectrum of the known type of flaw, the smaller the angle, the more likely it belongs to the same type, thereby achieving accurate distinction of the flaw type. These three methods work together to purify the data, extract the contour, and finally determine the type, gradually and deeply completing the analysis of the flaw features, providing a scientific and accurate basis for identifying suspected flaw areas.

[0108] As an embodiment, the multi-dimensional verification in S4 includes switching the light wavelength of the pulsed polarized light source group, adjusting the surface reflectivity of the gradient light absorption component, and if the deviation of the flaw features in multiple detections is ≤5%, it is determined as a real flaw. The flaw grading in S5 includes light intensity deviation amplitude, spectral shift, and edge gradient, and is divided into three levels: slight, moderate, and severe, each level corresponding to a specific parameter threshold range.

[0109] S4, when performing multi-dimensional verification, for suspected defect areas, first switch the light wavelength type of the pulsed polarized light source group, that is, alternate between green light and red light. At the same time, change the polarization direction, and alternately adjust between 0° and 90°. In addition, adjust the surface reflectivity of the gradient light absorption assembly by changing the voltage of the electrochromic material or selecting an optical-grade polymer with different reflectivity, so that the surface reflectivity changes within a set range. After adjusting the parameters each time, the hyperspectral-light intensity fusion detection device reacquires data, and the multi-modal data processing unit compares the defect features of multiple detections. If the deviation is ≤5%, it is determined to be a real defect.

[0110] S5, when performing defect grading, the multi-modal data processing unit first calculates the light intensity deviation amplitude of the defect area, that is, the proportion of the light intensity difference from the normal area in the dynamic reference library. Then analyze the spectral shift, that is, the wavelength shift degree of the spectral curve of the defect area and the reference spectral curve. At the same time, combined with the quantified edge gradient, according to the preset parameter threshold range, the defect is divided into three levels of slight, moderate and severe. Each level corresponds to a clear light intensity deviation amplitude, spectral shift and edge gradient range, ensuring the uniformity of the grading standard.

[0111] The principle of S4 multi-dimensional verification is based on the stability of the optical characteristics of real defects, while the characteristics of false defects will fluctuate significantly with the change of detection parameters. When switching the type of polarized light, green light is sensitive to surface defects, and red light is sensitive to deep defects. If the suspected area shows similar characteristics under both types of light, it is likely to be a real defect. Changing the polarization direction will affect the interaction between light and fiber. The light intensity and spectral change law of real defects is consistent under different polarization directions, while false defects do not have this law. When adjusting the surface reflectivity of the gradient light absorption assembly, the light intensity change amplitude of real defects will change proportionally with the reflectivity, and the change trend is stable, while the light intensity change of false defects is irregular. When the deviation of the defect features detected multiple times is ≤5%, it is indicated that the characteristics are not affected by parameter adjustment, which meets the optical characteristics of real defects, and therefore can be determined as a real defect.

[0112] As an embodiment, the textile is pretreated before detection, including surface cleaning and humidity adjustment, so that the textile moisture regain is maintained at 8%-15%, reducing environmental interference. The multi-modal data processing unit uses a convolutional neural network to fuse real-time data features, and the model training sample size is ≥1000 groups.

[0113] Before detection, the textile is pretreated. First, surface cleaning is performed to remove dust, fiber debris and other impurities on the surface of the textile by a dust removal device to avoid misjudgment of these impurities as defects. Then, humidity adjustment is performed to control the environmental humidity by using a constant temperature and humidity device to stabilize the moisture regain of the textile at 8%-15%. The moisture regain is monitored in real time by a humidity sensor to ensure that it reaches the preset range before entering the detection link. The multi-modal data processing unit uses a convolutional neural network for feature fusion. First, at least 1000 groups of textile sample data containing different types and degrees of defects are collected. These data cover multi-dimensional information such as spectral features and light intensity distribution. These sample data are input into the convolutional neural network for training. Through the calculation of multiple layers of neural networks, the model gradually learns the feature patterns of different defects, and finally achieves a defect recognition accuracy of ≥95%.

[0114] Example 2: Detection of surface stains on silk fabric

[0115] Detection object: thin silk fabric with a thickness of about 0.05 mm, containing surface stains such as oil stains and water stains with a diameter of 1-2 mm, which are difficult to identify with the naked eye.

