An intelligent cloud audio and video access and big data analysis method
Through intelligent cloud audio and video access and big data analysis methods, the audio and video streams are analyzed in real time and copyright metadata is dynamically adjusted, which solves the accuracy and compliance of copyright information dissemination in cross-border copyright management systems, and realizes cross-regional copyright Intelligent management and efficient management.
Patent Information
- Application Number
- CN202411219239.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-02
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-09-02
AI Technical Summary
In a cross-border copyright management system, when audio and video content is spread across different copyright areas, the accuracy and compliance of copyright information are difficult to guarantee, and there are technical challenges such as copyright verification, income distribution and information synchronization during cross-region transmission.
Using intelligent cloud audio and video access and big data analysis methods, extract copyright watermark information through real-time analysis of audio and video streams, generate copyright feature vectors, combine network routing tracking and geographical location database to determine copyright jurisdiction areas, dynamically adjust copyright metadata, and realize cross-regional Copyright licensing and profit distribution, and deploy lightweight copyright verification modules at the edge of the network for real-time control.
It realizes intelligent management of cross-regional audio and video copyrights, solves the problems caused by regional differences in copyright rules, improves copyright management efficiency, and ensures the accuracy and compliance of copyright information.
Smart Images

Figure CN118761047B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of information technology, and in particular to an intelligent cloud audio and video access and big data analysis method. Background Art
[0002] In a cross-border copyright management system, copyright owners can sell copyrights by region, but this leads to a complex technical challenge: how to ensure the accuracy and compliance of copyright information when audio and video content is transmitted across different copyright regions. When content is transmitted from one region to another, copyright information may need to be re-encoded multiple times to adapt to the regulations of different regions. At the same time, the system also needs to deal with factors such as copyright period and usage restrictions, which may vary in different regions. In addition, copyright verification, revenue distribution, and information synchronization during cross-regional transmission also bring challenges. Using distributed learning to optimize copyright identification models and rule reasoning engines to improve the system's ability to handle complex cross-regional copyright issues has become a key technical problem. Summary of the invention
[0003] The present invention provides an intelligent cloud audio and video access and big data analysis method, which mainly includes:
[0004] According to the granularity of copyright regional division, the audio and video streams are analyzed in real time, the copyright watermark information in the region is extracted and the copyright feature vector is generated. The feature vector is classified through the pre-trained copyright recognition model to obtain the preliminary copyright attribution result in the region; based on the preliminary copyright attribution result, the transmission path information of the cross-regional copyright authorized audio and video is obtained through network routing tracing, and the current copyright jurisdiction area is determined in combination with the IP address geographic location database, and the copyright rule database applicable to the area is queried to obtain the corresponding copyright period and use restriction rule information; if there is a difference between the current regional copyright rule and the preliminary attribution result, the copyright metadata structure is dynamically adjusted according to the differentiated configuration of the regional copyright rule, and the copyright information is re-encoded using the encryption and compression method corresponding to the area, and synchronized to the copyright management node in this area through the copyright data transmission protocol; the copyright management node is based on the copyright information in this area Parse the current regional copyright rules, parse the received copyright metadata, extract copyright ownership and term information, and if a copyright exclusivity conflict is found in the region, initiate the copyright dispute resolution mechanism, and give compliant copyright authorization suggestions through the rule reasoning engine; according to the authorization results of the copyright management node, use the unified copyright revenue distribution rules in the region to realize the distribution and settlement of copyright revenue, and synchronize the authorized copyright information to the copyright management nodes in the relevant regions; in the process of cross-regional transmission of copyright information, deploy a lightweight copyright verification module at the edge node of the network to perform copyright verification on the audio and video streams passing through the node to determine whether they have legal cross-regional authorization, and perform real-time control according to the regional copyright usage restriction rules; audio and video with legal cross-regional authorization are transmitted through multiple regions to coordinate training of copyright data distributed in different regions, and optimize the copyright recognition model and rule reasoning engine.
[0005] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:
[0006] The present invention discloses an intelligent cloud-based audio and video access and big data analysis method. In view of the cross-regional copyright management difficulties caused by differences in copyright rules in different regions, the present invention uses deep learning to analyze audio and video streams in real time to extract copyright information, combines network routing tracing to determine the copyright jurisdiction area, and dynamically adjusts copyright metadata to adapt to regional rule differences. The copyright management node parses metadata and handles copyright conflicts to achieve cross-regional copyright authorization and revenue distribution. Lightweight copyright verification is deployed at the edge of the network to perform real-time control of cross-regional audio and video streams. Through multi-region collaborative training, the copyright identification model and rule engine are continuously optimized to improve the system's adaptability to cross-regional copyright issues. The present invention realizes the intelligent management of cross-regional audio and video copyrights, effectively solves the problems caused by regional differences in copyright rules, and improves the efficiency of copyright management. BRIEF DESCRIPTION OF THE DRAWINGS
[0007] Figure 1 The present invention is a flow chart of an intelligent cloud audio and video access and big data analysis method.
[0008] Figure 2 A schematic diagram of an intelligent cloud audio and video access and big data analysis method of the present invention.
[0009] Figure 3 It is another schematic diagram of an intelligent cloud audio and video access and big data analysis method of the present invention. DETAILED DESCRIPTION
[0010] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0011] like Figure 1-3 In this embodiment, an intelligent cloud audio and video access and big data analysis method may specifically include:
[0012] Step S101, according to the granularity of copyright area division, the audio and video streams are analyzed in real time, the copyright watermark information in the area is extracted and a copyright feature vector is generated, and the feature vector is classified by a pre-trained copyright recognition model to obtain a preliminary copyright ownership result in the area.
[0013] According to the preset copyright area division rules, the audio and video streams are segmented into regions, the watermark features of the segmented regions are extracted, and the two-dimensional discrete Fourier transform results of the watermark features are obtained. By calculating the frequency domain amplitude spectrum of the two-dimensional discrete Fourier transform results, the preset high-frequency components are selected and determined as the copyright feature vector. The copyright feature vector is reduced in dimension by principal component analysis to obtain the feature components whose cumulative variance contribution rate reaches the preset threshold, and the copyright feature representation after dimension reduction is obtained. The most similar matching items to the copyright feature representation after dimension reduction are retrieved from the pre-established copyright feature database. If the similarity is higher than the preset threshold, it is determined that the match is successful, and the corresponding copyright ownership information is obtained. If the similarity is lower than the preset threshold, the copyright feature representation after dimension reduction is input into the pre-trained copyright classifier, and the copyright ownership of the region is determined by multi-category classification of the copyright feature representation.
