A multi-modal data-based national art dynamic digital protection system and method

Abstract

This invention discloses a dynamic digital preservation system and method for ethnic arts based on multimodal data, belonging to the field of big data management technology. The system includes an acquisition and control module, a data processing module, a rule engine, a machine learning engine, and a closed-loop optimization controller, effectively solving the industry problem of difficulty in quantifying, dynamizing, and adaptively optimizing quality assessment standards in the digitization of ethnic arts.

Description

Technical Field

[0001] This invention relates to the field of big data management technology, and more specifically to a dynamic digital protection system and method for ethnic art based on multimodal data. Background Technology

[0002] Ethnic art is an important carrier of China's outstanding traditional culture, encompassing various forms such as painting, sculpture, clothing, and handicrafts. Its digital preservation is a core pathway to achieving cultural inheritance and sustainable utilization. Currently, the field of digital preservation of ethnic art has gradually begun multimodal data collection; however, existing technologies still have many unresolved issues, making it difficult to meet the high-quality, intelligent preservation needs of complex ethnic artworks. First, the quality control system for digital acquisition has limitations. Existing systems mostly use static rules for quality verification, which can only perform basic checks on the physical parameters of digital assets. They lack cultural semantic level verification based on knowledge of ethnic art, which can easily lead to problems where the physical quality is qualified but the cultural logic is wrong, such as confusing the craft characteristics and color matching of different ethnic groups, causing digital assets to lose their cultural authenticity.

[0003] Secondly, the quality control rules lack adaptive optimization capabilities. Ethnic art is diverse in category and craftsmanship, and there are subtle differences in the artistic characteristics of different schools. The existing static rules cannot be dynamically adjusted with the accumulation of data and the deepening of understanding. As the collection task progresses, the classification accuracy is prone to decline and the quality control fails, making it difficult to adapt to the complex and ever-changing needs of digitizing ethnic art.

[0004] Furthermore, the utilization rate of multimodal data fusion is low. Existing technologies mostly process different modal data such as images, 3D point clouds, and audio separately, without achieving effective feature fusion. This results in insufficient semantic classification accuracy of ethnic artworks, making it difficult to accurately extract their cultural characteristics. At the same time, digital assets are mostly stored in a fragmented form, lacking structured semantic associations, which makes subsequent retrieval, research, and reuse more difficult, thus reducing the application value of digital assets.

[0005] Furthermore, the existing system lacks a data-driven closed-loop optimization mechanism. The classification results of machine learning are independent of the quality control rules, and the verification rules cannot be optimized in reverse through classification feedback. This makes it difficult to continuously improve the system's intelligence level and adaptability, which restricts the large-scale and high-quality advancement of the digital protection of ethnic arts.

[0006] Therefore, how to provide a dynamic digital protection system and method for ethnic arts with adaptive optimization, cultural semantic verification and structured management capabilities has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] In view of this, the present invention provides a dynamic digital preservation system and method for ethnic art based on multimodal data, in order to solve the technical problems existing in the prior art.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: A dynamic digital preservation system for ethnic art based on multimodal data includes: an acquisition and control module, a data processing module, a rule engine, a machine learning engine, and a closed-loop optimization controller; The acquisition and control module is used to acquire multimodal raw data of ethnic art works and output raw digital assets; The data processing module is connected to the acquisition control module and is used to preprocess the acquired raw digital assets and transmit them to the rule engine. The rule engine, connected to the data processing module, is used to perform automated quality checks on the preprocessed raw digital assets based on deterministic rules, and output the rule check results and anomaly type labels. The digital assets that pass the check are transmitted to the machine learning engine, and the digital assets that fail the check are fed back to the acquisition control module. The machine learning engine, connected to the rule engine, is used to extract and classify multi-dimensional semantic features of digital assets that have passed rule verification, generate semantic metadata and output classification confidence. The closed-loop optimization controller is bidirectionally connected to the rule engine and the machine learning engine, and is used to optimize the verification rules of the rule engine by monitoring the output of the machine learning engine.

