Tibetan costume process digital evaluation method and system based on image analysis

Abstract

The invention discloses a Tibetan costume process digital evaluation method and system based on image analysis, relates to the technical field of Tibetan costume image analysis, and aims to extract a stitch uniformity feature value, a pattern symmetry feature value and a color integrity feature value as surface quality features for preliminary compliance judgment. For the surface quality qualification, a time sequence quality vector sequence reflecting the long-term quality stability of the surface quality qualification is further extracted and constructed, the microscopic features and the time sequence are jointly input into a process stability prediction model, and a stability risk index is calculated, so that when the skills of the user are recessively degraded but the finished product is still qualified, the stability risk index is calculated. And a personalized pre-judgment correction scheme can be generated and executed in time. According to the method, accurate diagnosis and active intervention of hidden stability risks in the process inheritance process are realized, the spanning from single finished product quality inspection to long-term skill stability monitoring is realized, and potential operation flaws, which cause future quality fluctuation, of users can be actively found and corrected.

Description

Technical Field

[0001] This invention relates to the field of Tibetan clothing image analysis technology, and more specifically, this application relates to a digital evaluation method for Tibetan clothing craftsmanship based on image analysis. Background Technology

[0002] In application scenarios oriented towards large-scale teaching or personalized remote guidance, the traditional master-apprentice oral transmission model has revealed its limitations: the quality of transmission is difficult to guarantee a unified standard, and the valuable experience of outstanding craftsmen cannot be effectively preserved and replicated on a large scale.

[0003] To address the aforementioned problems, existing technologies have developed methods for quality assessment of finished garment products using digital image analysis. These methods primarily involve automatically detecting and measuring apparent attributes such as stitching, patterns, and colors by acquiring static images of the finished product. However, this assessment method, based on a single, static image of the finished product, has inherent limitations.

[0004] First, this kind of post-production evaluation cannot trace back and reveal the dynamic process that leads to the quality of the finished product. For example, it cannot capture the subtle tremors of the hand, the disorder of the needlework rhythm, or the lack of smooth transitions in the movements during the production process. These micro-operational characteristics are the key intrinsic factors affecting the stability of a craftsman's skills.

[0005] Secondly, this isolated, single-point assessment cannot depict the long-term evolution trend of learners' skill levels. Due to the lack of time-series modeling and analysis of historical process operation quality data, existing technologies struggle to construct a time-series quality vector sequence that reflects the learner's process stability. Consequently, they cannot conduct early diagnosis and quantitative assessment of potential risks in skill development, such as outputting a forward-looking stability risk index.

[0006] Therefore, existing technologies have only achieved preliminary automation of quality inspection, but have failed to build a digital assessment system that can deeply integrate process data and outcome data, and based on this, predict and intervene in the stability of learners' skills. This makes it difficult to fundamentally solve the core problem of the lack of targeted and predictive teaching guidance in the large-scale and personalized inheritance of Tibetan clothing craftsmanship in practice. Summary of the Invention

[0007] To address the aforementioned technical problems, this technical solution provides a digital evaluation method for Tibetan clothing craftsmanship based on image analysis, resolving the issues raised in the background section.

[0008] In a first aspect, embodiments of this application provide a digital evaluation method for Tibetan clothing craftsmanship based on image analysis, comprising the following steps: acquiring clothing quality images and time-series operation data of a target user; extracting features from the clothing quality images to obtain surface quality features; if all surface quality features are greater than or equal to a preset quality threshold, extracting features from the time-series operation data stream to obtain micro-operation features; constructing a time-series quality vector sequence based on the surface quality features of the target user; inputting the micro-operation features and the time-series quality vector sequence together into a pre-constructed process stability prediction model to output a stability risk index; and outputting a personalized correction scheme based on the stability risk index when the stability risk index exceeds a preset stability threshold.

[0009] Secondly, embodiments of this application provide a digital evaluation system for Tibetan clothing craftsmanship based on image analysis, comprising: a data acquisition module for acquiring clothing quality images of a target user through an image sensor and acquiring time-series operational data through a motion sensor; a quality feature set acquisition module for extracting features from the clothing quality images to obtain surface quality features; a micro-operational feature set acquisition module for extracting features from the time-series operational data stream to obtain micro-operational features if all features in the surface quality features are greater than or equal to a preset quality threshold; a vector sequence acquisition module for constructing a time-series quality vector sequence based on the surface quality features of the target user; a risk index calculation module for inputting the micro-operational features and the time-series quality vector sequence into a pre-constructed process stability prediction model and outputting a stability risk index; and a scheme output module for outputting a personalized correction scheme based on the stability risk index when the stability risk index exceeds a preset stability threshold.

[0010] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0011] 1. It not only checks the finished product's quality by examining surface quality characteristics, but more importantly, it analyzes micro-operational data such as jitter spectrum characteristics and rhythm consistency characteristics, and combines this with the long-term fluctuations reflected in the constructed time-series quality vector sequence to comprehensively assess the stability risk index. This allows for the prediction of potential future problems caused by trainees' poor operating habits from a single qualified operation, enabling early diagnosis of the most challenging hidden risks in the transmission process. It represents a leap from static compliance inspection to dynamic stability prediction, solving the problem that existing technologies cannot predict potential risks.