[0014] Specifically, the input audio and video streams are segmented according to the preset copyright area division rules, and the watermark features are extracted for each area using the ResNet-50 convolutional neural network, which contains five convolution blocks and a fully connected layer, and outputs a 2048-dimensional feature vector. The extracted watermark features are subjected to a two-dimensional discrete Fourier transform, and the frequency domain amplitude spectrum is calculated, from which the first 100 high-frequency components are selected as copyright feature vectors. The generated copyright feature vectors are subjected to dimensionality reduction processing, and the main feature components are extracted using the principal component analysis method. The covariance matrix of the feature vector is calculated, and its eigenvalues and eigenvectors are solved. The first k eigenvectors with a cumulative variance contribution rate of 95% are selected as principal components, and the original feature vectors are projected onto these k principal components to obtain the copyright feature representation after dimensionality reduction. The most similar matches to the copyright features of the current region are retrieved from the pre-established copyright feature database, which is constructed by extracting and storing features of known copyright content. The cosine similarity is used to calculate the similarity between feature vectors. The cosine similarity is defined as the dot product of two vectors divided by the product of their moduli, and the value range is [-1,1]. The similarity threshold is set to 0.9. If the similarity is higher than the threshold, it is determined to be a successful match and the corresponding copyright ownership information is obtained. For the successfully matched areas, the copyright ownership results are recorded. For the areas that failed to match successfully, their feature vectors are input into the pre-trained copyright classifier. The classifier is built based on the support vector machine and uses the radial basis function kernel. The training data comes from the feature vectors of known copyright content. By performing multi-category classification on the feature vectors, the probability distribution of each category is output, and the category with the highest probability is selected as the copyright ownership determination result of the area. If the highest probability is lower than the preset threshold of 0.7, the area is marked as "unknown copyright". In the copyright area division, the audio and video frames are processed based on the image segmentation method. Taking the video with a resolution of 1920x1080 as an example, each frame is divided into a 10x10 grid, and each grid is a copyright area. The ResNet-50 convolutional neural network is used to extract watermark features. The network input is a 224x224 image block. After five convolution blocks and a fully connected layer, a 2048-dimensional feature vector is output. The extracted watermark features are subjected to a two-dimensional discrete Fourier transform, the frequency domain amplitude spectrum is calculated, and the first 100 high-frequency components are selected as copyright feature vectors. In the dimensionality reduction process, principal component analysis is applied to the 2048-dimensional feature vector to calculate the eigenvalues and eigenvectors of the covariance matrix. The first 50 eigenvectors with a cumulative variance contribution rate of 95% are selected as principal components. The original eigenvectors are projected onto these 50 principal components to obtain the copyright feature representation after dimensionality reduction. The copyright feature database stores 10,000 feature vectors of known copyright content, each of which contains 50 principal component values after dimensionality reduction. The cosine similarity is used to calculate the similarity between feature vectors, and the similarity threshold is set to 0.9. If the similarity is higher than the threshold, it is determined to be a successful match.For the areas that failed to match successfully, the copyright classifier constructed by support vector machine was used for processing. The classifier used radial basis function kernel, kernel parameter γ was set to 0.1, penalty parameter C was set to 10, and the training data contained 100,000 feature vectors of known copyrighted content. The classifier outputs the probability distribution of each category, and selects the category with the highest probability as the copyright attribution determination result. If the highest probability is lower than the preset threshold of 0.7, the area is marked as "unknown copyright". Real-time analysis can be achieved under GPU acceleration. For 30fps video, 25-28 frames can be processed per second, meeting the real-time requirements.
[0015] Step S102, based on the preliminary copyright ownership results, obtain the transmission path information of cross-regional copyright-authorized audio and video through network routing tracking, determine the current copyright jurisdiction area in combination with the IP address geographic location database, and query the copyright rule database applicable to the area to obtain the corresponding copyright period and usage restriction rule information.
[0016] Perform application layer protocol identification on the audio and video data packets to obtain the source IP address and destination IP address of the audio and video data packets. Through a route tracking method that combines the ICMP protocol and the UDP protocol, obtain the transmission path node information of the data packet in the network, and the transmission path node information includes the node IP address and the router identifier. Query the geographic location information of each node in the transmission path node information from a pre-established IP address geographic location mapping database. If the geographic location information meets the preset copyright jurisdiction division rules, determine the copyright jurisdiction where the audio and video is currently located. According to the copyright jurisdiction, obtain the corresponding copyright rules from a pre-built copyright rule database, and the copyright rules include copyright term and usage restriction information. If cross-regional copyright authorization is involved, prioritize and resolve conflicts for the copyright rules of multiple regions to generate a comprehensive set of copyright rules, including generating a comprehensive score for usage restrictions according to the following formula: , U represents the comprehensive score of usage restriction, A i represents the usage restriction weight of the ith area, P i represents the priority of the ith region, and n represents the number of regions.
[0017] Specifically, the audio and video streams are marked according to the preliminary copyright attribution results, the nDPI library is used for application layer protocol identification, the source IP address and destination IP address of the audio and video data packets are extracted, and the transmission path node information of the data packets in the network is obtained through the routing tracking method combining the ICMP protocol and the UDP protocol, and the IP address and router identifier of each node are recorded. The geographical location information of each node in the transmission path is queried from the pre-established IP address geographical location mapping database. The database uses a prefix tree structure to store IP address segments and corresponding geographical information such as countries, regions, cities, etc. The geographical location of the IP address is quickly located through the prefix matching algorithm, and updated data is obtained from the authoritative geographical location data provider every week. Based on the queried geographical location information, combined with the preset copyright jurisdiction division rules, the copyright jurisdiction area where the audio and video is currently located is determined. The copyright jurisdiction division rules are constructed based on a multi-layer perceptron neural network, and the input includes multi-dimensional information such as country, region, longitude and latitude, and the output is a specific jurisdiction area identifier. According to the determined copyright jurisdiction, the pre-built copyright rule database is queried. The database adopts a key-value pair storage structure, with the jurisdiction as the key and the corresponding copyright rule as the value in JSON format, which contains hierarchical rule information such as copyright period and usage restrictions, and is quickly retrieved through hash index. For cross-regional copyright authorization, the copyright rules of multiple regions involved are checked, and the rule priority sorting and conflict resolution algorithm are used to generate a comprehensive set of copyright rules, and this information is associated with the preliminary copyright attribution results to form complete copyright management information. In actual applications, for a 720p resolution, 30fps video stream, the nDPI library is used for protocol identification, which can process about 5,000 packets per second. The extracted IP addresses are tracked by a route tracking method combining ICMP and UDP. It takes an average of 30 milliseconds for each IP address to complete the path tracking and obtain 10-15 intermediate node information. The IP address geographic location mapping database is stored in a 32-bit prefix tree, containing about 2 million IP address segment records, and a single query takes an average of 0.5 milliseconds. The copyright jurisdiction division rules use a 5-layer multilayer perceptron network with 128 neurons in each layer. The input dimension is 10, which includes country code, longitude and latitude, time zone, etc. The output is 1,000 different jurisdiction identifications, and the single inference time is about 2 milliseconds. The copyright rule database uses Redis storage, with the key being the jurisdiction identification and the value being a set of rules in JSON format, including information such as copyright period and usage restrictions. The average rule size for each region is 5KB, and the query response time is within 1 millisecond. For cross-regional copyright authorization, if a video is authorized in both regions A and B, the rules of these two regions are retrieved, and the timestamp-based priority sorting is used to select the latest rules. The rule merging method is used to handle conflicting items, such as taking the intersection of the copyright periods of the two regions and the union of the usage restrictions to generate a comprehensive rule set.From receiving the video stream to outputting complete copyright management information, the single-frame processing time is controlled within 50 milliseconds, meeting the real-time requirements. The IP geographic location data is automatically updated from the authoritative data source every week. The update process adopts an incremental update strategy, and an average of 100,000 records are updated each time, which takes about 5 minutes.