[0116] The bottom layer of the gradient light absorption assembly is a nano carbon black coating with an absorption rate of 96%. The porous adjustment surface layer is optical grade polyethylene with a thickness of 50 μm, a pore size of 3 μm, and a porosity gradient of 10%-30% along the detection direction.

[0117] The pulsed polarized light source group alternately emits 530 nm green light and 630 nm red light at a pulse frequency of 30 Hz, with a polarization direction of 0° / 90° switching, and the light intensity is adjusted to a low gear.

[0118] The hyperspectral-light intensity fusion detection device is at a 45° angle with the fabric, and the detection precision is 400-800 nm for the hyperspectral camera and 0.05 lux for the light intensity sensor array, with a sampling frequency of 100 fps.

[0119] The partitioned dynamic tension mechanism controls the distance between the fabric and the gradient light absorption assembly to be 1 mm, with a flatness deviation of <50 μm.

[0120] 30 groups of defect-free silk samples of different batches are collected to establish a baseline model under the conditions of 25°C and 50% humidity, and the reflection spectrum characteristics of the 530 nm green light band are extracted.

[0121] When the light source emits 530 nm green light, the reflectivity of the stain area is reduced by 15%-20% due to the presence of oil and water, and the amount of penetrating light increases.

[0122] The high porosity area of the gradient light absorption assembly has a 2-fold increase in the absorption of penetrating light compared to the normal area, and the light intensity sensor measures a light intensity deviation of 5 lux in this area.

[0123] After denoising by wavelet transform, the multi-modal unit extracts the stain profile by edge detection algorithm. The spectral angle of the stain region is >15° at 530nm waveband, which is determined as surface stain.

[0124] After switching the polarization direction by 90°, the detection is repeated. The deviation of the stain characteristics is <3%, confirming the authenticity. The light intensity gradient is quantified as 30lux / mm, and the edge gradient is 0.2mm, which is determined as a slight defect according to the standard.

[0125] The detection effect is 100% accurate in stain recognition, with a lower miss rate than traditional optical detection, and a detection speed of 30 frames / second.

[0126] Example 3: Detection of uneven yarn in denim

[0127] The detection object is thick denim with a thickness of 1.2mm, which has local yarn density unevenness, deep fiber stacking / sparse, and no obvious traces on the surface.

[0128] The bottom layer of the gradient light absorption assembly is black ceramic with a light absorption rate of 95%, and the porous adjustment surface layer is electrochromic material. The thickness of the electrochromic material is 200μm, the pore size of the pores is 10μm, and the porosity gradient is 30%-50%.

[0129] The pulse polarized light source group is mainly 630nm red light, with a pulse frequency of 50Hz, a polarization direction of 0°, and a light intensity adjusted to the high gear.

[0130] The hyperspectral-light intensity fusion detection device is at a 60° angle with the fabric, the detection accuracy of the hyperspectral camera is 600-1000nm, the accuracy of the light intensity sensor array is 0.1lux, and the sampling frequency is 100fps.

[0131] The partitioned dynamic tension mechanism controls the distance between the fabric and the assembly to be 5mm, and the tension is adjusted to a high level.

[0132] 30 groups of different color and depth of denim samples without defects are collected, and a baseline model is established at 28℃ and 60% humidity. The transmission light intensity distribution characteristics of the 630nm red light waveband are extracted.

[0133] When red light penetrates the denim, the transmission light amount in the yarn sparse area increases by 30%-50%. In the 50% porosity area of the porous adjustment surface layer, the amount of transmission light absorbed by the bottom layer is 40% higher than that in the normal area, resulting in a decrease in reflected light intensity by 2-3lux.

[0134] The multi-modal unit fuses the spectral data and light intensity distribution, and identifies the yarn uneven area through convolutional neural network, with an edge gradient quantization of 0.2mm.

[0135] Switch the polarization direction to 90°, repeat the detection of deviation <2%, confirm the authenticity; According to the light intensity deviation amplitude, it is judged as medium flaw.

[0136] Successfully identify deep yarn unevenness, detection accuracy of 0.3mm, adapt to denim production line speed 10m / min.

[0137] Example 4: wool fabric fiber break detection

[0138] The detection object is thick wool fabric with a thickness of 1mm, there are small clumps formed by local fiber breakage, the diameter of the small clumps is 0.5-1mm, and the internal structure is loose.

