Textile weighing data calibration system and method
By combining multimodal perception and deep learning algorithms, the accuracy problem of textile weighing equipment under easily deformable and environmentally changing conditions has been solved, achieving high-precision and automated weighing data calibration, thereby improving production efficiency and weighing accuracy.
Patent Information
- Application Number
- CN202511237701.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-01
- Publication Date
- 2025-11-11
AI Technical Summary
Existing textile weighing equipment struggles to achieve high-precision and stable weighing data calibration when faced with issues such as easy deformation, environmental changes, random stacking conditions, and the need for manual calibration.
Employing a multimodal sensing module and deep learning algorithms, combined with high-precision weighing sensors, penetrating sensors, visual acquisition components, and environmental parameter sensors, calibration coefficients are generated through multi-branch neural networks and reinforcement learning algorithms to automatically correct weighing data.
It achieves high-precision weighing under complex environments and material variations, automatically adapts to different textiles, reduces reliance on manual calibration, and improves production efficiency and weighing accuracy.
Smart Images
Figure CN120927115A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of textile weighing calibration technology, and in particular relates to a textile weighing data calibration system and method. Background Technology
[0002] With the rapid development of the textile industry, the accuracy of textile weighing has a significant impact on product quality control and production efficiency. Textile weighing is a crucial step in the textile production process, and its accuracy directly affects product quality and enterprise economic benefits. Currently, the textile industry widely uses electronic weighing equipment for textile weight measurement, but many technical challenges remain in practical applications: The existing technology has the following shortcomings: First, due to the deformable nature of textiles, it is difficult to manually control the weighing process, resulting in frequent occurrences of substandard products per unit area and causing economic losses. Second, the existing weighing system does not fully consider the weight changes of textiles under different temperature and humidity environments, leading to significant deviations in weighing results when environmental conditions vary greatly. Third, the stacking state of textiles on the weighing platform is random, with the fiber stacking density in wrinkled areas being greater than in flat areas, resulting in uneven force distribution on the pressure sensor, a problem that the existing technology cannot effectively solve. Fourth, traditional equipment requires professional personnel to periodically calibrate with standard weights, each calibration is time-consuming, and the calibration results are only applicable to specific environments and material combinations. Finally, when environmental parameters change drastically or unknown materials are introduced, the existing equipment cannot autonomously correct deviations, causing distortion of weighing data.
[0003] Therefore, there is an urgent need for a weighing data calibration system that can comprehensively consider the characteristics of textiles, environmental factors, and stacking conditions, and achieve high-precision calibration of textile weighing data through the fusion of multimodal sensing and intelligent algorithms.
[0004] Therefore, we need to design a textile weighing data calibration system and method to solve these problems. Summary of the Invention
[0005] To address the issue of weighing data deviation caused by material properties, morphological changes, and environmental influences during textile weighing, and to achieve high-precision, adaptive textile weighing data calibration, a textile weighing data calibration system and method are provided.
[0006] The technical solution adopted by this invention to solve its technical problem is as follows: A textile weighing data calibration system is provided, comprising a weighing module, a multimodal sensing module, a data processing module, a fusion calibration module, and an execution module. The weighing module includes a high-precision weighing sensor for collecting raw weight data of the textile. The multimodal sensing module includes a penetrating sensor, a visual acquisition component, and an environmental parameter sensor. The penetrating sensor is used to detect the internal density distribution of the textile, the visual acquisition component is used to capture the surface texture features of the textile, and the environmental parameter sensor is used to collect temperature, humidity, and air pressure parameters. The data processing module receives and preprocesses the output data from the weighing module and the multimodal sensing module, including filtering, spectral transformation, and feature extraction. The fusion calibration module uses a multi-branch neural network and reinforcement learning algorithm to fuse the preprocessed multi-source data to generate calibration coefficients. The execution module corrects the raw weight data according to the calibration coefficients and outputs the final result.
[0007] Preferably, the high-precision weighing sensor has a range adapted to the weighing requirements of textiles, is installed at the bottom of the weighing platform, and has a uniformly distributed force point.
[0008] Preferably, the penetrating sensor is installed at a preset height above the weighing platform, with the transmitting end and the receiving end arranged at a preset angle to meet the detection resolution requirements of the internal density distribution of the textile, and the signal penetration attenuation rate is controlled within a preset threshold.