[0018] Step S103: If there is a difference between the current regional copyright rules and the preliminary attribution results, the copyright metadata structure is dynamically adjusted according to the differentiated configuration of the regional copyright rules, the copyright information is re-encoded using the encryption and compression methods corresponding to the region, and synchronized to the copyright management node of this region through the copyright data transmission protocol.
[0019] Calculate the Jaccard similarity difference based on the acquired regional copyright rules and preliminary attribution results; determine whether the difference exceeds a preset threshold, and if so, select a metadata template that is adapted to the regional copyright rules; parse the regional copyright rules, extract key attributes in the regional copyright rules, map the key attributes to the metadata template, and generate a copyright metadata structure; encrypt the copyright metadata structure to obtain encrypted copyright data; compress the encrypted copyright data to obtain encoded copyright information; encapsulate the encoded copyright information through a copyright data transmission protocol to generate a copyright data packet; use the TCP protocol to send the copyright data packet to a copyright management node, and the copyright data packet includes a checksum and a timestamp.
[0020] Specifically, according to the current regional copyright rules and preliminary attribution results, the difference between the two is calculated by Jaccard similarity. If the difference exceeds the preset threshold of 0.7, the dynamic adjustment process of the copyright metadata structure is triggered, and the template structure that best suits the current regional rules is selected from the pre-defined metadata template library. The template library contains a collection of copyright metadata fields commonly used in different regions, such as copyright owner, scope of use, validity period, etc. The regional copyright rules are parsed using the JSON schema specification, key attributes are extracted and mapped to the selected metadata template, and a copyright metadata structure adapted to the current region is generated. The structure contains fields such as copyright owner, scope of use, validity period, etc. At the same time, the processing flow is dynamically adjusted according to the differentiated configuration, such as adjusting the encryption strength or compression level. The specified encryption algorithm AES-256 and compression method GZIP are obtained from the regional copyright rule configuration, and the generated copyright metadata is encrypted. The encrypted data is then compressed to obtain the final encoded copyright information. The encoded copyright information is encapsulated into a data packet through the copyright data transmission protocol, which defines the data packet format, transmission mode and error handling mechanism, adds checksum and timestamp, and uses the TCP protocol to send it to the copyright management node in this region. During the transmission process, the breakpoint-resume mechanism is used to ensure data integrity, and the error retransmission and data packet sequence reorganization functions are realized. In actual application, the current regional copyright rules and preliminary attribution results are first converted into a set form, such as {A, B, C, D} and {A, C, E, F}. The difference is 0.67 calculated by Jaccard similarity. Since it exceeds the preset threshold of 0.7, the dynamic adjustment process is triggered. A suitable template is selected from the metadata template library, which contains 100 templates from different regions, and each template contains an average of 20 fields. The copyright rules are parsed using the JSON mode, and key attributes such as "copyright owner", "scope of use", "validity period", etc. are extracted and mapped to the selected template. According to the differentiated configuration, the encryption strength is dynamically adjusted from AES-128 to AES-256, and the compression level is increased from the default level 6 of GZIP to level 9. The generated copyright metadata (about 2KB in size) is encrypted with AES-256, the key length is 256 bits, and the encryption mode is CBC. The encrypted data is compressed using the GZIP algorithm, and the compression ratio is about 40% of the original data. The copyright data transmission protocol defines the data packet format, including a 16-byte header, which contains a 4-byte checksum and an 8-byte timestamp and a variable-length data payload. The TCP protocol is used for transmission, and the maximum segment size (MSS) is set to 1460 bytes and the window size is 65535 bytes. The breakpoint-resume mechanism is implemented by recording the amount of data transmitted, and a checkpoint is set for every 100KB of data. The exponential backoff algorithm is used for error retransmission, with an initial retransmission interval of 100ms and a maximum number of retransmissions of 5 times.When data packets arrive out of order, the receiver uses a 1MB buffer to reassemble them and sort them according to the sequence number in the packet header. The processing takes an average of 200ms, meeting real-time transmission requirements.
[0021] By analyzing and comparing copyright rules in different regions, we can identify differences in the rules and generate a mapping conversion solution for copyright metadata based on the differences, ensuring that the structure and semantics of the converted metadata fully match the current regional rules.
[0022] The copyright rules of different regions are compared to obtain the word frequency-inverse document frequency vector of the rule text. Based on the word frequency-inverse document frequency vector, the similarity between the rules is calculated by cosine similarity. If the similarity between the rules is lower than the preset threshold, the rule difference point is identified. Based on the rule difference point, a rule difference tree structure is constructed. The nodes of the rule difference tree structure include difference type, difference content and region information. The rule difference tree is traversed by a depth-first search algorithm to extract difference features. The difference features are processed using a pre-trained long short-term memory neural network model. The long short-term memory neural network model includes an input layer, an LSTM layer and a fully connected output layer. Based on the difference features, metadata mapping conversion rules are generated. Based on the metadata mapping conversion rules, the original copyright metadata is restructured and semantically converted, including locating the corresponding fields in the metadata, and format conversion is performed through regular expressions to obtain a new copyright metadata structure that matches the current regional rules. The new copyright metadata structure is verified by a rule matching scoring mechanism.
