A multi-sensor fusion cloth online flaw identification method and system

Abstract

The application provides a multi-sensor fusion cloth online defect identification method and system, and relates to the technical field of cloth defect identification. The method comprises the following steps: acquiring specified graphs of the front and back surfaces of the current detection cloth; performing space-time alignment on the data in the specified graphs to obtain multi-source data after alignment; extracting key information from the multi-source data; determining the fusion weight of each data in the key information according to the cloth type parameter of the current detection cloth; performing weighted fusion on the data in the key information according to the fusion weight to generate a fusion feature vector; comparing the fusion feature vector with the fusion feature of normal cloth to determine and locate the color difference area. The method of the application aims at the cloth manufacturing production line and aims to solve the technical problem of color difference defect detection of cloth with double-sidedness, light transmission and after finishing process, and realizes more accurate identification and positioning of color difference defects of complex cloth.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cloth defect recognition, in particular to a cloth online defect recognition method and system based on multi-sensor fusion. BACKGROUND

[0002] In cloth manufacturing enterprises, after the cloth dyeing and finishing process is completed, automatic quality detection is an important link to ensure that the product meets the standards. The existing automatic detection system is usually deployed on the cloth production line to detect the cloth passing through the detection area in a continuous manner. This kind of system is often configured with visible light image sensors and spectral sensors to collect visual information and spectral information on the surface of the cloth, and the main goal is to identify whether the cloth has color difference defects.

[0003] However, in actual production, the types of cloth produced by enterprises are diverse, and a part of the cloth has double-sided characteristics, such as some functional fabrics or jacquard fabrics, which may have differences in color, texture or functional coating on the front and back sides. In order to ensure the overall quality of the cloth, the front and back sides of the cloth need to be detected simultaneously or step by step. This requires the detection system to be able to simultaneously collect image data and spectral data of the front and back sides of the cloth. For example, visible light image sensors and spectral sensor arrays can be configured above and below the cloth as it passes through the detection area, or the cloth can pass through a single-side detection station again after passing through a turning mechanism. Double-sided data collection doubles the amount of raw data, and the correlation between the front and back data in space and time needs to be considered, which increases the complexity of data processing.

[0004] In addition, part of the cloth has certain light transmittance, especially thin fabrics such as silk or light and thin chemical fiber fabrics. For such cloth, color performance not only depends on the reflection characteristics of its surface to light, but also is affected by the internal structure or fiber distribution after the light penetrates the cloth, as well as the characteristics of the light transmitted from the back of the cloth. Color difference may be reflected on reflected light, transmitted light, or both. In order to comprehensively evaluate the color quality of such cloth, the detection system needs to be able to collect the transmission spectral data of the cloth and possibly the image data combined with the transmission light illumination. This requires configuring a light source below the cloth and configuring a transmission spectral sensor and an image sensor above or below the cloth. Combining transmission data further increases the complexity of data sources and the difficulty of data fusion.

[0005] After dyeing and finishing, fabrics usually undergo various finishing processes such as softening, wrinkle resistance, water resistance, anti-static, or functional coating. These finishing processes use specific chemical additives or materials that adhere to the surface of the fabric fibers or penetrate into the fibers. These additives or materials themselves have specific optical properties such as refractive index, absorption spectrum or scattering characteristics. These optical properties will superimpose or change the spectral response of the original dye of the fabric. For example, certain waterproof coatings increase the gloss of the fabric surface, changing the reflection characteristics; certain anti-ultraviolet finishing agents have strong absorption in the ultraviolet band; fluorescent whitening agents emit light in the visible region under ultraviolet excitation. When the finishing process is locally uneven on the fabric, such as uneven coating thickness, uneven additive distribution, it will cause the optical properties of the local area of the fabric to be slightly different from the surrounding area. This local change in optical properties caused by uneven finishing may appear as a local gloss difference, slight color shift or brightness change in the visible light image, and as an abnormal deviation of the local area spectral curve in a specific wavelength range in the spectral data. This optical anomaly may be similar to the real color difference defects caused by uneven dye concentration or improper dyeing in terms of visual or spectral characteristics, or superimposed with real color difference, making it difficult to distinguish.

[0006] Traditional fabric color difference detection methods usually only focus on the reflection image and reflection spectrum data of the fabric surface, and mainly analyze based on the characteristics of the dye itself. These methods are difficult to effectively deal with fabrics with double-sidedness, light transmission and complex finishing processes. Simply splicing or fixed weight fusion of front and back data, reflection / transmission data, image / spectrum data is not enough to accurately distinguish real color difference from optical illusions or interference caused by fabric structure, light transmission or uneven finishing. For example, for light-transmitting fabrics, transmission spectra may provide key information about internal structure or deep dye distribution, while reflection spectra focus more on the surface. For finished fabrics, features that can strip the finishing influence and only reflect dye distribution differences need to be extracted from complex fusion data.

[0007] Therefore, innovation is needed in data acquisition, preprocessing, feature extraction and fusion strategy to fully utilize the complementary information of multi-source heterogeneous data while suppressing the influence of various interference factors. In addition, fabrics may experience slight positional shifts or deformations during high-speed movement, challenging the accurate spatial alignment of front and back, reflection / transmission data. How to complete these complex data processing and fusion in real time on high-throughput production lines and give accurate color difference judgment and positioning is the key to realizing robust fabric quality detection. SUMMARY

[0008] The purpose of the present application is to provide a multi-sensor fusion cloth online defect identification method and system, which aims to solve the technical problem of color difference defect detection for cloth with double-sidedness, light transmission and after-finishing process on the cloth manufacturing production line, and realize more accurate identification and positioning of complex cloth color difference defects.

[0009] In a first aspect, the present application provides a multi-sensor fusion cloth online defect identification method, comprising the following steps:

[0010] Obtaining the specified atlas of the front and back of the current detection cloth;

[0011] Aligning the data in the specified atlas in space and time to obtain multi-source data after alignment;

[0012] Extracting key information from the multi-source data;

[0013] Determining the fusion weight of each data in the key information according to the cloth type parameter of the current detection cloth;

[0014] Weighted fusion of the data in the key information according to the fusion weight to generate a fusion feature vector;

[0015] Comparing the fusion feature vector with the fusion feature of normal cloth, if the comparison result is deviation, then according to the intensity value of the after-finishing interference feature in the key information, adjusting the color difference judgment threshold to determine the color difference area and positioning.

[0016] The multi-sensor fusion cloth online defect identification method provided by the present application can efficiently collect and accurately align the reflection and transmission image and spectral data of the front and back of the cloth under the harsh conditions of high-speed running, data volume multiplication, complex data sources and multiple optical interferences (such as optical changes caused by structure, light transmission and uneven after-finishing), and through multi-source data hierarchical feature extraction and cloth type adaptive weighted fusion analysis, effectively removes or suppresses the optical interference introduced by non-dye factors, accurately identifies and locates the real color difference defects caused by dyes, and meets the real-time processing needs of the production line.

