Embroidery digitization full-process protection method integrating deep learning and block chain evidence storage

Abstract

The invention provides a full-process protection method for embroidery digitization fusing deep learning and block chain evidence storage. The method comprises the following steps: step 1, acquiring multi-modal data of an embroidery work by adopting a multi-modal data acquisition technology; 2, eliminating environmental noise from the multi-modal data acquired in the step 1 through an improved adaptive Kalman filtering algorithm, and compressing redundant time sequence information through a key frame extraction technology to obtain a standardized process data set; step 3, utilizing deep learning to drive stitch recognition, and realizing reproduction of an embroidery method through a convolutional neural network, a generative adversarial network and an edge detection algorithm; 4, linking through a block chain protocol, determining digital ownership, and managing resources in batches; and step 5, mobile terminal application deployment: constructing a mobile application through a lightweight cross-platform framework to realize embroidery digital protection. According to the method, the data basis is greatly improved, and rich and systematic data resources are provided for embroidery digital research and application.

Description

Technical Field

[0001] The present invention relates to the technical field of digital protection of intangible cultural heritage, and particularly relates to a full-process protection method for embroidery digitization that integrates deep learning and blockchain evidence storage. Background Art

[0002] Embroidery, as an important part of Chinese intangible cultural heritage, carries rich historical and cultural values and artistic connotations. However, with the rapid development of modern technology and the gradual decline of traditional handicrafts, the inheritance and protection of embroidery techniques are facing severe challenges. Currently, the digital protection of embroidery mainly relies on image acquisition and plane scanning technologies, specifically manifested in three ways: manual recording of stitches, static image archiving, and simple blockchain applications. Although these traditional methods have achieved digital preservation of embroidery works to a certain extent, they have obvious deficiencies in terms of efficiency, accuracy, restoration ability, and the depth of blockchain application, and are difficult to meet the needs of modern intangible cultural heritage protection.

[0003] The present invention proposes a full-process protection method for embroidery digitization that integrates deep learning and blockchain evidence storage. By introducing deep learning technology, accurate identification of embroidery stitches and dynamic reproduction of embroidery methods are achieved; through the in-depth application of blockchain technology, transparent management and effective protection of digital resources are ensured. Specifically, the present invention first obtains the three-dimensional structure, stitch trajectories, and material information of embroidery works through multi-dimensional data acquisition technologies (such as 3D laser scanning, nine-axis inertial measurement unit, and hyperspectral imaging); secondly, uses an improved adaptive Kalman filter algorithm and a deep learning model to process and analyze the data to achieve accurate identification of stitches and dynamic restoration of embroidery methods; finally, through blockchain technology, the digital information of embroidery works is deeply bound to the information of inheritors to ensure the immutability and transparent management of digital assets. The present invention can not only improve the efficiency and accuracy of embroidery digitization, but also provide a new technical support for the protection and inheritance of embroidery intangible cultural heritage. Summary of the Invention

[0004] Object of the Invention: To solve the problems in traditional embroidery digitization technologies such as low efficiency of manual stitch recording, inability to restore the three-dimensional structure by static image archiving, record distortion, low inheritance efficiency, and property right confusion caused by superficial blockchain applications, the present invention realizes multi-dimensional data acquisition, intelligent analysis, dynamic modeling, and blockchain evidence storage of embroidery works by constructing a full-process protection method for embroidery digitization that integrates deep learning and blockchain evidence storage, ensuring accurate recording, efficient inheritance, and transparent management of embroidery intangible cultural heritage. The present invention aims to break through the limitations of traditional technologies in time and space, provide an efficient, accurate, and secure embroidery digitization protection solution, and provide a new technical support for the protection and inheritance of embroidery intangible cultural heritage.

[0005] Technical solution: A full - process protection method for embroidery digitization that integrates deep learning and blockchain evidence storage of the present invention is generally divided into five layers: data acquisition layer, data processing layer, deep learning layer, blockchain evidence storage layer, and mobile application layer. The specific steps are as follows: Step 1, collect multimodal data: Use multimodal data acquisition technology to obtain multimodal data of embroidery works, including three - dimensional point cloud data, needle tip movement trajectories, and material characteristics, providing high - quality raw data for subsequent process data analysis and deep - learning modeling; Step 2, perform data pre - processing and feature extraction: Eliminate environmental noise from the multimodal data collected in Step 1 through an improved adaptive Kalman filtering algorithm, and compress redundant time - series information through key - frame extraction technology. Finally, obtain a standardized process data set and store it in the database; Step 3, perform stitch recognition and embroidery method reproduction driven by deep learning: Use deep - learning - driven stitch recognition, and through 3D convolutional neural network, generative adversarial network, and Canny edge detection algorithm, realize the reproduction of embroidery methods; Step 4, blockchain evidence storage and digital ownership management: Upload to the blockchain RC721 protocol to clarify digital ownership and manage resources in batches; Step 5, deploy mobile applications: Build mobile applications through a lightweight cross - platform framework to achieve digital protection of embroidery.