[0023] Specifically, the TF-IDF algorithm is used to compare copyright rules in different regions, and the word frequency-inverse document frequency vector of the rule text is calculated. The similarity between rules is calculated by cosine similarity, and the similarity threshold is set to 0.8. Rule paragraphs with similarity below the threshold are identified as potential rule differences. According to the identified rule differences, a rule difference tree structure is constructed. The nodes of the tree contain difference types, such as addition, deletion, modification, difference content and regional information. The hierarchical structure of the differences is represented by the parent-child relationship between nodes. The difference tree is traversed by the depth-first search algorithm to extract key difference features, including difference type frequency, difference content keywords and rule level where the differences are located. A pre-trained long short-term memory neural network model is used. The model includes an input layer, two LSTM layers, each of which has 128 neurons and a fully connected output layer. The training data comes from historical copyright rule differences and their corresponding mapping transformation instances. The extracted difference features are used as input to generate corresponding metadata mapping transformation rules. The transformation rules contain operation instructions such as field mapping, value conversion and structure adjustment. Based on the generated mapping conversion rules, the original copyright metadata is restructured and semantically converted. The JSONPath syntax is used, such as $.copyright.owner to represent the copyright owner field, to locate and modify specific fields in the metadata, and regular expressions such as \d{4}-\d{2}-\d{2} are used to match the date format and convert the value format, finally obtaining a new copyright metadata structure that matches the current regional rules. The converted metadata structure is verified by the rule matching scoring mechanism, and the field coverage and value format compliance of the new structure and the target regional rules are calculated. If the matching degree is lower than the preset threshold of 0.95, the feedback adjustment mechanism is triggered to re-execute the difference analysis and conversion rule generation process. In practical applications, the copyright rule texts of two different regions can be first processed by TF-IDF. Each text contains about 1,000 words, and a 10,000-dimensional term frequency-inverse document frequency vector is generated. The cosine similarity of the two vectors is calculated, and a result of 0.75 is obtained, which is lower than the preset threshold of 0.8, so the rule difference is identified. Then a rule difference tree is constructed, with a tree height of 5 layers, containing 50 nodes, and each node contains an average of 3 child nodes. A depth-first search traverses the difference tree and extracts 20 key difference features, including 5 difference type frequencies, 10 difference content keywords, and 5 rule level information of the difference. These features are input into a pre-trained long short-term memory neural network with 128 input neurons, two LSTM layers with 128 neurons each, and a fully connected output layer with 64 neurons to generate mapping transformation rules. 100 rounds of training were performed on 100,000 historical data, achieving an accuracy of 95%. The generated transformation rules are applied to the original metadata. The original JSON format data size is 5KB and contains 30 fields.Eight fields were modified using JSONPath syntax such as $.copyright.owner, and the formats of 12 values were converted using five regular expressions such as \d{4}-\d{2}-\d{2}. The size of the new metadata structure after conversion is 6KB, containing 35 fields. The rule matching scoring mechanism calculated that the field coverage rate is 97%, the value format compliance rate is 96%, and the overall matching degree is 96.5%, which exceeds the preset threshold of 95% and does not need to trigger the feedback adjustment mechanism.
[0024] Step S104, the copyright management node parses the current regional copyright rules based on the copyright information of the region, parses the received copyright metadata, extracts copyright ownership and term information, and if a copyright exclusivity conflict is found within the region, initiates the copyright dispute resolution mechanism, and gives compliant copyright authorization recommendations through the rule reasoning engine.
[0025] Receive copyright metadata, which is in JSON format and includes fields such as copyright owner, right type, authorization scope and validity period; parse the copyright metadata to obtain copyright information; obtain existing copyright records in the current area; compare the validity period of the copyright information with the existing copyright records, construct a time interval and perform an interval overlap query to determine whether there is a time overlap; if there is a time overlap, determine whether there is an intersection between the right type and the authorization scope, compare the binary representation of the right type through bit operations, and determine the intersection of the authorization scope; if an exclusive conflict is determined, trigger a copyright dispute resolution mechanism and call a rule reasoning engine based on a decision tree, wherein the decision tree includes nodes for right priority, time priority and geographical scope; and generate a copyright authorization suggestion based on the output result of the rule reasoning engine.
[0026] Specifically, according to the preset copyright information parsing rules, the received copyright metadata is parsed using a regular expression matching algorithm, where the metadata is in JSON format and contains fields such as copyright owner, right type, authorization scope and validity period, etc., using "(?<copyright_owner> .?)", "(?P<rights_type> .?)" and other patterns to extract key information, and store the extracted results in a relational database. Through database query operations, retrieve the existing copyright records in the current area, use the interval tree algorithm to compare the validity periods of the old and new copyright records, build a time interval tree and perform interval overlap queries. If time overlap is found, further determine whether there is an intersection between the right type and the authorization scope. Use bit operations to compare the binary representation of the right type, and use geographic coordinate overlap detection to determine the intersection of the authorization scope, thereby identifying potential exclusive conflicts. If an exclusive conflict is identified, the copyright dispute resolution mechanism is triggered, and a rule reasoning engine based on a decision tree is called. The decision tree contains nodes such as right priority, time priority, and geographical scope. Each node has a weight value and a threshold. The conflicting copyright record is used as input, and the decision tree nodes are traversed. According to The preset weight and priority rules evaluate and sort conflicts, while taking into account the impact of historical authorization records. Based on the output results of the rule reasoning engine, compliant copyright authorization suggestions are generated, including the recommended authorization subject, scope of rights, and term adjustment plan. A rule-based generation algorithm is used, taking into account the minimum rights adjustment principle and the fair distribution principle. For multi-party conflicts, the game theory model is used to calculate the optimal allocation plan, and the generated suggestions are sent to the relevant parties through the message queue. The processing process and results are recorded in the copyright management system, and the copyright information parsing rule library is updated to ensure that the rules are consistent with the latest laws and regulations. In actual applications, the copyright management node receives a 5KB JSON format copyright metadata containing 20 fields. Use the preset 10 regular expression patterns, such as "(?P<copyright_owner> .*?)" to extract the copyright ownership, with a processing time of 50 milliseconds. 15 valid fields were extracted and stored in the PostgreSQL database. Subsequently, SQL query statements were used to retrieve relevant copyrights from 1 million existing records, which took 200 milliseconds. The interval tree algorithm was used to build a time interval tree containing 10,000 nodes. The query of overlapping intervals took 30 milliseconds and 5 time overlapping records were found. The right types were compared through bit operations of 32-bit integers, and the overlap of authorization scopes was calculated using latitude and longitude coordinates, identifying 2 exclusive conflicts. The dispute resolution mechanism was triggered and the decision tree inference engine was called. The tree depth was 5, containing 30 nodes, and each node had an average of 3 branch conditions and 2 thresholds. It took 20 milliseconds to traverse the decision tree, and the influence of 3 historical authorization records was considered.When generating copyright authorization suggestions, a rule-based generation algorithm is used, which includes 15 priority rules, applies a game theory model to the 2-party conflict, and calculates the optimal solution after 100 iterations, which takes 100 milliseconds. The final generated authorization suggestion includes 3 adjustment plans, which are sent through the RabbitMQ message queue with an average transmission delay of 10 milliseconds. It takes a total of 450 milliseconds from receiving metadata to generating suggestions, which meets the real-time processing requirements. The copyright information parsing rule base is automatically updated once a week, with an average update of 50 rules to ensure consistency with the latest regulations.
[0027] After receiving the synchronized copyright metadata, the management nodes in different regions use the decompression method that matches the source region to decompress the data, and use the corresponding decryption method to decrypt the metadata, restoring the copyright metadata structure that adapts to the rules of different regions; the decrypted copyright metadata is stored according to the data format standards of different regions, and the copyright metadata is registered in the copyright management process of different regions through the interface of the copyright information management system and then parsed.
[0028] According to the copyright metadata packet header information, the source region identifier is identified, and the corresponding decompression method and decryption method are obtained from the preset algorithm mapping table; the algorithm mapping table adopts a key-value pair storage structure, which contains the corresponding relationship between the source region, the target region, the decompression method and the decryption method. The function library is called to perform data decompression and decryption operations to obtain the restored original copyright metadata; if the decompression or decryption fails, the error handling mechanism is triggered, the error log is recorded and the source region is requested to retransmit. The pre-built regional rule adapter is used to convert the restored copyright metadata according to the configuration file. The conversion process includes field mapping, value conversion and structural reorganization to generate new copyright metadata that meets the data format standard of this region. For the new copyright metadata, the storage method is selected according to the data storage strategy of the current region; wherein the storage selection criteria include data volume, query frequency, and associated complexity factors. After the new copyright metadata is stored through the data access layer, a unique identifier and index information are generated. The stored new copyright metadata is registered in the copyright management process of this region. The registration process includes data integrity verification, copyright information consistency check, copyright conflict detection and copyright status initialization.