[0009] Preferably, the visual acquisition component is installed at a preset height above the weighing platform, with the lens's field of view covering the weighing area. The texture contour is extracted using an edge detection algorithm, and a high-dimensional feature vector is generated using a feature extractor.
[0010] Preferably, the preprocessing of the data processing module further includes a data standardization step: normalizing various types of sensor data according to preset rules and mapping them uniformly to a preset numerical range.
[0011] Preferably, the multi-branch neural network includes: a first branch that processes weighing data and density distribution data through a convolutional neural network; a second branch that processes visual feature vectors through a fully connected network; and a fusion layer that dynamically allocates the weights of the features of each branch using an attention mechanism and outputs a fused feature vector.
[0012] Preferably, the reinforcement learning algorithm includes a state space and an action space. The state space contains fused feature vectors and environmental parameters, the action space is a calibration coefficient adjustment amount within a preset range, and the reward function is constructed based on the deviation between the calibration result and the standard value.
[0013] Preferably, the system further includes a self-learning update module, which triggers an incremental learning mode when the feature matching degree is detected to be lower than a preset threshold, and automatically collects data and updates the model parameters in a preset number of weighing cycles.
[0014] This invention also provides a method for calibrating textile weighing data, comprising the following steps: S1. Synchronously acquiring the original weight data, internal density distribution, surface texture features, and environmental parameters of the textile, controlling the acquisition time deviation within a preset range; S2. Preprocessing the acquired data, including filtering, spectral transformation, and feature extraction; S3. Fusing multi-source features through a multi-branch neural network to generate a fused feature vector; S4. Calculating calibration coefficients based on a reinforcement learning algorithm, wherein the algorithm uses the error between the calibrated weight and the standard weight as a reward signal; S5. Correcting the original weight data according to the calibration coefficients and outputting the final result.
[0015] Preferably, step S4 also includes a self-learning optimization process: by periodically and automatically performing standard part calibration, the deviation data is supplemented to the training set, and the model accuracy is verified through iterative optimization until the calibration error on the preset types of textiles stabilizes within the preset range.
[0016] The advantages and positive effects of this invention are: 1. This invention adopts a multi-dimensional data fusion and correlation modeling approach, which can accurately capture and eliminate various interferences caused by environmental changes, morphological differences, and different material properties, fundamentally improving the accuracy and stability of calibration and ensuring that weighing data remains reliable under various complex conditions.
[0017] 2. This invention, through visual feature extraction and multi-source data fusion, can automatically identify the differences in physical properties of different textile materials. It can automatically adapt to various common textile materials; and through a self-learning update module, it can quickly complete parameter optimization without a lengthy manual debugging process, significantly shortening the adaptation cycle for new materials. Furthermore, the adaptation effect is accurate and reliable, meeting diverse textile weighing needs.
[0018] 3. This invention employs a highly efficient synchronous acquisition mechanism and data processing link, enabling real-time tracking of morphological changes in textiles and fluctuations in environmental parameters during the weighing process. When environmental parameters undergo sudden changes, it can react quickly and adjust the calibration coefficients, maintaining stable calibration results even in complex and ever-changing environments. This effectively avoids batch weighing data deviations caused by environmental changes, ensuring data consistency and accuracy.
[0019] 4. This invention, through reinforcement learning algorithms and a self-learning update module, can automatically perform standard part calibration and update the model, greatly reducing reliance on manual calibration and lowering equipment maintenance costs and manpower investment. In flexible production scenarios, it can quickly adapt to changes in fabric types, significantly improving overall production efficiency. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0021] Figure 1 This is a schematic diagram of the calibration system module of the present invention; Figure 2 This is a schematic diagram of the multimodal sensing module of the present invention; Figure 3 This is a schematic diagram of the calibration method of the present invention.
[0022] The annotations in the attached figures are explained as follows: 100 Weighing module; 200 Multimodal sensing module; 210 Penetrating sensor; 220 Vision acquisition component; 230 Environmental parameter sensor; 300 Data processing module; 400 Fusion calibration module; 500 Execution module. Detailed Implementation
[0023] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.
[0024] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.
