Intelligent evaluation method suitable for shrinkage rate of complex pattern fabric

Abstract

The invention discloses an intelligent evaluation method suitable for the shrinkage rate of a complex pattern fabric. The method comprises the following steps: (1) acquiring a fabric image; (2) preprocessing the fabric image; (3) model design: designing an enhanced deep neural network target detection algorithm model which comprises a small target detection layer, a multi-scale attention module and a bidirectional feature pyramid network structure, and replacing a loss function with an MPDIOU regression loss function; (4) model training; and (5) detecting the shrinkage rate of the fabric: visually displaying the position of the mark point on the fabric image, and then calculating the position change of the mark point of the fabric image before and after washing to obtain the shrinkage rate. According to the invention, the shrinkage rate of the textile fabric before and after the treatment flows of washing, drying, ironing and the like can be accurately measured, the identification, measurement and data recording of sample marks can be automatically completed, the interference of human factors is reduced, and the detection speed and accuracy are improved, so that the quality detection requirements of the textile industry are better met.

Description

Technical Field

[0001] The invention belongs to the field of textile fabric detection, and particularly relates to an intelligent evaluation method for shrinkage rate of complex pattern fabrics. Background Art

[0002] Today, with the booming textile industry, as the global market continues to improve its requirements for textile quality, especially for complex pattern fabrics, the precise control of shrinkage has gradually become one of the key indicators for measuring product quality. Complex pattern fabrics, due to their diverse patterns and rich colors, have more complex and unpredictable dimensional changes during washing, ironing and drying, which directly affects the aesthetics, durability, and fit of finished products such as clothing, accessories, and home textiles. At present, testing institutions and enterprises mostly use manual measurement to evaluate the shrinkage of fabrics. On the one hand, manual operation and contact measurement with the fabric can easily destroy the surface morphology of the fabric, resulting in a decrease in accuracy; on the other hand, the visual method is affected by the pattern and color of the fabric, resulting in inaccurate judgment and large errors; in addition, manual measurement is time-consuming and labor-intensive, and it is difficult to meet the large-scale and wide-ranging testing needs. Therefore, there is an urgent need for a non-contact, fast, and accurate automatic measurement method to overcome the many difficulties in current shrinkage detection.

[0003] Traditional template matching methods work well in solid-color fabric detection, but for fabrics with complex patterns on the surface, the influence of lighting, and the small size of the markers, it is difficult to use effective similarity indicators and determine appropriate thresholds for judgment in template matching. Deep learning can overcome the limitations of traditional algorithms in complex environments by actively learning the feature representation in the data, and can accurately capture subtle changes on the fabric surface. This type of algorithm can not only process large amounts of data in real time, but also flexibly adjust model parameters according to different detection scenarios and task requirements, providing a solid technical foundation for the intelligent and automated detection of fabric shrinkage.

[0004] At present, the testing standards for fabric shrinkage, such as ISO 3759:2011, GB / T 8628-2013, GB / T38006-2019, FZ / T 20010-2012, BS EN ISO 5077:2008, and AATCC 135-20018, all stipulate that the testers should conduct manual measurements, and no objective testing methods have been adopted. They need to be further revised to meet the testing needs of the current industry. Among the published patents, in terms of shrinkage testing equipment design, the patent with announcement number CN219122209U discloses a fully automatic shrinkage tester with a built-in shock-absorbing mechanism combined with multiple sensors and intelligent control to achieve efficient and accurate testing, and has abnormal alarm and remote monitoring functions to improve the level of intelligence. The patent with the announcement number CN220399290U discloses an automatic fabric shrinkage measuring instrument, which uses a negative pressure fan and a visual camera to accurately measure fabrics without a fixture, and is equipped with a shading mechanism to optimize the visual environment and enhance the ranging accuracy. The patent with the announcement number CN117330349B discloses a shrinkage sampling device and method for textile production, which integrates cutting and sampling units for efficient sampling but does not involve shrinkage measurement. The patent with the publication number CN118134418A discloses a product intelligent mold parting method and its product intelligent mold parting system to improve the accuracy and efficiency of shrinkage detection in textile production and mold design. In terms of measurement methods, the patent with the publication number CN114739999A discloses a high-precision fabric shrinkage test method, which uses image contrast and rotation processing to reduce human interference and improve test accuracy. The patent with the publication number CN106767484A discloses a method, device and system for automatically detecting fabric size change rate, which uses the SIFT algorithm to automatically match fabric images and accurately calculate the size change rate. The patent with publication number CN117871832A discloses a textile shrinkage qualified rate detection method based on dual data model processing, which integrates dual data models, comprehensively evaluates fabric shrinkage, and improves detection accuracy and efficiency. The patent with publication number CN116008513A discloses a knitted fabric shrinkage prediction method, which uses a linear regression model to quickly predict shrinkage, shortens detection time, and meets production needs. These innovations have jointly promoted the intelligent and efficient process of textile shrinkage detection.