[0006] Step 1 includes the following steps: Step 1.1, use a 3D laser scanner (resolution ≥ 1600dpi) to perform non - contact scanning on the surface of the embroidery work, generate three - dimensional point cloud data with millimeter - level accuracy, and perform denoising processing through the neighborhood average filtering algorithm. The formula is: , where, represents the th original three - dimensional point cloud coordinate within the preset neighborhood radius centered on the target point, that is, each point represents the three - dimensional coordinates of a tiny area on the embroidery surface captured by the scanner ; is the number of valid points within the neighborhood, (it is necessary to satisfy to exclude isolated noise points); is the smoothed three - dimensional point cloud coordinate output after denoising; Step 1.2, real - time collect needle tip movement data through a nine - axis IMU sensor (model MPU - 9250, sampling frequency 200Hz) deployed on the embroidery needle, and perform smoothing processing on the needle tip movement data through the following formula: , where, is the movement speed of the needle tip at time t, which is calculated by a nine-axis IMU sensor deployed on the embroidery needle tool, and the unit can be m / s or mm / s; is the movement speed of the needle tip at time t after smoothing, that is, a more stable speed value after filtering, which is used to reduce instantaneous noise and make the movement trajectory smoother; is the smoothing coefficient, and the value range is , which is used to control the smoothing degree. If is larger (close to 1), the newly acquired speed data plays a dominant role, the speed is updated faster, and the filtering effect is weaker. If is smaller (close to 0), the smoothed speed data depends more on historical data , and the trajectory is more stable, but it may cause a lag effect; Step 1.3, analyze the spectral reflection characteristics of the embroidery thread material through a hyperspectral imaging system (wavelength range 400~2500nm) to generate a material spectral fingerprint, and normalize the spectral data through the following formula: , This formula is used to normalize the hyperspectral data and map the spectral data to the range of [0,1] for subsequent analysis and machine learning modeling. Among them, is the original spectral data at wavelength , that is, the spectral reflectance or intensity value corresponding to a certain wavelength measured by the hyperspectral imaging system, and the unit is usually reflectance (0~100%) or radiance (W / m²·sr·nm); and are the minimum and maximum values of the original spectral data respectively; is the normalized spectral data, which linearly transforms to between [0,1], removing the influence of lighting conditions or equipment differences on the data, and making it suitable for machine learning models or pattern recognition tasks; Step 1.4, identify the material composition and aging state, including the following steps: Step 1.4.1, use the Wiley General Spectral Database to analyze the composition of the embroidery material; Step 1.4.2, adopt a support vector machine (SVM) to train a classification model, input the material spectral fingerprint, and output the predicted material category (such as 100% silk, cotton and linen mixture, etc.); Step 1.4.3, calculate the spectral offset rate of the material: , where is the standard spectral data of the material to be measured; Reflect the spectral shift situation and be used to evaluate the aging degree.

[0007] Step 2 includes the following steps: Step 2.1, execute the improved adaptive Kalman filtering algorithm. The formula for the improved state transition matrix is: , where is the state transition matrix, is the control input matrix, is the control input vector, is the process noise; is the system state vector at time k, representing the key parameters of the tip movement: , where represents the three-dimensional coordinate position of the tip, measured by a 3D laser scanner, is the velocity of the tip in the direction calculated by a nine-axis IMU sensor; The state transition matrix reflects the change in the tip movement state and adopts a discrete motion model: , where is the sampling time interval, and the first three rows of the state transition matrix reflect the change of position over time: , and the last three rows reflect that the velocity remains unchanged: ; In some cases, the tip movement will be affected by external control inputs, such as the force exerted by the operator's hand. The control input matrix is: , The control input vector includes the accelerations of the tip in the x, y, and z directions measured by a nine-axis IMU sensor : , Obtain the influence on the tip movement, including position change and velocity change: Position change: ; Velocity change: ; where represents the velocity in the x-axis direction at the current moment, represents the velocity in the x-axis direction at the previous moment; The process noise Indicates the dynamic error of the system (such as hand vibration), which follows a Gaussian distribution : ; The covariance matrix is calculated from the tip acceleration noise and the time interval : : , where is the variance of the tip motion acceleration noise and can be obtained from the data sheet provided by the IMU sensor; Finally, the smoothed tip trajectory after adaptive Kalman filtering is output ; Step 2.2, extract key frames: Calculate the compression ratio through the following formula : , where is the multi-modal data collected in Step 1, is the amount of data after compression.