[0017] In a second aspect, the present application provides a multi-sensor fusion cloth online defect identification system, comprising:

[0018] An acquisition module for acquiring the specified atlas of the front and back of the current detection cloth;

[0019] An alignment module for aligning the data in the specified atlas in space and time to obtain multi-source data after alignment;

[0020] extracting a key information from the multi-source data;

[0021] determining a fusion weight of each data in the key information according to a fabric type parameter of the current detected fabric;

[0022] generating a fusion feature vector by weighted fusion of the data in the key information according to the fusion weight;

[0023] comparing the fusion feature vector with a fusion feature of normal fabric, and if the comparison result is deviation, determining a color difference region and positioning by adjusting a color difference judgment threshold according to an intensity value of the post-finishing interference feature in the key information.

[0024] As can be seen from the above, the multi-sensor fusion fabric online defect identification method provided by the application can effectively solve the challenges encountered by the existing fabric color difference detection system when processing complex fabrics with double-sidedness, light transmission and post-finishing process, especially in a high-throughput production line environment. Through multi-source data acquisition, accurate alignment, hierarchical feature extraction (including post-finishing interference feature) and weighted fusion and judgment logic based on fabric type adaptation, more accurate identification and positioning of complex fabric color difference defects are realized, the robustness and reliability of fabric quality detection are improved, and the high efficiency requirement of real-time processing of the production line is met.

[0025] Other features and advantages of the present application will be described in the following description, and some will become apparent from the description, or will be learned from practice of the application. The objectives and other advantages of the present application will be realized and attained by the structure particularly pointed out in the written description and the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 A flow chart of the multi-sensor fusion fabric online defect identification method provided by the embodiment of the present application.

[0027] Figure 2 A structural schematic diagram of the multi-sensor fusion fabric online defect identification system provided by the embodiment of the present application.

[0028] Label explanation:

[0029] 100, acquisition module; 200, alignment module; 300, extraction module; 400, determination module; 500, generation module; 600, comparison module. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. The components of the embodiments of the present application described and shown in the accompanying drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0031] It should be noted that: similar reference numbers and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. Meanwhile, in the description of the present application, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0032] With reference to the accompanying drawings, Figure 1 the present application provides a multi-sensor fusion cloth online defect identification method, comprising the following steps:

[0033] obtaining the specified atlas of the front and back of the current detected cloth; the specified atlas includes reflection image, reflection spectrum, transmission image and transmission spectrum;

[0034] using image registration technology and time synchronization technology, spatially aligning and time aligning the data in the specified atlas to obtain the aligned multi-source data;

[0035] extracting key information from the multi-source data; the key information includes color space component, local texture feature, local brightness feature, specific wavelength intensity ratio, spectrum curve slope, spectrum curve curvature, absorption peak position, absorption peak intensity, correlation feature of reflection spectrum and transmission spectrum, difference feature of reflection spectrum and transmission spectrum, post-finishing interference feature;

[0036] determining the fusion weight of each data in the key information according to the cloth type parameter of the current detected cloth;

[0037] weighting and fusing the data in the key information according to the fusion weight to generate a fusion feature vector;

[0038] comparing the fusion feature vector with the fusion feature of normal cloth, if the comparison result is deviation, then determining the color difference area and positioning by adjusting the color difference judgment threshold according to the intensity value of the post-finishing interference feature in the key information.

[0039] The designated atlas refers to a multi-modal data set used to describe the current detection of cloth optical properties, which can include reflection images, reflection spectra, transmission images and transmission spectra, which are mainly used to obtain comprehensive information of the front and back of the cloth under different light modes. Image registration technology and time synchronization technology refer to technologies used to correct the spatial position and time deviation of data collected by different sensors. Spatial alignment can be achieved by using feature matching-based image registration algorithm, frequency domain analysis-based image registration algorithm or machine learning-based image registration algorithm. Time alignment can be achieved by using timestamp alignment or hardware trigger synchronization. The main purpose is to ensure that the data from different sensors can accurately correspond to the same physical area and the same time of the cloth. Key information refers to a set of features extracted from the aligned multi-source data, which are used to represent the state and potential defects of the cloth. It can include color space components, local texture features, local brightness features, specific wavelength intensity ratios, spectral curve slopes, spectral curve curvatures, absorption peak positions, absorption peak intensities, correlation features of reflection spectra and transmission spectra, difference features of reflection spectra and transmission spectra, and post-finishing interference features. The main purpose is to extract discriminative information related to cloth defects, structural characteristics and post-finishing effects from multi-source data. Cloth type parameters refer to parameters that describe the inherent properties of the current detected cloth, which can include fiber type, yarn density, weaving method, finishing process, etc. The main purpose is to provide prior information of the cloth to guide the subsequent feature processing and judgment. Fusion weight refers to the relative importance coefficient of different key information data in the process of multi-source feature fusion. The main purpose is to dynamically adjust the contribution of each feature in the final judgment according to the cloth type and interference. Post-finishing interference features refer to features used to quantify the influence of cloth finishing process on optical properties. The main purpose is to identify and evaluate the optical artifacts caused by uneven finishing, so as to distinguish in defect judgment. Adjusting the color difference judgment threshold refers to dynamically changing the threshold for judging whether there is color difference according to certain conditions. The main purpose is to adaptively adjust the strictness of color difference judgment according to the intensity of post-finishing interference, to avoid misjudgment.

[0040] The working principle of the present application is as follows: first, through the sensors and light sources located above and below the cloth, the reflection images, reflection spectra, transmission images and transmission spectra of the front and back of the cloth are collected at high speed and synchronously. Then, the collected data are preprocessed, including accurate spatial alignment using image registration technology and time synchronization to ensure that different sources of data correspond to the same area and the same time of the cloth. Then, the hierarchical feature extraction stage is entered, and different levels of features are extracted from the original data, including basic optical intensity, color, texture, spectral shape and other middle-level features, as well as specially identified optical interference features related to post-finishing unevenness. In the feature fusion and judgment stage, according to the type of the cloth being detected, the preset parameters are consulted, the weights from different data sources (front / back, reflection / transmission, image / spectrum) and different feature types (color, texture, spectrum, post-finishing interference) are dynamically adjusted, the features are weighted and fused to form a comprehensive high-level feature vector. Finally, based on this fused feature vector, the feature mode of normal cloth is compared to determine whether there is a color difference defect. In the judgment process, the system will refer to the extracted post-finishing interference features. If the interference features are significant, the judgment logic or threshold will be adjusted to distinguish between real color differences caused by dyes and optical illusions caused by post-finishing unevenness, cloth structure or light transmission, thereby achieving accurate identification, positioning and reducing false positives of color difference defects. The entire process is processed in a pipeline manner in parallel to meet the real-time detection needs of the production line.

