Textile bobbin intelligent quality control image analysis system

Abstract

The invention discloses an intelligent quality control image analysis system for textile bobbins, particularly relates to the field of textile bobbin monitoring, is used for solving the problems of uneven winding and yarn breakage caused by static accumulation, and is characterized in that a fusion image is generated through a multispectral imaging technology; after the convolutional neural network analyzes the image, an electrostatic accumulation area is automatically identified, the intensity, the rate and the spatial distribution characteristics are quantized, time sequence data are fused through an optical flow algorithm, and the electrostatic intensity level is dynamically updated; when the electrostatic strength reaches a set value, capturing a dynamic track and tension change in the yarn winding and releasing process by using a corresponding sensor, extracting yarn offset data, and evaluating the influence of electrostatic interference on winding uniformity and releasing smoothness in combination with tension anomaly analysis; and finally, after the electrostatic strength and the interference analysis result are integrated, a random forest algorithm is adopted to generate a risk score, and an early warning mechanism is triggered, so that the spinning process is optimized, yarn breakage is reduced, the stability and the product quality are improved, and the production efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of textile bobbin monitoring, and more specifically, to an intelligent quality control image analysis system for textile bobbins. Background Art

[0002] The spinning process is a core link in the textile industry, and its quality directly determines the performance of the final yarn and the quality of the fabric. In this process, the bobbin, as a key component for yarn winding, is subjected to the dual effects of high-speed contact and friction between fibers and the spinning machinery. This high-speed friction not only promotes the accumulation of static electricity on the surface of the bobbin but also causes dynamic changes in the internal temperature and stress of the bobbin. The phenomenon of static electricity accumulation appears as local high-charge-density regions on the surface of the bobbin. These regions will interfere with the arrangement of fibers and the uniform winding of the yarn, resulting in mutual repulsion between yarns and even triggering static electricity discharge phenomena. Seriously, static electricity accumulation may lead to yarn breakage, uneven winding, and equipment failures, significantly reducing production efficiency and product quality. In addition, the non-uniformity and dynamic change characteristics of static electricity accumulation make it affected by various factors such as the surface material of the bobbin, fiber material, environmental humidity, temperature, and mechanical parameters (such as rotational speed, pressure), further increasing the complexity and challenge of static electricity management. In actual production scenarios, the static electricity accumulation regions usually show local concentrated distribution, which is closely related to the microscopic structure of the bobbin surface. At the same time, its accumulation rate is highly correlated with the changes in mechanical parameters during the spinning process. These multi-dimensional interactions make it difficult to accurately evaluate and effectively control static electricity accumulation through traditional single measurement means. To solve the above problems, a technical solution is provided herein. Summary of the Invention

[0003] To overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a textile bobbin intelligent quality control image analysis system. The present invention synchronously collects images of the bobbin surface through a high-resolution multi-spectral industrial camera, combines the humidity and temperature data recorded by the environmental sensor, and uses an adaptive Gaussian filter and an illumination equalization algorithm based on the Retinex theory to denoise and correct the illumination of the image, generating a clear multi-spectral fused image; Subsequently, a multi-spectral convolutional neural network model is used to automatically identify and segment the electrostatic accumulation area of the fused image, quantify the electrostatic intensity, accumulation rate and spatial distribution characteristics, and fuse the electrostatic characteristics with the time series data through an optical flow algorithm to dynamically update the electrostatic intensity level to match the changes during the yarn running process; When the electrostatic intensity level is medium or high, use high-frame-rate dynamic images and tension sensor data to synchronously capture the movement trajectory and tension changes of the yarn winding and unwinding, apply a multi-target motion tracking algorithm to extract the dynamic trajectory deviation of the yarn, and combine with the tension anomaly analysis to accurately quantify the actual interference degree of electrostatic accumulation on the winding uniformity and unwinding smoothness; Finally, integrate the electrostatic intensity level and the interference analysis results, generate a comprehensive electrostatic risk score based on the random forest algorithm, and trigger an early warning mechanism according to the score result, dynamically adjust production parameters such as humidity, speed and tension management through an intelligent control module, optimize the spinning process, and prevent potential quality problems to solve the problems proposed in the above background technology.

[0004] To achieve the above object, the present invention provides the following technical solutions: A textile bobbin intelligent quality control image analysis system, comprising: A bobbin image correction module, an electrostatic feature segmentation module, a trajectory tension recognition module and a risk score early warning module.

[0005] The bobbin image correction module collects high-resolution images of the bobbin surface through a multi-spectral industrial camera, applies a Gaussian filter algorithm to remove noise, and corrects the image light intensity using an illumination equalization algorithm based on the Retinex theory, outputting a fused image after illumination correction.

[0006] The electrostatic feature segmentation module receives the corrected fused image and environmental variable data, uses a multi-spectral convolutional neural network model to segment the electrostatic accumulation area of the fused image, quantifies the electrostatic intensity, accumulation rate and spatial distribution characteristics, and fuses the electrostatic characteristics with the time series data through an optical flow algorithm to update the electrostatic intensity level, outputting electrostatic feature data.

[0007] The trajectory tension recognition module, after obtaining the corresponding electrostatic intensity level, uses high-frame-rate dynamic images and a tension sensor to synchronously capture the movement trajectory and tension changes of the yarn winding and release, applies a multi-target motion tracking algorithm to extract the dynamic trajectory deviation of the yarn, and combines tension anomaly analysis to identify the actual interference degree of electrostatic accumulation on the winding uniformity and release smoothness, and outputs the interference level.