[0008] Step 3 includes the following steps: Step 3.1, use a 3D convolutional neural network (3D-CNN) to extract the spatial topological structure information of the stitches and capture the characteristics of different stitching methods. An improved 3D convolutional neural network is used to extract the spatial topological features of the stitches, and the formula is: , where is the input feature map, and the parameters of the input feature map include where is the number of input channels (such as feature channels for color, texture, etc.), respectively represent the depth, height, and width of the stitch structure; is the convolutional kernel weight, is the bias term, is the activation function, is the output feature map; Step 3.2, use the generative adversarial network GAN to generate the vector field of the stitch trajectory to improve the continuity and rationality of the trajectory; Step 3.3, perform edge detection and optimization on the generated stitch path to ensure the accuracy of the reproduced embroidery method.

[0009] Step 3.2 includes: The loss function of the generator of the generative adversarial network GAN is: , where the generator has an input of , the output is the stitch trajectory vector field; the discriminator is used to determine whether the generated stitch path conforms to the real embroidery method; is the physical constraint loss, ensuring that the generated path conforms to the embroidery physical rules; is the input random noise of the generator, following a Gaussian distribution ; represents the expected value of the result sampled from the random variable ; the hyperparameter is used to control the influence degree in the overall loss.

[0010] Step 3.3 includes: improving the Canny edge detection algorithm, and the formula is: , where represents the gradient magnitude, are the gradients in the horizontal and vertical directions respectively, and are calculated using the Sobel operator: , where is the stitch path image generated in Step 3.2, represents the convolution operation.

[0011] Perform edge detection and optimization on the gradient magnitude . For each pixel point , calculate the gradient direction ; according to the gradient direction, compare the current pixel with the neighboring pixels. If the gradient magnitude of the current pixel is not the local maximum, set it to 0, so that the edge noise can be removed and only the most obvious edges are retained.

[0012] Step 4 includes the following steps: Step 4.1, combine the tip trajectory extracted in Step 2, the compressed data volume and the stitch path image I in Step 3 to generate a digital twin, that is, combine the virtual representation of the physical embroidery with the process characteristics to obtain a unique digital representation; Calculate the hash value of the digital twin through the hash algorithm and bind it to the biometric information of the inheritor (such as identity, fingerprint or facial features, etc.) to form a standardized meta data packet to ensure the uniqueness and authenticity of the digital embroidery. The meta data packet is stored in a decentralized manner through IPFS (InterPlanetary File System) to ensure that the data content of the digital embroidery cannot be tampered with and has the function of permanent evidence preservation; Step 4.2: Use the ERC-721 non-fungible token (NFT) standard to mint a unique non-fungible token (NFT) for each digital embroidery, endowing it with an on-chain identity to ensure the uniqueness and irreplaceability of digital assets.

[0013] Step 5 includes: Deploying the pruned and optimized deep learning model using the integrated TensorFlow Lite engine to support real-time low-latency identification of at least 20 basic stitch types in an offline state; in addition, implementing augmented reality interaction through ARCore technology, enabling users to view and interact with virtual-displayed embroideries via a mobile device, enhancing the user experience; users can encapsulate the parameters of each embroidery into a unique non-fungible token (NFT) in ERC-1155 format with one click, and each unique non-fungible token (NFT) will contain its unique digital certificate and copyright information and conduct on-chain transactions through a built-in smart wallet; this process automatically enforces copyright rules through a smart contract to ensure the ownership, transaction, and copyright management of each digital embroidery. In addition, the generated NFTs will be synchronized on the cultural relics and museums platform to promote the digital inheritance of embroideries and the protection of cultural assets.