[0041] The core innovation of the present application is that by comprehensively utilizing the image and spectral data of the front and back of the cloth, reflection and transmission, multi-dimensional key information including post-finishing interference features is extracted, and the feature fusion weight and color difference judgment threshold are dynamically adjusted according to the type of the cloth and the degree of post-finishing interference, thereby effectively distinguishing between real color differences and optical interference caused by cloth structure, light transmission or post-finishing unevenness, improving the accuracy and robustness of cloth defect recognition.

[0042] Specifically, the method first acquires the specified atlas of the front and back of the current detected cloth, which contains reflection images, reflection spectra, transmission images and transmission spectra, providing comprehensive raw data. Then, the image registration technology and time synchronization technology are used to accurately align the multi-source data in space and time, ensuring that the subsequent processing is based on accurately corresponding data. Then, key information is extracted from the aligned multi-source data, which covers various features of images and spectra, especially the features reflecting the relationship between reflection and transmission data and the post-finishing interference features. On this basis, according to the cloth type parameters of the current detected cloth, the fusion weight of each data in the key information is determined, so that the feature fusion process can adapt to the characteristics of different cloths. Subsequently, the extracted key information is weighted and fused according to the determined fusion weight to generate a comprehensive fusion feature vector. Finally, the fusion feature vector is compared with the fusion feature of the normal cloth to determine whether there is deviation, and if there is deviation, the color difference judgment threshold is dynamically adjusted according to the post-finishing interference feature intensity value in the key information, so as to determine and locate the color difference area. The whole process forms a complete defect recognition system that can deal with complex cloth types and interference factors through multi-source data acquisition, accurate alignment, multi-dimensional feature extraction, adaptive fusion based on cloth type and dynamic judgment based on post-finishing interference.

[0043] As a preferred embodiment, the scheme of the present application is implemented as follows: visible light cameras, hyperspectral cameras, and transmission light sources and reflection light sources arranged above and below the fabric can be used to obtain the reflection images, reflection spectra, transmission images, and transmission spectra of the front and back surfaces of the fabric. Spatial alignment can use an image registration algorithm based on phase correlation to determine the deformation parameters by calculating the phase correlation of the images to be registered and the standard images in the frequency domain, and to correct; time synchronization can be achieved by attaching accurate time stamps to each sensor data and matching according to the time stamps when processing. The extraction of key information can be achieved by using image processing algorithms (such as color space conversion, local binary pattern LBP, gray scale statistics) and spectral analysis algorithms (such as spectral normalization, derivative calculation, peak detection), the correlation / difference features of reflection spectra and transmission spectra can be achieved by calculating the correlation coefficient, Euclidean distance or spectral difference between them, and the post-finishing interference features can be extracted by analyzing the spectral anomalies or image gloss changes in a specific wavelength range. Fabric type parameters can be pre-stored in the system and input by the operator at the beginning of detection or automatically identified by the sensor. The determination of the fusion weight can establish a lookup table of fabric type and initial weight, and adjust the initial weight by using a piecewise linear function or a nonlinear function according to the intensity value of the post-finishing interference feature. The generation of the fusion feature vector can use weighted average or weighted series. The comparison step can use the calculation of the Euclidean distance or Mahalanobis distance between the fusion feature vector and the normal sample feature vector, and compare with the set threshold value. When the comparison result exceeds the threshold value, according to the intensity value of the post-finishing interference feature, the final color difference judgment threshold value is dynamically calculated by consulting a preset threshold adjustment curve or formula, and the color difference area is judged and positioned according to the threshold value.

[0044] Through the above scheme, the present application can effectively utilize the image and spectral information of the front and back surfaces of the fabric, reflection and transmission, fully capture the optical characteristics of the fabric; through accurate data alignment, the effective fusion of multi-source data is ensured; by extracting multi-dimensional key information including post-finishing interference features, a basis is provided for distinguishing different types of optical changes; by dynamically adjusting the fusion weight and judgment threshold according to the fabric type and post-finishing interference, the adaptability to different fabric types and complex interference environments is improved, thereby significantly improving the accuracy and robustness of fabric defect recognition, especially when dealing with fabrics with double-sidedness, light transmission or complex post-finishing, the real color difference and optical false image caused by structure or post-finishing can be effectively distinguished, the false positive rate is reduced, and accurate judgment and positioning of defects are realized.

[0045] In some embodiments, the step of spatially aligning the data in the specified atlas comprises:

[0046] According to the reflection image or the transmission image in the specified atlas, a frequency domain feature of the image is extracted by Fourier transform to obtain a frequency domain feature map;

[0047] According to a preset deformation parameter range, a plurality of candidate deformation fields are generated; the deformation parameter range includes a rotation angle range, a scaling ratio range, and a translation amount range;

[0048] For each candidate deformation field, the frequency domain feature map is corrected using the deformation field to obtain a corrected frequency domain feature map, and a cross-correlation coefficient between the corrected frequency domain feature map and a standard frequency domain feature map is calculated;

[0049] The candidate deformation field with the maximum cross-correlation coefficient is selected as the optimal deformation field, and the reflection image and the transmission image in the specified atlas are spatially corrected using the optimal deformation field to obtain multi-source data after spatial alignment.

[0050] Fourier transform is a mathematical tool that can convert an image from a spatial domain to a frequency domain. In the frequency domain, global characteristics of an image, such as periodicity, directionality, edge information, etc., are represented in the form of specific frequency components. Using Fourier transform to extract a frequency domain feature map can convert the deformation (such as translation, rotation, scaling) of an image into corresponding changes in the frequency domain, facilitating subsequent analysis and correction. The preset deformation parameter range defines the boundaries of the types and degrees of deformation that an image may undergo when performing spatial alignment. It limits the possible value ranges of rotation angles, scaling ratios, and translation amounts. Setting this range can reduce the search space and improve alignment efficiency. A deformation field is a mathematical model that describes the transformation of an image from one state to another. According to the preset deformation parameter range, a series of specific deformation parameter combinations can be generated, each corresponding to a candidate deformation field. These candidate deformation fields represent various deformation conditions that an image may experience. The cross-correlation coefficient is a measure of the similarity between two signals or images. Here, it is used to evaluate the matching degree of the frequency domain feature map of the image corrected by a candidate deformation field and a pre-set standard frequency domain feature map. The standard frequency domain feature map can represent the frequency domain features of an image in an ideal, non-deformed state. The larger the cross-correlation coefficient, the more similar the corrected image is to the standard image, and the better the compensation effect of the candidate deformation field on the current image deformation. Among all the generated candidate deformation fields, the one that can make the cross-correlation coefficient between the corrected frequency domain feature map and the standard frequency domain feature map reach the maximum is selected as the optimal deformation field. This optimal deformation field is considered to most accurately describe the actual deformation of the current image to be aligned relative to the standard state.