[0005] Although the existing patented design technology has significantly improved the efficiency and accuracy in shrinkage detection, it still has defects such as reliance on manual operation, susceptibility to environmental interference, and some designs have not fully realized automated measurement and data processing. In addition, since the current detection technology still mainly involves manual marking and manual measurement, the entire process includes multiple steps such as marking, fabric processing, measurement and data recording, which takes a long time and affects the detection efficiency. At the same time, the accuracy of manual marking and measurement is greatly affected by the operator's experience and technical level. There may be large test differences between different personnel, which affects the accuracy and repeatability of the test results. Therefore, there is an urgent need for a fast, objective and intelligent detection method that can replace manual labor, which can automatically complete the identification, measurement and data recording of sample markings, reduce the interference of human factors, and improve the detection speed and accuracy, so as to better meet the quality inspection needs of the textile industry. Summary of the invention

[0006] The purpose of the present invention is to provide an intelligent evaluation method for the shrinkage rate of complex pattern fabrics, which aims to accurately measure the shrinkage rate of textile fabrics before and after washing, drying, ironing and other processing processes, and can automatically complete the identification, measurement and data recording of sample marks, reduce the interference of human factors, improve the detection speed and accuracy, so as to better meet the quality inspection needs of the textile industry.

[0007] In order to solve the above technical problems, the following technical solutions are adopted:

[0008] An intelligent evaluation method for shrinkage of complex pattern fabrics, characterized by comprising the following steps:

[0009] (1) Fabric image acquisition;

[0010] (2) Fabric image preprocessing;

[0011] (3) Model design: Design an enhanced deep neural network target detection algorithm model, which includes a small target detection layer, a multi-scale attention module, and a bidirectional feature pyramid network structure, and replaces the loss function with the MPDIoU regression loss function;

[0012] (4) Model training:

[0013] 1) Image labeling: label the acquired image dataset with marker points and divide the image dataset into training set, validation set and test set;

[0014] 2) Model training: Implement hyperparameter fine-tuning strategy, first adjust the training set and validation set data samples to the preset size specifications, then input them into the model for training, obtain the best performance model in the training process, and use the test set to verify the model detection effect;

[0015] (5) Fabric shrinkage detection: The position of the marking points is visually displayed on the fabric image. Then, the shrinkage rate is obtained by calculating the position change of the marking points on the fabric image before and after washing.

[0016] After optimization, the step (1) uses an image acquisition device to acquire fabric images, and the image acquisition device includes a cloth loading platform, a camera support frame, a digital camera and an LED light source. The cloth loading platform is used to place the fabric, the camera support frame is used to install the digital camera, the digital camera is connected to a computer via a connecting line, and the LED light source is used as a lighting source to surround the digital camera.

[0017] After optimization, the step (2) of fabric image preprocessing includes effective position segmentation and image enhancement.

[0018] Preferably, the effective position segmentation includes: batch cutting out target areas of the fabric image.

[0019] After optimization, image enhancement includes: first, Gaussian filtering to reduce image noise; then, sharpening processing to enhance the edges and details of the image; and finally, brightness enhancement to improve the brightness of the image.

[0020] After optimization, in step (3), in terms of feature extraction, the multi-scale attention module is seamlessly integrated into the deep neural network backbone network to enhance feature fusion within the cervical segment; and three multi-scale attention modules are integrated into the cervical network to process the input feature map.

[0021] After optimization, in step (3), in terms of receptive field, the small target detection layer, as a new detection level, is located at a shallower position in the deep neural network architecture and has a higher spatial resolution.

[0022] After optimization, in step (3), in terms of feature pyramid, a bidirectional feature pyramid network structure is adopted, by establishing a bidirectional connection between the top-down and bottom-up paths, allowing information between features of different scales to flow and merge more effectively, so that feature information can flow bidirectionally between different scales.

[0023] After optimization, in the step (3), in the post-processing, the MPDIoU regression loss function is introduced to perform regression adjustment on the detected bounding box; MPDIoU is used as the bounding box regression loss function, as shown in formula (1-3), by minimizing the minimum point distance between the prediction and the true bounding box:

[0024]

[0025] Where A and B represent two predicted bounding boxes, Indicates the coordinates of the upper left and lower right corners of A, Represents the coordinates of the upper left and lower right corners of B, h represents the height of the input image, and w represents the width of the input image.

