Method and system for detecting defects of super-soft knitted gauze based on image recognition

Abstract

The invention relates to the technical field of image recognition, in particular to a super-soft knitted gauze defect detection method and system based on image recognition, and aims at texture periodic interference and micro-defect sub-pixel level size features caused by a super-soft gauze honeycomb-shaped woven structure. By constructing an optical enhancement-feature decoupling-cross-modal perception three-level cooperative detection framework, accurate mapping of defect features from a physical space to a feature space is realized, so that the problem that low-contrast micro-defects are difficult to perceive and detect under a complex texture background is solved.

Description

Technical Field

[0001] The present application relates to the field of image recognition technology, and more specifically, to a method and system for detecting defects in ultra-soft knitted gauze fabric based on image recognition. Background Art

[0002] Ultra-soft knitted gauze fabric is a special textile material made of ultra-fine denier fibers (monofilament fineness ≤ 0.8 dtex) through warp knitting or weft knitting processes. Its unique "honeycomb" stitch structure and low gram weight characteristics (usually < 30 g / m 2 ) endow the fabric with excellent breathability, drapability and skin-friendly touch, and are widely used in high-end medical dressings, infant clothing, precision instrument packaging and other fields. In the medical scenario, micro-defects of 0.1 mm level (such as broken yarns, snagging) may lead to bacterial penetration; in the field of optical instrument packaging, yarn knot defects with a diameter < 0.3 mm may cause scratches on the mirror surface, which poses a detection requirement of sub-millimeter level accuracy for the defect detection system.

[0003] However, traditional defect detection schemes, such as the gray histogram segmentation method, have a high false detection rate in the detection of ultra-soft gauze because there is a significant overlapping area between the gray standard deviation of the normal texture and the micro-defects. At the same time, due to the characteristics of ultra-soft knitted gauze fabric, the defects themselves may be relatively small and have a low contrast with the normal texture, which increases the detection difficulty. Although the fine defects may not be obvious and are easily overlooked or submerged in the background texture, they still need to be detected in some application scenarios with high quality requirements. In addition, although the traditional frequency domain filtering method (such as the Gabor filter bank) can extract the gauze base texture, it cannot distinguish the structural defects caused by stitch offset from the normal texture fluctuations, and it is difficult to meet the actual requirements of detection accuracy when the production line speed is high.

[0004] Therefore, an optimized defect detection scheme for ultra-soft knitted gauze fabric is desired. Summary of the Invention

[0005] The present application provides a method and system for detecting defects in ultra-soft knitted gauze fabric based on image recognition. It can address the periodic interference of the texture caused by the honeycomb weaving structure of ultra-soft gauze and the sub-pixel size characteristics of micro-defects. By constructing a three-level collaborative detection framework of "optical enhancement - feature decoupling - cross-modal perception", it realizes the accurate mapping of defect features from the physical space to the feature space, thus breaking through the problem that low-contrast micro-defects are difficult to perceive and detect in a complex texture background. According to one aspect of the present application, a method for detecting defects in ultra-soft knitted gauze fabric based on image recognition is provided, including: obtaining an image of ultra-soft knitted gauze fabric collected by an image acquisition system;

[0006] Performing adaptive image preprocessing and contrast enhancement on the image of ultra-soft knitted gauze fabric to obtain an enhanced image of ultra-soft knitted gauze fabric;

[0007] Perform multi-scale feature analysis on the enhanced image of the ultra-soft knitted gauze fabric to obtain the texture features and depth features of the ultra-soft knitted gauze fabric;

[0008] Perform feature alignment and joint perception on the texture features and depth features of the ultra-soft knitted gauze fabric to obtain the multi-scale joint perception coding features of the ultra-soft knitted gauze fabric features, including: decouple the texture feature coding map and depth feature coding map of the ultra-soft knitted gauze fabric to obtain a set of texture feature coding matrices and a set of depth local feature coding matrices of the ultra-soft knitted gauze fabric; perform phase alignment and attention significant fusion on the set of texture feature coding matrices and the set of depth local feature coding matrices of the ultra-soft knitted gauze fabric to obtain the multi-scale joint perception coding features of the ultra-soft knitted gauze fabric features;

[0009] Based on the multi-scale joint perception coding features of the ultra-soft knitted gauze fabric features, determine whether there are defects in the ultra-soft knitted gauze fabric to be detected.

[0010] In the above method for detecting defects in ultra-soft knitted gauze fabric based on image recognition, the image acquisition system includes a high-resolution industrial camera and an illumination system.

[0011] In the above method for detecting defects in ultra-soft knitted gauze fabric based on image recognition, the illumination system uses a uniform diffuse light source. In the above method for detecting defects in ultra-soft knitted gauze fabric based on image recognition, perform adaptive image preprocessing and contrast enhancement on the ultra-soft knitted gauze fabric image to obtain the enhanced image of the ultra-soft knitted gauze fabric, including:

[0012] Perform adaptive Wiener filtering on the ultra-soft knitted gauze fabric image to obtain the adaptive preprocessed image of the ultra-soft knitted gauze fabric;

[0013] Perform multi-scale contrast enhancement on the adaptive preprocessed image of the ultra-soft knitted gauze fabric to obtain the enhanced image of the ultra-soft knitted gauze fabric.

[0014] In the above method for detecting defects in ultra-soft knitted gauze fabric based on image recognition, perform multi-scale feature analysis on the enhanced image of the ultra-soft knitted gauze fabric to obtain the texture features and depth features of the ultra-soft knitted gauze fabric, including: use a multi-scale convolutional neural network to perform multi-scale image feature extraction on the enhanced image of the ultra-soft knitted gauze fabric to obtain the texture feature coding map of the ultra-soft knitted gauze fabric as the texture features of the ultra-soft knitted gauze fabric and the depth feature coding map of the ultra-soft knitted gauze fabric as the depth features of the ultra-soft knitted gauze fabric.

[0015] In the above method for detecting defects in ultra-soft knitted gauze fabric based on image recognition, phase alignment and attention-based significant fusion are performed on the set of texture feature encoding matrices of the ultra-soft knitted gauze fabric and the set of depth local feature encoding matrices of the ultra-soft knitted gauze fabric to obtain the multi-scale joint perception encoding features of the ultra-soft knitted gauze fabric features, including:

[0016] Based on the feature phase alignment degree between any two texture feature encoding matrices and depth local feature encoding matrices in the set of texture feature encoding matrices of the ultra-soft knitted gauze fabric and the set of depth local feature encoding matrices of the ultra-soft knitted gauze fabric, feature phase dynamic search alignment is performed on the set of texture feature encoding matrices of the ultra-soft knitted gauze fabric and the set of depth local feature encoding matrices of the ultra-soft knitted gauze fabric to obtain a set of phase-aligned {texture feature encoding matrix of ultra-soft knitted gauze fabric, depth local feature encoding matrix of ultra-soft knitted gauze fabric} feature pairs; each phase-aligned {texture feature encoding matrix of ultra-soft knitted gauze fabric, depth local feature encoding matrix of ultra-soft knitted gauze fabric} feature pair in the set of phase-aligned {texture feature encoding matrix of ultra-soft knitted gauze fabric, depth local feature encoding matrix of ultra-soft knitted gauze fabric} feature pairs is input into the feature joint perception attention network to obtain a set of joint perception attention weights of the ultra-soft knitted gauze fabric features;

[0017] Based on the set of joint perception attention weights of the ultra-soft knitted gauze fabric features, attention-driven significant aggregation is performed on the set of phase-aligned {texture feature encoding matrix of ultra-soft knitted gauze fabric, depth local feature encoding matrix of ultra-soft knitted gauze fabric} feature pairs to obtain the multi-scale joint perception encoding map of the ultra-soft knitted gauze fabric features as the multi-scale joint perception encoding features of the ultra-soft knitted gauze fabric features.