[0029] Specifically, according to the received copyright metadata packet header information, the source region identifier is identified, and the corresponding decompression algorithm and decryption algorithm are searched from the preset algorithm mapping table. The mapping table contains the corresponding relationship between the source region, the target region, the decompression algorithm and the decryption algorithm, and adopts a key-value pair storage structure. The corresponding function library is called to perform data decompression and decryption operations to restore the original copyright metadata. If the decompression or decryption fails, the error handling mechanism is triggered, the error log is recorded and the source region is requested to retransmit. The pre-built regional rule adapter is used. The adapter implements the mapping rule engine based on the configuration file to convert the restored copyright metadata into a structure adapted to the current regional rules. The conversion process uses field mapping, value conversion and structure reorganization operations, including field name correspondence, data type conversion and nested structure adjustment, to generate new copyright metadata that meets the data format standard of this region. For the converted copyright metadata, according to the data storage strategy of the current region, a suitable storage method is selected. The storage selection criteria include factors such as data volume, query frequency, and association complexity. If the data volume is large and the query is frequent, a relational database is selected, and if the data structure is flexible, a document database is selected. The metadata is written into the storage system through the data access layer, and a unique identifier and index information are generated at the same time. The registration interface provided by the copyright information management system is called to register the stored copyright metadata into the copyright management process of this region. The registration process includes data integrity verification, copyright information consistency check, copyright conflict detection and copyright status initialization. For additional data processing requirements for specific regions, such as sensitive information filtering or special field supplementation, they are implemented through regional customized plug-ins. After registration is completed, the registration result and copyright identifier are returned. In actual application, the copyright management node receives a 10MB encrypted and compressed copyright metadata packet, and the header information shows that the source region identifier is "REG001". The corresponding information is searched from the preset algorithm mapping table, which contains the mapping relationship of 100 regions, and the search takes 5 milliseconds. The applicable LZMA2 decompression algorithm and AES-256 decryption algorithm are identified, and the corresponding function library is called for operation. The decompression rate reaches 60%, the decryption key length is 256 bits, and the entire decompression and decryption process takes 200 milliseconds to restore 50MB of original JSON format copyright metadata. The decompression and decryption success rate is 99.9%. If it fails, it will automatically retry 3 times, each with an interval of 500 milliseconds. The regional rule adapter processes the conversion of 100 fields based on 500 mapping rules, including 30 field renamings, 50 data type conversions, and 20 structural reorganizations, which takes 150 milliseconds. Based on the characteristics of 1 million queries per day in the current region and an average single data size of 100KB, PostgreSQL relational database storage is selected, a data table with 20 fields is created, 3 indexes are set, and writing takes 50 milliseconds.During the copyright registration process, 100 data integrity checks were performed, and the accuracy of copyright information consistency verification reached 99.8%. The conflict detection algorithm processed an average of 10,000 historical records in 100 milliseconds. The sensitive information filtering plug-in for specific areas processed 20 sensitive fields and added 5 additional area-specific fields. The entire registration process took 300 milliseconds. From reception to registration completion, the average processing time was controlled within 1 second, meeting real-time processing requirements.
[0030] Step S105, according to the authorization result of the copyright management node, the unified copyright revenue distribution rules in the region are adopted to realize the distribution and settlement of copyright revenue, and the authorized copyright information is synchronized to the copyright management node in the relevant region.
[0031] Obtain authorization-related information according to the authorization result generated by the copyright management node, the authorization-related information including the authorization object, authorization scope, authorization period and authorization fee; calculate the right holder's profit share using a preset profit distribution rule, the profit distribution rule including profit distribution to right holders with different contribution levels, and the profit distribution weight is determined based on the creative contribution level, market influence and historical profit ratio; perform structured processing on the authorized copyright information to obtain a complete copyright information data packet, the copyright information data packet including the authorization result, profit distribution status and copyright status update information; push the copyright information data packet to a pre-configured regional copyright management node, the regional copyright management node receives the copyright information data packet and performs verification, and if the verification passes, updates the copyright information in the local database.
[0032] Specifically, according to the authorization results generated by the copyright management node, the authorization-related information is extracted, including the authorized object, authorization scope, authorization period and authorization fee. The revenue share of each right holder is calculated through the preset revenue distribution rules, and the revenue is distributed to the right holders with different contributions using a nonlinear weighted algorithm. The weights are determined based on the creative contribution, market influence and historical revenue share. For multi-level right holders, a recursive allocation algorithm with a tree structure is used to calculate the revenue of each level of right holders from top to bottom. The revenue distribution results are executed using smart contracts on the Ethereum platform. The contract code contains the revenue calculation logic and transfer function. The contract execution is triggered by the blockchain node, and the calculated revenue amount is automatically transferred to the digital wallet of each right holder. If the transfer fails, the error log is recorded and the retry mechanism is started. A maximum of 3 retries are made, with an interval of 5 minutes each time. At the same time, the transaction details, including the allocation time, amount and recipient address, are recorded on the blockchain to ensure the transparency and immutability of the allocation process. The copyright information after authorization is structured, and the authorization results, revenue distribution, and copyright status update information are integrated into the original copyright metadata to form a complete copyright information data packet, and a unique version identifier is generated based on the current timestamp and the SHA-256 hash value combination of the data content. The updated copyright information data packet is pushed to the pre-configured relevant regional copyright management node through the distributed message queue system. The receiving node first verifies the digital signature of the data packet to ensure the reliability of the source, and then performs a content consistency check, including version number continuity and key field value rationality verification. After verification, the new version of the copyright information is updated to the local database to achieve real-time synchronization of copyright information across regions. The unified copyright revenue distribution rules within the region are formulated by the regional copyright management committee, and the rules are updated every quarter through a voting mechanism to ensure the consistency of the rules with market practices and laws and regulations. In actual applications, the copyright management node handles an authorization case involving 5 rights holders, and the authorization fee is 100,000 yuan. The nonlinear weighted algorithm uses a power function model, and the weight calculation formula is w=(c^0.5*m^0.3*h^0.2), where c is the creative contribution of 1-10 points, m is the market influence of 1-10 points, and h is the historical income ratio of 0-100%. The weights of the five rights holders are calculated as 10.5, 8.2, 6.7, 5.3 and 4.1 respectively. The tree-structured recursive allocation algorithm handles the three-level rights holder relationship, which takes 50 milliseconds. The Ethereum smart contract code size is 2KB, the Gas consumption is 50,000 Wei, and the contract execution time is 2 seconds. The digital wallet transfer success rate is 99.8%, and the success rate is increased to 99.99% after 3 retries. The copyright information data packet size is 200KB, and the SHA-256 hash value and Unix timestamp can be accurately spliced to generate a 32-byte version identifier to milliseconds. The distributed message queue uses Kafka, the number of topic partitions is 10, the replication factor is 3, and the message push delay is less than 100 milliseconds.The data packet verification uses the ECDSA signature algorithm, the signature length is 64 bytes, and the verification time is 10 milliseconds. The content consistency check includes 20 key fields and takes 30 milliseconds to check. The local database uses MongoDB, and the write operation takes an average of 5 milliseconds. The regional copyright management committee consists of 15 members. The rule update adopts a weighted majority voting mechanism, and the weight is based on the market share of the member units. The update process takes an average of 3 days. From the generation of the authorization result to the completion of cross-regional synchronization, the total time is controlled within 10 seconds, meeting the real-time processing requirements.