[0012] 2. By combining the micro-operational characteristics reflecting the process with the derived indicators of the time-series quality vector sequence reflecting the long-term stability of the results in a pre-defined two-dimensional space, a comprehensive risk feature vector is formed. This data processing logic profoundly reveals the intrinsic connection between the two, ensuring that the final risk assessment is not based on isolated operations or results, but on their correlation. This provides precise and root-cause-based adjustment directions for subsequent personalized correction schemes, establishing a causal relationship between process data and long-term quality results, and solving the problems of single diagnostic dimensions and inaccurate root cause identification in existing technologies.

[0013] 3. Based on the implementation effect of the personalized correction scheme, data is collected again and a stable risk index sequence curve is generated. Then, the model parameters are dynamically adjusted using this curve, enabling the system to adapt to the learning curves of different users and continuously optimize its prediction and intervention accuracy as data accumulates. It becomes an intelligent mentor that gets smarter with use, ultimately ensuring that the regulation of the personalized inheritance path sequence becomes more and more precise, significantly improving inheritance efficiency. This forms a self-evolving closed loop of evaluation, intervention, and re-evaluation, solving the pain points of low intelligence level and inability to continuously optimize existing systems. Attached Figure Description

[0014] Figure 1 A schematic diagram of the structure of the image analysis-based digital evaluation method for Tibetan clothing craftsmanship provided in this application embodiment;

[0015] Figure 2 This is a schematic diagram of the logic flow of the self-optimizing closed loop of the regression prediction model provided in the embodiments of this application.

[0016] Figure 3 A schematic diagram of a digital evaluation system for Tibetan clothing craftsmanship based on image analysis provided in an embodiment of this application. Detailed Implementation

[0017] This application's embodiments address the technical problem of insufficient accuracy in existing digital assessment systems that integrate process and result data and use this data to predict and intervene in learners' skill stability through an image analysis-based digital assessment method for Tibetan clothing craftsmanship.

[0018] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0019] like Figure 1The diagram shown is a structural schematic of the image analysis-based digital evaluation method for Tibetan clothing craftsmanship provided in this application embodiment. It includes the following steps: acquiring clothing quality images and time-series operational data from a target user; extracting features from the clothing quality images to obtain surface quality features; if all surface quality features are greater than or equal to a preset quality threshold, extracting features from the time-series operational data stream to obtain micro-operational features; constructing a time-series quality vector sequence based on the target user's surface quality features; inputting the micro-operational features and the time-series quality vector sequence into a pre-constructed process stability prediction model to output a stability risk index; and outputting a personalized correction scheme based on the stability risk index when the stability risk index exceeds a preset stability threshold.

[0020] Acquiring garment quality images and time-series operation data from the target user can be understood as collecting static and dynamic information related to garment production through various methods. For example, high-resolution images of finished garments can be captured using industrial cameras to obtain garment quality images, while accelerometers and gyroscopes in wearable devices can record the user's hand movement trajectories during the production process as time-series operation data.

[0021] The process for obtaining static parameters such as preset quality threshold, preset stability threshold, first slope threshold, and second slope threshold is as follows:

[0022] Statistical analysis was used to analyze the surface quality characteristics of a large number of qualified Tibetan clothing and handicrafts, and a certain probability distribution, such as the 10th percentile, was taken as the initial value of the preset quality threshold. Similarly, the preset stability threshold was determined by correlation analysis between the historical user stability risk index and substantive craftsmanship degradation events.

[0023] The first and second slope thresholds, among other control parameters, are selected during the model optimization phase using cross-validation techniques on the validation set to find the values ​​that optimize model performance. All these thresholds are configurable parameters in the system, allowing for fine-tuning based on specific application scenarios.

[0024] By integrating the surface quality characteristics of finished garments with the temporal dynamic data of the manufacturing process, an assessment system capable of quantifying skill stability has been constructed. Compared to existing technologies that rely solely on static images of finished products for post-process evaluation, this application not only traces back to the microscopic operational details of the manufacturing process but also depicts the long-term evolution trend of learners' skill levels, thereby enabling early diagnosis and quantitative assessment of skill risks. This improvement fundamentally solves the core problem of traditional methods lacking targeted and predictive teaching guidance in both large-scale and personalized skill transmission.

[0025] Furthermore, surface quality features are obtained, specifically including: surface quality features such as uniformity feature value of stitches, symmetry feature value of pattern, and color integrity feature value; edge detection algorithm is used to extract the stitch contours in the clothing quality image, and the reciprocal of the standard deviation of the distance between adjacent stitch contours is calculated as the uniformity feature value of stitches; the clothing quality image is mirrored along a preset symmetry axis, and the cross-correlation coefficient between the original pattern and the mirrored pattern is calculated, and the normalized cross-correlation coefficient is used as the pattern symmetry feature value; a region growing algorithm is used to segment the target color region in the clothing quality image, and the ratio of the area of ​​the target color region to the area of ​​the preset standard region is calculated as the color integrity feature value.

