A method and device for rating fuzz and pilling based on visual continuous regression

Abstract

The application discloses a pilling rating method and device based on visual continuous regression. The method comprises the following steps: S1, collecting a global image of a fabric sample to obtain an original fabric image; S2, rubbing the fabric sample, collecting a real-time local image of a rubbing contact area in the rubbing process to obtain a continuous local image sequence, and obtaining process evolution information according to the continuous local image sequence; S3, collecting an image of the rubbed fabric sample at a rubbing station after the rubbing is completed to obtain a rubbed fabric image; and S4, inputting the original fabric image, the process evolution information and the rubbed fabric image into a pilling evaluation model to obtain an ordered pilling grade. The pilling rating is full-automatic, objective, consistent and continuous.

Description

Technical Field

[0001] This invention relates to pilling and fuzzing rating, and in particular to a pilling and fuzzing rating method and apparatus based on visual continuous regression. Background Technology

[0002] The pilling performance of fabrics is a key indicator for evaluating the durability and appearance quality of textile products. The core challenge in its testing lies in the fact that pilling is not an instantaneous state, but a dynamic and continuous evolutionary process from the initial fabric state, through friction, to the final formation of pills. However, all current mainstream testing methods in the industry, whether relying on standard sample methods for manual comparison or automated methods based on computer vision, can only observe and compare two static endpoints before and after friction, and based on this, provide a discrete or coarse rating result.

[0003] The existing static "post-friction rating" model is essentially a "black box test," providing only the final result. Its limitations are as follows:

[0004] (1) Information is extremely scarce and the source cannot be traced.

[0005] We only know the final pilling level, but we have no idea how it developed (whether it deteriorated rapidly in the early stages or suddenly collapsed in the later stages), which makes it impossible to locate the key stage where the process defect occurred.

[0006] (2) The optimization direction is unclear and the cost of trial and error is high.

[0007] Process engineers can only adjust process parameters (such as yarn twist and fabric density) based on experience and the final grade. Due to the lack of process data, the adjustments are blind and often require multiple "guess-and-verify" cycles, which is inefficient and costly.

[0008] (3) Unable to identify "critical point" and "stability"

[0009] Some fabrics may perform well in the initial stages of friction, but their performance drops sharply after a certain critical point (i.e., a "jump"). Current technology cannot capture this critical phenomenon and may misclassify a fabric that is actually unstable as acceptable.

[0010] (4) Discrete ratings mask subtle changes

[0011] Discrete rating systems like the five-level system are not sensitive to minor improvements in the manufacturing process. Process adjustments may have substantially improved fabric performance, but if these improvements don't cross the rating threshold, they won't be reflected in the final result, thus discouraging optimization efforts. Summary of the Invention

[0012] To address the aforementioned problems, this application provides a method and apparatus for rating pilling and fuzzing based on visual continuous regression.

[0013] The present invention discloses the following solution: a method for rating pilling and fuzzing based on visual continuous regression, comprising the following steps:

[0014] S1. Perform global image acquisition on the fabric sample to obtain the original fabric image;

[0015] S2. When the fabric sample is rubbed, real-time local images of the friction contact area are acquired to obtain a continuous local image sequence; process evolution information is obtained based on the continuous local image sequence.

[0016] S3. After the friction is completed, the image of the rubbed fabric sample is acquired to obtain the image of the rubbed fabric.

[0017] S4. Input the original fabric image, process evolution information and friction fabric image into the pilling evaluation model to obtain the ordered pilling level.

[0018] Further, in step S2, the continuous local image sequence is input into the no-reference visual pilling evaluation model to obtain process evolution information;

[0019] The no-reference visual pilling assessment model is based on the visual Transformer architecture. It extracts and encodes multi-scale features from key frames in a continuous local image sequence and outputs continuous evaluation values ​​of the pilling process corresponding to each key frame.

[0020] Based on the continuous evaluation values ​​of the pilling process output in time series, process evolution information reflecting the dynamic evolution of pilling is generated. The process evolution information includes the dynamic evolution curve of pilling, as well as the initial deterioration point, performance jump, deterioration rate, final stability level, process uniformity assessment, and process parameter-performance response relationship obtained from the dynamic evolution curve of pilling.

[0021] Furthermore, in step S4, the working process of the pilling and fuzzing assessment model is as follows:

[0022] (1) Spatial alignment and multi-scale feature extraction layer

[0023] The friction fabric image is registered using a spatial transformation network. Then, the original fabric image and the registered friction fabric image are divided into multi-scale regions and multi-feature extraction is performed, including grayscale, spatial domain and frequency domain features, so as to construct the multi-scale feature sequence of the original fabric image and the multi-scale feature sequence of the friction fabric image.