[0139] Device parameter setting

[0140] The bottom layer of the gradient light absorption assembly is black ceramic with a light absorption rate of 95%, the porous adjustment surface layer is electrochromic material, the thickness of the electrochromic material is 100μm, the pore size of the pores is 5μm, and the porosity gradient is 20%-50%.

[0141] The pulse polarized light source group is mainly 650nm red light, the pulse frequency is 20Hz, the polarization direction is 0°, and the light intensity is adjusted to the middle gear.

[0142] The hyperspectral-light intensity fusion detection device is at an angle of 30° with the fabric, and the detection accuracy of the hyperspectral camera is 500-1000nm, and the sampling frequency is 100fps.

[0143] The partition dynamic tension mechanism controls the fabric and assembly spacing to be 3mm, and the flatness deviation is <80μm.

[0144] Detection process and effect

[0145] Collect 30 groups of flawless wool samples, extract the spectral features of the 650nm red light band.

[0146] The fiber breakage clump area has a loose structure, and the red light penetration increases by 25%. In the 30%-50% porosity area of the porous adjustment surface layer, the reflected light intensity changes by 35% higher than the normal area, and the local light intensity increases by 1.5-2lux.

[0147] The spectral angle matching displays the 650nm band spectral angle of the clump area >10°, the edge detection algorithm extracts the clump contour, and the edge gradient quantization is 0.25mm.

[0148] Adjust the light source intensity ±20%, repeat the detection deviation <4%, confirm the authenticity; According to the edge gradient and spectral offset, it is judged as a slight flaw.

[0149] The recognition rate of the small clumps is 98%, the missed detection rate is reduced by 80% compared with manual detection, and the detection speed is 30 frames / second.

[0150] Embodiment common conclusion

[0151] The signal amplification effect of the gradient light absorption assembly is achieved by the porosity gradient design, so that the light intensity variation of the flaw area is amplified by 2-4 times, and the problem of weak signal of small flaws is solved.

[0152] The multi-modal data fusion advantage: combined with spectral and light intensity data, the accuracy is improved by 20%-30% compared with single signal detection.

[0153] Through the partition dynamic tension mechanism and light source parameter adjustment, various fabrics from light and thin silk to thick and heavy wool can be adapted, and the detection precision is all within 0.3mm.

[0154] The detection speed is greater than or equal to 100fps, which meets the real-time detection demand of industrial production line, and the efficiency is improved by more than 10 times compared with manual detection.

[0155] It should be understood that for those skilled in the art, improvements or changes can be made according to the above description, and all these improvements and changes shall belong to the protection scope of the appended claims of the present application.

Claims

1. A textile defect detection apparatus, characterized in that, The gradient light absorption assembly is arranged below the textile and is composed of a bottom layer with light absorption effect and a porous adjusting surface layer with light reflection effect. The porous adjusting surface layer has a plurality of pores, and the porosity of the porous adjusting surface layer is gradiently distributed along the detection direction to amplify the local change of the penetration light intensity through the porosity difference. The pulse polarized light source group is arranged above the textile and can emit polarized light. Part of the polarized light is reflected by the textile to form reflected light, and the other part of the polarized light passes through the textile to form penetration light. The hyperspectral-light intensity fusion detection device is arranged above one side of the textile and is divided into two parts with the pulse polarized light source group. The hyperspectral-light intensity fusion detection device is used to collect the spectrum and light intensity distribution of the reflected light and the penetration light reflected by the porous adjusting surface layer. The partitioned dynamic tension mechanism is arranged on one side of the textile and is used to fix the textile and control the distance between the textile and the gradient light absorption assembly. The multi-modal data processing unit is arranged on one side of the textile close to the hyperspectral-light intensity fusion detection device and is connected to the hyperspectral-light intensity fusion detection device. The multi-modal data processing unit pre-stores a dynamic reference library constructed based on a flawless textile sample, and is used to compare and analyze and identify the defect area of the textile and quantify the edge gradient. When the textile has defects, the absorption amount of the penetration light on the bottom layer is different through the porosity difference of different areas of the porous adjusting surface layer, thereby causing different changes in the reflection light intensity of the porous adjusting surface layer.