[0026] In this embodiment, the uniformity feature value of the stitches refers to the feature value calculated based on the statistical distribution characteristics of the distance between adjacent stitch contours after the stitch contours in the clothing quality image are extracted by the edge detection algorithm. This can be achieved by calculating the standard deviation of the distance between adjacent stitch contours and taking its reciprocal, which is used to quantify the uniformity of the stitches.

[0027] Pattern symmetry feature values ​​can be understood as normalized cross-correlation coefficients obtained by mirroring the clothing quality image along a preset symmetry axis, based on the mathematical correlation between the pixels of the original pattern and the mirrored pattern. This is achieved by calculating the cross-correlation coefficients and performing normalization, which is used to accurately capture the geometric symmetry of the pattern and ensure that the feature values ​​are comparable on a uniform scale.

[0028] Color integrity feature value refers to the feature value calculated based on the ratio of the area of ​​the target color region to the area of ​​a preset standard region after segmenting the target color region using a region growing algorithm. It can be achieved through region connectivity analysis and by calculating the area ratio, and is used to accurately locate specific color regions.

[0029] By covering key visual dimensions of garment craftsmanship, including stitch uniformity, pattern symmetry, and color integrity, a comprehensive assessment was ensured. This multi-dimensional quantitative approach not only solved the problem of inaccurate feature extraction but also laid a solid foundation for subsequent prediction of stable risk indices, thereby improving the accuracy and reliability of the overall assessment system.

[0030] Furthermore, micro-operational features are obtained, specifically including: micro-operational features such as jitter spectrum feature values, rhythm consistency feature values, and operation smoothness feature values; a fast Fourier transform is performed on the hand movement trajectory signal in the time-series operation data to obtain the hand movement trajectory spectrum; a bandpass filter is used to extract the signal components of the preset frequency band of the hand movement trajectory spectrum to obtain the signal energy value, and the ratio of the signal energy value to the total signal energy value is calculated as the jitter spectrum feature value; the clothing process pressure sequence in the time-series operation data is obtained, its cross-correlation coefficient with the pre-stored standard pressure rhythm sequence is calculated and normalized; the maximum value of the normalized cross-correlation coefficient is obtained through a sliding window and used as the rhythm consistency feature value; the hand movement trajectory signal in the time-series operation data is segmented according to the historical average action cycle of the target user; the sample variance of the hand movement trajectory signal of the historical average action cycle is statistically analyzed using a sliding window, the sample variance is standardized, and the reciprocal of the standardized sample variance is used as the operation smoothness feature value.

[0031] In this embodiment, the jitter spectrum characteristic value refers to an index used to quantify the jitter intensity of the hand within a specific frequency range. It can be achieved by combining a fast Fourier transform with a bandpass filter to focus on the typical jitter frequency range in the process operation and eliminate high-frequency noise and low-frequency trend interference.

[0032] The rhythm consistency feature value can be understood as an indicator that measures the similarity between actual operational pressure and standard rhythm. It can be achieved by calculating the cross-correlation coefficient and combining it with the sliding window technique. It is used to dynamically capture the best matching moment between the pressure rhythm and the standard pattern, avoiding misjudgment due to short-term fluctuations.

[0033] Operational fluency features are indicators that reflect the continuity and stability of hand movements. They can be achieved by combining a segmentation method based on the historical average movement cycle with the variance of the sliding window statistical sample. This ensures that feature extraction is consistent with the user's own rhythm, while standardization eliminates the influence of individual differences.

[0034] Combining this process with acquiring images of clothing quality and time-series operational data from target users provides accurate input for process stability prediction models. Through multi-dimensional feature extraction from the time-series operational data, not only is a comprehensive characterization of hand tremors, pressure rhythm, and movement smoothness achieved, but a reliable foundation is also laid for the subsequent output of the stability risk index. This deep integration of process and outcome data effectively solves the problem of existing technologies failing to accurately reflect the inherent stability risks of the process.

[0035] Furthermore, the specific process of constructing the time-series quality vector sequence is as follows: a process quality space vector is constructed through surface quality features, wherein the uniformity feature value of the line trace, the symmetry feature value of the pattern, and the integrity feature value of the color are respectively used as the three orthogonal dimensions of the process quality space vector in the preset first coordinate system; several process quality space vectors corresponding to the target user at different times are obtained in time sequence to form a time-series quality vector sequence; the vector sequence discreteness is obtained by calculating the average Euclidean distance between adjacent process quality space vectors in the three-dimensional space of the preset first coordinate system.

[0036] In this embodiment, the process quality space vector refers to a mathematical representation that maps surface quality features to a multi-dimensional space, providing a spatial structure that can independently characterize quality changes in different process dimensions.