[0008] The risk score warning module integrates the electrostatic intensity level, the interference level result, and environmental data, generates an electrostatic risk score based on the random forest algorithm, and triggers an early warning mechanism according to the score result, and outputs the electrostatic risk score and warning signal.

[0009] In a preferred embodiment, the operation process of the yarn bobbin image correction module includes the following: S1.1, Synchronously collect high-resolution images of the surface of the yarn bobbin in multiple bands through a multi-spectral industrial camera.

[0010] S1.2, The environmental sensor synchronously records the humidity and temperature of the environment where the yarn bobbin is located.

[0011] S1.3, Perform noise removal processing on the collected three-channel images respectively, using an adaptive Gaussian filter, and dynamically adjust the size and weight of the filter kernel according to the intensity distribution around each pixel point.

[0012] S1.4, Adopt an illumination equalization algorithm based on the Retinex theory, decompose each channel image into a reflection component and an illumination component, and reconstruct the image to enhance the contrast of the electrostatic area.

[0013] S1.5, Weightedly fuse the images of the visible light, ultraviolet light, and infrared light bands. The images of each channel are fused according to specific weights, and the finally generated fused image .

[0014] In a preferred embodiment, the operation process of the electrostatic feature segmentation module includes the following: S2.1, Use a multi-spectral convolutional neural network to analyze the fused image, and identify and segment the electrostatic accumulation area on the surface of the yarn bobbin.

[0015] S2.2, After obtaining the binary mask of the electrostatic accumulation area, quantify the electrostatic intensity and the accumulation rate .

[0016] A1. Electrostatic intensity quantization: Gray value mapping: Map the gray value of the electrostatic area in the fused image to the electrostatic intensity, using a non-linear mapping function: , where represents the pixel point The light intensity value.

[0017] Local contrast enhancement: Apply adaptive histogram equalization within the electrostatic region to highlight subtle intensity differences and obtain the electrostatic intensity value: , Indicates adaptive histogram equalization.

[0018] A2. Accumulation rate calculation: Time series analysis: Use the optical flow algorithm to capture the change in electrostatic intensity between consecutive frames and calculate the electrostatic accumulation rate : , where is the time interval, representing the reciprocal of the frame rate.

[0019] In a preferred embodiment, S2.3, analyze the spatial distribution characteristics of the electrostatic accumulation region and identify high-accumulation hotspots and regional distribution patterns; B1. Cluster analysis: K-Means clustering: Perform K-Means clustering on the coordinates of the electrostatic accumulation region and the electrostatic intensity to identify electrostatic accumulation hotspots with different densities: , where is the th cluster center, is the th data point.

[0020] B2. Distribution pattern recognition: Morphological feature extraction: Extract the morphological features of each cluster center, , where represents the area of the electrostatic region, represents the perimeter of the electrostatic region, represents the boundary set of the electrostatic region, extracted by the boundary detection algorithm in the binary mask . Specifically defined as: If the pixel and at least one adjacent pixel , then the corresponding pixel belongs to the boundary.

[0021] Density calculation: Calculate the electrostatic density of each cluster region, defined as the ratio of the electrostatic intensity to the area of the region: , where represents the electrostatic density, representing the average electrostatic intensity within the cluster region , represents the electrostatic region 's area.

[0022] In a preferred embodiment, in S2.4, the static electricity accumulation feature is fused with the time series data through an optical flow algorithm to dynamically update the static electricity intensity level .

[0023] C1. Optical flow calculation: Motion vector extraction: Use the optical flow algorithm to calculate the motion vectors of the static electricity accumulation regions between consecutive frames , reflecting the dynamic changes of static electricity accumulation: , where The optical flow motion vector represents the displacement speed of pixel points between consecutive time frames.

[0024] C2. Update of static electricity intensity level: Dynamic adjustment mechanism: Based on the static electricity intensity and the accumulation rate , define the static electricity intensity level : , where, is the dynamic adjustment coefficient.

[0025] C3. Classification of static electricity levels: Threshold division: According to the distribution of the static electricity intensity level , divide it into three levels: low, medium, and high: , where, is the low-medium boundary value of the static electricity intensity, is the medium-high boundary value of the static electricity intensity.

[0026] In a preferred embodiment, the operation process of the trajectory tension recognition module includes the following: S3.1, when the static electricity intensity level is medium or high, synchronously capture the real-time dynamic process of yarn winding and releasing through a high-frame-rate industrial camera and a tension sensor.

[0027] S3.2, use the multi-target motion tracking algorithm to extract the motion trajectory of the yarn during the winding and releasing process.

[0028] S3.3, combine the motion trajectory of the yarn with the tension sensor data to analyze the actual interference degree of static electricity accumulation on the yarn winding and releasing process.

[0029] D1. Tension anomaly detection: Anomaly definition: Define that the yarn tension fluctuation exceeding the normal range is a tension anomaly, and the calculation process is: , where: : Tension anomaly flag, : Tension value at the current moment, : Historical average tension, : Historical tension standard deviation.

[0030] D2. Trajectory offset calculation: Offset definition: The degree of deviation between the yarn movement trajectory and the ideal winding trajectory, quantifying the interference of static electricity on the yarn movement. Calculation process: , where: : The trajectory offset at the current position, : The actual yarn position coordinates, : The ideal winding position coordinates.

[0031] D3. Quantification of interference degree: Interference index calculation: , where: : The interference index, reflecting the influence degree of static electricity accumulation on the smoothness of yarn winding and release.

[0032] D4. Interference level classification: According to the interference index , it is divided into three levels: low, medium, and high: , where, The demarcation value between low interference and medium interference, is the demarcation value between medium interference and high interference.