[0033] Step S106, during the cross-regional transmission of copyright information, a lightweight copyright verification module is deployed at the network edge node to perform copyright verification on the audio and video streams passing through the node to determine whether they have legal cross-regional authorization, and perform real-time control according to regional copyright usage restriction rules.
[0034] Perform real-time data packet detection on the audio and video data stream of the network edge node; perform discrete cosine transform according to the data packet detection result to obtain the frequency domain analysis result of the audio and video data stream; extract digital watermark information and metadata identifier from the frequency domain analysis result; obtain a pre-synchronized copyright authorization database; perform copyright matching on the digital watermark information and metadata identifier to obtain a copyright matching result; if the copyright matching result indicates that the audio and video data stream has a legal cross-regional authorization, obtain a preset regional copyright usage restriction rule; perform real-time control on the audio and video data stream according to the regional copyright usage restriction rule to obtain a control result; push the copyright matching result and the control result to a central management node.
[0035] Specifically, a lightweight copyright verification module based on FPGA is deployed at the edge node of the network. The FPGA chip configuration contains 100,000 logic units and the processing speed reaches 1Gbps. Real-time data packet detection of the audio and video streams passing through the node is realized through hardware acceleration. Discrete cosine transform (DCT) is used for frequency domain analysis to extract digital watermark information and metadata identifiers in the audio and video streams. The extracted copyright information is quickly matched with the pre-synchronized copyright authorization database to verify whether the audio and video streams have legal cross-regional authorization. The matching process uses Bloom filters for preliminary screening. The Bloom filter is set to 1,000,000 bits and uses 3 hash functions to reduce the number of database queries. The copyright authorization database is synchronized with the central node every 5 minutes through an incremental update mechanism to ensure the real-time nature of the data. According to the matching results, combined with the preset regional copyright usage restriction rules, the audio and video streams are controlled in real time. The control methods include release, bandwidth restriction, quality degradation or blocking. The control decision is implemented through the decision tree algorithm. The decision tree contains 10 decision nodes and 16 leaf nodes. The input includes features such as authorization status, usage area, time period, etc., and the output is a specific control operation. The copyright verification results and control behaviors are recorded in the distributed log system. The ELK (Elasticsearch, Logstash, Kibana) architecture is adopted. The Elasticsearch cluster is configured with 3 nodes. At the same time, the abnormal situation is pushed to the central management node in real time through the Apache Kafka message queue. Kafka is configured with 3 Broker server instances and the number of topic partitions is 6 for subsequent copyright tracking and rule optimization. In actual applications, the FPGA copyright verification module deployed at the edge node of the network processes a 1080p, 30fps H.264 encoded video stream with a data rate of 6Mbps. The FPGA chip adopts the Xilinx Virtex-7 series, contains 118,560 logic units, and has a processing speed of 1.2Gbps. DCT frequency domain analysis is performed on 8x8 pixel blocks to extract 32-bit digital watermarks embedded in the intermediate frequency coefficients, and 64-byte copyright identifiers are extracted from the video stream metadata. The entire extraction process takes no more than 2 milliseconds. The copyright authorization database contains 10 million authorization records. The Bloom filter is set to 12,000,000 bits, using 5 hash functions, and the false alarm rate is controlled below 0.1%. The initial screening process takes an average of 0.5 milliseconds. The database incremental update synchronizes about 1,000 new records each time, which takes no more than 2 seconds. The average time for the decision tree algorithm to process a single authorization judgment is 0.2 milliseconds, the tree depth is 4, and it contains 15 decision nodes and 16 leaf nodes. In the ELK log system, the Elasticsearch cluster processes 5,000 log records per second, and the data retention period is 30 days. The number of Topic partitions configured for the Kafka message queue is 12, and the replication factor is 3.The copyright verification and control process, from the entry of data packets to decision execution, has an end-to-end delay of less than 10 milliseconds, meeting the requirements of real-time processing. When processing 100Mbps network traffic, the CPU utilization rate is maintained at around 65%, and the memory occupies 4GB, effectively ensuring the stable operation of edge nodes.
[0036] Step S107, in which audio and video with legal cross-regional authorization are transmitted through multiple regions to perform collaborative training on copyright data distributed in different regions, thereby optimizing the copyright recognition model and rule reasoning engine.
[0037] Acquire audio and video transmission data and copyright information, and preprocess and extract features of the data through edge computing nodes; according to the preprocessing and feature extraction results, use a federated learning algorithm to train the copyright data, and use the FedAvg algorithm to aggregate model parameters; build a knowledge graph based on historical authorization data and copyright processing records to analyze the relationship between copyright entities in the knowledge graph; update the copyright identification model and rule inference engine online, and if the update time interval reaches a preset threshold, use an online gradient descent incremental learning method to update the model parameters; obtain copyright usage trend data, perform anomaly analysis on the copyright usage trend data, and determine whether there is potential copyright infringement.
[0038] Specifically, the audio and video transmission data and copyright information of multiple regions are collected through a distributed data collection network, and edge computing nodes are used for data preprocessing and feature extraction. The edge nodes use the ResNet-50 convolutional neural network to extract audio and video features, and the extracted feature data is uploaded to the cloud storage system through a secure multi-party computing protocol. The federated learning algorithm is used to conduct collaborative training on copyright data distributed in different regions, and the FedAvg algorithm is used to aggregate model parameters. Under the premise of protecting data privacy, the homomorphic encryption technology is used to encrypt and transmit model parameters, update the parameters of the copyright identification model, and improve the model's recognition accuracy for cross-regional copyright features. Based on historical authorization data and copyright processing records, a knowledge graph is constructed, and the relationship between copyright entities is analyzed through the GraphSAGE graph neural network algorithm. The decision logic of the rule reasoning engine is optimized to improve the reasoning ability of complex cross-regional copyright issues. The incremental learning method of online gradient descent is used to update the copyright identification model and rule reasoning engine online every hour, and the model parameters and rule weights are adjusted according to real-time feedback data to achieve the system's rapid adaptation and processing of emerging copyright issues. Analyze copyright usage trends through the big data analysis platform, use anomaly detection algorithms to identify potential copyright infringements, combine the updated results of the knowledge graph with the incremental learning model, and dynamically adjust feature weights and decision thresholds.