[0024] (2) Weight-sharing twin ViT encoder

[0025] The first ViT encoder and the second ViT encoder are used to process the multi-scale feature sequence of the original fabric image and the multi-scale feature sequence of the friction fabric image respectively, and output the high-level semantic features of the original fabric image and the high-level semantic features of the friction fabric image; the weights of the first ViT encoder and the second ViT encoder are the same.

[0026] (3) Differential Cross Attention Module

[0027] The high-level semantic features of the original fabric image, the process evolution information, and the high-level semantic features of the friction fabric image are input into the differential cross-attention module to obtain the final difference features.

[0028] (4) Adaptive gating module

[0029] The final difference features and the high-level semantic features of the friction fabric image are input into the adaptive gating module for feature fusion and weight allocation to obtain the pilling level.

[0030] (5) Input the pilling grade into the ordered regression head to obtain the ordered pilling grade.

[0031] Furthermore, the working process of the differential cross-attention module is as follows:

[0032] (1) Obtain the first key vector and the first value vector by linear projection of the high-level semantic features of the original fabric image;

[0033] The query vector is obtained by linearly projecting the high-level semantic features of the friction fabric image.

[0034] The process evolution information is obtained by linear projection to obtain the second key vector and the second value vector;

[0035] (2) Input the first key vector, the second key vector and the query vector into the attention model to calculate the attention score and obtain the differential attention weight that incorporates the process evolution;

[0036] (3) Perform feature weighting aggregation of the differential attention weights with the first value vector and the second value vector to obtain differential features;

[0037] (4) The high-level semantic features of the original fabric image, the high-level semantic features of the friction fabric image, and the process evolution information are spliced ​​together, and then nonlinear transformation is performed through a multilayer perceptron to obtain fused features;

[0038] (5) Perform residual connection between the differential features and the fusion features to obtain the final difference features.

[0039] Furthermore, the pilling and fuzzing assessment model is trained through the following steps:

[0040] (1) Obtain the training dataset, which includes several original fabric images, corresponding friction fabric images, and corresponding pilling and fuzzing grades;

[0041] (2) The training dataset was trained using the pilling and fuzzing assessment model;

[0042] If the difference between the output pilling grade and the corresponding pilling grade in the training dataset meets the requirements, then the output is a usable pilling evaluation model.

[0043] If the difference between the output pilling grade and the corresponding pilling grade in the training dataset does not meet the requirements, the parameters are optimized and retrained. If the requirements are still not met, the sample size is increased and training continues until the requirements are met.

[0044] Furthermore, after step S4 is completed, step S5 is performed, as follows:

[0045] Move the new fabric sample to the friction station and repeat steps S1-S4 to obtain process evolution information and ordered pilling level of the new fabric sample.

[0046] A pilling and fuzzing rating device based on visual continuous regression, comprising:

[0047] The first acquisition module is used to perform global image acquisition on the fabric sample and obtain the original fabric image.

[0048] The second acquisition module is used to acquire real-time local images of the friction contact area when the fabric sample is rubbed, and obtain a continuous local image sequence; and obtain process evolution information based on the continuous local image sequence.

[0049] The third acquisition module is used to acquire images of the fabric sample after friction, and obtain images of the friction fabric after friction.

[0050] The evaluation module is used to input the original fabric image, process evolution information, and friction fabric image into the pilling evaluation model to obtain the ordered pilling level.

[0051] An electronic device, comprising:

[0052] One or more processors;

[0053] Memory, used to store one or more programs;

[0054] When the one or more programs are executed by the one or more processors, the one or more processors perform the methods described above.

[0055] A computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of the method described above.

[0056] The beneficial effects of this invention are as follows:

[0057] (1) By acquiring real-time local images of the friction contact area during the friction process, a continuous local image sequence is obtained; based on the continuous local image sequence, process evolution information is obtained, providing more in-depth information for the improvement of fabric technology;

[0058] (2) By inputting the original fabric image and the friction fabric image into the pilling evaluation model, an ordered pilling level is obtained, realizing fully automatic, objective, consistent and continuous pilling rating;

[0059] (3) It has the ability to fuse multi-scale features and enhance differences, resulting in high rating accuracy;

[0060] (4) It has a clear structure, strong scalability, and is suitable for a variety of fabric types. Attached Figure Description

[0061] Figure 1 Here is a flowchart of the method described in this application;

[0062] Figure 2 This is a structural schematic diagram of the friction head at one angle of Embodiment 1 of this application;

[0063] Figure 3 This is a schematic diagram of the friction head from another angle in Embodiment 1 of this application;