[0018] In the above method for detecting defects in ultra-soft knitted gauze fabric based on image recognition, each phase-aligned {texture feature encoding matrix of ultra-soft knitted gauze fabric, depth local feature encoding matrix of ultra-soft knitted gauze fabric} feature pair in the set of phase-aligned {texture feature encoding matrix of ultra-soft knitted gauze fabric, depth local feature encoding matrix of ultra-soft knitted gauze fabric} feature pairs is input into the feature joint perception attention network to obtain a set of joint perception attention weights of the ultra-soft knitted gauze fabric features, including:

[0019] Calculate the correlation matrix between each phase-aligned {texture feature encoding matrix of ultra-soft knitted gauze fabric, depth local feature encoding matrix of ultra-soft knitted gauze fabric} feature pair in the set of phase-aligned {texture feature encoding matrix of ultra-soft knitted gauze fabric, depth local feature encoding matrix of ultra-soft knitted gauze fabric} feature pairs to obtain a set of surface-to-scale correlation matrices of the textile to be inspected;

[0020] Perform trace metric on each pair of surface-to-scale correlation matrices in the set of surface-to-scale correlation matrices of the textile to be inspected to obtain a set of trace metric values of the surface of the textile to be inspected;

[0021] Based on the set of trace metric values of the surface of the textile to be inspected, obtain a set of jointly perceived attention weights for the characteristics of the ultra-soft knitted gauze fabric.

[0022] In the above method for defect detection of ultra-soft knitted gauze fabric based on image recognition, obtaining a set of jointly perceived attention weights for the characteristics of the ultra-soft knitted gauze fabric based on the set of trace metric values of the surface of the textile to be inspected includes:

[0023] Based on the set of trace metric values of the surface of the textile to be inspected, perform matrix manifold optimization based on the number of biconnectivity on each pair of phase-aligned {ultra-soft knitted gauze fabric texture feature encoding matrix, ultra-soft knitted gauze fabric depth local feature encoding matrix} feature pairs in the set of phase-aligned {ultra-soft knitted gauze fabric texture feature encoding matrix, ultra-soft knitted gauze fabric depth local feature encoding matrix} feature pairs to obtain a set of phase-optimized aligned {ultra-soft knitted gauze fabric texture feature encoding matrix, ultra-soft knitted gauze fabric depth local feature encoding matrix} feature pairs;

[0024] Based on the set of phase-optimized aligned {ultra-soft knitted gauze fabric texture feature encoding matrix, ultra-soft knitted gauze fabric depth local feature encoding matrix} feature pairs, obtain a set of jointly perceived attention weights for the characteristics of the ultra-soft knitted gauze fabric.

[0025] According to another aspect of the present application, there is provided a defect detection system for ultra-soft knitted gauze fabric based on image recognition, including:

[0026] An ultra-soft knitted gauze fabric image acquisition module for acquiring an image of an ultra-soft knitted gauze fabric collected by an image acquisition system; an ultra-soft knitted gauze fabric image enhancement module for performing adaptive image preprocessing and contrast enhancement on the ultra-soft knitted gauze fabric image to obtain an enhanced image of the ultra-soft knitted gauze fabric;

[0027] An image multi-scale feature analysis module for performing multi-scale feature analysis on the enhanced image of the ultra-soft knitted gauze fabric to obtain the texture features and depth features of the ultra-soft knitted gauze fabric;

[0028] The image feature alignment and joint perception module is used to perform feature alignment and joint perception on the texture features and depth features of the ultra-soft knitted gauze fabric to obtain the multi-scale joint perception encoded features of the ultra-soft knitted gauze fabric features. Among them, the image feature alignment and joint perception module includes: an image feature decoupling module, which is used to decouple the texture feature encoded map and depth feature encoded map of the ultra-soft knitted gauze fabric to obtain a set of texture feature encoded matrices of the ultra-soft knitted gauze fabric and a set of depth local feature encoded matrices of the ultra-soft knitted gauze fabric; an image feature fusion and encoding module, which is used to perform phase alignment and attention significant fusion on the set of texture feature encoded matrices of the ultra-soft knitted gauze fabric and the set of depth local feature encoded matrices of the ultra-soft knitted gauze fabric to obtain the multi-scale joint perception encoded features of the ultra-soft knitted gauze fabric features;

[0029] The ultra-soft knitted gauze fabric detection result determination module is used to determine whether there are defects in the ultra-soft knitted gauze fabric to be detected based on the multi-scale joint perception encoded features of the ultra-soft knitted gauze fabric features.

[0030] In the above-mentioned ultra-soft knitted gauze fabric defect detection system based on image recognition, the image acquisition system includes a high-resolution industrial camera and an illumination system.

[0031] A method and system for detecting defects in ultra-soft knitted gauze fabric based on image recognition provided by this application can, aiming at the periodic interference of texture caused by the honeycomb weaving structure of ultra-soft gauze and the sub-pixel size characteristics of micro-defects, realize the accurate mapping of defect features from the physical space to the feature space by constructing a three-level collaborative detection framework of "optical enhancement - feature decoupling - cross-modal perception", thus breaking through the problem of difficult perception and detection of low-contrast micro-defects in complex texture backgrounds. Description of the Drawings

[0032] In order to more clearly illustrate the technical solutions of the embodiments of this application, the drawings of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below only relate to some embodiments of this application and do not limit this application.

[0033] Figure 1 It is a schematic flow chart of the method for detecting defects in ultra-soft knitted gauze fabric based on image recognition according to the embodiment of this application.

[0034] Figure 2 It is a schematic diagram of data flow of the method for detecting defects in ultra-soft knitted gauze fabric based on image recognition according to the embodiment of this application. Figure 3 It is a schematic flow chart of S2 in the method for detecting defects in ultra-soft knitted gauze fabric based on image recognition according to the embodiment of this application.

[0035] Figure 4Schematic flowchart of S4 in the method for detecting defects in ultra-soft knitted gauze fabric based on image recognition according to an embodiment of the present application.