[0033] In a preferred embodiment, the operation process of risk score warning includes the following: S4.1. First, integrate the static electricity intensity level and the interference level, combine with environmental parameters, and construct a comprehensive feature set for risk scoring; in order to capture the dynamic change trend, adopt the sliding time window technology to calculate the statistics of each feature in the past seconds; then, construct the interaction terms between features to capture the non-linear relationship; through multi-dimensional feature fusion, the recognition ability of the model for complex static electricity accumulation situations is enhanced, and all features are normalized.

[0034] S4.2. Apply the random forest algorithm to train and predict the normalized feature vectors; in the risk score generation stage, the random forest model outputs the risk probability at each time point to obtain the risk score; when the risk score exceeds the first threshold, trigger a medium warning and notify the operator to check; when the risk score exceeds the second threshold, trigger a high alarm and start the intelligent control module to adjust the production parameters.

[0035] The technical effects and advantages of the textile yarn bobbin intelligent quality control image analysis system of the present invention: The present invention synchronously acquires images of the surface of a yarn tube through a high-resolution multispectral industrial camera, combines the humidity and temperature data recorded by an environmental sensor, uses an adaptive Gaussian filter and an illumination equalization algorithm based on the Retinex theory to denoise and correct the illumination of the image, and generates a clear multispectral fusion image. Subsequently, a multispectral convolutional neural network model is used to automatically identify and segment the electrostatic accumulation area of the fusion image, quantify the electrostatic intensity, accumulation rate, and spatial distribution characteristics, and fuse the electrostatic characteristics with time series data through an optical flow algorithm to dynamically update the electrostatic intensity level to match the changes during the yarn running process. When the electrostatic intensity level is medium or high, the high-frame-rate dynamic images and the data of the tension sensor are used to synchronously capture the movement trajectory and tension change of the yarn winding and unwinding, apply a multi-target motion tracking algorithm to extract the dynamic trajectory deviation of the yarn, and combine the tension anomaly analysis to accurately quantify the actual interference degree of electrostatic accumulation on the winding uniformity and unwinding smoothness. Finally, the electrostatic intensity level and the interference analysis results are integrated, a comprehensive electrostatic risk score is generated based on the random forest algorithm, and an early warning mechanism is triggered according to the score result. The production parameters, such as humidity, rotation speed, and tension management, are dynamically adjusted through an intelligent control module to optimize the spinning process and prevent potential quality problems. Brief Description of the Drawings

[0036] Figure 1 It is a schematic structural diagram of the intelligent quality control image analysis system for textile yarn tubes of the present invention. Detailed Embodiments

[0037] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0038] Embodiment 1: Figure 1 The intelligent quality control image analysis system for textile yarn tubes of the present invention is given, including: a yarn tube image correction module, an electrostatic feature segmentation module, a trajectory tension identification module, and a risk score early warning module.

[0039] The yarn tube image correction module acquires high-resolution images of the surface of the yarn tube through a multispectral industrial camera, applies a Gaussian filter algorithm to remove noise, and corrects the image light intensity using an illumination equalization algorithm based on the Retinex theory, and outputs a fused image after illumination correction.

[0040] The electrostatic feature segmentation module receives the corrected fused image and environmental variable data, uses a multi-spectral convolutional neural network model to segment the electrostatic accumulation area in the fused image, quantifies the electrostatic intensity, accumulation rate, and spatial distribution characteristics, and fuses the electrostatic features with time series data through an optical flow algorithm to update the electrostatic intensity level and output electrostatic feature data.

[0041] The trajectory tension recognition module, after obtaining the corresponding electrostatic intensity level, uses high-frame-rate dynamic images and a tension sensor to synchronously capture the movement trajectory and tension changes of yarn winding and release, applies a multi-object motion tracking algorithm to extract the dynamic trajectory deviation of the yarn, and combines tension anomaly analysis to identify the actual interference degree of electrostatic accumulation on winding uniformity and release smoothness, and outputs the interference level.

[0042] The risk score warning module integrates the electrostatic intensity level, interference level results, and environmental data, generates an electrostatic risk score based on the random forest algorithm, and triggers a warning mechanism according to the score result, and outputs the electrostatic risk score and warning signal.

[0043] During the spinning process, the electrostatic accumulation generated by the yarn tube due to high-speed rotation and friction has dynamic characteristics, and the optical anomalies presented on its surface need to be captured by high-precision imaging technology. However, the complexity of the yarn tube surface texture and the change of environmental light conditions are likely to introduce noise and affect the image quality. In addition, the image features in different spectral bands are different. For example, visible light is used to capture the basic texture, ultraviolet light enhances the fluorescence reaction in the electrostatic area, and infrared light highlights the abnormal temperature distribution. Therefore, multi-spectral image acquisition and fusion preprocessing technology need to be adopted to provide a reliable data basis.

[0044] The operation process of the yarn tube image correction module includes the following: S1.1, synchronously collect high-resolution images of the yarn tube surface through a multi-spectral industrial camera in the visible light (400 - 700nm), ultraviolet light (300 - 400nm), and infrared light (700 - 1400nm) bands. For example, the camera parameters are fixed as follows: Resolution: 4000×3000 pixels.

[0045] Exposure time: Automatically adjustable range 10 - 50ms.

[0046] Frame rate: 50FPS.

[0047] Three spectral images are generated for each frame and stored in a three-channel form for subsequent fusion.

[0048] The data of the images collected by the camera and the data of the environmental sensor (recording humidity and temperature) are bound by a time stamp to ensure the accurate data correspondence between the two.