[0026] After optimization, the shrinkage calculation in step (5) includes:

[0027] 1) Shrinkage calculation: Calculate and output the shrinkage value based on the original distance of the marking point and the distance data after washing and protection; the calculation formula is:

[0028]

[0029] In the formula, is the shrinkage value, L1 is the initial distance before processing, L2 is the new distance after washing and care;

[0030] 2) Calculation of average value and standard deviation: Calculate the average value and standard deviation of the three shrinkage values;

[0031] 3) Shrinkage evaluation: The fabric shrinkage grade is evaluated according to the preset standards based on the fabric category.

[0032] The above technical solution has the following beneficial effects:

[0033] (1) Improve the efficiency and cost-effectiveness of fabric shrinkage detection: Traditional methods for evaluating fabric shrinkage often rely on cumbersome manual measurements and long natural or artificial accelerated tests, which are not only time-consuming and labor-intensive, but also accompanied by high costs. The present invention significantly reduces the detection time and cost through automated and intelligent image recognition technology. The method can quickly capture the dimensional changes of fabrics under different conditions, realize instant analysis and feedback, and greatly improve the detection efficiency. At the same time, due to the reduction of manual intervention, the repeatability and consistency of the evaluation process are ensured, thereby reducing the overall cost.

[0034] (2) Ensure the objectivity and stability of fabric shrinkage assessment, regardless of fabric appearance differences: When evaluating fabric shrinkage, traditional methods may be interfered by complex appearance factors such as fabric color, pattern, and texture, resulting in subjectivity and instability in the assessment results. However, the present invention uses deep learning technology to deeply analyze fabric images and effectively eliminate the interference of fabric appearance features, ensuring that no matter how bright the fabric color, how complex the pattern, or how varied the texture, accurate and stable mark point calibration can be achieved, providing a more objective and reliable basis for fabric quality control. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The present invention will be further described below in conjunction with the accompanying drawings:

[0036] Figure 1 It is a flowchart of intelligent detection of fabric shrinkage;

[0037] Figure 2 It is a structural schematic diagram of a fabric image acquisition device;

[0038] Figure 3 for Figure 2 Schematic diagram of the dotted line part;

[0039] Figure 4 A diagram of the network structure for enhancing the deep neural network object detection algorithm;

[0040] Figure 5 This is the structural diagram of the multi-scale attention module;

[0041] Figure 6 This is the structural diagram of the small target detection layer;

[0042] Figure 7 It is a bidirectional feature pyramid network structure diagram;

[0043] Figure 8 Schematic diagram of intelligent positioning of marker points.

[0044] The accompanying drawings are marked as follows: cloth carrier 1, camera support frame 2, LED light source 3, digital camera 4, connecting line 5. DETAILED DESCRIPTION

[0045] The technical solution of the present invention is described in detail below in conjunction with specific embodiments:

[0046] An intelligent evaluation method for shrinkage rate of complex pattern fabrics, comprising the following steps:

[0047] (1) Fabric image acquisition: A set of optimized image acquisition devices is used to focus on the clear capture of fabric surface morphology and marking points. The device mainly includes the following components:

[0048] 1) Fabric carrier 1: Use high-precision materials, such as flat glass, acrylic plate or special plate, to maintain the high flatness and shape stability of the fabric during the data collection stage to prevent any deformation.

[0049] 2) Adjustable camera support 2: The main body is made of metal and supports dual adjustment of height and angle to ensure that the camera lens is precisely vertical to the fabric surface. The bottom is integrated with an efficient shockproof pad to reduce the potential interference of external vibration on the imaging quality.

[0050] 3) High-performance digital camera 4: Equipped with a high resolution of more than 5 million pixels, it can show rich details of fabrics. It integrates automatic focus and exposure optimization technology to flexibly cope with various lighting environments.

[0051] 4) Connection cable 5: Through the USB high-speed interface or stable Ethernet cable, the camera and computer can be seamlessly connected to achieve instant transmission and efficient storage management of image data.

[0052] 5) Intelligent dimming LED light source 3: Uses brightness-adjustable LED as the lighting source around the camera head, aiming to create a uniform lighting environment without blind spots and suppress shadows and reflections.

[0053] (2) Fabric image preprocessing;

[0054] The acquired images are effectively segmented and enhanced to ensure image clarity and improve subsequent processing efficiency.