[0037] The preset first coordinate system can be understood as a reference frame for defining the process quality space vector. Specifically, it can be implemented through a Cartesian coordinate system or other orthogonal coordinate systems, with the aim of ensuring the orthogonality and independence between the dimensions.

[0038] Specifically, vector sequence dispersion is a quantitative indicator that measures the fluctuation range of a time-series quality vector sequence. It can be achieved by calculating the Euclidean distance between adjacent vectors and is used to reflect the stability of process quality over time.

[0039] By constructing a process quality space vector using the uniformity feature value of the line trace, the symmetry feature value of the pattern, and the integrity feature value of the color as three orthogonal dimensions, the quality changes of different process dimensions can be represented on independent spatial axes, avoiding mutual interference between features, thus solving the problem that a simple time series cannot reflect the spatial relationship of quality features.

[0040] Based on this, the process quality spatial vectors of the target user are obtained sequentially over time and a time-series quality vector sequence is constructed. This process transforms static quality assessment into a dynamic quality evolution trajectory, effectively capturing the continuous change pattern of process quality over time. Furthermore, the vector sequence dispersion is obtained by calculating the average Euclidean distance between adjacent process quality spatial vectors in three-dimensional space. This quantitative indicator objectively reflects the geometric amplitude of process quality fluctuations, providing a reliable basis for subsequent stability risk index prediction.

[0041] It not only solves the problem of lack of multi-dimensional spatial representation and sequence discretization mechanism in the construction of time-series quality vector sequences, but also provides dynamic geometric characteristics and measurability support for process stability assessment.

[0042] Furthermore, constructing the time-series quality vector sequence also includes: performing cluster analysis on the time-series quality vector sequence to divide several process quality space vectors into several clusters; calculating the centroid vector of each cluster and calculating the average distance between all centroid vectors as the quality fluctuation range; normalizing the quality fluctuation range and the vector sequence dispersion; if the normalized quality fluctuation range is greater than or equal to the normalized vector sequence dispersion, then the normalized quality fluctuation range and the normalized vector sequence dispersion are weighted and summed to obtain the first input value of the process stability prediction model; if the normalized quality fluctuation range is greater than or equal to the normalized vector sequence dispersion, then the absolute value of the weighted subtraction of the normalized quality fluctuation range and the normalized vector sequence dispersion is taken to obtain the second input value of the process stability prediction model.

[0043] In this embodiment, cluster analysis refers to the process of grouping data points with similar characteristics using a specific algorithm. It can be implemented using methods such as K-means clustering, hierarchical clustering, or density clustering to reveal the inherent grouping patterns of process quality data.

[0044] The centroid vector refers to the center point of all vectors in each cluster. It can be calculated by the arithmetic mean or weighted average of all vectors in the cluster and is used to quantify the overall difference between different quality levels.

[0045] Normalization refers to the process of converting data with different dimensions into a unified standard. It can be achieved by methods such as min-max normalization, Z-score standardization, or decimal scaling normalization. It is used to eliminate differences in dimensions and scale effects, making the two comparable under a unified standard.

[0046] By introducing cluster analysis, the process quality spatial vector is divided into several clusters, thereby identifying the stable state intervals of different quality levels and avoiding the shortcomings of relying solely on the distance between adjacent points while ignoring global fluctuation characteristics. Based on this, by calculating the centroid vector of each cluster and defining the quality fluctuation range based on the average distance between all centroid vectors, the overall difference in quality levels is directly quantified.

[0047] This fluctuation range complements the vector sequence dispersion: the dispersion focuses on the details of local changes, while the fluctuation range reflects the intensity of macroscopic fluctuations. The combination of the two can comprehensively characterize the stability dimension of process quality.

[0048] By normalizing the quality fluctuation range and the vector sequence dispersion, the differences in dimensions and the influence of scale are eliminated, laying the foundation for subsequent dynamic decision-making. Furthermore, the conditional judgment mechanism dynamically generates the input values ​​for the process stability prediction model based on the relative magnitudes of the normalized fluctuation range and dispersion: when the fluctuation range is relatively large, a weighted summation is used to highlight the dominant influence of overall fluctuation on stability.

[0049] When the fluctuation range is relatively small, the detailed features of local dispersion are enhanced by taking the absolute value of the weighted subtraction. This adaptive combination strategy ensures that the input value can accurately respond to the actual fluctuation characteristics of the sequence, significantly improving the sensitivity and diagnostic accuracy of the process stability prediction model to skill risks.

[0050] It effectively compensates for the limitations of a single dispersion index, and provides richer input features for process stability prediction, ultimately achieving a more reliable stability risk index output.

[0051] Furthermore, the specific output process of the stability risk index is as follows: The jitter spectrum feature value, rhythm consistency feature value, and operation smoothness feature value in the micro-operational features are normalized to obtain normalized jitter spectrum feature values, normalized rhythm consistency feature values, and normalized operation smoothness feature values, which are then used to form a three-dimensional feature sample vector. This feature sample vector is input into a pre-constructed principal component analysis model. The principal component analysis model projects the feature sample vector onto the principal component directions trained from historical user data, and the resulting scalar value is used as the comprehensive micro-feature value. This comprehensive micro-feature value is used as a vector on the horizontal axis in a two-dimensional space of a preset second coordinate system, and the first input value or the second input value of the process stability prediction model is used as a vector on the vertical axis in the same two-dimensional space. These two values ​​are then combined to obtain a comprehensive risk feature vector. This comprehensive risk feature vector is then input into a pre-constructed regression prediction model to output the stability risk index.