[0014] The present invention also provides an electronic device, including a processor and a memory, where the memory stores program code, and when the program code is executed by the processor, the processor is caused to execute the steps of the method.

[0015] The present invention also provides a storage medium storing a computer program or instruction, and when the computer program or instruction runs on a computer, it executes the steps of the method.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. High efficiency of data collection and processing: Traditional embroidery digitization technologies mainly rely on manual records and static image archiving. Manual records are inefficient and prone to distortion, and static image archiving cannot restore the three-dimensional structure and material texture of embroidery works, resulting in insufficient integrity and accuracy of data collection. The present invention realizes the all-round capture and structured processing of embroidery process data by integrating multi-dimensional sensing devices, greatly improving the data foundation and providing rich and systematic data resources for embroidery digitization research and applications.

[0017] 2. Precision of stitch identification and embroidery method reproduction: Existing stitch identification technologies mostly rely on manual records or simple image analysis and are difficult to accurately identify complex stitch structures, especially the layer-by-layer order and stitch density distribution of multi-layer embroidery threads. The deep learning-driven method of the present invention can more accurately identify complex stitch structures and restore the three-dimensional structure and process of embroidery works through dynamic reproduction technology, significantly improving the accuracy and restoration ability of embroidery digitization.

[0018] 3. Security of blockchain-based evidence storage: Most existing blockchain applications in digital embroidery protection only stay at the stage of simply uploading images to the blockchain, failing to deeply bind with the craft data of embroidered products, information of inheritors, etc., resulting in unclear ownership of digital resources and making it difficult to achieve effective intellectual property protection. The blockchain-based evidence storage technology of the present invention can not only clarify digital ownership, but also realize transparent management and effective protection of digital resources, completely solving the core problems of data islands, ambiguous ownership, and high rights protection costs in traditional evidence storage.

[0019] 4. Convenience of mobile applications: Traditional digital embroidery technologies lack convenient mobile application support, and users cannot access and manage digital embroidered products anytime and anywhere, resulting in a poor user experience. The mobile application of the present invention not only provides a more convenient user experience, but also expands the application scenarios of digital embroidery through augmented reality technology and blockchain-based evidence storage function, significantly improving the operation convenience and interaction experience of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is the hierarchical diagram of the method of the present invention.

[0021] Figure 2 is the flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] As Figure 1 shown, the embodiment of the present invention provides a full-process protection method for digital embroidery integrating deep learning and blockchain-based evidence storage, which is generally divided into five layers from the bottom layer to the top layer, namely: Data acquisition layer: used to acquire multi-modal data of embroidery craft data, including 3D laser scanning, IMU motion trajectory acquisition, and hyperspectral imaging.

[0023] Data processing layer: used for preprocessing and feature extraction of data, including adaptive Kalman filtering algorithm and key frame extraction technology.

[0024] Deep learning layer: used for stitch recognition and embroidery method reproduction, including 3D convolutional neural network and generative adversarial network.

[0025] Blockchain-based evidence storage layer: used for digital ownership management and NFT minting, including blockchain RC721 protocol and IPFS distributed storage.

[0026] Mobile application layer: used for mobile application deployment, including React Native framework and TensorFlow Lite engine.

[0027] As Figure 2 shown, the method includes the following steps: Step 1, data acquisition and preprocessing; Step 1.1, Before building the system, the embroidery work to be digitized needs to be placed within the scanning area of the 3D laser scanner, ensuring that the surface of the embroidery is flat to avoid wrinkles or obstructions. The operator wears or fixes a nine-axis inertial measurement unit device on the embroidery tool or the operator's limb to ensure that the device can accurately capture the movement trajectory of the needle tip.

[0028] Step 1.2, Start the 3D laser scanner, IMU device, and hyperspectral imaging system through the mobile application or console to begin collecting multimodal data of the embroidery work. The 3D laser scanner generates three-dimensional point cloud data with millimeter-level accuracy, the IMU device real-time collects the movement trajectory data of the needle tip, and the hyperspectral imaging system analyzes the spectral reflection characteristics of the embroidery thread material.

[0029] Step 1.3, Eliminate environmental noise from the collected raw data through an improved adaptive Kalman filtering algorithm, and compress redundant temporal information through key frame extraction technology. The preprocessed data can be viewed through the mobile application to ensure that the data quality meets the requirements of subsequent analysis.