[0009] Furthermore, the semantic metadata includes semantic metadata that includes ethnicity, craftsmanship, pattern, and color tags.

[0010] Furthermore, it also includes: The knowledge graph module is used to store the cultural semantic relationships between ethnic groups, crafts, patterns, and colors. The semantic metadata generated by the machine learning engine, which includes ethnic, craft, pattern, and color tags, will be automatically associated and written into the knowledge graph to form a traceable and inferable digital cultural asset network.

[0011] Furthermore, the rule engine has a built-in art quality rule library, which includes at least color rules, composition rules, and craft consistency rules based on prior knowledge of ethnic art; wherein, the craft consistency rules are used to verify whether the craft characteristics reflected in different parts of the same work conform to the traditional norms of that ethnic school.

[0012] Furthermore, the automated quality verification of the rule engine includes: The physical specification verification unit checks whether the data format, resolution, and color space meet archival standards. The art rules verification unit determines whether digital assets have logical anomalies that violate relevant national art common sense, based on color rules, composition rules, and process consistency rules in the rule base.

[0013] Furthermore, the closed-loop optimization controller includes: When the average semantic classification confidence score of N consecutive collection tasks for an artwork belonging to category C is lower than a preset threshold, the controller generates a rule enhancement instruction. The instruction causes the rule engine to increase the weight of rules related to category C. The increase formula is as follows:

[0014] in, Adjust the strength coefficient for the rules. The confidence level warning threshold is... Let N be the mean confidence score of category C over the last N tasks. and These are the rule weights before and after the adjustment.

[0015] Furthermore, when the machine learning engine extracts multi-dimensional semantic features, it employs a multi-modal feature fusion method, the formula of which is:

[0016] in, , , These are depth feature vectors extracted from images, 3D point clouds, and process audio, respectively. This represents the vector normalization operation. , , Let be the fusion weight coefficients for each modality, and .

[0017] Furthermore, the closed-loop optimization controller includes: Suppose that category C is the classification confidence sequence in the most recent M valid tasks; Calculate the statistics for the classification confidence sequence; If the mean of the statistic is below the first threshold or the standard deviation of the statistic is above the second threshold, it is judged as a persistent anomaly.

[0018] A method for dynamic digital preservation of ethnic art based on multimodal data includes: S1: Perform multimodal digital acquisition on the target ethnic group's artworks to obtain the original digital assets; S2: Preprocess the collected raw digital assets and transmit them to the rules engine; S3: Call the art quality rule library in the rule engine to perform automated quality verification on the original digital assets; if the verification fails, generate collection re-suggestions and return to S1; if the verification passes, proceed to S4. S4: Call the machine learning engine to extract and classify multi-dimensional semantic features of digital assets that have passed the rule verification, generate semantic metadata and output classification confidence; S5: Record the semantic metadata and classification confidence scores output by this task, along with the corresponding artwork category information, into the task history database; S6: The closed-loop optimization controller analyzes historical task data. When it determines that the classification confidence of a specific art category is continuously abnormal, it generates a rule optimization instruction to drive the rule engine to adjust the parameters or logic of the corresponding category's verification rules in its rule base. By iteratively executing S1 to S6, the data-driven quality control rules can be adaptively evolved.

[0019] As can be seen from the above technical solution, compared with the prior art, the present invention provides a dynamic digital protection system and method for ethnic art based on multimodal data, effectively solving the industry problem of difficulty in quantifying, dynamizing, and adaptively optimizing quality assessment standards in the digitization of ethnic art. Specifically, it includes: (1) Traditional quality inspection system rules are static. This invention introduces a closed-loop optimization controller and uses the confidence level identified by the backend machine learning as an evaluation indicator of the frontend rule engine performance. When the identification confidence level suggests that the current rule may not be able to effectively identify quality defects of a certain type of artwork, the system can automatically trigger rule adjustment, so that the quality control standards can continuously evolve with the accumulation of data and the deepening of understanding.