[0052] In this embodiment, the pre-built principal component analysis model is constructed as follows:

[0053] Collect a large-scale dataset of micro-operational features from a historical user group. This dataset contains jitter spectrum features, rhythm consistency features, and operation smoothness features for each historical user.

[0054] The three feature values ​​in the dataset are normalized to eliminate the influence of dimensions and form a standardized historical feature matrix.

[0055] Principal component analysis (PCA) is performed on this historical feature matrix. The specific steps include: calculating the covariance matrix of the matrix; performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors; arranging the eigenvalues ​​in descending order and selecting the eigenvector corresponding to the largest eigenvalue as the first principal component direction.

[0056] Saving this first principal component direction vector constitutes the pre-constructed principal component analysis model. When new data is input, simply perform a dot product operation between the feature sample vector and this direction vector to obtain the comprehensive micro-feature values.

[0057] The pre-built regression prediction model is constructed as follows:

[0058] A training sample set is constructed based on historical data. The input of each sample is a comprehensive risk feature vector, which is synthesized from the comprehensive micro-feature values ​​of historical users and the first or second input value of the process stability prediction model; the output of each sample is a corresponding stability label value, which is evaluated by experts or calculated based on the long-term quality decay rate.

[0059] Support vector regression (SVR) is used as the model framework. The model is trained using a prepared training sample set, and the optimal model parameters are found through a sequence minimization algorithm to minimize the error between the model's predicted values ​​and the true stability label values.

[0060] The model's predictive performance is validated using a reserved test set to ensure its generalization ability. Finally, the trained, fixed-parameter vector regression model is saved as a pre-built regression prediction model for online prediction of the stability risk index of new users.

[0061] Normalization refers to the process of mapping eigenvalues ​​with different dimensions and ranges to a unified numerical interval through mathematical transformations. It is used to eliminate weight imbalances caused by differences in dimensions. In practical applications, minimum-maximum normalization, Z-score standardization, or other linear transformation methods can be used. The feature sample vector is a multidimensional vector composed of multiple normalized eigenvalues, which provides a standardized data foundation for subsequent principal component analysis.

[0062] Principal component analysis (PCA) is a dimensionality reduction algorithm based on statistical principles. Its core lies in projecting high-dimensional data into a low-dimensional space through linear transformation while preserving the main variability information. Typically, this model can be implemented using methods such as covariance matrix decomposition or singular value decomposition. Based on this, the comprehensive micro-eigenvalues ​​refer to the single scalar values ​​extracted after PCA, which are used to focus on key operational factors affecting process stability.

[0063] Vector synthesis refers to combining two vectors according to certain rules to form a new vector, which is used to fuse dynamic operational details with static quality results. This can be achieved using vector addition or other linear combination methods. The comprehensive risk feature vector is a multi-dimensional vector obtained through vector synthesis that simultaneously reflects operational stability and quality fluctuation trends.

[0064] The surface quality characteristics and micro-operational characteristics form an organic whole. By combining the normalized micro-operational characteristics with the input values ​​of the process stability prediction model constructed based on time-series quality vector sequences, accurate prediction of process stability is achieved. This design not only overcomes the one-sidedness of assessment caused by relying on a single data source, but also ensures the objectivity and operability of risk assessment, providing a reliable basis for subsequent intervention.

[0065] Furthermore, the personalized correction scheme based on the stable risk index also includes: forming a personalized correction scheme according to the stable risk index and a preset mapping rule; processing the clothing quality image and time-series operation data of the target user when completing the personalized correction scheme to obtain the stable risk index and combining them according to the number corresponding to the personalized correction scheme to obtain the stable risk index sequence curve; and optimizing and adjusting the regression prediction model based on the stable risk index sequence curve.

[0066] In this embodiment, the stability risk index is a quantitative indicator that integrates micro-operational characteristics with the dynamic correlation of time-series quality vectors. It can be calculated by multi-dimensional feature weighted fusion to ensure that the correction scheme can accurately locate the root cause of stability problems in individual operations.

[0067] The preset mapping rules are constructed as follows:

[0068] Based on the experience of domain experts, an initial mapping rule base was established. This rule base defines the correspondence between different stable risk index ranges and a series of specific training task identifiers, such as anti-interference training tasks, rhythm synchronization training tasks, motion connection training tasks, and quality stability consolidation tasks. For example, when the index is in the medium-to-high risk range and the jitter spectrum characteristic value is abnormal, it is mapped to an anti-interference training task.