[0030] Step 2, Stitch recognition and embroidery method reproduction; Step 2.1, Select the stitch recognition function through the mobile application, and the system automatically calls an improved 3D convolutional neural network to extract the spatial topological features of the stitches. The results of stitch recognition can be viewed in the application, and the system will generate a stitch path vector field and the physical properties of the embroidery thread.

[0031] Step 2.2, Select the embroidery method reproduction function, and the system generates a high-fidelity digital twin in real-time through the physics engine, including a 4K / 60fps process restoration animation, a physical simulation model based on PBR materials, and a lightweight AR / VR compatible format. The three-dimensional effect of the embroidery method reproduction can be viewed through AR / VR devices or the mobile application to experience the dynamic process of the embroidery work.

[0032] Step 3, Blockchain certification and digital ownership management; Step 3.1, Select the blockchain certification function through the mobile application, and the system automatically binds the three-dimensional geometric model, the process parameters of the deep learning analysis module, the hash value (SHA-3-512) of the digital twin generated by the dynamic modeling module, and the biometric characteristics of the inheritor (such as finger vein recognition, FAR < 0.0001%) to form a standardized metadata package (JSON-LD format). After confirming the content of the metadata package, the system ensures data traceability and anti-censorship through distributed storage in the InterPlanetary File System (IPFS).

[0033] Step 3.2: Select the NFT minting function. The system uses the ERC-721 protocol to mint unique NFTs for each digital embroidery, supporting multi-level ownership nesting, and the smart contract automatically enforces copyright rules. The NFT can be uploaded to the blockchain for trading through the built-in smart wallet and is synchronously updated to the cultural heritage platform in real time, ensuring the transparent management and effective protection of digital embroideries.

[0034] Step 4: User operation on the mobile application; Step 4.1: The user downloads and installs the "Embroidery Intelligence Connect" mobile application and logs in to the system through identity verification. The user can view their personal data space in the application and manage their digitized embroidery works and related metadata.

[0035] Step 4.2: The user selects the stitch recognition function through the application. The system calls the deep learning model in real time to recognize the stitches and generates a stitch path vector field and the physical properties of the embroidery thread. The stitch path guiding line and force feedback prompt can be viewed through the AR function to adjust the embroidery operation in real time.

[0036] Step 4.3: The user selects the blockchain deposit function through the application. The system automatically uploads the metadata package of the digital embroidery to the blockchain and mints a unique NFT. The NFT can be uploaded to the blockchain for trading through the built-in smart wallet, and the transaction records and ownership information can be viewed to ensure the transparent management and effective protection of digital embroideries.

[0037] Step 4.4: After the user completes the experiment, the system automatically releases the virtual machine resources and clears the experimental data and environment. The user can view the scores and feedback of the experimental results in the application and obtain learning suggestions and improvement opinions.

[0038] In this embodiment, the intangible cultural heritage embroidery work "Peony Picture of Suzhou Embroidery" (size approximately 120cm × 60cm) is taken as the specific implementation object, and the methods of this embodiment are combined to digitally protect and apply this work: 1. Use a FARO Focus S350 3D laser scanner with a point cloud accuracy of ±1mm, supporting the acquisition of millions of point clouds, to obtain high-precision three-dimensional texture data of the "Peony Picture of Suzhou Embroidery". The Xsens MVN IMU inertial measurement system with an error of less than 0.5° is used to achieve high-fidelity restoration of texture details and stitch features. The total amount of three-dimensional texture data generated by single-piece work acquisition is approximately 2.3GB, including approximately 31 million vertices and approximately 250 million pixels of texture maps, containing complete texture and geometric information. Compared with traditional two-dimensional picture archiving, this method completely retains the three-dimensional structure and material information of the embroidery, providing a data basis for subsequent identification and display.

[0039] 2. Feature extraction and stitch recognition are performed on the collected data using deep learning algorithms. A self-built stitch image library of 12,000 images is established, including flat stitches, random stitches, satin stitches, joining stitches, braid stitches, etc. Stitch feature extraction is carried out based on a deep convolutional network, and the recognition accuracy reaches 93.6%. The system dynamically demonstrates the stitch structure, and the average response time for dynamically demonstrating the construction path of a single stitch is less than 0.5 s. It supports users to gradually view the stitch structure and combination logic, assisting in the inheritance and learning of embroidery techniques, and reflecting the advantages of technology-assisted teaching.