[0036] Figure 5 Schematic flowchart of S42 in the method for detecting defects in ultra-soft knitted gauze fabric based on image recognition according to an embodiment of the present application.

[0037] Figure 6 Schematic block diagram of the system for detecting defects in ultra-soft knitted gauze fabric based on image recognition according to an embodiment of the present application. Detailed implementation manners

[0038] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only partial embodiments of the present application, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts also belong to the scope of protection of the present application.

[0039] In response to the above technical problems, in the technical solution of the present application, a method for detecting defects in ultra-soft knitted gauze fabric based on image recognition is proposed. It can address the periodic interference of textures caused by the honeycomb weaving structure of ultra-soft gauze and the sub-pixel size characteristics of micro-defects. By constructing a three-level collaborative detection framework of "optical enhancement - feature decoupling - cross-modal perception", it realizes the accurate mapping of defect features from the physical space to the feature space, thus breaking through the problem of difficult perception and detection of low-contrast micro-defects in a complex texture background.

[0040] Specifically, the technical concept of this application is as follows: At the spatial domain level, through the collaborative design of an optical imaging system and an adaptive preprocessing module, a physical separation substrate for defect features and background textures is established. A uniform diffused light source combined with a specific incident angle configuration utilizes the anisotropic reflection characteristics of the surface microstructure of the ultra-soft gauze to convert the geometric distortion of the defect into a differential response of the light intensity distribution, forming a physical enhancement effect of the defect. The adaptive filtering algorithm dynamically adjusts the denoising intensity according to the local texture spectrum, constructs a gradient protection domain for the defect edge while suppressing high-frequency noise, and avoids detail loss caused by traditional global filtering. At the frequency domain feature modeling level, a heterogeneous multi-scale convolutional network architecture is adopted to achieve two-channel decoupling of the substrate texture and defect features. The shallow network branch captures the microscopic topological features of the yarn interweaving nodes through dense convolutional kernels to construct a phase map of periodic textures; the deep branch extracts long-range correlation patterns across yarn scales with dilated convolutions to form a semantic encoding of abnormal features. Through a dynamic phase alignment mechanism, the texture phase map and the deep semantic encoding are sub-pixel level feature registered in the frequency domain space to eliminate the feature misalignment interference caused by fabric elastic deformation, establish a cross-scale joint perception model for defect responses, improve the signal-to-noise separation ability of sub-millimeter defects in dynamic detection scenarios, and provide a basis for defect detection of ultra-soft knitted gauze fabrics.

[0041] Specifically, Figure 1 is a schematic flowchart of a method for defect detection of ultra-soft knitted gauze fabrics based on image recognition according to an embodiment of this application. Figure 2 is a schematic diagram of data flow of a method for defect detection of ultra-soft knitted gauze fabrics based on image recognition according to an embodiment of this application. As Figure 1 and Figure 2 shown, the method for defect detection of ultra-soft knitted gauze fabrics based on image recognition includes: S1, acquiring an image of an ultra-soft knitted gauze fabric collected by an image acquisition system; S2, performing adaptive image preprocessing and contrast enhancement on the image of the ultra-soft knitted gauze fabric to obtain an enhanced image of the ultra-soft knitted gauze fabric; S3, performing multi-scale feature analysis on the enhanced image of the ultra-soft knitted gauze fabric to obtain the texture features and depth features of the ultra-soft knitted gauze fabric; S4, performing feature alignment joint perception on the texture features and depth features of the ultra-soft knitted gauze fabric to obtain a multi-scale joint perception encoded feature of the ultra-soft knitted gauze fabric features; S5, based on the multi-scale joint perception encoded feature of the ultra-soft knitted gauze fabric features, determining whether there are defects in the ultra-soft knitted gauze fabric to be detected.

[0042] Exemplarily, in step S1, an image of a super-soft knitted gauze fabric collected by an image acquisition system is obtained. In one embodiment, the image acquisition system includes a high-resolution industrial camera and an illumination system. The illumination system uses a uniform diffuse light source. That is to say, at the optical imaging level, a uniform diffuse light source is used in combination with a high-resolution industrial camera. By controlling the incident light angle (recommended 15°±2° low-angle illumination), the three-dimensional characterization ability of the surface micro-topography is enhanced, so that linear defects such as snagging and broken yarns produce a directional shadow effect, and the gray-scale contrast between the defects and the substrate texture is increased.

[0043] In one embodiment, as Figure 3 shown, in S2, the super-soft knitted gauze fabric image is subjected to adaptive image preprocessing and contrast enhancement to obtain a super-soft knitted gauze fabric enhanced image, including: S21, performing adaptive Wiener filtering on the super-soft knitted gauze fabric image to obtain a super-soft knitted gauze fabric adaptively preprocessed image; S22, performing multi-scale contrast enhancement on the super-soft knitted gauze fabric adaptively preprocessed image to obtain a super-soft knitted gauze fabric enhanced image. It should be understood that since the substrate texture of the super-soft knitted gauze fabric has periodic high-frequency components, and micro-defects (such as 0.1 mm broken yarns) only occupy a 5-15 pixel area in the image, traditional global filtering algorithms will cause gradient dispersion at the defect edges, further exacerbating the frequency-domain aliasing between the defect features and the background texture. Adaptive Wiener filtering dynamically perceives the spectral characteristics of the local texture (using a sliding window of 7×7 to 15×15 pixels), constructs a real-time estimation model of the noise variance and signal power in the frequency-domain space, and realizes the dynamic balance between high-frequency noise (σ<0.3) suppression and defect edge retention (sharpness loss rate <8%), overcoming the artifact interference generated by traditional fixed-parameter filters in the gauze texture mutation area. In addition, the multi-scale contrast enhancement process is carried out based on the difference in the frequency-domain energy distribution between the defects and the background: histogram equalization is used in the low-frequency component (scale factor λ = 8) to enhance the overall contrast of the substrate texture, while non-linear gain amplification is implemented in the high-frequency component (scale factor λ = 2) to focus on enhancing the local gradient response of the micro-defects. This strategy reconstructs the luminance channel in the HSV color space, enhances the edge gradient intensity of the defect area, and compresses the background texture fluctuation amplitude at the same time, effectively solving the texture artifact problem caused by over-enhancement in traditional algorithms for gauze detection.

[0044] Exemplarily, in step S3, multi-scale feature analysis is performed on the enhanced image of the ultra-soft knitted gauze fabric to obtain the texture features and depth features of the ultra-soft knitted gauze fabric. It should be understood that due to the honeycomb weaving structure of the ultra-soft gauze presenting periodic high-frequency textures in the image, and micro-defects (such as 0.3 mm yarn knots) only occupying an area of about 15×15 pixels under high-resolution imaging, it is difficult for the fixed receptive field of the traditional single-scale convolutional network to simultaneously capture the microscopic deformation of the yarn nodes and the long-range structural anomalies across the yarns.