[0069] The system continuously collects stable risk index sequence curves of users after implementing different personalized correction schemes. By analyzing this data, the initial rule base is optimized and adjusted, such as correcting the boundaries of the index intervals and adding new task combinations, so that the mapping relationship can more effectively guide users' skill improvement. The final stable and effective set of rules constitutes the preset mapping rules.

[0070] A stability risk index sequence curve is a time-series curve composed of multiple stability risk indices arranged in chronological order. It can be generated by sliding window algorithm or time series analysis method and is used to reveal the trend of the impact of a specific correction scheme on process stability.

[0071] By combining the stable risk index sequence curve with the optimization process of the regression prediction model, not only is the problem of model adaptive adjustment solved, but the accuracy and sustainability of the correction scheme are also improved. This method is particularly suitable for skill training scenarios that require long-term tracking and optimization, and can effectively support the needs of large-scale and personalized inheritance of Tibetan clothing craftsmanship.

[0072] Furthermore, the regression prediction model is optimized and adjusted based on the stability risk index sequence curve. Specifically, this includes selecting several monitoring points within a preset time window for the stability risk index sequence curve. Using the sequence number of the monitoring points at each time point, a stable risk index sequence is obtained, arranged according to the monitoring points at each time point. Calculate the instantaneous slope and acceleration of the stability risk index at the current moment; update the learning rate of the regression prediction model based on the instantaneous slope and acceleration to obtain the updated regression prediction model.

[0073] In this embodiment, Figure 2 This is a schematic diagram of the logic flow of the self-optimizing closed loop of the regression prediction model provided in the embodiments of this application.

[0074] The stability risk index sequence curve refers to a discrete data sequence formed by sampling the stability risk index of a target user at different times and arranging it in chronological order. It can be implemented using sliding window technology or fixed time interval sampling and is used to capture the dynamic evolution of process stability risk.

[0075] The instantaneous slope refers to the ratio of the difference in the stable risk index between two adjacent monitoring points to the time interval. It can be achieved through differential calculation or linear fitting methods and is used to reflect the instantaneous rate of risk change.

[0076] The change acceleration refers to the trend of instantaneous slope change, which can be achieved through quadratic difference or polynomial fitting methods to capture the acceleration or deceleration characteristics of the rate of risk change.

[0077] By constructing a stable risk index sequence curve, the dynamic evolution of process stability risk is discretized into calculable time-series data, thus providing a foundation for subsequent analysis. Based on this, the calculation of the instantaneous slope is used to quantify the immediate rate of risk change, while the introduction of the acceleration of change further reveals the acceleration or deceleration trend of this rate. Together, they constitute a dual quantitative characterization of the dynamic characteristics of risk evolution.

[0078] Based on the results of instantaneous slope and acceleration, the learning rate of the regression prediction model is adaptively adjusted through a piecewise function mechanism: when the risk rises rapidly, the learning rate is increased to accelerate the model response, and when the risk changes steadily, the learning rate is decreased to maintain model stability.

[0079] This dynamic control method ensures that the pace of model updates is strictly synchronized with the trend of risk evolution, thereby solving the problem of model update lag caused by the lack of a refined quantification mechanism.

[0080] Combined with the aforementioned content, this approach can more comprehensively depict the long-term evolution trend of learners' skill levels and enhance the model's ability to capture early signs of process stability risks by dynamically adjusting the learning rate, thereby avoiding prediction bias caused by update lag and significantly improving the timeliness and accuracy of process stability prediction.

[0081] Furthermore, the specific process of updating the learning rate of the regression prediction model based on the instantaneous slope and the changing acceleration is as follows: when When the instantaneous slope is at a certain moment, the average value of the historical data is directly used as the instantaneous slope of the monitoring point at that moment; when... Calculate the instantaneous slope. The formula is: ,in, Indicates the first Stability risk index at each monitoring point at any given time. Indicates the first The stability risk index of each monitoring point at any given time; when When the historical data of the changing acceleration is used, the average value is directly taken as the changing acceleration at the monitoring point at this moment; when Calculate the changing acceleration. The formula is: ,in, Indicates the first The instantaneous slope of the monitoring point at each moment; when When, the original learning rate is not updated; when At that time, the updated learning rate is obtained. The piecewise function formula is: ,in, This represents the original learning rate. This represents the preset first threshold slope. This represents the preset second threshold slope. Indicates the slope as the first weight. This indicates the second weight of the slope. Indicates the third weight of the slope. For acceleration weights, and satisfying .

[0082] In this embodiment, the instantaneous slope refers to the ratio of the change in the stability risk index at adjacent moments to the time interval. It can be implemented by a difference algorithm to capture the short-term fluctuation trend of the stability risk index.

[0083] The acceleration of change can be understood as the rate of change of the instantaneous slope, which is obtained through two consecutive difference operations and aims to reveal the dynamic characteristics of the trend of change of the stable risk index.

[0084] Piecewise function mechanism is a conditional learning rate adjustment strategy that can adaptively adjust the learning rate according to different changing scenarios, thereby improving the model's ability to respond to changes in risk.