[0064] Figure 4 This is a schematic diagram of the modules of the pilling and fuzzing evaluation model of this application;

[0065] Figure 5 This is a schematic diagram of the differential cross-attention module of this application;

[0066] Figure 6 This is a schematic diagram of the modules of the no-reference visual pilling assessment model in Embodiment 1 of this application;

[0067] Figure 7 The images shown are actual pictures of the pilling test of fabric sample 1 in Example 1. Among them, a is the original picture without wear, b is the picture after 100 pilling rubs, c is the picture after 200 pilling rubs, d is the picture after 300 pilling rubs, e is the picture after 400 pilling rubs, f is the picture after 500 pilling rubs, and g is the picture after 600 pilling rubs.

[0068] Figure 8 This is a line graph showing the number of naps and the wear level of fabric sample 1 in Example 1;

[0069] Figure 9 The images shown are actual pictures of the pilling test of fabric sample 2 in Example 1. Among them, a is the original picture without wear, b is the picture after 100 pilling rubs, c is the picture after 200 pilling rubs, d is the picture after 300 pilling rubs, e is the picture after 400 pilling rubs, f is the picture after 500 pilling rubs, and g is the picture after 600 pilling rubs.

[0070] Figure 10 This is a line graph showing the number of naps and the wear level of fabric sample 2 from Example 1.

[0071] Figure 11 This is a schematic diagram of the module of the device of this application. Detailed Implementation

[0072] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0073] In this application, fabric samples include, but are not limited to, flexible materials such as textiles, knitted fabrics, woven fabrics, and nonwoven fabrics that require pilling performance testing. Ordered pilling grades include continuous evaluation values ​​(such as floating-point numbers between 0 and 5) and ordered discrete evaluation values ​​(such as refined grades 1-5 conforming to GB / T4802.1). Continuous evaluation values ​​during the pilling process are the continuous representation of the ordered pilling grade.

[0074] The embodiments of the present invention will be further described below with reference to several examples.

[0075] Example 1

[0076] like Figure 1 A method for rating pilling and fuzzing based on visual continuous regression includes the following steps:

[0077] S1. Perform global image acquisition on the fabric sample to obtain the original fabric image;

[0078] In this embodiment, a clamp is used to press down the edge of the fabric sample, exposing the middle part of the fabric sample for friction;

[0079] In this embodiment, a Berton GX696 electron microscope was used for image acquisition; the original fabric image was acquired using a 60-megapixel lens.

[0080] S2. When the fabric sample is rubbed, real-time local images of the friction contact area are acquired to obtain a continuous local image sequence; process evolution information is obtained based on the continuous local image sequence.

[0081] In this embodiment, a 3-megapixel lens is used to acquire local images.

[0082] In this embodiment, a circular friction head is used to rub the fabric. A circular observation window is provided in the center of the circular friction head. The diameter of the observation window is 1 / 3 of the size of the friction surface, and the observation window is covered with a transparent material, which is sapphire, to ensure its optical clarity and mechanical durability during continuous friction. Real-time local image acquisition of the friction contact area is performed through the observation window. In some embodiments, the transparent material can also be high-hardness optical glass. In some embodiments, the observation window can also be other shapes, with a size smaller than 1 / 3 of the size of the friction surface.

[0083] like Figures 2-3 The friction head includes a friction part A1 for contacting and rubbing against the fabric, an observation window A2 opened on the friction part A1 for observing the friction of the fabric, and a support part A3 for supporting the friction part A1. In this embodiment, the observation window A2 is covered with transparent frosted glass to press down the fabric at the position of the observation window A2 during friction. The support part A3 is connected to a three-coordinate gantry frame, and the movement of the three-coordinate gantry frame drives the friction head to move, thereby achieving friction on the fabric.

[0084] In step S2, the continuous local image sequence is input into the no-reference visual pilling evaluation model to obtain the process pilling level;

[0085] The no-reference visual pilling assessment model is based on the visual Transformer architecture. It extracts and encodes multi-scale features from a single key frame and directly outputs a continuous evaluation value of the pilling process at that moment.

[0086] Based on the continuous evaluation values ​​of the process output in time series, process evolution information reflecting the dynamic evolution of pilling and fuzzing is generated. The process evolution information includes the dynamic evolution curve of pilling and fuzzing, as well as the initial deterioration point, performance jump, deterioration rate, final stability level, process uniformity assessment, and process parameter-performance response relationship obtained from the dynamic evolution curve of pilling and fuzzing.