[0020] (2) Color rules, composition rules, and craft consistency rules based on prior knowledge of ethnic art are embedded in the rule engine. This enables the system to discover deep-seated problems that are physically qualified but have cultural logic errors (for example, typical patterns of Dai brocade appear on Miao silver ornaments), realizing a leap in quality control from the physical layer to the cultural semantic layer.

[0021] (3) The classification confidence index is transformed into the adjustment amount of rule weight, forming a specific and implementable closed-loop feedback path, which gives the system configurable optimization sensitivity and robustness.

[0022] (4) The semantic metadata generated by intelligent recognition is automatically linked to the knowledge graph of ethnic art. This not only completes the digital archiving, but also builds a structured, associative, and reasonable digital cultural asset system in one step, which greatly enhances the subsequent research and application value of the data. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the system structure provided by the present invention; Figure 2 This is a schematic diagram of the method flow provided by the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Example 1: See Figure 1 Embodiment 1 of this invention discloses a dynamic digital protection system for ethnic art based on multimodal data, comprising: an acquisition and control module, a data processing module, a rule engine, a machine learning engine, and a closed-loop optimization controller; The acquisition and control module is used to acquire multimodal raw data of ethnic art works and output raw digital assets; The data processing module is connected to the acquisition control module and is used to preprocess the acquired raw digital assets and transmit them to the rule engine. The rule engine, connected to the data processing module, is used to perform automated quality checks on the preprocessed raw digital assets based on deterministic rules, and output the rule check results and anomaly type labels. The digital assets that pass the check are transmitted to the machine learning engine, and the digital assets that fail the check are fed back to the acquisition control module. The machine learning engine, connected to the rule engine, is used to extract and classify multi-dimensional semantic features of digital assets that have passed rule verification, generate semantic metadata and output classification confidence. The closed-loop optimization controller is bidirectionally connected to the rule engine and the machine learning engine, and is used to optimize the verification rules of the rule engine by monitoring the output of the machine learning engine.

[0027] In one specific embodiment, the semantic metadata includes semantic metadata containing ethnic, craft, pattern, and color tags.

[0028] In one specific embodiment, it further includes: The knowledge graph module is used to store the cultural semantic relationships between ethnic groups, crafts, patterns, and colors. The semantic metadata generated by the machine learning engine, which includes ethnic, craft, pattern, and color tags, will be automatically associated and written into the knowledge graph to form a traceable and inferable digital cultural asset network.

[0029] In one specific embodiment, the rule engine has a built-in art quality rule library, which includes at least color rules, composition rules, and craft consistency rules based on prior knowledge of ethnic art; wherein, the craft consistency rules are used to verify whether the craft characteristics reflected in different parts of the same work conform to the traditional norms of that ethnic school.

[0030] In one specific embodiment, the automated quality verification of the rules engine includes: The physical specification verification unit checks whether the data format, resolution, and color space meet archival standards. The art rules verification unit determines whether digital assets have logical anomalies that violate relevant national art common sense, based on color rules, composition rules, and process consistency rules in the rule base.

[0031] This invention embeds color, composition, and craft consistency rules based on prior knowledge of ethnic art into the rule engine. It breaks through the limitations of existing technologies that only focus on physical parameter verification, and can accurately identify problems that are physically qualified but have cultural logic errors. This ensures that digital assets fully restore the cultural characteristics and craft specifications of ethnic art, and safeguards the cultural bottom line of digital protection of ethnic art.

[0032] In one specific embodiment, the closed-loop optimization controller includes: When the average semantic classification confidence score of N consecutive collection tasks for an artwork belonging to category C is lower than a preset threshold, the controller generates a rule enhancement instruction. The instruction causes the rule engine to increase the weight of rules related to category C. The increase formula is as follows:

[0033] in, Adjust the strength coefficient for the rules. The confidence level warning threshold is... Let N be the mean confidence score of category C over the last N tasks. and These are the rule weights before and after the adjustment.