[0045] In one embodiment, multi-scale feature analysis is performed on the enhanced image of the ultra-soft knitted gauze fabric to obtain the texture features and depth features of the ultra-soft knitted gauze fabric, including: using a multi-scale convolutional neural network to perform multi-scale image feature extraction on the enhanced image of the ultra-soft knitted gauze fabric to obtain a texture feature encoded map of the ultra-soft knitted gauze fabric as the texture features of the ultra-soft knitted gauze fabric and a depth feature encoded map of the ultra-soft knitted gauze fabric as the depth features of the ultra-soft knitted gauze fabric. In the technical solution of the present application, the multi-scale convolutional neural network is further used to perform multi-scale image feature extraction on the enhanced image of the ultra-soft knitted gauze fabric to obtain a texture feature encoded map of the ultra-soft knitted gauze fabric and a depth feature encoded map of the ultra-soft knitted gauze fabric. In the texture feature encoding branch, pixel-level phase parsing of the gauze substrate texture is performed through densely stacked small convolutional kernels. For the anomaly in the local phase distribution caused by the needle offset defect of the ultra-soft gauze, the topological continuity interruption feature caused by yarn breakage can be accurately captured. In the depth feature encoding branch, a semantic perception field across the yarn scale is constructed through dilated convolution, and the yarn tension imbalance feature caused by snagging defects is mainly extracted.

[0046] Exemplarily, in step S4, feature alignment joint perception is performed on the texture features and depth features of the ultra-soft knitted gauze fabric to obtain the multi-scale joint perception encoded features of the ultra-soft knitted gauze fabric features. It should be understood that due to the honeycomb weaving structure of the ultra-soft gauze being prone to elastic deformation during the dynamic detection process, there is a sub-pixel level spatial misalignment between the texture feature encoded map and the depth feature encoded map. Traditional feature fusion methods (such as channel splicing or weighted averaging) cannot eliminate the feature response deviation caused by such geometric distortions. Therefore, in the technical solution of this application, further feature alignment joint perception is performed on the texture feature encoded map and the depth feature encoded map of the ultra-soft knitted gauze fabric to obtain the multi-scale joint perception encoded map of the ultra-soft knitted gauze fabric features. Specifically, by decoupling the texture feature encoded map (focusing on the phase information of yarn nodes) and the depth feature encoded map (representing cross-yarn semantic associations) of the ultra-soft knitted gauze fabric into a set of local feature encoding matrices, the feature phase alignment degree (the distribution map smoothness measure calculated based on the number of biconnected components) is used to dynamically search for the optimal matching pairs, breaking through the dependence of traditional spatial alignment methods on the rigid transformation assumption. During the phase alignment process, the embedded manifold compactification technology reduces the geometric association error of the feature pairs at the pixel level by optimizing the trace metric operation of the discrete manifold representation, significantly improving the matching robustness under long-range deformation. In particular, the processing process of feature alignment joint perception models the interaction relationship of the phase-aligned feature pairs, realizes the enhancement of the saliency of the defect response region and the collaborative suppression of background texture interference, while reducing the misjudgment rate of structural defects caused by stitch offset, breaking through the "high missed detection - high false detection" double-high dilemma caused by feature misalignment in traditional methods.

[0047] In one embodiment, as Figure 4 shown, S4 includes: S41, decoupling the texture feature encoded map and the depth feature encoded map of the ultra-soft knitted gauze fabric to obtain a set of texture feature encoding matrices of the ultra-soft knitted gauze fabric and a set of depth local feature encoding matrices of the ultra-soft knitted gauze fabric; S42, performing phase alignment and attention saliency fusion on the set of texture feature encoding matrices of the ultra-soft knitted gauze fabric and the set of depth local feature encoding matrices of the ultra-soft knitted gauze fabric to obtain the multi-scale joint perception encoded features of the ultra-soft knitted gauze fabric features.

[0048] Specifically, in step S41, decoupling the texture feature encoded map and the depth feature encoded map of the ultra-soft knitted gauze fabric to obtain a set of texture feature encoding matrices of the ultra-soft knitted gauze fabric and a set of depth local feature encoding matrices of the ultra-soft knitted gauze fabric, specifically, this process can be expressed by the formula as:

[0049] Decouple(F1)={M 11 ,M 12 ,...,M1i ,...,M 1n}

[0050] Decouple(F2) = {M 21 , M 22 ,..., M 2j ,..., M 2n}

[0051] Among them, F1 is the texture feature encoding map of the ultra-soft knitted gauze fabric, F2 is the depth feature encoding map of the ultra-soft knitted gauze fabric, Decouple(·) is the operation of feature decoupling, M 11 , M 12 , M 1i and M 1n are respectively the 1st, 2nd, ith, and nth ultra-soft knitted gauze fabric texture feature encoding matrices in the set of ultra-soft knitted gauze fabric texture feature encoding matrices, M 21 , M 22 , M 2j and M 2n are respectively the 1st, 2nd, jth, and nth ultra-soft knitted gauze fabric depth local feature encoding matrices in the set of ultra-soft knitted gauze fabric depth local feature encoding matrices.

[0052] It should be understood that by decomposing the texture feature encoding map (mainly focusing on the microscopic topological features of yarn nodes) and the depth feature encoding map (focusing on the long-range correlation patterns across yarn scales) of the ultra-soft knitted gauze fabric into multiple local feature encoding matrices, a more refined analysis can be performed on the complex periodic texture background of the ultra-soft knitted gauze fabric. This decomposition enables each local feature encoding matrix to focus on describing the subtle structural changes within a specific region, such as the specific manifestation forms of defects like stitch offset and snagging, thereby improving the ability to capture these detailed features. For example, by decoupling the texture feature encoding map of the ultra-soft knitted gauze fabric, a local feature encoding matrix reflecting the topological continuity interruption feature caused by yarn breakage can be obtained; similarly, the local feature encoding matrix extracted from the depth feature encoding map can accurately capture the yarn tension imbalance feature caused by the snagging defect.