[0039] In actual applications, the distributed data collection network deployed 1,000 edge computing nodes, each equipped with an NVIDIA Jetson Xavier NX processor, using the ResNet-50 model to process 30 frames of 1080p video per second and extract 2048-dimensional feature vectors. The secure multi-party computing protocol uses Shamir's Secret Sharing to divide the feature data into 5 copies and upload them to the cloud through different routes. The FedAvg algorithm performs federated learning on 100 regional nodes. Each round of training involves 10 million pieces of copyright data. The CKKS homomorphic encryption scheme is used to protect model parameters. A single round of training takes 30 minutes, and the model accuracy increases by 5% after 10 rounds. The knowledge graph contains 100 million entities and 1 billion relationships. The GraphSAGE algorithm uses a 3-layer network structure with 128 hidden units in each layer, and the single inference time is controlled within 10 milliseconds. The incremental learning uses the Adam optimizer, with a learning rate set to 0.001. The model is updated once an hour, and 100,000 new data are processed each time. The big data analysis platform uses a Spark cluster with 50 working nodes and processes 100TB of data every day. The anomaly detection algorithm is based on Isolation Forest, with an anomaly threshold of 0.95, and detects 10 million copyright usage behaviors per hour with an accuracy rate of 98%. When the entire system processes 1,000 concurrent copyright queries, the average response time is controlled within 100 milliseconds, the CPU utilization rate is maintained at 75%, and the memory occupies 200GB, effectively balancing processing efficiency and resource consumption.
[0040] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.
Claims
1. An intelligent cloud audio and video access and big data analysis method, It is characterized in that The method comprises: According to the granularity of copyright regional division, the audio and video streams are analyzed in real time, the copyright watermark information in the region is extracted and the copyright feature vector is generated. The feature vector is classified through the pre-trained copyright recognition model to obtain the preliminary copyright attribution result in the region; based on the preliminary copyright attribution result, the transmission path information of the cross-regional copyright authorized audio and video is obtained through network routing tracing, and the current copyright jurisdiction area is determined in combination with the IP address geographic location database, and the copyright rule database applicable to the area is queried to obtain the corresponding copyright period and use restriction rule information; if there is a difference between the current regional copyright rule and the preliminary attribution result, the copyright metadata structure is dynamically adjusted according to the differentiated configuration of the regional copyright rule, and the copyright information is re-encoded using the encryption and compression method corresponding to the area, and synchronized to the copyright management node in this area through the copyright data transmission protocol; the copyright management node is based on the copyright information in this area Parse the current regional copyright rules, parse the received copyright metadata, extract copyright ownership and term information, and if a copyright exclusivity conflict is found in the region, initiate the copyright dispute resolution mechanism, and give compliant copyright authorization suggestions through the rule reasoning engine; according to the authorization results of the copyright management node, use the unified copyright revenue distribution rules in the region to achieve the distribution and settlement of copyright revenue, and synchronize the authorized copyright information to the copyright management nodes in the relevant regions; in the process of cross-regional transmission of copyright information, deploy a lightweight copyright verification module at the edge node of the network to perform copyright verification on the audio and video streams passing through the node to determine whether they have legal cross-regional authorization, and perform real-time control according to the regional copyright use restriction rules; audio and video with legal cross-regional authorization are transmitted through multiple regions to conduct collaborative training on copyright data distributed in different regions, and optimize the copyright recognition model and rule reasoning engine; The method of dividing the copyright area into granularities, performing real-time analysis on the audio and video streams, extracting the copyright watermark information in the area and generating a copyright feature vector, and classifying the feature vector through a pre-trained copyright recognition model to obtain a preliminary copyright ownership result in the area includes: According to the preset copyright area division rules, the audio and video streams are segmented, the watermark features of the segmented areas are extracted, and the two-dimensional discrete Fourier transform results of the watermark features are obtained; By calculating the frequency domain amplitude spectrum of the two-dimensional discrete Fourier transform result, a preset number of high-frequency components are selected and determined as the copyright feature vector; Using principal component analysis to perform dimensionality reduction processing on the copyright feature vector, obtaining feature components whose cumulative variance contribution rate reaches a preset threshold, and obtaining a copyright feature representation after dimensionality reduction; Retrieving the most similar matching item to the copyright feature representation after dimensionality reduction from a pre-established copyright feature database, and determining that the match is successful if the similarity is higher than a preset threshold, and obtaining the corresponding copyright ownership information; If the similarity is lower than a preset threshold, the copyright feature representation after dimension reduction is input into a pre-trained copyright classifier, and the copyright ownership of the region is determined by performing multi-category classification on the copyright feature representation; Among them, based on the preliminary copyright ownership results, the transmission path information of cross-region copyright authorized audio and video is obtained through network routing tracking, the current copyright jurisdiction area is determined in combination with the IP address geographic location database, and the copyright rule database applicable to the area is queried to obtain the corresponding copyright period and use restriction rule information, including: Perform application layer protocol identification on audio and video data packets to obtain the source IP address and destination IP address of the audio and video data packets; By using a routing tracking method combining ICMP protocol and UDP protocol, the transmission path node information of the data packet in the network is obtained, wherein the transmission path node information includes a node IP address and a router identifier; Querying the geographical location information of each node in the transmission path node information from a pre-established IP address geographical location mapping database; If the geographic location information satisfies the preset copyright jurisdiction division rule, then determining the copyright jurisdiction where the audio or video is currently located; According to the copyright jurisdiction, obtaining corresponding copyright rules from a pre-built copyright rule database, wherein the copyright rules include copyright term and usage restriction information; If cross-regional copyright authorization is involved, the copyright rules of multiple regions are prioritized and conflicts are resolved to generate a comprehensive set of copyright rules, including a comprehensive score for usage restrictions based on the following formula: U represents the comprehensive score of the usage restriction, Ai represents the usage restriction weight of the ith area, Pi represents the priority of the ith area, and n represents the number of areas; If the current regional copyright rules differ from the preliminary attribution results, the copyright metadata structure is dynamically adjusted according to the differentiated configuration of the regional copyright rules, the copyright information is re-encoded using the encryption and compression methods corresponding to the region, and synchronized to the copyright management node of the region through the copyright data transmission protocol, including: Calculate the Jaccard similarity difference based on the obtained regional copyright rules and preliminary attribution results; Determining whether the difference exceeds a preset threshold, and if the difference exceeds the preset threshold, selecting a metadata template adapted to the regional copyright rule; Parsing the regional copyright rules, extracting key attributes in the regional copyright rules, mapping the key attributes to the metadata template, and generating a copyright metadata structure; Encrypting the copyright metadata structure to obtain encrypted copyright data; compressing the encrypted copyright data to obtain encoded copyright information; Encapsulating the encoded copyright information through a copyright data transmission protocol to generate a copyright data packet; The copyright data packet is sent to the copyright management node using the TCP protocol, wherein the copyright data packet includes a checksum and a timestamp; the method further includes: identifying the differences in the rules by analyzing and comparing the copyright rules of different regions, and generating a mapping conversion scheme for the copyright metadata based on the differences, so as to ensure that the structure and semantics of the converted metadata completely match the current regional rules; The copyright management node analyzes the current regional copyright rules based on the copyright information of the region, analyzes the received copyright metadata, extracts copyright ownership and term information, and activates the copyright dispute resolution mechanism if a copyright exclusivity conflict is found in the region. The rule inference engine gives corresponding compliant copyright authorization suggestions, including: Receive copyright metadata, which is in JSON format and includes fields such as copyright owner, right type, authorization scope, and validity period; Parsing the copyright metadata to obtain copyright information; Get the copyright records that exist in the current region; Compare the copyright information with the validity period of the existing copyright record, construct a time interval and perform an interval overlap query to determine whether there is a time overlap; If there is time overlap, determine whether the right type and the authorization scope have an intersection, and compare the binary representation of the right type through bit operations to determine the intersection of the authorization scope; If an exclusive conflict is determined, the copyright dispute resolution mechanism is triggered, and a rule-based reasoning engine based on a decision tree is called, wherein the decision tree includes nodes for rights priority, time priority, and geographical scope; Generate copyright authorization suggestions based on the output results of the rule reasoning engine; also include: after the management nodes in different regions receive the synchronized copyright metadata, use the decompression method that matches the source region to decompress the data, and use the corresponding decryption method to decrypt the metadata, restore the copyright metadata structure that adapts to the rules of different regions, store the decrypted copyright metadata according to the data format standards of different regions, and register the copyright metadata to the copyright management process of different regions through the interface of the copyright information management system and then parse it.