[0087] like Figure 4 The working process of the no-reference visual pilling assessment model is as follows:

[0088] After preprocessing, the keyframe Ik' is used to obtain the preprocessed keyframe Sk'. The preprocessed keyframe Sk' is then input into the third ViT encoder to obtain the high-level semantic features Ek' of the friction fabric image during the friction process. The high-level semantic features Ek' of the friction fabric image during the friction process are then input into the ordered regression head to obtain the pilling level of the process.

[0089] The training method for a no-reference visual pilling assessment model includes the following steps:

[0090] (1) Obtain a training dataset, which includes a sequence of continuous keyframes collected from the friction process of different fabrics, and a label of the process pilling level corresponding to each keyframe; the process pilling level is a continuous numerical value or an ordered discrete level.

[0091] (2) Input the key frame into the no-reference visual pilling assessment model for training. The model is based on the visual Transformer architecture, extracts multi-scale features from a single key frame and encodes them, and outputs the predicted value of the process pilling level corresponding to that moment through an ordered regression head.

[0092] (3) Calculate the difference between the predicted value of the process pilling level output by the model and the corresponding labeled level in the training dataset. If the difference is within the preset allowable range, the model is determined to meet the accuracy requirements and the output is a usable no-reference visual pilling evaluation model.

[0093] (4) If the difference does not meet the preset allowable range, the model parameters are optimized and the model is retrained;

[0094] (5) If the requirements are still not met after retraining, the sample size of the training dataset is increased, and more key frame samples of fabric type, friction stage and environmental conditions are added. Steps (2) to (4) are repeated until the model output meets the accuracy requirements.

[0095] The dynamic evolution curve of pilling and fuzzing visually displays the complete deterioration trajectory from start to finish. It determines whether the fabric deteriorates slowly and evenly or fails suddenly and rapidly, allowing for the matching of different process improvement strategies for different failure modes.

[0096] The initial deterioration point is the time when detectable pilling and fuzzing begin to appear on the fabric. This evaluates the initial abrasion resistance of the fabric and guides improvements to its initial surface structure.

[0097] A performance jump is the point in time when the slope of the rating curve changes abruptly. This may correspond to the wear and tear of the finishing agent protecting the fibers, the large-scale shedding of fibers with poor cohesion, or the irreversible loosening of the fabric structure.

[0098] Degradation rate, the slope of the curve at different stages. Quantifying the rate of degradation helps compare the durability differences between different formulations or processes.

[0099] The final stability level determines whether the fabric has completely deteriorated to a state of equilibrium.

[0100] Process uniformity assessment, by analyzing keyframe images at different time points, allows observation of whether pilling and fuzzing occur uniformly or locally concentrated on the sample surface. Localized concentration may indicate unevenness defects in spinning, weaving, or finishing processes, providing precise direction for production line process adjustments.

[0101] The process parameter-performance response relationship involves correlating the aforementioned dynamic parameters (such as jump time and final stable value) with the fabric production process parameters (such as twist, density, and additive concentration). Data models can be established to predict the impact of process adjustments on the dynamic curve, thereby enabling data-driven, precise reverse engineering and optimization of the process, rather than blind trial and error.

[0102] The non-reference visual pilling assessment model supports dynamic rating during continuous friction processes and is suitable for real-time monitoring on production lines.

[0103] S3. After the friction is completed, the image of the rubbed fabric sample is acquired to obtain the image of the rubbed fabric.

[0104] In this embodiment, a 60-megapixel lens is used to acquire images of the friction fabric.

[0105] S4. Input the original fabric image, process evolution information and friction fabric image into the pilling evaluation model to obtain the ordered pilling level.

[0106] like Figure 5 The working process of the pilling and fuzzing assessment model is as follows:

[0107] (1) Spatial alignment and multi-scale feature extraction layer

[0108] The friction fabric image is registered using a spatial transformation network. Then, the original fabric image and the registered friction fabric image are divided into multi-scale regions and multi-feature extraction is performed, including grayscale, spatial domain and frequency domain features, so as to construct the multi-scale feature sequence of the original fabric image and the multi-scale feature sequence of the friction fabric image.

[0109] (2) Weight-sharing twin ViT encoder

[0110] The first ViT encoder and the second ViT encoder are used to process the multi-scale feature sequence of the original fabric image and the multi-scale feature sequence of the friction fabric image respectively, and output the high-level semantic features of the original fabric image and the high-level semantic features of the friction fabric image; the weights of the first ViT encoder and the second ViT encoder are the same.

[0111] (3) Differential Cross Attention Module

[0112] The high-level semantic features of the original fabric image, the process evolution information, and the high-level semantic features of the friction fabric image are input into the differential cross-attention module to obtain the final difference features.