2. The textile defect detection device according to claim 1, wherein: The bottom layer of the gradient light absorption assembly is a high light absorption material layer with a light absorption rate of ≥95%. The pore size of the porous adjusting surface layer is microns, and the gradient distribution range of the porosity along the detection direction is 10%-50%. The bottom layer is a nano-carbon black coating or a black ceramic material, the porous adjusting surface layer is an electrochromic material or an optical-grade polymer, and the thickness of the porous adjusting surface layer is 50-200μm.

3. The textile defect detection device according to claim 1, wherein: The pulse polarized light source group can emit at least two wavelengths of polarized light, has pulse modulation function and polarization direction switching function, and the light source intensity can be dynamically adjusted. The pulse polarized light source group emits polarized light including 500-550nm green light and 600-650nm red light, the pulse frequency is 10-50Hz, and the polarization direction can be alternately switched between 0° and 90°.

4. The textile defect detection device according to claim 1, wherein: The hyperspectral-light intensity fusion detection device includes a hyperspectral camera and a light intensity sensor array, and is arranged at an angle of 30°-60° with the textile surface to synchronously collect the spectrum and light intensity distribution data of the reflected light and the penetration light. The spectral range of the hyperspectral camera is 400-1000nm, the detection accuracy of the light intensity sensor array is ≤0.1lux, and the sampling frequency of both is ≥100fps.

5. The textile defect detection device according to claim 1, wherein: ​ The partition dynamic tension mechanism is composed of multiple servo motors independently driven clamping units, which can adjust the tension of the textile in different partitions, control the distance between the textile and the gradient light absorption component to be 2-5mm, and the flatness deviation to be less than 100μm; The multi-modal data processing unit fuses spectral features and light intensity deviation through a convolutional neural network, and the edge gradient quantization accuracy of the defect is less than or equal to 0.3mm.

6. A method for textile defect detection, for use with any of the textile defect detection apparatuses of claims 1-5, comprising: It comprises: S1: Constructing a dynamic reference library, fixing a sample of flawless textile by a partition dynamic tension mechanism, irradiating and collecting reflected spectrum and light intensity distribution data by using a pulsed polarized light source group, and establishing a reference feature model; S2: Real-time spectrum and light intensity collection, repeating the irradiation and collection operation of S1 for the textile to be tested to obtain real-time detection data; S3: Defect feature analysis, comparing real-time data with the dynamic reference library through a multi-modal data processing unit, combining the amplified light intensity changes of the gradient light absorption component, and identifying suspected defect areas; S4: Multi-dimensional verification, repeating the detection of suspected areas by switching detection parameters to verify the consistency of defect features; S5: Outputting defect information, quantifying the edge gradient of the defect area and grading the output.

7. The textile defect detection method according to claim 6, wherein: In S1, at least 30 groups of sample data of flawless samples under different material and environmental conditions are collected, and common reference features are extracted by principal component analysis when constructing the dynamic reference library; In S2, the pulsed polarized light source group alternately emits polarized light of different wavelengths, and the hyperspectral-light intensity fusion detection device synchronously records the multi-band spectrum and light intensity distribution matrix during real-time spectrum and light intensity collection; The polarized light of different wavelengths in S2 includes green light and red light, the green light is used to enhance the spectral contrast of surface defects, and the red light is used to amplify the light intensity difference of deep fiber defects.

8. The textile defect detection method according to claim 6, wherein: In S3, the light intensity data is denoised by wavelet transform, the contour features of the defect area are extracted by edge detection algorithm, and the defect types are distinguished by combining spectral angle matching.

9. The textile defect detection method according to claim 6, wherein: In S4, multi-dimensional verification includes switching the light wavelength, polarization direction of the pulsed polarized light source group, and adjusting the surface reflectivity of the gradient light absorption component, and if the defect feature deviation is less than or equal to 5% after multiple detections, it is determined to be a real defect; In S5, the defect grading is based on light intensity deviation amplitude, spectral shift and edge gradient, and is divided into three levels: slight, moderate and severe, each level corresponds to a specific parameter threshold range.

10. The textile defect detection method according to claim 6, wherein: Before detection, the textile is pretreated, including surface cleaning and humidity adjustment, so that the moisture regain of the textile is maintained at 8%-15%, and the environmental interference is reduced; The multi-modal data processing unit uses a convolutional neural network to fuse the features of the real-time data, and the model training sample size is greater than or equal to 1000 groups.

Images