[0053] In one embodiment, as Figure 5As shown in the figure, phase alignment and attention-based significant fusion are performed on the set of texture feature encoding matrices of the ultra-soft knitted gauze fabric and the set of depth local feature encoding matrices of the ultra-soft knitted gauze fabric to obtain the multi-scale joint perception encoding features of the ultra-soft knitted gauze fabric features, including: S421, based on the feature phase alignment degree between any two texture feature encoding matrices of the ultra-soft knitted gauze fabric and depth local feature encoding matrices of the ultra-soft knitted gauze fabric in the set of texture feature encoding matrices of the ultra-soft knitted gauze fabric and the set of depth local feature encoding matrices of the ultra-soft knitted gauze fabric, perform feature phase dynamic search alignment on the set of texture feature encoding matrices of the ultra-soft knitted gauze fabric and the set of depth local feature encoding matrices of the ultra-soft knitted gauze fabric to obtain a set of phase-aligned {ultra-soft knitted gauze fabric texture feature encoding matrix, ultra-soft knitted gauze fabric depth local feature encoding matrix} feature pairs; S422, input each phase-aligned {ultra-soft knitted gauze fabric texture feature encoding matrix, ultra-soft knitted gauze fabric depth local feature encoding matrix} feature pair in the set of phase-aligned {ultra-soft knitted gauze fabric texture feature encoding matrix, ultra-soft knitted gauze fabric depth local feature encoding matrix} feature pairs into the feature joint perception attention network to obtain a set of ultra-soft knitted gauze fabric feature joint perception attention weights; S423, based on the set of ultra-soft knitted gauze fabric feature joint perception attention weights, perform attention-driven significant aggregation on the set of phase-aligned {ultra-soft knitted gauze fabric texture feature encoding matrix, ultra-soft knitted gauze fabric depth local feature encoding matrix} feature pairs to obtain the multi-scale joint perception encoding map of the ultra-soft knitted gauze fabric features as the multi-scale joint perception encoding features of the ultra-soft knitted gauze fabric features.

[0054] Specifically, in step S421, the process of feature phase dynamic search alignment can be expressed by the formula as follows:

[0055]

[0056] P ik ={M 1i ,M 2k} pair

[0057] where T is the transpose operation, ‖·‖ F is to calculate the Frobenius norm, d P (M 1i ,M 2j ) is the feature phase alignment degree between M 1i and M 2j , is the j value corresponding to the returned maximum value, k is the position of searching for the maximum approximate matching value in the set of depth local feature encoding matrices of the ultra-soft knitted gauze fabric, {M 1i ′,M 2k ′} pairIt is a feature pair of phase-optimized alignment {ultra-soft knitted gauze fabric texture feature encoding matrix, ultra-soft knitted gauze fabric depth local feature encoding matrix}, P ik represents the feature pair.

[0058] It should be understood that the dynamic search for feature phase alignment emphasizes finding the feature pair with the highest phase alignment degree in the feature set, rather than simply relying on static spatial position correspondence. This enables the model to flexibly handle the elastic deformation that may occur during the dynamic detection of ultra-soft knitted gauze fabric within a certain range, thereby ensuring accurate matching even in the presence of sub-pixel spatial misalignment. By optimizing the trace metric operation of the discrete manifold representation, the geometric correlation error is further reduced, and the matching robustness under long-range deformation is improved. This precise feature alignment provides high-quality basic data for subsequent joint feature perception, contributing to the improvement of the quality of the finally generated joint perception feature map. In addition, the dynamic search for feature phase alignment not only improves the accuracy of feature alignment but also enhances the model's ability to identify complex backgrounds and minor defects. In the detection of ultra-soft knitted gauze fabric, even minor deformation may affect the final detection result, so this strategy is crucial for ensuring high-precision defect detection. Through the effective measurement of feature phase alignment degree and the dynamic search strategy, the model can efficiently find the best matching pairs in the set of local feature encoding matrices, overcoming problems such as spatial misalignment, deformation, or occlusion that may exist in the feature map, and finally generating a set of phase-aligned feature pairs. After being processed by the subsequent attention mechanism, these feature pairs can adaptively learn the importance weights of different feature pairs, further optimizing the feature aggregation process, highlighting the contributions of important features, suppressing noise interference, and enhancing the interpretability and generalization ability of the model.

[0059] In one embodiment, in step S422, each phase-aligned {ultra-soft knitted gauze fabric texture feature encoding matrix, ultra-soft knitted gauze fabric depth local feature encoding matrix} feature pair in the set of phase-aligned {ultra-soft knitted gauze fabric texture feature encoding matrix, ultra-soft knitted gauze fabric depth local feature encoding matrix} feature pairs is input into the feature joint perception attention network to obtain a set of ultra-soft knitted gauze fabric feature joint perception attention weights, including: calculating the correlation matrix between each phase-aligned {ultra-soft knitted gauze fabric texture feature encoding matrix, ultra-soft knitted gauze fabric depth local feature encoding matrix} feature pair in the set of phase-aligned {ultra-soft knitted gauze fabric texture feature encoding matrix, ultra-soft knitted gauze fabric depth local feature encoding matrix} feature pairs to obtain a set of correlation matrices of the surface of the textile to be inspected with respect to scale; performing trace metric on each correlation matrix of the surface of the textile to be inspected with respect to scale in the set of correlation matrices of the surface of the textile to be inspected with respect to scale to obtain a set of trace metric values of the surface of the textile to be inspected. Specifically, this process can be represented by the formula:

[0060] ai = Tr({M 1i , M 2k}) pair ) = Tr(M 1i T M 2k )

[0061] where Tr(·) is the trace metric value of the matrix, and a i is the trace metric value of the surface of the textile to be inspected between M 1i and M 2k .

[0062] Based on the set of trace metric values of the surface of the textile to be inspected, a set of joint perception attention weights of the characteristics of the super-soft knitted gauze fabric is obtained. Specifically, this process can be expressed by the formula as follows:

[0063]

[0064] where A net (·) is the calculation of the joint perception attention weight of the characteristics, M 1i T ′ is the optimized phase-optimized alignment super-soft knitted gauze fabric texture feature coding matrix of M 1i , M 2l ′ is the optimized phase-optimized alignment super-soft knitted gauze fabric depth local feature coding matrix of M 2k , a i ′ is the optimized trace metric value of the surface of the textile to be inspected between M 1i and M 2k , sigmoid is the normalization function, and w i is the joint perception attention weight of the characteristics of the super-soft knitted gauze fabric between M 1i and M 2k .

[0065] It should be understood that in the context of the complex honeycomb weaving structure of the ultra-soft knitted gauze fabric, traditional feature fusion methods often struggle to fully utilize the complementary information between different modalities or levels, and may even introduce noise interference. However, through the application of the feature joint perception attention network, these problems can be effectively solved. First of all, the function of the feature joint perception attention network is not limited to predicting weights. More importantly, it is a key component to achieve joint perception ability. Although the phase alignment step has initially established the corresponding relationship of structurally similar features, the importance of different feature pairs and their complex interaction patterns still need to be further modeled and learned. Through the attention mechanism, the network can assign appropriate importance weights to each phase-aligned feature pair and learn the interaction relationship between feature pairs. This adaptive learning method enables the model to selectively fuse complementary feature information in the subsequent feature aggregation process, while suppressing redundant or irrelevant information, thereby improving the quality of the final feature representation.