[0049] S1.2, The environmental sensor synchronously records the humidity and temperature of the environment where the yarn bobbin is located. Humidity affects the accumulation rate of static electricity, and temperature changes can indirectly reflect the static electricity intensity. These data are used for subsequent adjustment of image processing parameters. For example, high humidity may cause the weakening of fluorescence signals, and the enhancement intensity of the ultraviolet light channel image needs to be dynamically adjusted.

[0050] The humidity data is used to evaluate the possibility of static electricity accumulation: the lower the humidity, the easier it is for static electricity to form; The temperature data is used to exclude infrared abnormal signals unrelated to the environment: the abnormal area of local temperature rise corresponds to the static electricity distribution.

[0051] During the image acquisition process, noise may be generated due to mechanical vibration, surface reflection, or environmental light interference, affecting the detail performance of the static electricity area. Adaptive noise removal is performed on the images of each channel respectively.

[0052] Define the neighborhood range of each pixel point in the image, and calculate the dynamic weight through the pixel value distribution within the neighborhood; For each pixel point, its value is adjusted to a weighted value proportional to the degree of difference from the neighboring pixels; The key to noise removal is to retain high-contrast details while smoothing the background area and eliminating texture interference.

[0053] If the value of a certain pixel point is significantly higher than that of its surrounding pixels, it may be caused by noise. Through the weighted adjustment of the surrounding pixels, its value is made closer to the neighborhood mean, but the high-contrast edges still remain prominent.

[0054] For example, the following method can be adopted for disposal: Perform noise removal processing on the three-channel images collected respectively, using an adaptive Gaussian filter, and dynamically adjust the size and weight of the filter kernel according to the intensity distribution around each pixel point.

[0055] The adaptive filtering formula is as follows: ; Where: : Pixel intensity after denoising.

[0056] : Pixel intensity within the neighborhood window.

[0057] : Neighborhood window (dynamically adjusted in size, initially 3×3, and maximally extended to 7×7).

[0058] : Weight, defined as: .

[0059] : Standard deviation of image grayscale difference (automatically calculated based on the pixel intensity difference within the neighborhood).

[0060] The three-channel image generated after denoising is , , .

[0061] S1.4. Since the surface of the bobbin may be affected by uneven illumination (such as local overbright or shadow areas), it is necessary to correct the interference of illumination on the image brightness.

[0062] Use the Gaussian blur algorithm to estimate the background light intensity distribution (i.e., the illumination component) of the entire image; Subtract the illumination component at the corresponding position from each pixel value of the original image and perform normalization to restore the true reflection component; Readjust the contrast and details on the illumination-corrected image to make the boundaries of the electrostatic areas clearer.

[0063] After correction, each pixel value more reflects the true characteristics of the bobbin surface rather than the change in the intensity of external light. For example, in the ultraviolet light image, the fluorescent spots caused by static electricity are more prominent after correction and will not be masked by the light and dark.

[0064] For example, the following methods can be adopted for disposal: To eliminate the influence of uneven environmental illumination, use the illumination equalization algorithm based on the Retinex theory to decompose each channel image into the reflection component and the illumination component , and reconstruct the image to enhance the contrast of the electrostatic areas.

[0065] The calculation process is as follows: 1). Initial illumination component estimation: , where .

[0066] 2). Reflection component calculation: , where is a regulation parameter to prevent the denominator from being zero, with a value of .

[0067] 3). Reconstruct the equalized image: , and the three-channel image generated after equalization correction is , , .

[0068] S1.5. Fuse the images of the visible light, ultraviolet light, and infrared light bands to make the characteristics of the electrostatic areas more prominent.

[0069] The visible light image provides the basic structural framework and is evenly added to the final image; The fluorescence signal of the ultraviolet light image is enhanced through logarithmic transformation for highlighting the electrostatic region; The local high-temperature region of the infrared light image is smoothed by square root transformation to avoid overemphasizing background temperature fluctuations; The images of each channel are fused according to specific weights, and finally a multi-spectral enhanced image that can comprehensively present electrostatic characteristics is formed.

[0070] For example, through a specific fusion function, the three-channel images are synthesized into a multi-spectral enhanced image to enhance the characteristics of the electrostatic region.

[0071] The fusion formula is as follows: , where: , , : channel weight factors, with values of 0.5, 0.3, and 0.2 respectively.

[0072] The fusion strategy for each channel highlights its specific spectral characteristics: the visible light channel enhances the basic texture; the ultraviolet light channel highlights the electrostatic fluorescence characteristics; the infrared light channel highlights the temperature anomaly. The finally generated fused image is used as the input for subsequent steps.

[0073] For example, in areas with severe electrostatic accumulation, strong fluorescence signals will appear in the ultraviolet light image, the surface texture of the bobbin shown by visible light helps with positioning, and the infrared light further verifies the thermal effect caused by static electricity in this area. These pieces of information can provide multi-dimensional electrostatic manifestations after fusion.

[0074] The bobbin image correction module collects three-channel images of the bobbin surface through a multi-spectral industrial camera, and combines the humidity and temperature data synchronously recorded by the environmental sensor to provide multi-dimensional input for electrostatic detection and analysis. In the data preprocessing stage, noise removal, illumination equalization correction, and multi-spectral fusion are respectively performed on the images to enhance the optical characteristics of the electrostatic region. Through dynamic weight adjustment and specific fusion strategies between each step, the high quality and rich multi-dimensional information of the collected data are ensured, laying a reliable foundation for subsequent electrostatic identification and impact analysis.