[0051] The working principle of the method is that the input reflection image or transmission image is first converted to the frequency domain, because certain spatial transformations (such as translation, rotation, scaling) have specific corresponding relationships in the frequency domain, which is convenient for analysis and correction. Next, according to the estimation of the possible deformation of the cloth, a search range of deformation parameters is set, and a series of possible deformation models, i.e. candidate deformation fields, are generated in this range. Then, for each candidate deformation field, it is applied to the frequency domain feature map of the image for correction, simulating the compensation of the influence of the deformation. By calculating the cross-correlation coefficient of the corrected frequency domain feature map and a standard frequency domain feature map, the alignment effect of the deformation field can be quantified. The cross-correlation coefficient, especially the phase correlation based method, has good robustness to the brightness change and complex texture of the image, and is suitable for the alignment of cloth images. By comparing the cross-correlation coefficients of all candidate deformation fields, the optimal deformation field is selected, which is considered to most accurately reflect the actual deformation of the image. Finally, the optimal deformation field is used to perform spatial transformation on the original reflection image and transmission image, realizing accurate alignment. The application of this accurate spatial alignment method to the multi-sensor fusion cloth defect recognition process can ensure that the data from different sensors (such as reflection image sensors and transmission image sensors) and the front and back of the cloth are accurately corresponding in space. This accurate correspondence is the basis for subsequent extraction of key information from multi-source data and weighted fusion, which can effectively avoid false positives or missed detection caused by alignment errors, especially when dealing with cloth with complex texture, easy deformation or light transmission, significantly improving the accuracy and reliability of defect recognition, especially color difference detection.

[0052] As a specific implementation, the Fourier transform can adopt a fast Fourier transform (FFT) algorithm, which can be efficiently implemented on a digital signal processor (DSP) or a graphics processing unit (GPU) to meet the speed requirement of online detection. The preset deformation parameter range can be empirically set according to the type and motion state of the cloth on the actual production line, for example, the rotation angle range can be set to several degrees in positive and negative directions, the scaling ratio range can be set to a certain interval close to 1, and the translation amount range can be set according to the field of view of image acquisition and the cloth offset degree. Generating multiple candidate deformation fields can be achieved by discretely sampling within these parameter ranges. Applying the candidate deformation fields to the correction of the frequency domain feature map can involve corresponding phase adjustment, scaling or rotation operations in the frequency domain. The calculation of the cross-correlation coefficient can utilize a phase correlation technique by calculating the inverse Fourier transform of the normalized cross-power spectrum of the two frequency domain feature maps, and the peak position indicates the translation amount and the peak size reflects the similarity. For rotation and scaling, the correlation calculation can be performed on the frequency domain amplitude spectrum in the logarithmic polar coordinate. The standard frequency domain feature map can be obtained in advance by processing a large number of normal cloth sample images and averaging. The selection of the optimal deformation field is to find the parameter combination corresponding to the maximum cross-correlation coefficient. The spatial correction of the original image using the optimal deformation field can adopt a standard image affine transformation or perspective transformation algorithm to resample the image pixels according to the parameters of the optimal deformation field.

[0053] By adopting the spatial alignment method based on Fourier transform and deformation field search, the rotation, scaling and translation and other deformations of the cloth during high-speed motion can be effectively compensated, and the accurate spatial correspondence between the reflection image and the transmission image is ensured. This method uses the frequency domain features of the image for alignment, which has better robustness to complex textures or patterns on the cloth surface compared to traditional methods that rely on spatial domain feature points, and can find accurate alignment relationship under such challenging conditions. Thus, an accurate spatial basis is provided for subsequent feature extraction and fusion of multi-source data (reflection image, transmission image, etc.), significantly improving the accuracy and reliability of online defect identification of the cloth, especially color difference detection.

[0054] In some embodiments, the step of calculating the cross-correlation coefficient between the corrected frequency domain feature map and the standard frequency domain feature map comprises:

[0055] The phase correlation between the corrected frequency domain feature map and the standard frequency domain feature map is calculated to obtain a cross-correlation coefficient; if it is determined according to a preset texture judgment algorithm that there is a periodic texture on the cloth surface, the cross-correlation coefficient is weighted to reduce the influence of the periodic texture on the cross-correlation coefficient, and the weighted cross-correlation coefficient is taken as the final cross-correlation coefficient.

[0056] Wherein, the phase correlation refers to an image similarity measurement method based on Fourier transform, which determines the relative translation between two images by calculating the cross-correlation between the phase spectrum of the Fourier transform of the two images, which can be achieved by calculating the peak position and intensity of the inverse Fourier transform of the conjugate product of the Fourier transform of the two images.

[0057] Wherein, the preset texture judgment algorithm refers to an algorithm for analyzing image content and identifying whether there is a repeated or periodic structure, which can be achieved by using a method based on frequency domain analysis or based on spatial domain analysis.

[0058] Wherein, the weighting processing of the cross-correlation coefficient refers to the process of modifying or adjusting the calculated cross-correlation coefficient value according to specific conditions or rules, which can be achieved by multiplying a coefficient less than 1, or adjusting the weight distribution according to the intensity or direction information of the periodic texture.

[0059] Based on the above technical features, the working principle of the cross-correlation coefficient calculation method of the present application is as follows: when the data in the specified atlas is spatially aligned, the cross-correlation coefficient between the corrected frequency domain feature map and the standard frequency domain feature map needs to be calculated to evaluate the pros and cons of different candidate deformation fields. On this basis, the present scheme first uses the phase correlation to obtain the preliminary cross-correlation coefficient value, and uses the robustness of phase correlation to image translation, rotation and other deformations to improve the accuracy of preliminary evaluation. Further, considering the possible interference of periodic texture on the frequency domain feature map, after the preliminary cross-correlation coefficient is calculated, the preset texture judgment algorithm is used to detect whether there is a periodic texture on the surface of the cloth. If it is determined that there is a periodic texture, the preliminary calculated cross-correlation coefficient is weighted. This weighting process aims to weaken the excessive influence of the strong peak generated by the periodic texture in the frequency domain on the cross-correlation coefficient, so that the cross-correlation coefficient can more accurately reflect the alignment degree of the non-periodic content (i.e. the deformation of the cloth itself). In this way, even in the presence of periodic texture, the cross-correlation coefficient can more accurately evaluate the spatial correction effect. Finally, the cross-correlation coefficient after weighting processing (if there is a periodic texture) or without processing (if there is no periodic texture) is used as the final cross-correlation coefficient, which is used to select the candidate deformation field with the maximum cross-correlation coefficient as the optimal deformation field. The improved cross-correlation coefficient calculation method is applied to the spatial alignment process, which can more accurately determine the optimal deformation field, thereby improving the accuracy of the spatial alignment of the cloth image data, especially when dealing with cloth with periodic texture, effectively avoiding the misjudgment caused by texture interference, and providing more reliable alignment data for subsequent defect identification.