[0066] Secondly, the feature joint perception attention network can significantly enhance the model's ability to identify complex backgrounds and tiny defects. In the detection of ultra-soft knitted gauze fabric, texture features and depth features provide different information respectively: the texture feature encoding map mainly captures the microscopic topological features of yarn nodes, while the depth feature encoding map focuses on the long-range correlation patterns across the yarn scale. Through the feature joint perception attention network, the weights of these feature pairs can be dynamically adjusted to ensure that important features are fully emphasized, while secondary features or noise are effectively suppressed. For example, when detecting a broken yarn defect caused by stitch offset, the attention network can assign higher weights to the relevant texture features, and when detecting the tension imbalance caused by snagging, it increases the weights of the depth features, thereby improving the accuracy of defect identification.

[0067] In addition, the feature joint perception attention network also enhances the model's robustness to geometric deformation and occlusion. Since the ultra-soft knitted gauze fabric may undergo elastic deformation or there may be some areas occluded in practical applications, traditional methods are prone to performance degradation in such cases. However, through the attention mechanism, the model can automatically correct the deformation to a certain extent and suppress the influence of the occluded area. Specifically, the feature joint perception attention network can identify and highlight those feature pairs that contain key information, ensuring the stability and reliability of the detection results even in the presence of slight deformation or occlusion.

[0068] In one embodiment, based on a set of surface trace measurement values of a textile to be inspected, a set of joint perception attention weights of super-soft knitted gauze fabric features is obtained, including: based on the set of surface trace measurement values of the textile to be inspected, performing matrix manifold optimization based on the number of biconnectivity on each phase-aligned {super-soft knitted gauze fabric texture feature encoding matrix, super-soft knitted gauze fabric depth local feature encoding matrix} feature pair in the set of phase-aligned {super-soft knitted gauze fabric texture feature encoding matrix, super-soft knitted gauze fabric depth local feature encoding matrix} feature pairs to obtain a set of phase-optimized aligned {super-soft knitted gauze fabric texture feature encoding matrix, super-soft knitted gauze fabric depth local feature encoding matrix} feature pairs; based on the set of phase-optimized aligned {super-soft knitted gauze fabric texture feature encoding matrix, super-soft knitted gauze fabric depth local feature encoding matrix} feature pairs, obtaining a set of joint perception attention weights of super-soft knitted gauze fabric features.

[0069] Specifically, here, for each M for which matrix trace metric operation is performed 1i T and M 2k , let M a = M 1i T and M b = M 2k , and m ai ∈ M a , m bi ∈ M b , by calculating the number of eigenvalues that satisfy the L1 distance d ai (m bi , m L1 ) < ε and the L2 distance d ai (m bi , m L2 ) < ε between the corresponding m ai and m bi , the biconnectivity numbers α and β between the matrices M a and M b are obtained, that is, the distribution smoothness quantity between the super-soft knitted gauze fabric texture feature encoding matrix and the super-soft knitted gauze fabric depth local feature encoding matrix used to represent phase alignment.

[0070] Then, in the discrete manifold representation of the super-soft knitted gauze fabric texture feature encoding matrix and the super-soft knitted gauze fabric depth local feature encoding matrix, perform embedding optimization on the trace metric operation of the super-soft knitted gauze fabric texture feature encoding matrix and the super-soft knitted gauze fabric depth local feature encoding matrix through embedding manifold compactification based on the number of biconnectivity, so as to improve the calculation accuracy of the joint matrix trace metric on the basis of enhancing the long-range correlation connectivity robustness:

[0071] M1i T ′ = exp[-(α - β + a i ) ⊙ M 1i T

[0072] M 2k ′ = exp[-(β - α + a i ) ⊙ M 2k

[0073] Among them, in the case of calculating a 1i T and M 2k based on the initial M i , substitute a i , α, β, the initial M 1i T and M 2k for optimization, and then calculate a i ′ = Tr(M 1i T ′M 2k ′) to calculate the optimized trace metric.

[0074] Specifically, in step S423, the process of attention-driven saliency aggregation can be expressed by the formula as:

[0075]

[0076] F f = [M1′, M2′,..., M i ′,..., M n ′]

[0077] Among them, W1 and W2 are the first joint perception weight matrix and the second joint perception weight matrix, is subtraction by position points, is addition by position points, M′ i is the joint super-soft knitted gauze fabric feature joint perception matrix of M 1i and M 2k , that is, the i-th super-soft knitted gauze fabric feature joint perception matrix in the set of super-soft knitted gauze fabric feature joint perception matrices, M′1, ′2 and M′ n are the 1st, 2nd, and nth super-soft knitted gauze fabric feature joint perception matrices in the set of super-soft knitted gauze fabric feature joint perception matrices respectively, F f ​​It is a multi-scale joint perception coding map of the characteristics of the super soft knitted gauze fabric. Here, it should be known to those skilled in the art that the first joint perception weight matrix, the second joint perception weight matrix and other network parameters such as weight matrices or biases of this application are obtained through training. At the initial stage of training, the parameters in these weight matrices are randomly initialized. As the training process progresses, optimization strategies such as the backpropagation algorithm and the gradient descent method are used to gradually adjust these weight values according to the feedback information provided by the loss function. The design of the loss function aims to measure the gap between the model prediction result and the actual label, and optimize the network parameters by minimizing this gap.

[0078] It should be understood that the attention-driven saliency aggregation process aims to highlight the contributions of important features, rather than simply performing averaging or splicing operations. By applying the feature joint perception attention weights to the set of phase-aligned feature pairs, it can ensure that those feature pairs containing key information dominate in the final feature map, while secondary features or noise are suppressed. This selective aggregation method enables the model to more clearly identify target objects, key regions or important feature relationships in complex backgrounds, thereby improving the performance of subsequent tasks. For example, when detecting subtle defects such as broken yarns or snagged threads in the super soft knitted gauze fabric, the attention-driven saliency aggregation can focus on these key regions and provide a more accurate and reliable feature representation. Secondly, the saliency aggregation not only improves the quality of the feature map, but also enhances the robustness of the model to geometric deformations and occlusions. Since the super soft knitted gauze fabric may undergo elastic deformations or there are partially occluded regions in practical applications, traditional methods are prone to performance degradation in such cases. However, through the attention mechanism, the model can automatically correct deformations to a certain extent and suppress the influence of occluded regions. Specifically, during the saliency aggregation process, the attention weights guide the model to preferentially focus on those unoccluded feature pairs containing key information, ensuring the stability and reliability of the detection results even in the presence of slight deformations or occlusions. In addition, the saliency aggregation further optimizes the effect of feature fusion and realizes the effective integration of multi-scale information. The texture features and depth features of the super soft knitted gauze fabric provide different information: the texture feature coding map mainly captures the microscopic topological features of yarn nodes, while the depth feature coding map focuses on the long-range correlation patterns across yarn scales. Through saliency aggregation, the model can organically combine these different-scale information to form a more comprehensive and robust feature representation. For example, when detecting broken yarn defects caused by stitch offsets, the saliency aggregation can comprehensively consider the relevant information of texture features and depth features to ensure that the finally generated feature map can more accurately reflect these subtle changes, thereby improving the accuracy of defect detection.