[0075] During the spinning process, the static electricity accumulation on the surface of the bobbin not only affects the winding uniformity and tension stability of the yarn, but may also cause yarn breakage and equipment failures. Due to the high spatial and temporal dynamics of static electricity accumulation, its characteristics are manifested as tiny and variable optical anomaly regions in multi-spectral fusion images. Therefore, accurately identifying these static electricity accumulation regions and quantifying their intensity and distribution characteristics are crucial for realizing real-time monitoring and risk warning. The bobbin image correction module has provided a high-quality multi-spectral fusion image as input through multi-spectral image acquisition and preprocessing. The goal of the static electricity feature segmentation module is to extract the key information of static electricity accumulation from it using advanced deep learning and image processing techniques, and dynamically update the static electricity intensity level to adapt to the changes during the yarn running process.

[0076] The operation process of the static electricity feature segmentation module includes the following: S2.1, Analyze the fusion image using a multi-spectral convolutional neural network to automatically identify and segment the static electricity accumulation regions on the surface of the bobbin.

[0077] Model architecture: Input layer: Receive the fusion image with a size of 4000×3000 pixels.

[0078] Convolutional layer: Multiple convolutional operations are performed to extract the spatial and spectral features of the image. The size of each convolutional kernel is 3×3, the stride is 1, and padding is used to maintain the image size.

[0079] Pooling layer: Alternately apply max pooling (2×2) to reduce the computational complexity and extract higher-level features.

[0080] Skip connection: Establish skip connections between the shallow layer and the deep layer to retain the detailed features and enhance the segmentation accuracy of the static electricity regions.

[0081] Output layer: Use the Sigmoid activation function to generate a binary mask of the static electricity accumulation region , where 1 represents the static electricity accumulation region and 0 represents the non-static electricity region.

[0082] Training process: Dataset: Use a pre-annotated multi-spectral image dataset containing samples with different degrees and distribution patterns of static electricity accumulation.

[0083] Loss function: Adopt a custom weighted cross-entropy loss function to balance the unbalanced distribution between the static electricity region and the background region.

[0084] Optimization algorithm: Use the Adam optimizer with dynamically adjusted learning rate to accelerate convergence and improve the model generalization ability.

[0085] S2.2, After obtaining the binary mask of the static electricity accumulation region, further quantify the static electricity intensity and accumulation rate 。

[0086] A1. Electrostatic intensity quantization: Gray value mapping: Map the gray values of the electrostatic regions in the fused image to electrostatic intensity using a non-linear mapping function: , where represents the light intensity value of pixel point , and the function is used to compress high gray values to prevent overemphasis on strong electrostatic regions.

[0087] Local contrast enhancement: Apply adaptive histogram equalization within the electrostatic regions to highlight subtle intensity differences and obtain electrostatic intensity values: , where represents adaptive histogram equalization, which is an image enhancement technique used to improve the local contrast of an image, especially in cases of uneven illumination or dark local details. By adjusting the contrast around each pixel point, the detailed features of the image become more obvious.

[0088] A2. Accumulation rate calculation: Time series analysis: Use the optical flow algorithm to capture the changes in electrostatic intensity between consecutive frames and calculate the electrostatic accumulation rate : , where is the time interval, representing the reciprocal of the frame rate.

[0089] S2.3. Analyze the spatial distribution characteristics of the electrostatic accumulation regions to identify high-accumulation hotspots and regional distribution patterns.

[0090] B1. Cluster analysis: K-Means clustering: Perform K-Means clustering on the coordinates of the electrostatic accumulation regions and the electrostatic intensity to identify electrostatic accumulation hotspots with different densities: , where is the th cluster center, is the th data point.

[0091] B2. Distribution pattern recognition: Morphological feature extraction: Extract the morphological features of each cluster center, such as area, perimeter, compactness, etc.: , where represents the area of the electrostatic region, represents the perimeter of the electrostatic region, represents the boundary set of the electrostatic region, which is obtained from the binary mask Extracted by the boundary detection algorithm (such as Canny edge detection) in it, and the specific definition is: if the pixel and at least one adjacent pixel , then the corresponding pixel belongs to the boundary.

[0092] Density calculation: Calculate the electrostatic density of each clustering region, which is defined as the ratio of the electrostatic intensity to the area of the region: , where represents the electrostatic density, and represents the clustering region The average value of the electrostatic intensity inside, with the unit of intensity value / pixel, represents the electrostatic region area.

[0093] S2.4. Through the optical flow algorithm, fuse the electrostatic accumulation characteristics with the time series data to dynamically update the electrostatic intensity level .

[0094] C1. Optical flow calculation: Motion vector extraction: Use the optical flow algorithm to calculate the motion vector of the electrostatic accumulation region between consecutive frames , reflecting the dynamic change of electrostatic accumulation: , where Optical flow motion vector, which represents the displacement speed of pixel points between consecutive time frames and reflects the dynamic change of the electrostatic region.

[0095] C2. Electrostatic intensity level update: Dynamic adjustment mechanism: Based on the electrostatic intensity and the accumulation rate , define the electrostatic intensity level : ; where, is the dynamic adjustment coefficient, which is automatically optimized according to the material characteristics of the bobbin and the yarn to adapt to the electrostatic changes under different production conditions.

[0096] C3. Electrostatic level classification: Threshold division: According to the distribution of the electrostatic intensity level , divide it into three levels: low, medium, and high: , where, is the low-medium boundary value of the electrostatic intensity, is the medium-high boundary value of the electrostatic intensity.