[0040] 3. The recognized embroidery stitch data and core information such as the process parameters of the work are packaged to generate a digital copyright information package, which is stored distributedly through IPFS and stored on the chain to achieve the immutability and credible traceability of the data. The node redundant backup is greater than or equal to three copies, and the single-file storage upper limit is 5 GB. Further using NFT technology, a digital collectible with a unique identifier is generated for the "Peony Picture Embroidered in Suzhou Style" (TokenID: 0xF3A9D3...E0B2). The data size of the digital copyright information package is about 180 MB, including core data such as process flow videos, stitch structure files, 3D model files, etc., and supports online display and copyright trading.

[0041] 4. Users can scan physical objects or view the embroidery digital model through the mobile App to experience the interactive display with high definition, multi-angle, and multi-stitch details. They can view the copyright information, production process, and cultural background of the work, enhancing the immersive experience of digital display.

[0042] 5. Experimental effects and superiority are demonstrated; Experiments show that the method of this embodiment has the following advantages in practical applications: High data integrity: 3D high-precision modeling plus stitch recognition, with a realistic restoration effect; Strong ownership security: Blockchain deposit and NFT issuance ensure the security of digital assets; Good interactive experience: Supports AR dynamic display and user independent interaction; Wide application expansion: It can serve multiple scenarios such as digital cultural relics, cultural and creative product development, intangible cultural heritage education, and copyright protection.

[0043] The present invention provides a full-process protection method for embroidery digitization that integrates deep learning and blockchain deposit. There are many methods and ways to specifically implement this technical solution. The above description is only the preferred implementation manner of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the principle of the present invention, several improvements and retouches can be made, and these improvements and retouches should also be regarded as the protection scope of the present invention. Each component not clearly defined in this embodiment can be implemented using existing technologies.

Claims

1. A full-process protection method for embroidery digitization that integrates deep learning and blockchain evidence storage, characterized in that: The following steps are involved: Step 1, collecting multimodal data: using multimodal data collection technology to obtain multimodal data of embroidery works, including three-dimensional point cloud data, needle tip motion trajectory and material characteristics; Step 2, perform data preprocessing and feature extraction: the multimodal data collected in step 1 is subjected to an improved adaptive Kalman filter algorithm to eliminate environmental noise, and redundant time series information is compressed by key frame extraction technology, and finally a standardized process data set is obtained and stored in a database; Step 3: Perform deep learning-driven needlework recognition and embroidery reproduction: Use deep learning to drive needlework recognition and reproduce embroidery techniques through 3D convolutional neural networks, generative adversarial networks, and Canny edge detection algorithms; Step 4: Blockchain evidence storage and digital ownership management: Through the blockchain RC721 protocol, the digital ownership is clarified and resources are managed in batches; Step 5, mobile application deployment: Build mobile applications through a lightweight cross-platform framework to achieve digital protection of embroidery.

2. The method according to claim 1, characterized in that Step 1 includes the following steps: Step 1.1: Use a 3D laser scanner to perform non-contact scanning on the surface of the embroidery work, generate three-dimensional point cloud data with millimeter-level accuracy, and perform denoising through the neighborhood average filtering algorithm. The formula is: ， in, Indicates the first The original 3D point cloud coordinates, that is, each point Represents the 3D coordinates of an area of ​​the embroidery surface captured by the scanner ; is the number of valid points in the neighborhood, ; The smoothed 3D point cloud coordinates output after denoising; Step 1.2: The nine-axis IMU sensor deployed on the embroidery needle collects the needle tip motion data in real time, and the needle tip motion data is smoothed using the following formula: ， in, is the movement speed of the needle tip at time t; is the moving speed of the needle tip at time t after smoothing; is the smoothing coefficient, the value range is ; Step 1.3, analyze the spectral reflectance characteristics of the embroidery thread material through the hyperspectral imaging system, generate the material spectral fingerprint, and normalize the spectral data using the following formula: ， in, The wavelength The original spectral data at and are the minimum and maximum values ​​of the original spectral data, respectively; is the normalized spectral data; Step 1.4, identifying material composition and aging status, includes the following steps: Step 1.4.1, use Wiley Universal Spectral Database to analyze the composition of embroidery materials; Step 1.4.2, use support vector machine (SVM) to train the classification model, input the material spectrum fingerprint, and output the predicted material category; Step 1.4.3, calculate the spectral shift rate of the material : ， in It is the standard spectrum data of the material being tested.