[0079] Exemplarily, in step S5, based on the multi-scale joint perception coding features of the ultra-soft knitted gauze fabric, it is determined whether there are defects in the ultra-soft knitted gauze fabric to be detected. In particular, in a specific example of the present application, the multi-scale joint perception coding map of the ultra-soft knitted gauze fabric features is detected by a fabric defect detector based on a classifier to determine whether there are defects in the ultra-soft knitted gauze fabric to be detected. It should be understood that the output layer of the classifier usually adopts a fully connected layer and a softmax function, which are used to map the extracted high-level features to the specific category probability distribution. First, the multi-scale joint perception coding map of the ultra-soft knitted gauze fabric features is unfolded into a multi-scale joint perception coding vector of the ultra-soft knitted gauze fabric features. The fully connected layer converts the multi-scale joint perception coding vector of the ultra-soft knitted gauze fabric features into a vector representation of a fixed length, and the softmax function maps these vectors to the probability values of each category. In practical applications, two categories can be defined: normal area and defect area. The model selects the category with the highest probability as the final prediction result by comparing the probability values of each category.

[0080] In summary, the method and system for detecting defects in ultra-soft knitted gauze fabric based on image recognition according to the embodiments of the present application are elucidated. It can address the problems of texture periodic interference caused by the honeycomb weaving structure of ultra-soft gauze and the sub-pixel size characteristics of micro-defects. By constructing a three-level collaborative detection framework of "optical enhancement - feature decoupling - cross-modal perception", it realizes the accurate mapping of defect features from the physical space to the feature space, thus breaking through the problem of difficult perception and detection of low-contrast micro-defects in complex texture backgrounds.

[0081] Figure 6 It is a schematic block diagram of a system for detecting defects in ultra-soft knitted gauze fabric based on image recognition according to an embodiment of the present application. As Figure 6 shown, the system 10 for detecting defects in ultra-soft knitted gauze fabric based on image recognition includes: an ultra-soft knitted gauze fabric image acquisition module 11, configured to acquire an ultra-soft knitted gauze fabric image collected by an image acquisition system; an ultra-soft knitted gauze fabric image enhancement module 12, configured to perform adaptive image preprocessing and contrast enhancement on the ultra-soft knitted gauze fabric image to obtain an enhanced ultra-soft knitted gauze fabric image; an image multi-scale feature analysis module 13, configured to perform multi-scale feature analysis on the enhanced ultra-soft knitted gauze fabric image to obtain the texture features and depth features of the ultra-soft knitted gauze fabric; an image feature alignment and joint perception module 14, configured to perform feature alignment and joint perception on the texture features and depth features of the ultra-soft knitted gauze fabric to obtain the multi-scale joint perception coding features of the ultra-soft knitted gauze fabric features; and an ultra-soft knitted gauze fabric detection result determination module 15, configured to determine whether there are defects in the ultra-soft knitted gauze fabric to be detected based on the multi-scale joint perception coding features of the ultra-soft knitted gauze fabric features.

[0082] In one embodiment, the image feature alignment and joint perception module includes: an image feature decoupling module, configured to perform feature decoupling on the texture feature encoding map of the ultra-soft knitted gauze fabric and the depth feature encoding map of the ultra-soft knitted gauze fabric to obtain a set of texture feature encoding matrices of the ultra-soft knitted gauze fabric and a set of depth local feature encoding matrices of the ultra-soft knitted gauze fabric; an image feature fusion and encoding module, configured to perform phase alignment and attention-based significant fusion on the set of texture feature encoding matrices of the ultra-soft knitted gauze fabric and the set of depth local feature encoding matrices of the ultra-soft knitted gauze fabric to obtain multi-scale joint perception encoded features of the ultra-soft knitted gauze fabric features.

[0083] In one embodiment, the image acquisition system includes a high-resolution industrial camera and an illumination system.

[0084] Here, those skilled in the art can understand that the specific operations of each module in the above-mentioned ultra-soft knitted gauze fabric defect detection system based on image recognition have been described in detail above with reference to Figures 1 to 5 the description of the ultra-soft knitted gauze fabric defect detection method based on image recognition, and therefore, the repeated description thereof will be omitted.

[0085] Finally, it should be noted that the above-described embodiments are some embodiments of the present invention, rather than all embodiments. The detailed description of the embodiments of the present invention is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

Claims

1. A method for detecting defects in ultra-soft knitted yarn fabric based on image recognition, characterized in that: include: Acquire an image of the ultra-soft knitted yarn fabric acquired by an image acquisition system; Performing adaptive image preprocessing and contrast enhancement on the super-flexible knitting yarn fabric image to obtain a super-flexible knitting yarn fabric enhanced image; Performing multi-scale feature analysis on the enhanced image of the super-flexible knitted yarn fabric to obtain texture features and depth features of the super-flexible knitted yarn fabric; The texture features of the super-flexible knitting yarn fabric and the depth features of the super-flexible knitting yarn fabric are feature-aligned and jointly perceived to obtain multi-scale joint perception coding features of the super-flexible knitting yarn fabric features, including: feature-decoupling the super-flexible knitting yarn fabric texture feature coding map and the super-flexible knitting yarn fabric depth feature coding map to obtain a set of super-flexible knitting yarn fabric texture feature coding matrices and a set of super-flexible knitting yarn fabric depth local feature coding matrices; phase-aligning and attention-significantly fusing the set of super-flexible knitting yarn fabric texture feature coding matrices and the set of super-flexible knitting yarn fabric depth local feature coding matrices to obtain multi-scale joint perception coding features of super-flexible knitting yarn fabric features; Based on the multi-scale joint perception coding features of the super-flexible knitting yarn fabric characteristics, it is determined whether the super-flexible knitting yarn fabric to be inspected has defects.

2. The method for detecting defects of ultra-soft knitted yarn fabric based on image recognition according to claim 1, characterized in that: The image acquisition system includes a high-resolution industrial camera and a lighting system.

3. The method for detecting defects of ultra-soft knitted yarn fabric based on image recognition according to claim 2, characterized in that: The lighting system adopts a uniform diffuse reflection light source.

4. The method for detecting defects of ultra-soft knitted yarn fabric based on image recognition according to claim 3, characterized in that: The method includes performing adaptive image preprocessing and contrast enhancement on the super-flexible knitted yarn fabric image to obtain a super-flexible knitted yarn fabric enhanced image, comprising: Performing adaptive Wiener filtering on the super-flexible knitting yarn fabric image to obtain an adaptive pre-processed image of the super-flexible knitting yarn fabric; The ultra-flexible knitting yarn fabric adaptive pre-processed image is subjected to multi-scale contrast enhancement processing to obtain the ultra-flexible knitting yarn fabric enhanced image.