[0034] In one specific embodiment, when the machine learning engine extracts multi-dimensional semantic features, it employs a multi-modal feature fusion method, the formula of which is:

[0035] in, , , These are depth feature vectors extracted from images, 3D point clouds, and process audio, respectively. This represents the vector normalization operation. , , Let be the fusion weight coefficients for each modality, and .

[0036] In one specific embodiment, the closed-loop optimization controller includes: Suppose that category C is the classification confidence sequence in the most recent M valid tasks; Calculate the statistics for the classification confidence sequence; If the mean of the statistic is below the first threshold or the standard deviation of the statistic is above the second threshold, it is judged as a persistent anomaly.

[0037] This invention uses a closed-loop optimization controller to link the rule engine and the machine learning engine. It uses the time-series change of classification confidence as a feedback indicator and adaptively adjusts the weight of the verification rules through a quantitative formula. This allows the quality control rules to continuously evolve with the accumulation of data, significantly improving the system's adaptability to different categories and styles of ethnic artworks and ensuring the quality and stability of digital assets.

[0038] See Figure 2 Embodiment 1 of the present invention also discloses a method for dynamic digital preservation of ethnic art based on multimodal data, comprising: S1: Perform multimodal digital acquisition on the target ethnic group's artworks to obtain the original digital assets; S2: Preprocess the collected raw digital assets and transmit them to the rules engine; S3: Call the art quality rule library in the rule engine to perform automated quality verification on the original digital assets; if the verification fails, generate collection re-suggestions and return to S1; if the verification passes, proceed to S4. S4: Call the machine learning engine to extract and classify multi-dimensional semantic features of digital assets that have passed the rule verification, generate semantic metadata and output classification confidence; S5: Record the semantic metadata and classification confidence scores output by this task, along with the corresponding artwork category information, into the task history database; S6: The closed-loop optimization controller analyzes historical task data. When it determines that the classification confidence of a specific art category is continuously abnormal, it generates a rule optimization instruction to drive the rule engine to adjust the parameters or logic of the corresponding category's verification rules in its rule base. By iteratively executing S1 to S6, the data-driven quality control rules can be adaptively evolved.

[0039] Example 2: Embodiment 2 of this invention discloses a dynamic digital preservation system for ethnic arts based on multimodal data, including an acquisition and control module, a data processing module, a rule engine, a machine learning engine, and a closed-loop optimization controller. These modules work together sequentially to form a complete digital preservation and quality control system. The specific logical relationships are as follows: The acquisition control module is used to execute the preset digital acquisition procedure, collect multimodal raw data from ethnic art works, and output raw digital assets. The data processing module, connected to the acquisition control module, is used to preprocess the acquired raw digital assets, remove noise, unify the data format, and then transmit them to the rule engine. The rules engine has a built-in configurable art quality rule library and connects to the data processing module. It is used to perform automated quality checks on the pre-processed raw digital assets based on deterministic rules, and output the rule check results and anomaly type labels. Digital assets that pass the check are transmitted to the machine learning engine, while digital assets that fail the check are fed back to the acquisition control module. The machine learning engine, connected to the rule engine, is used to extract and classify multi-dimensional semantic features of digital assets that have passed rule verification, generate semantic metadata and output classification confidence scores. The closed-loop optimization controller, which connects the rule engine and the machine learning engine bidirectionally, is used to optimize the validation rules of the rule engine by monitoring the output of the machine learning engine. Specifically, it executes the following closed-loop logic: (1) Monitor and analyze the time series change data of classification confidence output by the machine learning engine in real time, and establish a confidence change trend model; (2) A preset classification confidence threshold is set. When the classification confidence for a specific type of artwork is detected to deviate from the expected threshold, a corresponding rule optimization instruction is generated. (3) The rule optimization instruction drives the rule engine to adaptively adjust the verification rules associated with specific types of artworks in its art quality rule library, and optimize the rule parameters and verification logic; By optimizing the feedback adjustment of the closed-loop controller, dynamic optimization of the rule base based on data-driven approaches is achieved, thereby improving the overall quality control accuracy of the system for complex and ever-changing ethnic artworks and ensuring the integrity and accuracy of digital assets.