[0097] The electrostatic feature segmentation module automatically identifies and segments the electrostatic regions in the multispectral fusion image through a deep convolutional neural network, accurately quantifying the electrostatic intensity, accumulation rate, and spatial distribution characteristics. By using the optical flow algorithm to fuse the electrostatic features with the time series data, the electrostatic intensity level is dynamically updated, thereby realizing the real-time matching and evaluation of the smoothness of yarn running under different electrostatic accumulation degrees. The electrostatic intensity level during the processing is further transmitted to the trajectory tension recognition module as a key input for the analysis of the smoothness of winding and releasing, ensuring the data consistency and logical coherence of the entire system in electrostatic detection and risk warning.

[0098] During the spinning process, the electrostatic accumulation on the surface of the bobbin not only affects the winding uniformity and tension stability of the yarn, but may also lead to yarn breakage and equipment failures. The electrostatic feature segmentation module identifies and quantifies the intensity levels (low, medium, high) of electrostatic accumulation through a multispectral convolutional neural network, providing a key input for the subsequent analysis of the smoothness of winding and releasing. When the electrostatic intensity level is medium or high, the yarn is prone to being interfered by static electricity during the winding and releasing processes, manifested as uneven winding, tension fluctuations, and even yarn breakage. Therefore, the goal of the trajectory tension recognition module is to analyze in detail the specific impact of electrostatic accumulation on the smoothness of yarn winding and releasing through high-frame-rate dynamic images and tension sensor data, quantify the degree of interference on the production process, and provide a quantitative basis for risk warning.

[0099] The operation process of the trajectory tension recognition module includes the following: S3.1, when the electrostatic intensity level is medium or high, synchronously capture the real-time dynamic process of yarn winding and releasing through a high-frame-rate industrial camera and a tension sensor. The high-frame-rate video frames and the tension data at the corresponding time points are bound by accurate timestamps to ensure that each frame image corresponds to the corresponding tension data.

[0100] S3.2, use the multi-object motion tracking algorithm to extract the motion trajectory of the yarn during the winding and releasing processes, and analyze the correlation between the trajectory deviation and the tension change.

[0101] 1). Motion tracking algorithm: The multi-object tracking (Kalman-Hungarian Tracking, KHT) algorithm based on the Kalman filter and the Hungarian algorithm.

[0102] Kalman filter: Predict the position of the yarn in the next frame, update the position by combining the observation data, and smooth the yarn trajectory.

[0103] Hungarian algorithm: Solve the multi-object data association problem to ensure the trajectory continuity and accuracy of each yarn.

[0104] 2). Trajectory extraction: Feature point recognition: By means of edge detection and morphological processing, the feature points of the yarn are recognized as the tracking targets.

[0105] Trajectory recording: An independent motion trajectory record is established for each yarn, including position coordinates, velocity vectors, and acceleration vectors.

[0106] S3.3. Combine the motion trajectory of the yarn and the data of the tension sensor to analyze the actual interference degree of electrostatic accumulation during the winding and unwinding processes of the yarn.

[0107] D1. Tension anomaly detection: Anomaly definition: Define that the tension fluctuation of the yarn exceeds the normal range (for example, the fluctuation amplitude exceeds , which is the standard deviation of the historical tension data) as a tension anomaly. Calculation process: , where: : Tension anomaly flag (1 for anomaly, 0 for normal), : Tension value at the current moment. : Average historical tension, : Standard deviation of historical tension.

[0108] D2. Calculation of trajectory offset: Offset definition: The degree of deviation between the motion trajectory of the yarn and the ideal winding trajectory, quantifying the interference of static electricity on the motion of the yarn. Calculation process: , where: : Trajectory offset at the current position, : Actual yarn position coordinates, : Ideal winding position coordinates.

[0109] D3. Quantification of interference degree: Calculation of interference index: , where: : Interference index, reflecting the influence degree of electrostatic accumulation on the smoothness of yarn winding and unwinding.

[0110] D4. Classification of interference levels: According to the interference index , it is divided into three levels: low, medium, and high: , where, is the low-medium boundary value of electrostatic intensity, is the medium-high boundary value of electrostatic intensity.

[0111] When the electrostatic intensity level is medium or high, the trajectory tension recognition module uses high-frame-rate dynamic images and tension sensor data to precisely capture the movement trajectory and tension changes during the yarn winding and unwinding processes. The multi-object motion tracking algorithm is used to extract the dynamic trajectory deviation of the yarn, and combined with the analysis of tension anomalies, the actual interference degree of electrostatic accumulation on the winding uniformity and unwinding smoothness is quantified. Through the comprehensive interference index, a quantitative basis is provided for risk warning.

[0112] During the spinning process, the electrostatic accumulation on the surface of the bobbin has a significant impact on the winding and unwinding smoothness of the yarn. The electrostatic feature segmentation module uses a multi-spectral convolutional neural network model to identify and quantify the electrostatic intensity level (low, medium, high), and the trajectory tension recognition module further analyzes the interference degree of electrostatic accumulation on the yarn winding and unwinding processes. When the electrostatic intensity level is medium or high, and the interference level is also medium or high, the spinning process faces a relatively high quality risk. Therefore, the goal of the risk scoring and warning module is to integrate the analysis results of the electrostatic feature segmentation module and the trajectory tension recognition module, use the random forest algorithm to generate an electrostatic risk score, and trigger the warning mechanism according to the scoring result to dynamically adjust the production parameters to optimize the spinning process and prevent potential quality problems.