5. The method for detecting defects of ultra-soft knitted yarn fabric based on image recognition according to claim 4, characterized in that: The method comprises: performing multi-scale feature analysis on the super-flexible knitted yarn fabric enhanced image to obtain super-flexible knitted yarn fabric texture features and super-flexible knitted yarn fabric depth features, comprising: using a multi-scale convolutional neural network to perform multi-scale image feature extraction on the super-flexible knitted yarn fabric enhanced image to obtain a super-flexible knitted yarn fabric texture feature encoding map as the super-flexible knitted yarn fabric texture features and a super-flexible knitted yarn fabric depth feature encoding map as the super-flexible knitted yarn fabric depth features.

6. The method for detecting defects of ultra-soft knitted yarn fabric based on image recognition according to claim 5, characterized in that: The set of texture feature encoding matrices of super-flexible knitted yarn fabric and the set of deep local feature encoding matrices of super-flexible knitted yarn fabric are phase aligned and significantly fused to obtain multi-scale joint perception encoding features of super-flexible knitted yarn fabric features, including: Based on the feature phase alignment between any two super-flexible knitting yarn fabric texture feature coding matrices and super-flexible knitting yarn fabric depth local feature coding matrices in the set of the super-flexible knitting yarn fabric texture feature coding matrices and the set of the super-flexible knitting yarn fabric depth local feature coding matrices, the set of the super-flexible knitting yarn fabric texture feature coding matrices and the set of the super-flexible knitting yarn fabric depth local feature coding matrices are subjected to feature phase dynamic search alignment to obtain a set of phase-aligned {super-flexible knitting yarn fabric texture feature coding matrix, super-flexible knitting yarn fabric depth local feature coding matrix} feature pairs; Input each phase-aligned {super-flexible knitting yarn fabric texture feature encoding matrix, super-flexible knitting yarn fabric depth local feature encoding matrix} feature pair in the set of phase-aligned {super-flexible knitting yarn fabric texture feature encoding matrix, super-flexible knitting yarn fabric depth local feature encoding matrix} feature pairs into a feature joint perception attention network to obtain a set of super-flexible knitting yarn fabric feature joint perception attention weights; Based on the set of joint perceptual attention weights of the super-flexible knitting yarn fabric features, the set of the phase-aligned {super-flexible knitting yarn fabric texture feature encoding matrix, super-flexible knitting yarn fabric depth local feature encoding matrix} feature pairs is subjected to attention-driven significant aggregation to obtain a super-flexible knitting yarn fabric feature multi-scale joint perceptual coding map as the super-flexible knitting yarn fabric feature multi-scale joint perceptual coding feature.

7. The method for detecting defects of ultra-soft knitted yarn fabric based on image recognition according to claim 6, characterized in that: Input each phase-aligned {super-flexible knitting yarn fabric texture feature encoding matrix, super-flexible knitting yarn fabric depth local feature encoding matrix} feature pair in the set of the phase-aligned {super-flexible knitting yarn fabric texture feature encoding matrix, super-flexible knitting yarn fabric depth local feature encoding matrix} feature pairs into a feature joint perception attention network to obtain a set of super-flexible knitting yarn fabric feature joint perception attention weights, including: Calculating the correlation matrix between each phase-aligned {super-flexible knitted yarn fabric texture feature coding matrix, super-flexible knitted yarn fabric depth local feature coding matrix} feature pair in the set of phase-aligned {super-flexible knitted yarn fabric texture feature coding matrix, super-flexible knitted yarn fabric depth local feature coding matrix} feature pairs to obtain a set of scale correlation matrices of the surface of the textile to be inspected; Performing trace measurement on each of the textile surface pair scale association matrices to be inspected in the set of textile surface pair scale association matrices to obtain a set of textile surface trace measurement values ​​to be inspected; Based on the set of surface trace measurement values ​​of the textile to be inspected, a set of joint perceptual attention weights of the ultra-soft knitted yarn fabric features is obtained.

8. The method for detecting defects of ultra-soft knitted yarn fabric based on image recognition according to claim 7, characterized in that: Based on the set of surface trace values ​​of the textile to be inspected, a set of joint perceptual attention weights of the ultra-soft knitted yarn fabric features is obtained, including: Based on the set of surface trace measurement values ​​of the textile to be inspected, each phase-aligned {super-flexible knitting yarn fabric texture feature coding matrix, super-flexible knitting yarn fabric depth local feature coding matrix} feature pair in the set of phase-aligned {super-flexible knitting yarn fabric texture feature coding matrix, super-flexible knitting yarn fabric depth local feature coding matrix} feature pairs is subjected to matrix manifold optimization based on the number of dual connections to obtain a set of phase-optimized aligned {super-flexible knitting yarn fabric texture feature coding matrix, super-flexible knitting yarn fabric depth local feature coding matrix} feature pairs; Based on the phase optimization alignment of the set of {super-flexible knitting yarn fabric texture feature encoding matrix, super-flexible knitting yarn fabric depth local feature encoding matrix} feature pairs, a set of super-flexible knitting yarn fabric feature joint perception attention weights is obtained.

9. A super-soft knitted yarn fabric defect detection system based on image recognition, characterized in that: include: A super-soft knitting yarn fabric image acquisition module is used to acquire the super-soft knitting yarn fabric image acquired by the image acquisition system; A super-flexible knitted yarn fabric image enhancement module, used for performing adaptive image preprocessing and contrast enhancement on the super-flexible knitted yarn fabric image to obtain a super-flexible knitted yarn fabric enhanced image; An image multi-scale feature analysis module, used for performing multi-scale feature analysis on the super-flexible knitting yarn fabric enhanced image to obtain the texture feature and depth feature of the super-flexible knitting yarn fabric; An image feature alignment joint perception module is used to perform feature alignment and joint perception on the texture feature of the super-flexible knitting yarn fabric and the depth feature of the super-flexible knitting yarn fabric to obtain a multi-scale joint perception coding feature of the super-flexible knitting yarn fabric feature, wherein the image feature alignment and joint perception module includes: an image feature decoupling module, which is used to perform feature decoupling on the super-flexible knitting yarn fabric texture feature coding map and the super-flexible knitting yarn fabric depth feature coding map to obtain a set of super-flexible knitting yarn fabric texture feature coding matrices and a set of super-flexible knitting yarn fabric depth local feature coding matrices; an image feature fusion coding module, which is used to perform phase alignment and attention saliency fusion on the set of super-flexible knitting yarn fabric texture feature coding matrices and the set of super-flexible knitting yarn fabric depth local feature coding matrices to obtain a multi-scale joint perception coding feature of super-flexible knitting yarn fabric features; The super-flexible knitting yarn fabric detection result determination module is used to determine whether the super-flexible knitting yarn fabric to be detected has defects based on the multi-scale joint perception coding features of the super-flexible knitting yarn fabric characteristics.

10. The ultra-soft knitted yarn fabric defect detection system based on image recognition according to claim 9, characterized in that: The image acquisition system includes a high-resolution industrial camera and a lighting system.

Images