[0055] 1) Effective position segmentation: crop the target area in batches to reduce calculation and error and improve efficiency.

[0056] 2) Image enhancement: First, Gaussian filtering effectively reduces image noise; then, sharpening is performed to enhance the edges and details of the image, making the image clearer; finally, brightness enhancement further increases the brightness of the image and improves the visual effect. By combining Gaussian filtering denoising, image sharpening, and brightness enhancement technology, the acquired fabric image is fully optimized, while maintaining the characteristics of the fabric itself to the greatest extent, the accuracy of marker point recognition is improved.

[0057] (3) Model design: In view of the specific needs of fabric shrinkage detection, the deep neural network target detection algorithm is deeply customized and optimized, which is suitable for high-precision marker point positioning of complex printed and embroidered fabrics. An enhanced deep neural network target detection algorithm model (YOLOv8) is designed, which includes a small target detection layer (P2), a multi-scale attention module (EMA) and a bidirectional feature pyramid network structure (BiFPN), and the loss function is replaced by the MPDIoU regression loss function;

[0058] 1) In terms of feature extraction, EMA is seamlessly integrated into the deep neural network backbone network to enhance feature fusion within the neck segment, enabling the backbone network to prioritize channel features within the target area after extracting a large number of image features, while improving the ability to extract key features. In addition, three EMAs are integrated into the neck network to further process the input feature map. These EMAs emphasize the key information in the feature map and predict the detection results of features of different sizes, further enhancing the robustness and accuracy of landmark detection.

[0059] 2) In terms of receptive field, the target detection mechanism of the traditional YOLOv8 model focuses on the three deep feature maps P3, P4, and P5. In order to overcome the limited performance of marker point detection during downscaling detection, the P2 layer is introduced as a new detection layer, such as Figure 6As shown in the figure, the P2 layer is located at a shallow position in the network architecture and has a higher spatial resolution. This feature enables it to capture subtle features in the image more sensitively, especially for the detection of small-scale and deformed targets, which can significantly improve the accuracy and reliability of detection.

[0060] 3) In terms of feature pyramid, the bidirectional feature pyramid network structure (BiFPN) is used to replace the traditional path aggregation network (PANet), such as Figure 7 As shown in the figure, by establishing a bidirectional connection between the top-down and bottom-up paths, the information between features of different scales is allowed to flow and fuse more effectively, so that the feature information can flow bidirectionally between different scales, which enhances the effect of feature fusion, realizes multi-scale fusion and enhancement of features, and improves the model's detection performance for markers of different scales.

[0061] 4) In terms of post-processing, the MPDIoU regression loss function is introduced to regress the detected bounding box. As an innovative bounding box regression loss function, MPDIoU, as shown in formula (1-3), optimizes the accuracy and efficiency of bounding box regression by minimizing the minimum point distance between the predicted and true bounding boxes, making up for the deficiency of the traditional CIoU loss function that does not consider the influence of the center point distance and aspect ratio, and is expected to greatly improve the accuracy of marker point detection and model robustness.

[0062]

[0063]

[0064] Where A and B represent two predicted bounding boxes, Indicates the coordinates of the upper left and lower right corners of A, Represents the coordinates of the upper left and lower right corners of B, h represents the height of the input image, and w represents the width of the input image.

[0065] (4) Model training: Using a fabric dataset with different surface printing types and different marker point types, the enhanced deep neural network model is subjected to hyperparameter tuning and iterative training to ensure that the model can accurately identify and track key marker points on the fabric.

[0066] 1) Image labeling: annotate the acquired image dataset with marker points and divide the image dataset into training set, validation set and test set in a ratio of 7:1:2;

[0067] 2) Model training: After optimizing the model network architecture, implement the hyperparameter fine-tuning strategy, then adjust the training set and validation set data samples to the preset size specifications, and then input them into the improved network model for training, obtain the best performance model in the training process, and use the test set to verify the model detection effect.

[0068] (5) Fabric shrinkage detection: The position of the marking points is intuitively displayed on the image. Then, the shrinkage rate is obtained by calculating the position change of the marking points before and after washing. The average and standard deviation of three measurements are taken to ensure accuracy, realizing intelligent measurement of fabric shrinkage rate, including:

[0069] 1) Shrinkage calculation: Calculate and output the shrinkage value based on the original distance of the marking point and the distance data after washing and protection; the calculation formula is:

[0070]

[0071] In the formula, is the shrinkage value, L1 is the initial distance before processing, L2 is the new distance after washing and care;

[0072] 2) Calculation of average value and standard deviation: Calculate the average value and standard deviation of the three shrinkage values;

[0073] 3) Shrinkage evaluation: The fabric shrinkage grade is evaluated according to the preset standards based on the fabric category.