[0113] The operation process of risk scoring and warning includes the following: S4.1, First, integrate the electrostatic intensity level identified in the electrostatic feature segmentation module and the interference level analyzed in the trajectory tension recognition module, and combine the environmental parameters (humidity) and (temperature) to construct a comprehensive feature set for risk scoring. Specifically, the constructed feature vector contains the following elements: , To capture the dynamic change trend, the sliding time window technique is used to calculate the statistics of each feature within the past seconds, such as the mean , variance and trend slope . For example, for the mean of the electrostatic intensity level is , the variance is , and its slope is . Similarly, statistical calculations are performed on the interference level, humidity, and temperature to form a complete feature vector : , Subsequently, interaction terms between features are constructed to capture non-linear relationships, such as: , Through multi-dimensional feature fusion, the recognition ability of the model for complex electrostatic accumulation situations is enhanced. All features are normalized.

[0114] S4.2, apply the random forest algorithm to train and predict the normalized feature vectors. A random forest consists of multiple decision trees. Each tree samples from the training dataset through Bootstrap sampling and randomly selects some features at each splitting node for optimal splitting to increase the diversity and robustness of the model. During the training process, the model learns how to predict the electrostatic risk level of the yarn tubes based on the feature set.

[0115] In the risk score generation stage, the random forest model outputs the risk probability at each time point to obtain the risk score. For example, the prediction probabilities of each decision tree can be combined into the final risk score through the following complex non-linear formula : , where is the number of decision trees in the random forest, is the th decision tree's prediction probability at time , is a non-linear adjustment parameter that is dynamically adjusted according to historical data to enhance the sensitivity of high risk probabilities.

[0116] When the risk score exceeds the first threshold, a medium-level warning is triggered to notify the operator for inspection; when the risk score exceeds the second threshold, a high-level alarm is triggered to automatically start the intelligent control module to adjust the production parameters. The intelligent control module can respond through the following mechanisms: 1). Humidity control: Automatically adjust the humidifying device to increase the ambient humidity and reduce static electricity accumulation.

[0117] 2). Rotation speed optimization: Dynamically adjust the rotation speed of the yarn tubes to reduce the friction coefficient and inhibit static electricity generation.

[0118] 3). Tension management: Optimize the tension control system to stabilize the yarn tension and prevent uneven winding and yarn breakage caused by static electricity interference.

[0119] 4). Maintenance instructions: In the case of a high-level alarm, automatically generate instructions for repairing the antistatic coating on the surface of the yarn tube or start the equipment maintenance process to ensure the normal operation of the static electricity management system of the yarn tube and the spinning equipment.

[0120] The risk scoring and warning module realizes the precise assessment and dynamic response of electrostatic risks by integrating multi-dimensional features and a complex non-linear scoring mechanism. First, the random forest algorithm is used to capture the complex relationship between electrostatic accumulation and interference in the multi-dimensional feature space, and the sensitivity of high-risk areas is enhanced through a custom non-linear scoring formula. Second, the dynamic threshold setting mechanism ensures that the warning system can adapt to different production environments and process requirements, improving the accuracy and flexibility of warnings. In addition, the automated response mechanism of the intelligent control module, combined with the optimized adjustment of multiple parameters such as humidity, rotation speed, and tension, realizes the full-process closed-loop control from risk detection to production parameter adjustment, significantly improving the stability and quality control level of the spinning process.

[0121] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0122] Only some exemplary embodiments of the present invention have been described by way of illustration above. Undoubtedly, for those of ordinary skill in the art, the described embodiments can be modified in various different ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the protection scope of the claims of the present invention.

[0123] It should be noted that in this text, if there are relational terms such as first and second, they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article or device including the element.

[0124] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A textile bobbin intelligent quality control image analysis system, characterized in that, It includes: a bobbin image correction module, an electrostatic feature segmentation module, a trajectory tension recognition module, and a risk score warning module; The bobbin image correction module collects high-resolution images of the bobbin surface through a multispectral industrial camera, applies a Gaussian filtering algorithm to remove noise, and uses an illumination equalization algorithm based on the Retinex theory to correct the image light intensity, and outputs a fused image with corrected illumination; The electrostatic feature segmentation module receives the corrected fused image and environmental variable data, uses a multispectral convolutional neural network model to segment the electrostatic accumulation area of the fused image, quantifies the electrostatic intensity, accumulation rate, and spatial distribution characteristics, and fuses the electrostatic features with time series data through an optical flow algorithm to update the electrostatic intensity level, and outputs electrostatic feature data; The trajectory tension recognition module, after obtaining the corresponding electrostatic intensity level, uses high-frame-rate dynamic images and a tension sensor to synchronously capture the movement trajectory and tension changes of the yarn winding and release, applies a multi-object motion tracking algorithm to extract the dynamic trajectory offset of the yarn, and combines the tension anomaly analysis to identify the actual interference degree of electrostatic accumulation on the winding uniformity and release smoothness, and outputs the interference level; The risk score warning module integrates the electrostatic intensity level, interference level results, and environmental data, generates an electrostatic risk score based on the random forest algorithm, and triggers an early warning mechanism according to the score result, and outputs the electrostatic risk score and warning signal.

2. The intelligent quality control image analysis system for textile bobbins according to claim 1, wherein: The operation process of the bobbin image correction module includes the following: S1.1, synchronously collect high-resolution images of the bobbin surface in multiple bands through a multispectral industrial camera; S1.2, the environmental sensor synchronously records the humidity and temperature of the environment where the bobbin is located; S1.3, perform noise removal processing on the collected three-channel images respectively, use an adaptive Gaussian filter, and dynamically adjust the size and weight of the filter kernel according to the intensity distribution around each pixel point; S1.4, use an illumination equalization algorithm based on the Retinex theory to decompose each channel image into a reflection component and an illumination component, and reconstruct the image to enhance the contrast of the electrostatic area; S1.5, weighted fusion of images in three bands of visible light, ultraviolet light, and infrared light. The images in each channel are fused according to specific weights, and the finally generated fused image .