[0085] A piecewise function mechanism is used to dynamically adjust the learning rate: when a rapid and accelerating increase in the risk index is detected, the learning rate is significantly increased using high-weight parameters to ensure that the model can quickly respond to the trend of risk deterioration.

[0086] When the risk index drops rapidly but the downward trend weakens, the learning rate is appropriately reduced by using medium-weight parameters to avoid over-adjustment of the model and causing oscillations. In other stable or complex change scenarios, low-weight parameters are used for conservative fine-tuning to ensure the robustness of the model optimization process.

[0087] It not only solves the problem of slow or overly sensitive response to risk changes during model optimization, but also highlights the priority of risk prevention in process stability assessment by setting reasonable weight parameters, thereby accurately adapting to the complex characteristics of dynamic risk evolution in Tibetan clothing craftsmanship stability assessment.

[0088] like Figure 3 The diagram shown is a schematic of a digital evaluation system for Tibetan clothing craftsmanship based on image analysis provided in this application embodiment. This application embodiment provides a digital evaluation system for Tibetan clothing craftsmanship based on image analysis, comprising: a data acquisition module: used to acquire clothing quality images of a target user through an image sensor and time-series operational data through a motion sensor; a quality feature set acquisition module: used to extract features from the clothing quality images to obtain surface quality features; a micro-operational feature set acquisition module: used to extract features from the time-series operational data stream to obtain micro-operational features if all features in the surface quality features are greater than or equal to a preset quality threshold; a vector sequence acquisition module: used to construct a time-series quality vector sequence based on the surface quality features of the target user; a risk index calculation module: used to input the micro-operational features and the time-series quality vector sequence into a pre-constructed process stability prediction model and output a stability risk index; and a scheme output module: used to output a personalized correction scheme based on the stability risk index when the stability risk index exceeds a preset stability threshold.

[0089] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0090] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, are implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0091] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0092] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0093] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0094] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for digital evaluation of Tibetan costume crafts based on image analysis, characterized in that, The method comprises the following steps: obtaining clothing quality images and time-series operation data of a target user; extracting features from the clothing quality images to obtain surface quality features; if the surface quality features are all greater than or equal to a preset quality threshold, then extracting features from the time-series operation data stream to obtain microscopic operation features; constructing a time-series quality vector sequence based on the surface quality features of the target user; inputting the microscopic operation features and the time-series quality vector sequence into a pre-constructed process stability prediction model to output a stability risk index; when the stability risk index exceeds a preset stability threshold, outputting a personalized correction scheme based on the stability risk index.

2. The image analysis-based digital evaluation method of Tibetan costume crafts according to claim 1, characterized in that, The surface quality features are obtained by: the surface quality features include stitch uniformity feature values, pattern symmetry feature values, and color integrity feature values; an edge detection algorithm is used to extract stitch contours in the clothing quality images, and the reciprocal of the standard deviation of the distance between adjacent stitch contours is calculated as the stitch uniformity feature value; the clothing quality image is flipped along a preset symmetry axis, the cross-correlation coefficient of the original pattern and the mirror image is calculated, and the normalized cross-correlation coefficient is used as the pattern symmetry feature value; a region growing algorithm is used to segment the target color region in the clothing quality image, and the area ratio of the target color region to the preset standard region is calculated as the color integrity feature value.

3. The image analysis-based digital evaluation method of Tibetan costume crafts according to claim 1, characterized in that, The microscopic operation features are obtained by: the microscopic operation features include jitter spectrum feature values, rhythm consistency feature values, and operation fluency feature values; the hand movement trajectory signal in the time-series operation data is subjected to fast Fourier transform to obtain a hand movement trajectory spectrum, a bandpass filter is used to extract signal components in a preset frequency band of the hand movement trajectory spectrum, the signal energy value is obtained and the ratio to the total signal energy value is calculated as the jitter spectrum feature value; the clothing process pressure sequence in the time-series operation data is obtained, the cross-correlation coefficient with the pre-stored standard pressure rhythm sequence is calculated and normalized; the maximum value of the normalized cross-correlation coefficient is obtained by sliding window sampling, and is taken as the rhythm consistency feature value; the hand movement trajectory signal in the time-series operation data is segmented according to the historical average motion period of the target user; the sample variance of the hand movement trajectory signal of the historical average motion period is calculated by sliding window statistics, and the reciprocal of the standardized sample variance is taken as the operation fluency feature value.

4. The image analysis-based digital evaluation method of Tibetan costume crafts according to claim 2, characterized in that, The specific process of constructing the time-series quality vector sequence is: constructing a process quality space vector through the surface quality features, wherein the stitch uniformity feature value, the pattern symmetry feature value, and the color integrity feature value are taken as three orthogonal dimensions of the process quality space vector in a preset first coordinate system; obtaining a plurality of process quality space vectors corresponding to different time instants of the target user in chronological order to form a time-series quality vector sequence; obtaining the vector sequence dispersion by calculating the average Euclidean distance between adjacent process quality space vectors in the three-dimensional space of the preset first coordinate system.