[0113] (4) Adaptive gating module

[0114] The final difference features and the high-level semantic features of the friction fabric image are input into the adaptive gating module for feature fusion and weight allocation to obtain the pilling level.

[0115] (5) Input the pilling grade into the ordered regression head to obtain the ordered pilling grade.

[0116] like Figure 6 The working process of the differential cross-attention module is as follows:

[0117] (1) Feature stitching: The high-level semantic features of the original fabric image, the high-level semantic features of the friction fabric image, and the process evolution information are stitched together to obtain stitched features;

[0118] (2) Linear projection: The spliced ​​features are mapped to query vector, key vector and value vector respectively through a linear projection layer;

[0119] (3) Multi-head attention calculation: The query vector, key vector and value vector are input into the multi-head attention model for calculation to obtain attention output features; the multi-head attention model realizes differentiated interaction and fusion between features by parallel calculation of multiple attention heads and aggregating their results;

[0120] (4) Residual connection: The attention output feature and the splicing feature are residually connected to obtain the final difference feature.

[0121] In the differential cross-attention module, a multi-head attention mechanism is employed, and its structure includes:

[0122] (1) Feature stitching: The high-level semantic features of the original fabric image, the high-level semantic features of the friction fabric image, and the process evolution information are stitched together to obtain stitched features;

[0123] (2) Linear projection: The spliced ​​features are mapped to query vector Q, key vector K and value vector V through a linear projection layer.

[0124] (3) The attention calculation process can be expressed as:

[0125] Attention(Q,K,V)=softmax(QK T / ((d k ) 0.5 ))V;

[0126] in:

[0127] d k is the dimension of the key vector, used for scaling to avoid gradient vanishing;

[0128] Softmax is used to normalize attention weights;

[0129] The output is the weighted aggregated feature representation.

[0130] To enhance the model's representational power, multi-head attention can be employed.

[0131] MultiHead(Q,K,V)=Concat(head1,head2,……,head i ,……,head h );

[0132] Among them, head i The result of the calculation for the i-th attention head is given, and h is the number of attention heads (usually set to 8 or 16).

[0133] Each attention head is calculated independently:

[0134] head i =Attention(QW(i,Q),KW(i,K),VW(i,V));

[0135] in:

[0136] W(i,Q), KW(i,K), and VW(i,V) are learnable projection matrices.

[0137] The differential attention model is not trained independently; instead, the differential attention module is trained jointly with the pilling and fuzzing assessment model.

[0138] The pilling and fuzzing assessment model is trained through the following steps:

[0139] (1) Obtain the training dataset, which includes several original fabric images, corresponding friction fabric images, and corresponding pilling and fuzzing grades;

[0140] (2) The training dataset was trained using the pilling and fuzzing assessment model;

[0141] If the difference between the output pilling grade and the corresponding pilling grade in the training dataset meets the requirements, then the output is a usable pilling evaluation model.

[0142] If the difference between the output pilling grade and the corresponding pilling grade in the training dataset does not meet the requirements, the parameters are optimized and retrained. If the requirements are still not met, the sample size is increased and training continues until the requirements are met.

[0143] In step S2, the continuous local image sequence is input into the no-reference visual pilling evaluation model to obtain process evolution information;

[0144] The number of lint balls on the fabric sample after 30 seconds of friction was calculated to be 12, with an average diameter of 0.3 mm and an area ratio of 5%.

[0145] In this embodiment, the fabric sample used is a knitted cotton fabric sample (compliant with GB / T 4802.1 standard). The friction head rubs the knitted cotton fabric sample at a set pressure and speed, and the friction cycle is set to 60s. GB / T 4802.1 specifies that the pressure the sample withstands is 9 kPa (for synthetic fiber fabrics) or as specified in the product standard.

[0146] In this embodiment, the fabric sample used is a knitted cotton fabric sample (compliant with GB / T 4802.1 standard). The friction head is circular with a diameter of 25 mm, and its friction surface is covered with standard wool fabric. The friction head is pressed against the sample surface with a constant normal pressure F_normal = 4.42 N (corresponding to pressure P = 9 kPa), and driven to move at a constant speed along a circular trajectory with a fixed diameter of 40 mm, with a linear velocity v = 0.126 m / s (orbital speed n = 60 rpm). The total cycle of the friction treatment is set to t = 60 s.

[0147] In this embodiment, the illumination brightness is 3000 lux and the exposure time for image acquisition is 0.5 s;

[0148] The pilling detector method based on this invention achieves 8 hours of continuous observation (288 snapshots in total), can identify pills ≥0.1mm in size (verified by ISO-16665 resolution plate), and the friction interference rate (pressure fluctuation) caused by observation is <±0.7% (better than ASTM-D3512 requirements). The data analysis module's processing speed meets real-time requirements, with a single image processing time of <0.3s.