[0074] The above are only specific embodiments of the present invention, but the technical features of the present invention are not limited thereto. Any simple changes, equivalent replacements or modifications made based on the present invention to solve basically the same technical problems and achieve basically the same technical effects are all included in the protection scope of the present invention.

Claims

1. An intelligent evaluation method for shrinkage of complex pattern fabrics, characterized in that The steps include: (1) Fabric image acquisition; (2) Fabric image preprocessing; (3) Model design: Design an enhanced deep neural network target detection algorithm model, which includes a small target detection layer, a multi-scale attention module, and a bidirectional feature pyramid network structure, and replaces the loss function with the MPDIoU regression loss function; (4) Model training: 1) Image labeling: label the acquired image dataset with marker points and divide the image dataset into training set, validation set and test set; 2) Model training: Implement hyperparameter fine-tuning strategy, first adjust the training set and validation set data samples to the preset size specifications, then input them into the model for training, obtain the best performance model in the training process, and use the test set to verify the model detection effect; (5) Fabric shrinkage detection: The position of the marking points is visually displayed on the fabric image. Then, the shrinkage rate is obtained by calculating the position change of the marking points on the fabric image before and after washing.

2. The intelligent evaluation method for shrinkage of complex pattern fabrics according to claim 1, characterized in that: In the step (1), an image acquisition device is used to acquire a fabric image. The image acquisition device includes a fabric carrier, a camera support frame, a digital camera and an LED light source. The fabric carrier is used to place the fabric, the camera support frame is used to install the digital camera, the digital camera is connected to a computer via a connecting line, and the LED light source is used as an illumination source and surrounds the digital camera.

3. The intelligent evaluation method for shrinkage of complex pattern fabrics according to claim 1, characterized in that: The step (2) of fabric image preprocessing includes effective position segmentation and image enhancement.

4. The intelligent evaluation method for shrinkage of complex pattern fabrics according to claim 3, characterized in that: The effective position segmentation includes: cutting out target areas of the fabric image in batches.

5. The intelligent evaluation method for shrinkage of complex pattern fabrics according to claim 3, characterized in that: Image enhancement includes: first, Gaussian filtering to reduce image noise; then, sharpening to enhance the edges and details of the image; finally, brightness enhancement to increase the brightness of the image.

6. The intelligent evaluation method for shrinkage of complex pattern fabrics according to claim 1, characterized in that: In the step (3), in terms of feature extraction, the multi-scale attention module is seamlessly integrated into the deep neural network backbone network to enhance the feature fusion within the cervical segment; and three multi-scale attention modules are integrated into the cervical network to process the input feature map.

7. The intelligent evaluation method for shrinkage of complex pattern fabrics according to claim 1, characterized in that: In step (3), in terms of receptive field, the small target detection layer, as a new detection level, is located at a shallower position in the deep neural network architecture and has a higher spatial resolution.

8. The intelligent evaluation method for shrinkage of complex pattern fabrics according to claim 1, characterized in that: In the step (3), in terms of the feature pyramid, a bidirectional feature pyramid network structure is adopted, by establishing a bidirectional connection between the top-down and bottom-up paths, allowing information between features of different scales to flow and merge more effectively, so that feature information can flow bidirectionally between different scales.

9. The intelligent evaluation method for shrinkage of complex pattern fabrics according to claim 1, characterized in that: In the step (3), in terms of post-processing, the MPDIoU regression loss function is introduced to perform regression adjustment on the detected bounding box; MPDIoU is used as the bounding box regression loss function, as shown in formula (1-3), by minimizing the minimum point distance between the prediction and the true bounding box: Where A and B represent two predicted bounding boxes, Indicates the coordinates of the upper left and lower right corners of A, Represents the coordinates of the upper left and lower right corners of B, h represents the height of the input image, and w represents the width of the input image.

10. The intelligent evaluation method for shrinkage of complex pattern fabrics according to claim 1, characterized in that: The step (5) of calculating the shrinkage rate comprises: 1) Shrinkage calculation: Calculate and output the shrinkage value based on the original distance of the marking point and the distance data after washing and protection; the calculation formula is: In the formula, is the shrinkage value, L1 is the initial distance before processing, L2 is the new distance after washing and care; 2) Calculation of average value and standard deviation: Calculate the average value and standard deviation of the three shrinkage values; 3) Shrinkage evaluation: The fabric shrinkage grade is evaluated according to the preset standards based on the fabric category.

Images