[0060] In order to more clearly illustrate the technical solutions of the present application, a specific embodiment is described in detail below. In a specific embodiment, the phase correlation between the corrected frequency domain feature map and the standard frequency domain feature map can be calculated by first performing inverse Fourier transform on the two frequency domain feature maps to obtain corresponding spatial domain images. Then, the product of the Fourier transform of one of the images and the conjugate of the Fourier transform of the other image is calculated, and the result is subjected to inverse Fourier transform to obtain a phase correlation plane. The peak position in the phase correlation plane indicates the relative translation of the images, and the peak intensity can be used as the cross-correlation coefficient. At the same time, a preset texture judgment algorithm based on frequency domain analysis can be used to determine whether the surface of the fabric has periodic texture. For example, the power spectrum of the original image can be calculated to analyze whether there are obvious, concentrated, and far-from-origin peaks in the power spectrum. These peaks correspond to the periodic structure in the image. If these peaks are detected, it is determined that the surface of the fabric has periodic texture. If it is determined that there is periodic texture, the cross-correlation coefficient calculated is subjected to weighting processing. For example, the weighting coefficient can be determined according to the intensity of the periodic peak detected in the power spectrum. The stronger the periodic peak, the greater the interference of the periodic texture on the frequency domain feature map, and therefore a smaller weighting coefficient can be used to multiply the preliminary cross-correlation coefficient, or a nonlinear function can be used to adjust the cross-correlation coefficient according to the peak intensity. Through such weighting processing, the influence of periodic texture on the cross-correlation coefficient is reduced, and the weighted cross-correlation coefficient can more accurately reflect the alignment degree of the non-periodic content. Finally, the weighted cross-correlation coefficient is used as the basis for evaluating the spatial correction effect.

[0061] By using the above technical solutions, the present application can achieve the following technical effects: by calculating the phase correlation as the basis for the cross-correlation coefficient, the robustness of the deformation is utilized. Further, by judging the periodic texture and performing weighting processing, the interference of the periodic texture on the calculation of the cross-correlation coefficient is effectively reduced. This makes the cross-correlation coefficient more accurately reflect the alignment degree of the deformation of the fabric itself, and improves the accuracy and robustness of the spatial alignment on the fabric with periodic texture. Accurate spatial alignment provides a reliable basis for subsequent multi-sensor data fusion and defect recognition, which helps to improve the accuracy of defect recognition and reduce false positives and false negatives.

[0062] In some embodiments, if it is determined that the surface of the fabric has periodic texture, the step of weighting the cross-correlation coefficient includes:

[0063] obtaining direction information and period information of the periodic texture;

[0064] determining a frequency component perpendicular to the direction of the periodic texture according to the direction information;

[0065] calculating the energy value of the frequency component according to the period information;

[0066] The weighting coefficient is determined according to the energy value, the greater the energy value, the smaller the weighting coefficient, and the initial cross-correlation coefficient is weighted by the weighting coefficient to obtain the weighted cross-correlation coefficient.

[0067] The direction information and period information of the periodic texture refer to parameters describing the arrangement direction and repetition interval of the periodic structure on the fabric surface, which can be obtained by performing two-dimensional Fourier transform on the image and analyzing the position of the energy concentration area in the frequency domain. The frequency component perpendicular to the direction of the periodic texture refers to the frequency component in the frequency domain that is orthogonal to the main direction of the periodic texture, which usually corresponds to the change of the texture in the vertical direction, and can be selected in the frequency domain through coordinate transformation or sector filtering according to the obtained texture direction. The energy value of the frequency component refers to the sum of the intensity or amplitude square of the image in the frequency domain within a certain frequency range, which reflects the strength of the corresponding frequency component, and can be calculated by squaring and summing the frequency domain coefficients corresponding to the selected frequency component. The weighting coefficient refers to a multiplicative factor used to adjust the initial cross-correlation coefficient value, and the size of the factor is inversely related to the calculated frequency component energy value, which can be determined according to the energy value through a preset mapping function or lookup table. The initial cross-correlation coefficient refers to the cross-correlation coefficient value calculated by phase correlation before the periodic texture weighting process. The weighted cross-correlation coefficient refers to the final cross-correlation coefficient value obtained by multiplying the initial cross-correlation coefficient by the determined weighting coefficient.

[0068] The scheme of the present application is based on the in-depth analysis of the frequency domain characteristics of periodic textures. By obtaining the direction and period information of periodic textures, the most relevant frequency components in the frequency domain to the texture interference can be accurately locked. These frequency components are perpendicular to the texture direction, and their frequencies are closely related to the texture period, which are the key areas of texture energy concentration. By calculating the energy values of these specific frequency components, the strength of periodic textures can be quantified. The higher the energy value, the more significant the periodic texture, and the greater the interference with the cross-correlation calculation. Therefore, according to the energy value, the weighting coefficient is determined, and the greater the energy value, the smaller the weighting coefficient, which ensures that the cross-correlation coefficients affected by strong periodic textures are more suppressed. Using the weighting coefficient thus determined to weight the initial cross-correlation coefficients can specifically weaken the weight of periodic textures in the cross-correlation calculation, so that the weighted cross-correlation coefficients can more accurately reflect the alignment degree of the non-periodic characteristics of the image itself. This is different from the method of uniform weighting, which cannot distinguish the characteristics and strength of the texture and may over or under suppress the interference. Through this adaptive weighting based on the characteristics and strength of the texture, the present scheme can more effectively reduce the influence of periodic textures on the accuracy of spatial alignment, thereby improving the reliability of the selection of the optimal deformation field and the final spatial correction accuracy. The scheme is combined with the basic method of extracting frequency domain features by Fourier transform, calculating cross-correlation coefficients, and performing spatial alignment, which can significantly improve the overall spatial alignment performance in the complex cloth image processing scene with periodic textures.