[0040] In one specific embodiment, the art quality rule base in the rule engine is constructed based on prior knowledge in the field of ethnic art, and includes at least color rules, composition rules, and craft consistency rules. The specific limitations of each rule are as follows: (1) Color rules: Based on the typical color gamut characteristics of different ethnic art schools, the standard color range, color matching ratio and color gradient rules of various ethnic artworks are defined to verify whether the color reproduction of digital assets conforms to the traditional norms of the corresponding ethnic art. (2) Composition rules: Combining the traditional composition paradigms of various ethnic art forms, the main layout, element proportions, and symmetry / balance requirements of various ethnic artworks are clarified to verify whether the composition restoration of digital assets conforms to the aesthetic characteristics of the corresponding ethnic art. (3) Craftsmanship consistency rule: This rule is set for the traditional production process of various ethnic art forms. It is used to verify whether the craftsmanship characteristics (such as texture details and technique traces) reflected in different parts of the same digital asset are consistent and conform to the traditional craftsmanship standards of the ethnic art school, so as to avoid contradictions in craftsmanship logic.

[0041] In one specific embodiment, the logic for the closed-loop optimization controller to perform rule adjustment is as follows, and the rule weights are precisely adjusted through a preset quantization formula: The category of artwork to be monitored is set as C. A threshold N for the number of consecutive data collection tasks is preset (N is a positive integer, 3 ≤ N ≤ 10, which can be configured according to the actual application scenario), and a classification confidence warning threshold is also preset. ( For decimals, 0.7 ≤ ≤0.85 (calibrated based on historical global classification data), rule adjustment intensity coefficient ( For decimals, 0.1 ≤ ≤0.5, used to control the adjustment range of rule weights to avoid over- or under-adjustment. When the average semantic classification confidence score of N consecutive data collection tasks for artwork of category C is lower than a preset threshold, At that time, the closed-loop optimization controller determines that the rule base verification rules corresponding to the current category C are insufficient and generates a rule enhancement instruction; This rule enhances the instruction-driven rule engine by increasing the weight W of all validation rules associated with category C. The weight adjustment formula is as follows:

[0042] in, The weights of the rules related to category C after adjustment have a range of (0,1]. The weights of the relevant rules for category C before adjustment are set in the range of (0,1], with the initial values ​​set based on the experience of experts in the field of ethnic arts. The strength coefficient for rule adjustment is configured by the user according to system optimization needs, with a value of 0.1≤α≤0.5. The larger α is, the greater the weight adjustment range. The classification confidence warning threshold is set to 0.7≤T≤0.85, which is used to determine whether the classification result is abnormal. It can be dynamically calibrated based on historical classification data. The mean semantic classification confidence score of category C in the last N collection tasks is calculated as the arithmetic mean of the classification confidence scores of the last N tasks, and the value range is [0,1].

[0043] This invention employs a multimodal feature fusion method, integrating the feature advantages of different modal data such as images, 3D point clouds, and audio. Through normalization processing and weight allocation, it generates accurate fusion feature vectors, effectively improving the semantic classification confidence of ethnic artworks and providing support for the precise management of digital assets.

[0044] In one specific embodiment, when the machine learning engine extracts multi-dimensional semantic features, it employs a multi-modal feature fusion method to integrate the feature advantages of different modalities, thereby improving classification accuracy. The formula for calculating the fused feature vector F is as follows:

[0045] In the formula, This is a multimodal fusion feature vector used for subsequent semantic classification and metadata generation. The dimension is a preset fixed value (which can be configured according to the type of ethnic art). , , These are the fusion weight coefficients for each modality feature, all with values ​​of 0 ≤ 0. , , ≤1, and satisfy The weighting is based on the modal importance of different ethnic artworks (e.g., visual artworks have higher β and β weights, while audio-related artworks have higher β weights). This is a vector normalization operation used to normalize the feature vectors of each modality to the same numerical range [0,1], thereby eliminating the influence of dimensional differences on the fusion result. The depth feature vectors extracted from the two-dimensional image modality of digital assets are extracted using a convolutional neural network (CNN), and include visual features such as color, texture, and contour. The depth feature vectors extracted from the 3D point cloud modality of digital assets are extracted using the PointNet network and include 3D features such as spatial structure, morphological contour, and surface details. The deep feature vectors (such as instrumental music audio and handcrafted sound effects) extracted from the craft audio modalities of digital assets are obtained through Mel frequency cepstral coefficients (MFCC), which contain audio features such as frequency, amplitude, and rhythm.