2. The method according to claim 1, in, The method analyzes and compares the copyright rules of different regions, identifies the rule differences, and generates a mapping conversion scheme for copyright metadata based on the differences to ensure that the converted metadata structure and semantics fully match the current regional rules, including: comparing the copyright rules of different regions to obtain the word frequency-inverse document frequency vector of the rule text; calculating the similarity between the rules through cosine similarity based on the word frequency-inverse document frequency vector; identifying the rule differences if the similarity between the rules is lower than a preset threshold; constructing a rule difference tree structure based on the rule differences; the nodes of the rule difference tree structure contain the difference type, difference content and regional information. information; traverse the rule difference tree through a depth-first search algorithm to extract difference features; use a pre-trained long short-term memory neural network model to process the difference features; the long short-term memory neural network model includes an input layer, an LSTM layer and a fully connected output layer; generate metadata mapping conversion rules according to the difference features; based on the metadata mapping conversion rules, restructure and semantically convert the original copyright metadata, including locating the corresponding fields in the metadata, and converting the format through regular expressions to obtain a new copyright metadata structure that matches the current regional rules; verify the new copyright metadata structure through a rule matching degree scoring mechanism.
3. The method according to claim 1, in, After receiving the synchronized copyright metadata, the management nodes of the different regions use a decompression method that matches the source region to decompress the data, and use a corresponding decryption method to decrypt the metadata, restore the copyright metadata structure that adapts to the rules of different regions, store the decrypted copyright metadata according to the data format standards of different regions, register the copyright metadata in the copyright management process of different regions through the interface of the copyright information management system, and then parse it, including: identifying the source region identifier according to the copyright metadata packet header information, and obtaining the corresponding decompression method and decryption method from a preset algorithm mapping table; The algorithm mapping table adopts a key-value pair storage structure, including the corresponding relationship between the source area, the target area, the decompression method and the decryption method; Call the function library to perform data decompression and decryption operations to obtain the restored original copyright metadata; if the decompression or decryption fails, the error handling mechanism is triggered, the error log is recorded, and retransmission is requested from the source area; Using a pre-built regional rule adapter, the restored copyright metadata is converted according to the configuration file. The conversion process includes field mapping, value conversion and structure reorganization to generate new copyright metadata that conforms to the data format standard of this region. For the new copyright metadata, a storage method is selected according to the data storage strategy of the current region. The storage selection criteria include data volume, query frequency, and associated complexity factors; after storing the new copyright metadata through the data access layer, a unique identifier and index information are generated; the stored new copyright metadata is registered in the copyright management process of this area, and the registration process includes data integrity verification, copyright information consistency check, copyright conflict detection, and copyright status initialization.
4. The method according to claim 1, in, The method of using a unified copyright revenue distribution rule in the region to distribute and settle copyright revenue according to the authorization result of the copyright management node, and synchronizing the authorized copyright information to the copyright management node in the relevant region, includes: Acquire authorization related information according to the authorization result generated by the copyright management node, wherein the authorization related information includes the authorization object, authorization scope, authorization period and authorization fee; The profit share of the right holder is calculated using the preset profit distribution rules, which include distributing profits to right holders with different contribution levels, and the profit distribution weight is determined based on the creative contribution, market influence and historical profit ratio; Performing structural processing on the authorized copyright information to obtain a complete copyright information data packet, wherein the copyright information data packet includes authorization results, revenue distribution information, and copyright status update information; The copyright information data packet is pushed to a pre-configured regional copyright management node, and the regional copyright management node receives the copyright information data packet and performs verification. If the verification passes, the copyright information in the local database is updated.
5. The method according to claim 1, in, In the process of cross-regional transmission of copyright information, a lightweight copyright verification module is deployed at the network edge node to perform copyright verification on the audio and video streams passing through the node to determine whether they have legal cross-regional authorization, and perform real-time control according to regional copyright usage restriction rules, including: Perform real-time packet detection on audio and video data streams at network edge nodes; Perform discrete cosine transform according to the data packet detection result to obtain the frequency domain analysis result of the audio and video data stream; Extracting digital watermark information and metadata identifier from the frequency domain analysis result; Obtain a pre-synchronized copyright authorization database; Performing copyright matching on the digital watermark information and the metadata identifier to obtain a copyright matching result; If the copyright matching result indicates that the audio and video data stream has legal cross-regional authorization, then obtaining a preset regional copyright usage restriction rule; Controlling the audio and video data stream in real time according to the regional copyright usage restriction rules to obtain a control result; The copyright matching result and the control result are pushed to a central management node.
6. The method according to claim 1, in, The audio and video with legal cross-regional authorization is transmitted through multiple regions to conduct collaborative training on copyright data distributed in different regions, and optimize the copyright identification model and rule reasoning engine, including: Obtain audio and video transmission data and copyright information, and perform data preprocessing and feature extraction through edge computing nodes; According to the preprocessing and feature extraction results, the copyright data is trained using a federated learning algorithm, and the model parameters are aggregated using a FedAvg algorithm; Building a knowledge graph based on historical authorization data and copyright processing records to analyze the relationship between copyright entities in the knowledge graph; The copyright identification model and rule inference engine are updated online. If the update time interval reaches a preset threshold, the model parameters are updated using an online gradient descent incremental learning method. Obtain copyright usage trend data, perform anomaly analysis on the copyright usage trend data, and determine whether there is potential copyright infringement.