[0149] The method of this invention runs on an industrial computer (configured with an Intel Core i7 processor and 8GB of memory) and employs an OpenCV-based image processing algorithm, including functions such as image preprocessing, pom-pom recognition, and feature extraction.

[0150] The above method was used to rub fabric sample 1 (twill brushed fabric) and fabric sample 2 (coarse woolen fabric) respectively, and process evolution information and orderly pilling level were obtained respectively.

[0151] Figure 7 The images shown are actual pictures of the pilling test of fabric sample 1. Among them, a is the original picture without wear, b is the pilling friction after 100 times and wear level 4, c is the pilling friction after 200 times and wear level 4, d is the pilling friction after 300 times and wear level 4, e is the pilling friction after 400 times and wear level 4, f is the pilling friction after 500 times and wear level 4, and g is the pilling friction after 600 times and wear level 4.

[0152] Depend on Figure 8 It can be seen that the failure mode of fabric sample 1 is slow and uniform deterioration; the initial deterioration point is when the pilling occurs 100 times, and the final stability level is wear level 4.

[0153] Fabric Sample 1 Process Uniformity Assessment: Based on keyframe image analysis, the fabric showed uniform pilling throughout the entire wear process with no obvious local concentration, indicating good uniformity in spinning, weaving, and finishing processes.

[0154] Fabric sample 1 deteriorated slowly and uniformly. The fabric exhibited good uniformity and stable abrasion resistance. After initial deterioration, it remained at abrasion level 4, with no obvious accelerated failure.

[0155] like Figure 9 The images shown are actual pictures of the pilling test of fabric sample 2. Among them, a is the original picture without wear, b is the pilling friction after 100 cycles and wear level 3.3, c is the pilling friction after 200 cycles and wear level 3.5, d is the pilling friction after 300 cycles and wear level 3.5, e is the pilling friction after 400 cycles and wear level 3.5, f is the pilling friction after 500 cycles and wear level 3, and g is the pilling friction after 600 cycles and wear level 3.

[0156] Depend on Figure 10 It can be seen that the failure mode of fabric sample 2 is: sudden accelerated failure; the performance jump point is between 400-500 cycles of pilling; the final stability level is wear level 3.

[0157] Fabric Sample 2 Process Uniformity Assessment: Based on the data, it is inferred that the fabric exhibits a performance jump between 400-500 napping cycles, with the abrasion level dropping from 3.5 to 3, suggesting possible localized concentrated abrasion or wear through the protective layer.

[0158] Fabric sample 2 showed a performance jump between 400-500 napping cycles, with the abrasion level dropping from 3.5 to 3, and stabilizing at 3 after 600 cycles, suggesting possible uneven processing or wear through the protective layer.

[0159] Different improvement strategies are matched for different failure modes: fabric sample 1 needs to improve initial abrasion resistance, and fabric sample 2 needs to prevent performance jumps and improve uniformity.

[0160] The analysis of the process parameters-performance response relationship for fabric sample 1 and fabric sample 2 is shown in Table 1.

[0161] Table 1. Analysis of the Relationship between Process Parameters and Performance Response

[0162] Dynamic parameters Fabric Sample 1 Values Fabric Sample 2 Values Related process parameters Recommendations for process improvement Initial deterioration point 100 times of napping Raised 300 times ago Finishing agent concentration, fiber cohesion Increasing the initial finishing agent concentration enhances fiber cohesion. Performance breakpoint No obvious jump Fuzzing takes between 400-500 times Finishing agent durability and fabric structure stability Optimize finishing agent formulation to improve fabric structure density rate of deterioration 0.0 grade / 100 times of napping 0.5 level / 100 times of fuzzing (jump stage) yarn twist, fabric density Appropriately increase yarn twist and fabric density Final stability level Wear level 4 Wear level 3 Fiber strength, finishing process High-strength fibers are selected, and the finishing process is optimized.

[0163] Improvement strategies for "slow and uniform deterioration":

[0164] Optimize fiber material selection to improve initial abrasion resistance; improve the uniformity of finishing processes to prevent local defects; enhance fiber cohesion to delay fuzzing; monitor the initial deterioration point and adjust process parameter thresholds.

[0165] Improvement strategies for the "sudden acceleration failure" mode:

[0166] Improve the durability of finishing agents to prevent the protective layer from wearing through; optimize the fabric structure to prevent irreversible loosening; strengthen the control of process uniformity to avoid localized concentrated wear; set performance jump warnings to intervene in the process in advance.