3. The intelligent quality control image analysis system for textile bobbins according to claim 2, wherein: The operation process of the electrostatic feature segmentation module includes the following: S2.1, use a multispectral convolutional neural network to analyze the fused image, and identify and segment the electrostatic accumulation area on the bobbin surface; S2.2, after obtaining the binary mask of the static charge accumulation region, quantify the static charge intensity and the accumulation rate ; A1. Electrostatic intensity quantification: Gray value mapping: Map the gray value of the electrostatic region in the fused image to the electrostatic intensity, using a non-linear mapping function: , where represents the light intensity value of the pixel point ; Local contrast enhancement: Apply adaptive histogram equalization within the electrostatic region to highlight subtle intensity differences and obtain electrostatic intensity values: , denotes adaptive histogram equalization; A2. Accumulation rate calculation: Time series analysis: Using the optical flow algorithm to capture the change in electrostatic intensity between consecutive frames and calculate the electrostatic accumulation rate : , where is the time interval, representing the reciprocal of the frame rate.

4. The intelligent quality control image analysis system for textile bobbins according to claim 3, wherein: S2.3, analyze the spatial distribution characteristics of the electrostatic accumulation area, and identify high-accumulation hotspots and regional distribution patterns; B1. Cluster analysis: K-Means Clustering: Coordinates of the Electrostatic Accumulation Region and Electrostatic Intensity Perform K-Means clustering to identify hotspots of electrostatic accumulation with different densities: , where is the th clustering center, is the th data point; B2. Distribution pattern recognition: Morphological feature extraction: Extract the morphological features of each cluster center, , where represents the area of the electrostatic region, represents the perimeter of the electrostatic region, represents the boundary set of the electrostatic region, which is extracted by the boundary detection algorithm in the binary mask . Specifically defined as: if the pixel and at least one adjacent pixel , then the corresponding pixel belongs to the boundary; Density calculation: Calculate the electrostatic density of each clustering region, defined as the ratio of the electrostatic intensity to the area of the region: , where represents the electrostatic density and represents the clustering region is the average value of the electrostatic intensity within, represents the electrostatic region is the area of.

5. The intelligent quality control image analysis system for textile bobbins according to claim 4, wherein: S2.4, fuse the electrostatic accumulation feature with the time series data through the optical flow algorithm to dynamically update the electrostatic intensity level ; C1. Optical flow calculation: Motion vector extraction: Calculate the motion vectors of the electrostatic accumulation regions between consecutive frames using an optical flow algorithm , reflecting the dynamic changes in electrostatic accumulation: , where The optical flow motion vector represents the displacement speed of pixel points between consecutive time frames; C2. Electrostatic intensity level update: Dynamic adjustment mechanism: Based on the static electricity intensity and the accumulation rate , define the static electricity intensity level : , where is the dynamic adjustment coefficient; C3. Electrostatic level classification: Threshold division: According to the distribution of the static electricity intensity level it is divided into three levels: low, medium, and high: , where is the demarcation value between low and medium static electricity intensity, is the demarcation value between medium and high static electricity intensity.

6. The intelligent quality control image analysis system for textile bobbins according to claim 5, wherein: The operation process of the trajectory tension recognition module includes the following: S3.1, when the electrostatic intensity level is medium or high, synchronously capture the real-time dynamic process of yarn winding and unwinding through a high-frame-rate industrial camera and a tension sensor; S3.2, use the multi-target motion tracking algorithm to extract the motion trajectory of the yarn during winding and unwinding; S3.3, combine the motion trajectory of the yarn with the tension sensor data to analyze the actual interference degree of electrostatic accumulation on the yarn winding and unwinding process; D1. Tension anomaly detection: Abnormality definition: Defining that the fluctuation of yarn tension exceeding the normal range is tension abnormality. Calculation process: , where: : Tension abnormality flag, : Tension value at the current moment, : Historical average tension, : Historical standard deviation of tension; D2. Trajectory offset calculation: Offset definition: The degree of deviation between the yarn movement trajectory and the ideal winding trajectory, quantifying the interference of static electricity on the yarn movement. Calculation process: , where: : The trajectory offset at the current position, : The actual yarn position coordinates, : The ideal winding position coordinates; D3. Interference degree quantification: Interference index calculation: , where: : Interference index, reflecting the influence degree of static electricity accumulation on the smoothness of yarn winding and unwinding; D4. Interference level classification: According to the interference index , it is divided into three levels: low, medium, and high: , where is the demarcation value between low interference and medium interference, is the demarcation value between medium interference and high interference.

7. The textile yarn tube intelligent quality control image analysis system according to claim 6, characterized in that: The operation process of the risk score warning includes the following: S4.

1. First, integrate the static electricity intensity level and the interference level data, and combine with environmental parameters to construct a comprehensive feature set for risk scoring. To capture the dynamic change trend, use the sliding time window technique to calculate the statistics of each feature in the past seconds; Subsequently, construct interaction terms between features to capture non-linear relationships; Through multi-dimensional feature fusion, the recognition ability of the model for complex electrostatic accumulation situations is enhanced, and all features are normalized; S4.2, apply the random forest algorithm to train and predict the normalized feature vectors; In the risk score generation stage, the random forest model outputs the risk probability at each time point to obtain the risk score; When the risk score exceeds the first threshold, trigger a medium warning and notify the operator to check; when the risk score exceeds the second threshold, trigger a high alarm and start the intelligent control module to adjust the production parameters.

Images