[0069] In one embodiment, obtaining the direction information and period information of periodic textures can use two-dimensional Fourier transform on the input cloth image to obtain its power spectrum. In the power spectrum, periodic textures will appear as discrete energy peaks away from the origin. By detecting the positions of these energy peaks, the main direction and period of the texture can be determined. For example, the angle of the peak value relative to the origin corresponds to the texture direction, and the inverse of the distance of the peak value to the origin is related to the texture period. According to the obtained texture direction information, the frequency range perpendicular to the direction can be determined. For example, if the texture direction is close to horizontal, the frequency components close to the vertical direction in the frequency domain are selected. According to the period information, specific frequency points or frequency intervals corresponding to the texture period can be determined in the selected vertical direction frequency range. The energy value of the frequency component can be the integral or sum of the amplitude square of the corresponding frequency domain coefficients in these specific frequency points or frequency intervals. According to the calculated energy value, an inverse proportional function can be used to determine the weighting coefficient, for example, the weighting coefficient can be set to a certain constant divided by (1 plus a proportional factor multiplied by the energy value), ensuring that the greater the energy value, the smaller the weighting coefficient. Finally, the calculated weighting coefficient is multiplied by the initial cross-correlation coefficient to obtain the weighted cross-correlation coefficient, which is used for subsequent optimal deformation field selection.

[0070] By the above method, the present application can perform targeted weighting processing on the cross-correlation coefficient when there is a periodic texture on the surface of the cloth. Based on the quantitative analysis of the direction, period and energy of the periodic texture, this processing method can more accurately identify and suppress the interference of the texture on the spatial alignment calculation. This makes it possible to rely more on the non-periodic features in the image that reflect the true structure or defects when selecting the optimal deformation field, thereby improving the accuracy and robustness of spatial alignment, and is particularly suitable for high-precision online detection of cloth with complex periodic texture.

[0071] In some embodiments, the step of determining the fusion weight of each data in the key information according to the cloth type parameter of the currently detected cloth comprises:

[0072] According to the cloth type parameter of the currently detected cloth, the corresponding initial fusion weight is found from a pre-established weight lookup table; the cloth type parameter includes fiber type, yarn density, weaving method, finishing process; the weight lookup table contains the mapping relationship between cloth type parameter, key information data type and fusion weight;

[0073] According to the intensity value of the finishing interference feature in the key information, a piecewise function is used to dynamically adjust the initial fusion weight, and the adjusted fusion weight is determined as the final fusion weight of each data in the key information.

[0074] The cloth type parameter refers to technical indicators that describe the inherent properties and processing state of the cloth, which can be represented by specific classification information such as fiber type, yarn density, weaving method, finishing process, etc. The weight lookup table refers to a data structure that stores the mapping relationship between cloth type parameters and key information data types and corresponding initial fusion weights, which can be implemented using a database, configuration file or in-memory lookup table. The intensity value of the finishing interference feature in the key information refers to a numerical value that quantifies the degree of uneven influence of the finishing process on the cloth, which can be represented by a numerical value obtained by analyzing and calculating the extracted finishing interference feature. The piecewise function refers to a mathematical function with different expressions in different input intervals, which can be implemented using a series of conditional judgment statements or lookup table methods.

[0075] The scheme improves the step of determining the fusion weight of key information according to the cloth type parameter, aiming to solve the problem that the fixed weight set by the cloth type parameter alone cannot fully adapt to the dynamic changes of optical characteristics introduced by the cloth finishing process in the production process under the environment of various cloth types and frequent switching on the high-speed cloth production line, resulting in poor fusion effect and affecting the accuracy of color difference detection. The scheme dynamically determines the fusion weight by combining the cloth type parameter and the intensity value of the finishing interference feature, realizes fine adjustment of the initial weight, can effectively adapt to the dynamic changes of optical characteristics introduced by the cloth finishing process in the production process, improve the fusion effect, and thus improve the accuracy of color difference detection. Specifically, first, according to the cloth type parameter of the current detected cloth, the corresponding initial fusion weight is found from the pre-established weight lookup table. The cloth type parameter, such as fiber type, yarn density, weaving method, and finishing process, reflects the inherent characteristics of the cloth, which determines the relative importance of different types of key information (such as image features, spectral features, and reflection / transmission features) in identifying specific defects. The weight lookup table stores the mapping relationship between the cloth type parameter and the key information data type and the fusion weight, so that the system can quickly obtain a preliminary weight configuration suitable for the type of cloth based on the prior knowledge of the cloth, providing a basis for subsequent dynamic adjustment. Secondly, and also the key improvement point of the scheme, the initial fusion weight is dynamically adjusted using a piecewise function according to the intensity value of the finishing interference feature in the key information. The unevenness of the finishing process is an important interference source that affects the optical characteristics of the cloth and may lead to misjudgment. By extracting and quantifying the intensity value of the finishing interference feature, the system can real-time perceive the degree of interference of the current detection area. Dynamic adjustment using a piecewise function means that the weight adjustment strategy will change according to different intervals of the interference intensity, so that the fusion weight can adaptively respond to the actual interference situation. For example, when the finishing interference is strong, the weight of the interference feature can be reduced, and the weight of the feature that reflects the interference itself or is not easily interfered can be increased, so that the influence of the finishing interference can be effectively suppressed during feature fusion, and the fusion feature vector can more accurately reflect the true defect information. Finally, the fusion weight after dynamic adjustment is determined as the final fusion weight of each data in the key information. This final weight will be used in the subsequent weighted fusion step to generate a fusion feature vector that is more robust and accurate, thereby improving the accuracy and reliability of cloth defect recognition, especially in the face of complex cloth types and finishing interference.By combining the prior information provided by the cloth type parameters and the real-time information provided by the post-finishing interference feature strength, and using a piecewise function for fine adjustment, the scheme can generate a fusion weight that is more in line with the actual state of the cloth, thereby more effectively integrating multi-source key information in the subsequent weighted fusion step, highlighting features related to real defects and suppressing interference caused by uneven post-finishing, ultimately improving the discrimination ability of the fusion feature vector, and thus improving the accuracy of overall defect recognition.

[0076] In one embodiment, according to the currently detected cloth, its cloth type parameters can be determined as "cotton, plain, waterproof treatment". The system queries a pre-constructed weight lookup table according to these parameters. The lookup table can be a structure stored in a file or database, which records the initial fusion weights of various key information such as color space components, local texture features, spectral curve slopes, and post-finishing interference features under different combinations of cloth type parameters. For example, for a cloth of "cotton, plain, waterproof treatment" type, the lookup table may specify that the initial weight of color space components is 0.4, the local texture features is 0.3, the post-finishing interference features is 0.2, and the remaining weight is allocated to other features. After extracting the key information, the system obtains the intensity value of the post-finishing interference feature of the current detection area, which can be a value between 0 and 1, representing the degree of interference. The system then calls a piecewise function, which takes the initial fusion weight and the post-finishing interference feature strength value as input. The piecewise function can define multiple intensity intervals and set different weight adjustment rules for each interval. For example, when the intensity value is low, the initial weight is fine-tuned; when the intensity value is high, the weight of the post-finishing interference feature is significantly increased, while the weight of other features susceptible to interference is reduced. After the calculation of the piecewise function, a set of adjusted fusion weights is obtained, for example, if the post-finishing interference intensity value is high, the weight of the post-finishing interference feature may be adjusted to 0.5, while the weights of color space components and local texture features are correspondingly reduced. Finally, this set of adjusted weights is determined as the final weight for weighted fusion.