[0046] In one specific embodiment, the present invention further includes a knowledge graph module that works bidirectionally with the machine learning engine to achieve semantic association and structured management of digital assets. Specifically: The knowledge graph module, with knowledge in the field of ethnic art as its core, stores the cultural and semantic relationships between ethnic groups, art schools, craft techniques, pattern elements, and color symbolism, and constructs a structured knowledge network of ethnic art. The semantic metadata (including ethnic tags, craft tags, pattern tags, color tags, etc.) generated by the machine learning engine will be automatically associated with and written into the knowledge graph module, and matched with the corresponding nodes in the knowledge graph. Specifically, through the knowledge graph module, a traceable, reasonable, and associative digital cultural asset network is formed, enabling semantic retrieval and association analysis of digital assets, while providing domain knowledge support for rule optimization in the rule engine.

[0047] This invention stores semantic relationships in the field of ethnic arts through a knowledge graph module, and associates semantic metadata generated by machine learning with knowledge graph nodes to form a traceable and reasonable digital cultural asset network. This solves the problems of fragmented storage and poor reusability of existing digital assets, facilitates subsequent semantic retrieval, association analysis and academic research, and maximizes the value of digital assets.

[0048] On the other hand, Embodiment 2 of the present invention discloses a method for digital quality control of ethnic art based on rule-driven and data-driven optimization, applied to a dynamic digital protection system for ethnic art based on multimodal data. The method includes the following steps, each step being executed sequentially and forming a closed-loop iteration: S1: Multimodal Data Acquisition: The acquisition control module acquires multimodal raw data of the target ethnic art works according to the preset digital acquisition procedure. The acquired modalities include at least two-dimensional images, three-dimensional point clouds, and craft audio. After the acquisition is completed, the raw digital assets are output. The digital acquisition procedure presets acquisition parameters (resolution, sampling rate, color depth), and the parameter configuration conforms to the archival-level digital asset storage standard. S2: Preprocessing: Preprocess the collected raw digital assets and transmit them to the rules engine; S3: Quality Verification and Screening: Call the art quality rule library in the rule engine to perform automated quality verification on the original digital assets after preprocessing by the data processing module. The verification process is divided into two levels: physical specification verification and art rule verification. If the verification fails, a targeted collection re-sampling suggestion (including re-sampling parameter adjustment, collection area calibration, etc.) is generated and returned to S1 to re-execute the collection task. If the verification passes, the digital asset is transferred to S4. S4: Semantic Classification and Metadata Generation: The machine learning engine is invoked to extract multi-dimensional semantic features from the verified digital assets using a multimodal feature fusion method. Based on the extracted fusion feature vectors, ethnic art types are classified, generating semantic metadata containing information such as ethnicity, art style, craftsmanship, pattern, and color. The confidence score of this classification is then output. S5: Data Recording and Archiving: The semantic metadata and classification confidence scores output by this data collection task, along with the corresponding artwork category information, collection parameters, and verification results, are recorded in the task history database to form a complete task data archive, providing data support for subsequent closed-loop optimization. S6: Closed-loop rule optimization: The closed-loop optimization controller retrieves data from the task history database, analyzes the temporal change trend of the classification confidence of each category of artwork, and when it is determined that the classification confidence of a specific category of artwork is continuously abnormal, it generates a rule optimization instruction to drive the rule engine to adaptively adjust the parameters or logic of the corresponding category of the verification rules in its rule base. Iteratively execute S1 to S6 to achieve adaptive evolution of data-driven quality control rules, continuously improving the system's accuracy and adaptability in quality control of various ethnic artworks.