3. The method according to claim 2, characterized in that Step 2 includes the following steps: Step 2.1, execute the improved adaptive Kalman filter algorithm, the improved state transfer matrix formula is: ， in, is the state transition matrix, is the control input matrix, is the control input vector, is the process noise; is the system state vector at time k: ， in, represents the three-dimensional coordinate position of the needle tip, The needle tip is calculated by the nine-axis IMU sensor. Direction speed; State transition matrix To reflect the change of the needle tip motion state, a discrete motion model is used: ， in, is the sampling time interval, the state transfer matrix The first three rows show the change of position over time: , the last three lines show that the speed remains unchanged: ; Control Input Matrix for: ， Control input vector Including the nine-axis IMU sensor to measure the acceleration of the needle tip in the x, y, and z directions : ， Get the influence of needle tip motion, including position change and speed change: Position changes: ; Speed ​​Change: ; in Indicates the speed in the x-axis direction at the current moment, Indicates the velocity in the x-axis direction at the previous moment; Process noise Represents the dynamic error of the system, which obeys Gaussian distribution : ; The tip acceleration noise and time interval Calculate the covariance matrix : ， in is the variance of the acceleration noise of the needle tip motion; Finally, the smooth needle tip trajectory processed by adaptive Kalman filtering is output ; Step 2.2, extract key frames: Calculate the compression rate using the following formula: : ， in, is the multimodal data collected in step 1, The amount of data after compression.

4. The method according to claim 3, characterized in that Step 3 includes the following steps: Step 3.1: Use the improved 3D convolutional neural network to extract the topological features of the stitch space. The formula is: ， in, is the input feature map. The parameters of the input feature map include ,in is the number of input channels, Respectively represent the depth, height and width of the stitch structure; is the convolution kernel weight, is the bias term, is the activation function, is the output feature map; Step 3.2, using the generative adversarial network GAN to generate the vector field of the stitch trajectory; Step 3.3, edge detection and optimization of the generated stitch path to ensure the accuracy of the reproduced embroidery method.

5. The method according to claim 4, characterized in that Step 3.2 includes: Generate the loss function of the generator of the adversarial network GAN for: ， Among them, the generator The input is , the output is the stitch trajectory vector field; the discriminator Used to determine whether the generated stitch path conforms to the real embroidery method; Loss of physical restraint; is the input random noise of the generator, which follows a Gaussian distribution ; Represents a random variable Expected value of the sampled result; hyperparameters For control The extent of its impact on the overall loss.

6. The method according to claim 5, characterized in that Step 3.3 includes: improving the Canny edge detection algorithm, the formula is: ， in represents the gradient amplitude, They are the gradients in the horizontal and vertical directions, respectively, and are calculated using the Sobel operator: ， in is the stitch path image generated in step 3.2, represents the convolution operation; Gradient Amplitude Perform edge detection and optimization for each pixel , calculate the gradient direction ; According to the gradient direction, compare the current pixel with the neighboring pixels. If the gradient amplitude of the current pixel is not the local maximum, it is set to 0.

7. The method according to claim 6, characterized in that Step 4 includes the following steps: Step 4.1: Extract the needle tip trajectory from step 2 , compressed data volume Combined with the stitch path image I in step 3 to generate a digital twin, a unique digital representation is obtained; The hash value of the digital twin is calculated by a hash algorithm and bound to the inheritor's biometric information to form a standardized metadata package to ensure the uniqueness and authenticity of the digital embroidery. The metadata package is decentralized and stored through the Interstellar File System IPFS. Step 4.2: Use the ERC-721 non-fungible token NFT standard to mint a unique non-fungible token NFT for each digital embroidery and give it an on-chain identity.

8. The method according to claim 7, characterized in that Step 5 includes: using the integrated TensorFlow Lite engine to deploy a pruned and optimized deep learning model that supports real-time recognition of at least 20 basic stitches in an offline state; using ARCore technology to achieve augmented reality interaction, users can view and interact with virtually displayed embroidery through mobile devices; users can encapsulate the parameters of each embroidery into a unique non-fungible token NFT in the ERC-1155 format. Each unique non-fungible token NFT will contain a unique digital certificate and copyright information, and can be traded on the chain through a built-in smart wallet.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores program codes, and when the program codes are executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 8.

10. A storage medium, characterized in that: A computer program or instruction is stored, and when the computer program or instruction is run on a computer, the steps of the method according to any one of claims 1 to 8 are executed.

Images