[0077] By obtaining the initial fusion weight according to the cloth type parameters and dynamically adjusting it using a piecewise function combined with the intensity value of the post-finishing interference feature, the scheme can make the fusion weight of each data in the key information more accurately reflect the actual state of the current cloth and the degree of interference. This overcomes the limitation of relying solely on fixed weights that cannot adapt to dynamic changes in cloth post-finishing processes, improves the effectiveness of key information weighted fusion, thereby enhancing the representation ability of the fusion feature vector for real defects, and thus improving the accuracy and robustness of cloth color difference detection.

[0078] In some embodiments, the step of dynamically adjusting the initial fusion weight according to the intensity value of the post-finishing interference feature in the key information includes:

[0079] If the intensity value is greater than a preset first threshold value, the initial fusion weight corresponding to the post-finishing interference feature is adjusted to a maximum value, and the initial fusion weights corresponding to the color space component, the local texture feature, and the local brightness feature are correspondingly reduced;

[0080] If the intensity value is less than a preset second threshold value, the initial fusion weight corresponding to the post-finishing interference feature is adjusted to a minimum value, and the initial fusion weights corresponding to the color space component, the local texture feature, and the local brightness feature are correspondingly increased; the second threshold value is less than the first threshold value;

[0081] If the intensity value is between the first threshold value and the second threshold value, the initial fusion weights corresponding to the post-finishing interference feature and the color space component, the local texture feature, and the local brightness feature are linearly adjusted according to the relative distance between the intensity value and the first threshold value and the relative distance between the intensity value and the second threshold value.

[0082] The intensity value refers to a quantitative representation of the post-finishing interference feature extracted from the multi-source data, reflecting the influence degree of the post-finishing unevenness on the optical characteristics of the fabric. It can be obtained by analyzing the amplitude, energy, or difference degree of the post-finishing interference feature compared with the corresponding feature of the normal fabric. The preset first threshold value and the preset second threshold value are boundary values for dividing the post-finishing interference intensity interval, used to guide the weight adjustment strategy under different intensities, which can be determined according to the statistical analysis of the post-finishing interference feature intensity distribution of a large number of fabric samples or expert experience. The maximum value and the minimum value refer to the upper limit and the lower limit that the initial fusion weights corresponding to the post-finishing interference feature and the color space component, the local texture feature, and the local brightness feature can reach after adjustment, which can be set according to the theoretical importance of different features to the final defect judgment and the actual test effect. The relative distance refers to the difference or proportional relationship between the currently detected post-finishing interference feature intensity value and the first threshold value or the second threshold value, used for linear interpolation calculation when the intensity value is between the two threshold values, which can be obtained by simple difference calculation or normalized difference calculation. Linear adjustment refers to using linear interpolation or other linear function relationships to calculate the adjusted fusion weight when the intensity value is between the first threshold value and the second threshold value, so that the weight changes smoothly with the intensity value, which can be realized by linear interpolation calculation based on the position of the intensity value in the [second threshold value, first threshold value] interval.

[0083] According to the intensity value of the post-finishing interference feature, the initial fusion weight is dynamically adjusted in a piecewise function manner, so as to realize adaptive processing of different degrees of post-finishing interference. Specifically, when the detected post-finishing interference feature intensity value is high (greater than a preset first threshold value), it indicates that the post-finishing interference has a greater impact on the optical properties of the cloth. At this time, the weight of the post-finishing interference feature itself is adjusted to the maximum value, and the weights of the color space component, local texture feature and local brightness feature affected by the post-finishing interference are correspondingly reduced. This is done to focus more on identifying and quantifying the interference itself in a strong interference environment, and to reduce the weight of other features affected by the interference, so as to avoid misjudgment caused by feature distortion due to interference. On the contrary, when the post-finishing interference feature intensity value is low (less than a preset second threshold value), it indicates that the post-finishing interference has a smaller impact on the optical properties of the cloth. At this time, the weight of the post-finishing interference feature is adjusted to the minimum value, and the weights of the color space component, local texture feature and local brightness feature are correspondingly increased. This is done to mainly rely on the color, texture and brightness features that can better reflect the properties of the cloth itself for judgment in a weak interference environment, so as to reduce the noise influence that may be introduced by the weak post-finishing interference feature. When the post-finishing interference feature intensity value is between the first threshold value and the second threshold value, it indicates that the influence degree of the post-finishing interference is moderate. At this time, according to the relative distance of the intensity value and the two threshold values, the weights of the post-finishing interference feature and the color space component, local texture feature and local brightness feature are smoothly adjusted in a linear interpolation manner. This linear adjustment makes the weight change more continuous and fine, and can more accurately reflect the relative importance of each feature under different intensity interference, avoiding the sudden change of the weight near the threshold value. This strategy of dynamically adjusting the weight according to the post-finishing interference intensity is a further optimization based on the determination of the initial weight according to the cloth type, so that the weight not only considers the inherent properties of the cloth, but also considers the actual detected local interference degree. In this way, the system can intelligently adjust the fusion weight of different key information according to the actual detected post-finishing interference intensity, so that the finally generated fusion feature vector can more effectively strip the interference caused by post-finishing unevenness, and more accurately reflect the real defect information of the cloth.

[0084] For example, in one embodiment, the preset first threshold value can be set as 0.8, and the preset second threshold value can be set as 0.2. The maximum value of the post-finishing interference feature weight can be set as 1.0, and the minimum value can be set as 0.0. The sum of the initial weights of the color space component, the local texture feature, and the local brightness feature can be set as W. When the detected post-finishing interference feature intensity value is 0.9, since it is greater than the first threshold value 0.8, the system can adjust the weight of the post-finishing interference feature to 1.0, and correspondingly reduce the weights of the color space component, the local texture feature, and the local brightness feature. When the detected intensity value is 0.1, since it is less than the second threshold value 0.2, the system can adjust the weight of the post-finishing interference feature to 0.0, and correspondingly increase the weights of the color space component, the local texture feature, and the local brightness feature. When the detected intensity value is 0.5, since it is between 0.2 and 0.8, the system can perform linear interpolation according to the relative distance between the intensity value 0.5 and the threshold values 0.8 and 0.2. For example, the weight of the post-finishing interference feature can be adjusted to 0.0+(1.0-0.0)*(0.5-0.2) / (0.8-0.2)=0.5. Then the remaining weight (W-0.5) is proportionally distributed to the color space component, the local texture feature, and the local brightness feature. Through this specific numerical setting and adjustment logic, fine-grained weight control of different intensity post-finishing interference can be achieved.