[0049] The automated quality verification in step S2 is divided into two levels: physical specification verification and artistic rule verification. The specific process is as follows: S2.1: Physical Specification Verification: The focus is on checking the physical storage and acquisition quality of digital assets to ensure compliance with archival-level digital asset standards. Specific verification content includes: data format (must conform to the preset standard format, such as TIFF format for images, PLY format for 3D point clouds, and WAV format for audio), resolution (image resolution not less than 300dpi, 3D point cloud resolution not less than the preset threshold), color space (unified to Adobe RGB color space to ensure accurate color reproduction), and data integrity (no missing frames, no damaged data). S2.2: Art Rule Verification: Based on the color rules, composition rules, and craft consistency rules in the rule engine, the digital assets that have passed the physical specification verification are verified at the cultural and semantic level to determine whether there are any logical anomalies in the digital assets that violate the common sense of relevant ethnic art. Specifically, this includes whether the color matching conforms to the corresponding ethnic norms, whether the composition layout conforms to the corresponding ethnic aesthetics, and whether the craft details are consistent and conform to traditional standards. If any anomaly exists, the verification is deemed to have failed.

[0050] The condition for determining a persistently abnormal classification confidence level for a specific art category in step S5 is achieved by comparing a preset statistic with a threshold, specifically as follows: Let the category of the artwork to be judged be C, and set the number of historical task statistics M (M is a positive integer, 5≤M≤20, which can be configured according to the actual application scenario) and the confidence mean threshold. ( For decimals, 0.75 ≤ ≤0.85 (calibrated based on historical global classification data), confidence standard deviation threshold. ( For decimals, 0.1 ≤ ≤0.2 (used to determine whether the confidence level fluctuation is abnormal). The classification confidence sequence of category C in the most recent M valid data collection tasks is as follows: , Where P (i=1,2,...,M) is the classification confidence of the i-th task, and its value range is [0,1]. Calculate the two statistics for this confidence sequence: the mean μ and the standard deviation σ, as follows:

[0051] The specific definitions of each parameter are as follows: This represents the number of most recent valid data collection tasks, with a value of 5 ≤ M ≤ 20, and is configurable. Let be the classification confidence score of the i-th valid data collection task for category C, with a value of [0,1]. Let M be the average classification confidence score of the most recent M valid tasks for category C, with a value of [0,1]. Let be the standard deviation of the classification confidence of the most recent M valid tasks for category C, with a value of [0,1]. The larger the standard deviation, the greater the fluctuation of the confidence. The confidence level mean threshold is set to 0.75 ≤ T ≤ 0.85, and can be calibrated based on historical global data. The confidence level standard deviation threshold is set to 0.1 ≤ T ≤ 0.2, and can be calibrated based on historical global data.

[0052] The classification confidence of category C is considered to be persistently abnormal when any of the following conditions are met: mean Below the threshold This indicates that the classification accuracy of category C remains consistently low, suggesting that there are deficiencies in the corresponding rule base validation rules; Standard deviation Above the threshold This indicates that the classification results of category C fluctuate too much and lack stability, suggesting that the corresponding rule base validation rules have poor adaptability.

[0053] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0054] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-modal data based dynamic digital protection system for ethnic arts, characterized by, include: The system includes a data acquisition and control module, a data processing module, a rules engine, a machine learning engine, and a closed-loop optimization controller. The acquisition and control module is used to acquire multimodal raw data of ethnic art works and output raw digital assets; The data processing module is connected to the acquisition control module and is used to preprocess the acquired raw digital assets and transmit them to the rule engine. The rule engine, connected to the data processing module, is used to perform automated quality checks on the preprocessed raw digital assets based on deterministic rules, and output the rule check results and anomaly type labels. The digital assets that pass the check are transmitted to the machine learning engine, and the digital assets that fail the check are fed back to the acquisition control module. The machine learning engine, connected to the rule engine, is used to extract and classify multi-dimensional semantic features of digital assets that have passed rule verification, generate semantic metadata and output classification confidence. The closed-loop optimization controller is bidirectionally connected to the rule engine and the machine learning engine, and is used to optimize the verification rules of the rule engine by monitoring the output of the machine learning engine.