[0167] Data modeling and process optimization recommendations:

[0168] Based on the correlation analysis between dynamic parameters and process parameters, the following data model can be established:

[0169] Process parameters - performance response model:

[0170] Performance grade = F(twist, density, additive concentration, fiber strength, ...);

[0171] By establishing a mathematical model through regression analysis, the impact of process adjustments on dynamic curves can be predicted, enabling precise reverse engineering of processes.

[0172] Process optimization steps:

[0173] (1) Identify failure modes (slow and uniform deterioration / sudden accelerated failure);

[0174] (2) Determine the key dynamic parameters (initial deterioration point, jump point, deterioration rate);

[0175] (3) Associate relevant process parameters (spinning, weaving, finishing);

[0176] (4) Establish a data model to predict the effects of process adjustments;

[0177] (4) Implement process optimization and verify the improvement effect.

[0178] In some embodiments, after step S4 is completed, step S5 is performed, as follows:

[0179] The new fabric sample is moved to the friction station and the steps S1-S4 are repeated to obtain the process evolution information and orderly pilling level of the new fabric sample, thereby automating and continuously performing friction detection on multiple fabric samples.

[0180] Example 2

[0181] like Figure 11 A pilling and fuzzing rating device based on visual continuous regression, comprising:

[0182] The first acquisition module 1 is used to perform global image acquisition on the fabric sample and obtain the original fabric image.

[0183] The second acquisition module 2 is used to acquire real-time local images of the friction contact area while the fabric sample is being rubbed, thereby obtaining a continuous local image sequence; and to obtain process evolution information based on the continuous local image sequence.

[0184] The third acquisition module 3 is used to acquire images of the fabric sample after friction at the friction station after the friction is completed, and obtain the friction fabric image.

[0185] Evaluation module 4 is used to input the original fabric image, process evolution information and friction fabric image into the pilling evaluation model to obtain the ordered pilling level.

[0186] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0187] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0188] Accordingly, this application also provides an electronic device, including:

[0189] One or more processors;

[0190] Memory, used to store one or more programs;

[0191] When the one or more programs are executed by the one or more processors, the one or more processors perform the methods described above.

[0192] Accordingly, this application also provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of any of the above methods.

[0193] In the embodiments provided in this application, it should be understood that the disclosed methods and systems can also be implemented in other ways. The method and system embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0194] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0195] On the other hand, a computer-readable storage medium stores computer instructions thereon, which, when executed by a processor, implement the steps of the above-described method. When the computer program is executed by the processor, it implements the method as described in any of the first aspects above. If the function is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0196] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method of pilling rating based on visual continuous regression, characterized in that, Includes the following steps: S1. Perform global image acquisition on the fabric sample to obtain the original fabric image; S2. When the fabric sample is rubbed, real-time local images of the friction contact area are acquired to obtain a continuous local image sequence; process evolution information is obtained based on the continuous local image sequence. S3. After the friction is completed, the image of the rubbed fabric sample is acquired to obtain the image of the rubbed fabric. S4. Input the original fabric image, process evolution information and friction fabric image into the pilling evaluation model to obtain the ordered pilling level. In step S2, the continuous local image sequence is input into the no-reference visual pilling evaluation model to obtain process evolution information; The no-reference visual pilling assessment model is based on the visual Transformer architecture. It extracts and encodes multi-scale features from key frames in a continuous local image sequence and outputs continuous evaluation values ​​of the pilling process corresponding to each key frame. Based on the continuous evaluation values ​​of the pilling process output in time series, process evolution information reflecting the dynamic evolution of pilling is generated. The process evolution information includes the dynamic evolution curve of pilling, as well as the initial deterioration point, performance jump, deterioration rate, final stability level, process uniformity assessment, and process parameter-performance response relationship obtained from the dynamic evolution curve of pilling. In step S4, the working process of the pilling and fuzzing evaluation model is as follows: (1) The friction fabric image is registered by a spatial transformation network. Then, the original fabric image and the registered friction fabric image are divided into multi-scale regions and multi-feature extraction is performed, including gray-scale, spatial domain and frequency domain features, so as to construct the multi-scale feature sequence of the original fabric image and the multi-scale feature sequence of the friction fabric image. (2) The first ViT encoder and the second ViT encoder are used to process the multi-scale feature sequence of the original fabric image and the multi-scale feature sequence of the friction fabric image respectively, and output the high-level semantic features of the original fabric image and the high-level semantic features of the friction fabric image; the weights of the first ViT encoder and the second ViT encoder are the same. (3) Input the high-level semantic features of the original fabric image, the process evolution information, and the high-level semantic features of the friction fabric image into the differential cross-attention module to obtain the final difference features; (4) Input the final difference features and the high-level semantic features of the friction fabric image into the adaptive gating module for feature fusion and weight allocation to obtain the pilling level; (5) Input the pilling grade into the ordered regression head to obtain the ordered pilling grade.