[0085] By dynamically adjusting the initial fusion weight in a piecewise function manner according to the intensity value of the post-finishing interference feature, the scheme can intelligently adjust the fusion weight of different key information according to the actually detected post-finishing interference intensity, so that the finally generated fusion feature vector can more effectively strip the interference caused by post-finishing unevenness and more accurately reflect the real defect information of the fabric, especially improving the accuracy and robustness of color difference judgment.

[0086] Reference is made to the accompanying drawings Figure 2 The present application provides a multi-sensor fusion fabric online defect identification system, comprising:

[0087] The acquisition module 100 is used for acquiring the specified atlas of the front and back of the current detection fabric;

[0088] The alignment module 200 is used for spatially and temporally aligning the data in the specified atlas to obtain the aligned multi-source data;

[0089] The extraction module 300 is used for extracting key information from the multi-source data;

[0090] The determination module 400 is used for determining the fusion weight of each data in the key information according to the fabric type parameter of the current detection fabric;

[0091] The generating module 500 is configured to generate a fusion feature vector by weighting and fusing the data in the key information according to the fusion weight;

[0092] The comparing module 600 is configured to compare the fusion feature vector with the fusion feature of the normal cloth, and if the comparison result is deviation, determine the color difference region and position by adjusting the color difference judgment threshold according to the intensity value of the finishing interference feature in the key information.

[0093] In this article, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations.

[0094] The above only describes the embodiments of the present application and is not used to limit the protection scope of the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A multi-sensor fusion cloth online defect identification method, characterized in that, The method comprises the following steps: obtaining a specified atlas of the front and back of the current detected cloth; spatially and temporally aligning the data in the specified atlas to obtain multi-source data after alignment; extracting key information from the multi-source data after alignment; determining the fusion weight of each data in the key information according to the cloth type parameter of the current detected cloth; performing weighted fusion on the data in the key information according to the fusion weight to generate a fusion feature vector; comparing the fusion feature vector with the fusion feature of normal cloth, and if the comparison result is a deviation, determining a color difference region and positioning by adjusting the color difference judgment threshold according to the intensity value of the post-finishing interference feature in the key information; the step of spatially aligning the data in the specified atlas comprises: extracting the frequency domain features of the images from the reflection images or transmission images in the specified atlas by Fourier transform to obtain a frequency domain feature map; generating a plurality of candidate deformation fields according to a preset deformation parameter range; for each candidate deformation field, correcting the frequency domain feature map using the deformation field to obtain a corrected frequency domain feature map, and calculating the cross-correlation coefficient between the corrected frequency domain feature map and a standard frequency domain feature map; selecting the candidate deformation field with the largest cross-correlation coefficient as the optimal deformation field, and using the optimal deformation field to spatially correct the reflection images and transmission images in the specified atlas to obtain multi-source data after spatial alignment.

2. The multi-sensor fusion based on-line fabric defect identification method according to claim 1, wherein, The specified atlas comprises reflection images, reflection spectra, transmission images and transmission spectra.

3. The multi-sensor fusion based on-line fabric defect identification method according to claim 1, wherein, The deformation parameter range comprises a rotation angle range, a scaling ratio range and a translation amount range.

4. The multi-sensor fusion based on-line fabric defect identification method according to claim 1, wherein, The step of calculating the cross-correlation coefficient between the corrected frequency domain feature map and the standard frequency domain feature map comprises: calculating the phase correlation between the corrected frequency domain feature map and the standard frequency domain feature map to obtain a cross-correlation coefficient; if it is determined that the cloth surface has a periodic texture, the cross-correlation coefficient is weighted, and the weighted cross-correlation coefficient is used as the final cross-correlation coefficient.

5. The multi-sensor fusion based on-line fabric defect identification method according to claim 4, wherein, If it is determined that the cloth surface has a periodic texture, the step of weighting the cross-correlation coefficient comprises: obtaining the direction information and period information of the periodic texture; determining the frequency component perpendicular to the direction of the periodic texture according to the direction information; calculating the energy value of the frequency component according to the period information; determining a weighting coefficient according to the energy value, and weighting the initial cross-correlation coefficient using the weighting coefficient to obtain a weighted cross-correlation coefficient.

6. The multi-sensor fusion based on-line fabric defect identification method according to claim 2, wherein, The key information comprises color space components, local texture features, local brightness features, wavelength intensity ratios, spectral curve slopes, spectral curve curvatures, absorption peak positions, absorption peak intensities, correlation features of reflection spectra and transmission spectra, difference features of reflection spectra and transmission spectra, and post-finishing interference features.

7. The multi-sensor fusion based on-line fabric defect identification method according to claim 6, wherein, The step of determining the fusion weight of each data in the key information according to the cloth type parameter of the current detected cloth comprises: determining the initial fusion weight corresponding to the cloth type parameter of the current detected cloth from a pre-established weight lookup table; According to the intensity value of the post-finishing interference feature in the key information, the initial fusion weight is dynamically adjusted, and the adjusted fusion weight is determined as the final fusion weight of each data in the key information.

8. The multi-sensor fusion based on-line fabric defect identification method according to claim 7, wherein, The step of dynamically adjusting the initial fusion weight according to the intensity value of the post-finishing interference feature in the key information comprises: If the intensity value is greater than a preset first threshold value, the initial fusion weight corresponding to the post-finishing interference feature is adjusted to a maximum value, and the initial fusion weights corresponding to the color space component, the local texture feature and the local brightness feature are correspondingly reduced; If the intensity value is less than a preset second threshold value, the initial fusion weight corresponding to the post-finishing interference feature is adjusted to a minimum value, and the initial fusion weights corresponding to the color space component, the local texture feature and the local brightness feature are correspondingly increased; the second threshold value is less than the first threshold value; If the intensity value is between the first threshold value and the second threshold value, the initial fusion weights of the post-finishing interference feature and the color space component, the local texture feature and the local brightness feature are linearly adjusted according to the relative distance of the intensity value to the first threshold value and the relative distance of the intensity value to the second threshold value.

9. A multi-sensor fusion fabric online defect recognition system employing the multi-sensor fusion fabric online defect recognition method according to any one of claims 1 to 8, characterized in that, Comprise: An acquisition module is configured to acquire a specified atlas of a front and back surface of a current detection fabric; An alignment module is configured to perform spatial alignment and time alignment on data in the specified atlas to obtain multi-source data after alignment; An extraction module is configured to extract key information from the multi-source data after alignment; A determination module is configured to determine fusion weights of each data in the key information according to a fabric type parameter of the current detection fabric; A generation module is configured to perform weighted fusion on data in the key information according to the fusion weights to generate a fusion feature vector; A comparison module is configured to compare the fusion feature vector with a fusion feature of a normal fabric, and if the comparison result is a deviation, adjust a color difference judgment threshold according to an intensity value of a post-finishing interference feature in the key information to determine a color difference region and positioning.

Images