2. The dynamic digital preservation system for ethnic art based on multimodal data according to claim 1, characterized in that, The semantic metadata includes semantic metadata tags for ethnicity, craftsmanship, pattern, and color.

3. The dynamic digital preservation system for ethnic art based on multimodal data according to claim 2, characterized in that, Also includes: The knowledge graph module is used to store the cultural and semantic relationships between ethnic groups, crafts, patterns, and colors. The semantic metadata generated by the machine learning engine, which includes ethnic, craft, pattern, and color tags, will be automatically associated with and written into the knowledge graph to form a traceable and inferable digital cultural asset network.

4. The dynamic digital preservation system for ethnic art based on multimodal data according to claim 1, characterized in that, The rule engine has a built-in art quality rule library, which includes at least color rules, composition rules, and craft consistency rules based on prior knowledge of ethnic art. The craft consistency rules are used to verify whether the craft characteristics reflected in different parts of the same work conform to the traditional norms of that ethnic school.

5. A dynamic digital preservation system for ethnic art based on multimodal data according to claim 1, characterized in that, The automated quality verification of the rule engine includes: The physical specification verification unit checks whether the data format, resolution, and color space meet archival standards. The art rule verification unit determines whether digital assets have logical anomalies that violate relevant national art common sense, based on color rules, composition rules, and process consistency rules in the rule base.

6. A dynamic digital preservation system for ethnic art based on multimodal data according to claim 1, characterized in that, The closed-loop optimization controller includes: When the average semantic classification confidence score of N consecutive collection tasks for an artwork belonging to category C is lower than a preset threshold, the controller generates a rule enhancement instruction. The instruction causes the rule engine to increase the weight of rules related to category C. The increase formula is as follows: in, Adjust the strength coefficient for the rules. The confidence level warning threshold is... Let N be the mean confidence score of category C over the last N tasks. and These are the rule weights before and after the adjustment.

7. A dynamic digital preservation system for ethnic art based on multimodal data according to claim 1, characterized in that, When the machine learning engine extracts multi-dimensional semantic features, it employs a multi-modal feature fusion method, the formula of which is: in, , , These are depth feature vectors extracted from images, 3D point clouds, and process audio, respectively. This represents the vector normalization operation. , , Let be the fusion weight coefficients for each modality, and .

8. A dynamic digital preservation system for ethnic art based on multimodal data according to claim 6, characterized in that, The closed-loop optimization controller includes: Suppose that category C is the classification confidence sequence in the most recent M valid tasks; Calculate the statistics for the classification confidence sequence; If the mean of the statistic is below the first threshold or the standard deviation of the statistic is above the second threshold, it is judged as a persistent anomaly.

9. A method for utilizing a dynamic digital preservation system for ethnic arts based on multimodal data as described in any one of claims 1 to 8, characterized in that, include: S1: Perform multimodal digital acquisition on the target ethnic group's artworks to obtain the original digital assets; S2: Preprocess the collected raw digital assets and transmit them to the rules engine; S3: Calls the art quality rule library in the rule engine to perform automated quality verification on the original digital assets; If the verification fails, a collection re-suggestion is generated and returned to S1; If the verification passes, proceed to S4; S4: Call the machine learning engine to extract and classify multi-dimensional semantic features of digital assets that have passed the rule verification, generate semantic metadata and output classification confidence; S5: Record the semantic metadata and classification confidence scores output by this task, along with the corresponding artwork category information, into the task history database; S6: The closed-loop optimization controller analyzes historical task data. When it determines that the classification confidence of a specific art category is continuously abnormal, it generates a rule optimization instruction to drive the rule engine to adjust the parameters or logic of the corresponding category's verification rules in its rule base. By iteratively executing S1 to S6, the data-driven quality control rules can be adaptively evolved.

Images