5. The image analysis-based digital evaluation method of Tibetan costume crafts according to claim 4, characterized in that, The construction of the time-series quality vector sequence further comprises: performing cluster analysis on the time-series quality vector sequence to divide the plurality of process quality space vectors into a plurality of clusters; Calculate the centroid vector of each cluster, and calculate the average distance between all centroid vectors as the mass fluctuation range quantity; Normalize the mass fluctuation range quantity and the vector sequence dispersion, if the normalized mass fluctuation range quantity is greater than or equal to the normalized vector sequence dispersion, then the normalized mass fluctuation range quantity and the normalized vector sequence dispersion are weighted and summed to obtain the first input value of the process stability prediction model; If the normalized mass fluctuation range quantity is greater than or equal to the normalized vector sequence dispersion, then the absolute value of the result of the weighted subtraction of the normalized mass fluctuation range quantity and the normalized vector sequence dispersion is obtained as the second input value of the process stability prediction model.

6. The image analysis-based digital evaluation method of Tibetan costume crafts according to claim 4, characterized in that, The specific output process of the stability risk index is: The jitter spectrum feature value, the rhythm consistency feature value and the operation fluency feature value in the micro-operation feature are normalized to obtain the normalized jitter spectrum feature value, the normalized rhythm consistency feature value and the normalized operation fluency feature value, and a three-dimensional feature sample vector is formed accordingly; The feature sample vector is input into the pre-constructed principal component analysis model, and the feature sample vector is projected onto the principal component direction trained by the historical user data, and the obtained scalar value is taken as the comprehensive micro-feature value; The comprehensive micro-feature value is taken as a vector on the horizontal coordinate in the two-dimensional space of the pre-set second coordinate system, and the first input value of the process stability prediction model or the second input value of the process stability prediction model is taken as a vector on the vertical coordinate in the two-dimensional space of the pre-set second coordinate system, and the two are vector synthesized to obtain a comprehensive risk feature vector; The comprehensive risk feature vector is input into the pre-constructed regression prediction model, and the stability risk index is output.

7. The image analysis-based digital evaluation method of Tibetan costume crafts according to claim 6, characterized in that, Based on the stability risk index, an individual correction scheme is output, further comprising: According to the stability risk index, an individual correction scheme is formed according to a pre-set mapping rule; According to the clothing quality image and the time sequence operation data of the target user when completing the individual correction scheme, the stability risk index is processed again to obtain a stability risk index sequence curve according to the number combination corresponding to the individual correction scheme, and the regression prediction model is optimized and adjusted according to the stability risk index sequence curve.

8. The image analysis-based digital evaluation method of Tibetan costume crafts according to claim 7, characterized in that, The optimization and adjustment of the regression prediction model according to the stability risk index sequence curve specifically includes: For the stable risk index sequence curve, several time monitoring points are taken in a preset time window, is the serial number of the time monitoring point, and the stable risk index sequence arranged according to the time monitoring point is obtained ; Calculate the instantaneous slope of the stability risk index at the current time and its change acceleration; According to the instantaneous slope and the change acceleration, the learning rate of the regression prediction model is updated to obtain an updated regression prediction model.

9. The image analysis-based digital evaluation method of Tibetan costume crafts according to claim 8, characterized in that, The specific process of updating the learning rate of the regression prediction model according to the instantaneous slope and the change acceleration is: When the average value of the history data of the instantaneous slope is directly taken as the instantaneous slope of the monitoring point at this moment; When the instantaneous slope is calculated by the formula: wherein, represents the stability risk index of the monitoring point at the time point, represents the stability risk index of the monitoring point at the time point; When the average value of the history data of the change acceleration is directly taken as the change acceleration of the monitoring point at this moment; When the formula for the change acceleration is wherein, represents the instantaneous slope of the monitoring point at the nth time instant; When the original learning rate is not updated; When the updated learning rate is obtained, the piecewise function formula is: wherein, denotes the original learning rate, denotes a preset slope first threshold value, denotes a preset slope second threshold value, denotes a slope first weight, denotes a slope second weight, denotes a slope third weight, is an acceleration weight, and satisfies .

10. A digital evaluation system for Tibetan costume crafts based on image analysis, characterized in that, It includes: A data acquisition module for acquiring clothing quality images of a target user through an image sensor and acquiring time sequence operation data through a motion sensor; A quality feature set acquisition module for extracting features from the clothing quality images to obtain surface quality features; An operation feature set acquisition module for extracting features from the time sequence operation data stream to obtain micro-operation features if all features in the surface quality features are greater than or equal to a pre-set quality threshold. The vector sequence acquisition module is configured to construct a time-series quality vector sequence based on the surface quality characteristics of the target user. The risk index calculation module is configured to input the micro-operation characteristics and the time-series quality vector sequence into a pre-constructed process stability prediction model, and output a stability risk index. The scheme output module is configured to output a personalized correction scheme based on the stability risk index when the stability risk index exceeds a preset stability threshold.

Images