2. The method of claim 1, wherein, The working process of the differential cross-attention module is as follows: (1) Obtain a first key vector and a first value vector by linear projection of the high-level semantic features of the original fabric image; obtain a query vector by linear projection of the high-level semantic features of the friction fabric image; obtain a second key vector and a second value vector by linear projection of the process evolution information; (2) Input the first key vector, the second key vector and the query vector into the attention model to calculate the attention score and obtain the differential attention weight that incorporates the process evolution; (3) Perform feature weighting aggregation of the differential attention weights with the first value vector and the second value vector to obtain differential features; (4) The high-level semantic features of the original fabric image, the high-level semantic features of the friction fabric image, and the process evolution information are spliced ​​together, and then nonlinear transformation is performed through a multilayer perceptron to obtain fused features; (5) Perform residual connection between the differential features and the fusion features to obtain the final difference features.

3. The method of claim 1, wherein, The pilling and fuzzing assessment model is trained through the following steps: (1) Obtain the training dataset, which includes several original fabric images, corresponding friction fabric images, and corresponding pilling and fuzzing grades; (2) Use the pilling assessment model to train the training dataset; if the difference between the output pilling level and the corresponding pilling level in the training dataset meets the requirements, the output is a usable pilling assessment model; if the difference between the output pilling level and the corresponding pilling level in the training dataset does not meet the requirements, the parameters are optimized and retrained. If the requirements are still not met, the sample size is increased for training until the requirements are met.

4. The method of claim 1, wherein, After step S4 is completed, proceed to step S5, as follows: Move the new fabric sample to the friction station and repeat steps S1-S4 to obtain process evolution information and ordered pilling level of the new fabric sample.

5. A pilling rating device based on visual continuous regression, characterized by, include: The first acquisition module is used to perform global image acquisition on the fabric sample and obtain the original fabric image. The second acquisition module is used to acquire real-time local images of the friction contact area when the fabric sample is rubbed, and obtain a continuous local image sequence; and obtain process evolution information based on the continuous local image sequence. A continuous sequence of local images is input into a no-reference visual pilling assessment model to obtain information on the process evolution. The no-reference visual pilling assessment model is based on the visual Transformer architecture. It extracts and encodes multi-scale features from key frames in a continuous local image sequence and outputs continuous evaluation values ​​of the pilling process corresponding to each key frame. Based on the continuous evaluation values ​​of the pilling process output in time series, process evolution information reflecting the dynamic evolution of pilling is generated. The process evolution information includes the dynamic evolution curve of pilling, as well as the initial deterioration point, performance jump, deterioration rate, final stability level, process uniformity assessment, and process parameter-performance response relationship obtained from the dynamic evolution curve of pilling. The third acquisition module is used to acquire images of the fabric sample after friction, and obtain images of the friction fabric after friction. The evaluation module is used to input the original fabric image, process evolution information and friction fabric image into the pilling evaluation model to obtain the ordered pilling level; The working process of the pilling and fuzzing assessment model is as follows: (1) The friction fabric image is registered by a spatial transformation network. Then, the original fabric image and the registered friction fabric image are divided into multi-scale regions and multi-feature extraction is performed, including gray-scale, spatial domain and frequency domain features, so as to construct the multi-scale feature sequence of the original fabric image and the multi-scale feature sequence of the friction fabric image. (2) The first ViT encoder and the second ViT encoder are used to process the multi-scale feature sequence of the original fabric image and the multi-scale feature sequence of the friction fabric image respectively, and output the high-level semantic features of the original fabric image and the high-level semantic features of the friction fabric image; the weights of the first ViT encoder and the second ViT encoder are the same. (3) Input the high-level semantic features of the original fabric image, the process evolution information, and the high-level semantic features of the friction fabric image into the differential cross-attention module to obtain the final difference features; (4) Input the final difference features and the high-level semantic features of the friction fabric image into the adaptive gating module for feature fusion and weight allocation to obtain the pilling level; (5) Input the pilling grade into the ordered regression head to obtain the ordered pilling grade.

6. An electronic device, comprising: include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-4.

7. A computer readable storage medium having stored thereon computer instructions, wherein, When executed by the processor, this instruction implements the steps of the method as described in any one of claims 1-4.

Images