[0025] The present invention will be further described below with reference to the accompanying drawings: Example 1: Reference Figures 1-2 As shown, a textile weighing data calibration system includes a weighing module 100, a multimodal sensing module 200, a data processing module 300, a fusion calibration module 400, and an execution module 500.
[0026] The weighing module 100 includes a high-precision load cell for collecting raw weight data of textiles. The high-precision load cell has a measuring range specifically adapted to the weighing requirements of textiles, is installed at the bottom of the weighing platform, and has a uniformly distributed force distribution. This design ensures the accuracy and stability of the weighing data and avoids measurement errors caused by uneven force distribution.
[0027] The multimodal sensing module 200 includes a penetrating sensor 210, a visual acquisition component 220, and an environmental parameter sensor 230. The penetrating sensor 210 is mounted at a preset height above the weighing platform, with the transmitter and receiver arranged at a preset angle to meet the detection resolution requirements of the internal density distribution of the textile, and the signal penetration attenuation rate is controlled within a preset threshold. This arrangement ensures that the sensor acquires complete information about the internal structure of the textile, providing density distribution data for subsequent calibration. The visual acquisition component 220 is mounted at a preset height above the weighing platform, with its lens field of view covering the weighing area. It extracts texture contours using an edge detection algorithm and generates high-dimensional feature vectors using a feature extractor. This design enables the system to capture the surface texture features of the textile, providing visual information support for the calibration process. The environmental parameter sensor 230 is used to collect temperature, humidity, and air pressure parameters, which have a significant impact on textile weight measurement. By monitoring environmental changes in real time, the system can adjust calibration parameters accordingly.
[0028] The data processing module 300 receives and preprocesses the output data from the weighing module 100 and the multimodal sensing module 200. Preprocessing includes filtering, spectral transformation, and feature extraction. The preprocessing in the data processing module 300 also includes a data standardization step: normalizing the data from various sensors according to preset rules and mapping them uniformly to a preset numerical range. This standardization process eliminates dimensional differences between different sensor data, making subsequent fusion processing more efficient and accurate.
[0029] The fusion calibration module 400 employs a multi-branch neural network and reinforcement learning algorithms to fuse preprocessed multi-source data to generate calibration coefficients. The reinforcement learning algorithm uses the error between the calibrated weight and the standard weight as the reward signal. The multi-branch neural network includes: a first branch that processes weighing data and density distribution data through a convolutional neural network; a second branch that processes visual feature vectors through a fully connected network; and a fusion layer that dynamically allocates the weights of features from each branch using an attention mechanism, outputting a fused feature vector. This network structure design fully utilizes the characteristics of different types of data to improve the fusion effect. The reinforcement learning algorithm includes a state space and an action space. The state space contains the fused feature vector and environmental parameters, while the action space contains the calibration coefficient adjustment within a preset range. The reward function is constructed based on the deviation between the calibration result and the standard value. This algorithm design enables the system to gradually optimize the calibration coefficients and improve weighing accuracy through continuous trial and learning.
[0030] The execution module 500 corrects the original weight data according to the calibration coefficients and outputs the final result. The execution module 500 applies the calibration coefficients generated by the fusion calibration module 400 to the original data collected by the weighing module 100, performs correction calculations through a specific mathematical model, and finally outputs the calibrated weight data of the textile.
[0031] The system also includes a self-learning update module. When the detected feature matching degree is lower than a preset threshold, an incremental learning mode is triggered, automatically collecting data and updating model parameters in a preset number of weighing cycles. This self-learning mechanism enables the system to adapt to the weighing needs of different types of textiles and continuously optimizes the calibration effect as it is used, improving the system's versatility and long-term stability.
[0032] In actual system operation, when the textile is placed on the weighing platform, the weighing module 100 first collects the raw weight data, while the multimodal sensing module 200 collects the textile's density distribution, surface texture features, and environmental parameters. The data processing module 300 preprocesses this raw data, including filtering, spectral transformation, feature extraction, and standardization. The fusion calibration module 400 uses a multi-branch neural network and reinforcement learning algorithms to comprehensively analyze this multi-source data and generate calibration coefficients suitable for the current textile and environmental conditions. The execution module 500 applies these calibration coefficients to correct the raw weight data and outputs the final calibration result. The entire process achieves high-precision measurement of the textile's weight, effectively eliminating measurement errors caused by textile characteristics and environmental factors.
[0033] Example 2: Reference Figure 3 As shown, this embodiment provides a method for calibrating textile weighing data. This method achieves accurate calibration of textile weight data through multi-source data acquisition and deep learning technology. The specific implementation steps are as follows: S1: Synchronously collect the original weight data, internal density distribution, surface texture characteristics and environmental parameters of textiles, and control the collection time deviation within the preset range.
[0034] In this step, a high-precision electronic weighing sensor is used to collect the raw weight data of the textile. Simultaneously, an X-ray density scanner is used to obtain the internal density distribution map of the textile, a high-resolution optical scanner is used to collect the surface texture features of the textile, and environmental parameters such as temperature, humidity, and air pressure are collected using temperature and humidity sensors, barometers, and other devices. To ensure data synchronization, the system uses a unified clock source to control the acquisition sequence of each sensor, keeping the acquisition time deviation within 5 milliseconds to ensure time consistency of multi-source data.
[0035] S2: Preprocess the acquired data, including filtering, spectral transformation, and feature extraction.
[0036] For the original weight data, a Kalman filter algorithm is used to remove random noise and interference signals; wavelet transform is performed on the internal density distribution data to extract multi-scale density features; Fourier spectrum transform is performed on the surface texture features to extract texture frequency features; and environmental parameter data is normalized. In the feature extraction stage, temporal statistical features are extracted from the original weight data, spatial distribution features are extracted from the internal density distribution, texture complexity and directionality features are extracted from the surface texture, and temperature and humidity gradient features are extracted from the environmental parameters.
[0037] S3: Generate a fused feature vector by fusing multi-source features through a multi-branch neural network.
[0038] A deep neural network architecture with four parallel branches is constructed, each processing a different data source: the weight branch uses a three-layer fully connected network to process the raw weight features; the density branch uses a convolutional neural network to process density distribution features; the texture branch uses a residual network to process texture features; and the environment branch uses a bidirectional LSTM network to process temporal features of environmental parameters. The output features of the four branches are weighted and fused through an attention mechanism to generate a 128-dimensional fused feature vector, which contains complex correlation information between textile weight and various influencing factors.
[0039] S4: Calculate the calibration coefficient based on the reinforcement learning algorithm, which uses the error between the calibrated weight and the standard weight as the reward signal.
[0040] A Deep Q-Network (DQN) is employed as the reinforcement learning framework, using fused feature vectors as state inputs and calibration coefficients as action outputs. The system maintains an experience replay buffer to store historical calibration data. During training, the system randomly selects standard calibration pieces for weighing, and the error between the calibrated weight and the standard weight is used as a reward signal fed back to the reinforcement learning algorithm. The reward function is designed as a negative exponential function of the error; the smaller the error, the higher the reward. Through continuous iterative optimization, the reinforcement learning algorithm gradually learns the optimal calibration strategy, generating calibration coefficient matrices for different types of textiles and environmental conditions.
[0041] S4 also includes a self-learning optimization process: by periodically and automatically performing standard part calibration, deviation data is added to the training set, and the model accuracy is verified through iterative optimization until the calibration error on preset types of textiles stabilizes within a preset range. Specifically, the system automatically performs a standard calibration process every 24 hours, using 10 standard calibration parts of different materials and weight grades for testing. For each calibration test, the system records the original weight, the calibrated weight, the standard weight, and its deviation, and automatically adds this data to the training dataset. The system uses cross-validation to evaluate model performance. When the calibration error of all preset types of textiles stabilizes within ±0.1% in 5 consecutive calibration tests, the model is considered to have reached convergence. In this embodiment, the preset types of textiles include cotton, linen, silk, wool, chemical fibers, and their blends. If the convergence condition is not met, the system will automatically adjust the learning rate and network parameters and continue the iterative optimization process.
[0042] S5: Correct the original weight data according to the calibration coefficient and output the final result.
[0043] The calibration coefficient calculated in step S4 is applied to the original weight data. The calculation formula is: Calibrated weight = Original weight × (1 + Calibration coefficient). The calibration coefficient is a correction value that comprehensively considers the material properties, internal structure, surface characteristics, and environmental factors of the textile, and is usually within the range of ±5%. The system integrates and outputs the calibrated weight data, calibration coefficient, calibration confidence level, and other information, and generates a calibration report, which includes a comparison of the weight before and after calibration, a calibration accuracy assessment, and the calibration validity period.
[0044] Through the above steps, this textile weighing data calibration method can effectively eliminate measurement errors caused by material characteristics, structural inhomogeneity and environmental factors during the textile weighing process, significantly improve weighing accuracy, and the error of the calibrated weight data can be controlled within 0.1%, meeting the high-precision textile metrology requirements.
[0045] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.
Claims
1. A textile weighing data calibration system, characterized in that, include: The weighing module includes a high-precision weighing sensor for collecting raw weight data of textiles. The multimodal sensing module includes a penetrating sensor, a visual acquisition component, and an environmental parameter sensor, wherein: The penetrating sensor is used to detect the internal density distribution of textiles; The visual acquisition component is used to capture the surface texture features of textiles; The environmental parameter sensor is used to collect temperature, humidity, and air pressure parameters; The data processing module receives and preprocesses the output data from the weighing module and the multimodal sensing module. The preprocessing includes filtering, spectrum transformation, and feature extraction. The fusion calibration module uses a multi-branch neural network and reinforcement learning algorithm to fuse preprocessed multi-source data to generate calibration coefficients. The reinforcement learning algorithm uses the error between the calibrated weight and the standard weight as a reward signal. The execution module corrects the original weight data according to the calibration coefficient and outputs the final result.
2. The system according to claim 1, characterized in that, The high-precision weighing sensor has a range adapted to the weighing requirements of textiles, is installed at the bottom of the weighing platform, and has a uniformly distributed force point.
3. The system according to claim 1, characterized in that, The penetrating sensor is installed at a preset height above the weighing platform, with the transmitting end and receiving end arranged at a preset angle to meet the detection resolution requirements of the internal density distribution of textiles, and the signal penetration attenuation rate is controlled within a preset threshold.
4. The system according to claim 1, characterized in that, The visual acquisition component is installed at a preset height above the weighing platform, and the lens's field of view covers the weighing area. It extracts texture contours through an edge detection algorithm and generates high-dimensional feature vectors using a feature extractor.
5. The system according to claim 1, characterized in that, The preprocessing of the data processing module also includes a data standardization step: normalizing various types of sensor data according to preset rules and mapping them uniformly to a preset numerical range.
6. The system according to claim 1, characterized in that, The multi-branch neural network includes: The first branch processes weighing data and density distribution data using a convolutional neural network. The second branch processes visual feature vectors through a fully connected network. The fusion layer uses an attention mechanism to dynamically allocate the weights of each branch feature and outputs a fused feature vector.
7. The system according to claim 1, characterized in that, The reinforcement learning algorithm includes a state space and an action space. The state space contains fused feature vectors and environmental parameters, and the action space is a calibration coefficient adjustment amount within a preset range. The reward function is constructed based on the deviation between the calibration result and the standard value.
8. The system according to claim 1, characterized in that, It also includes a self-learning update module, which triggers an incremental learning mode when the feature matching degree is detected to be lower than a preset threshold, and automatically collects data and updates the model parameters in a preset number of weighing cycles.
9. A method for calibrating textile weighing data, characterized in that, Includes the following steps: S1. Synchronously collect the original weight data, internal density distribution, surface texture characteristics and environmental parameters of textiles, and control the collection time deviation within the preset range; S2. Preprocess the collected data, including filtering, spectral transformation and feature extraction; S3. Multi-source features are fused using a multi-branch neural network to generate a fused feature vector; S4. Calculate the calibration coefficient based on the reinforcement learning algorithm, wherein the algorithm uses the error between the calibrated weight and the standard weight as the reward signal; S5. Correct the original weight data according to the calibration coefficient and output the final result.
10. The method according to claim 9, characterized in that, Step S4 also includes a self-learning optimization process: by periodically and automatically performing standard part calibration, deviation data is added to the training set, and the model accuracy is verified through iterative optimization until the calibration error on the preset types of textiles stabilizes within the preset range.