12 results about "Approximate matching" patented technology

The invention provides a multi-target accurate retrieval method and system in a security video, and relates to the field of interdisciplinary application, and the method is characterized in that the method comprises the following steps: carrying out the standardized preprocessing of an original security video, and deploying an improved YOLOv8 model on a preprocessed key frame to carry out the high-precision multi-target detection. The method has the advantages that the appearance, motion and semantic complementary features of the target are synchronously extracted by improving the YOLOv8 model, cross-frame association and trajectory modeling are realized by adopting an enhanced DeepSORT tracker, and a compact feature index database and a three-level progressive retrieval mechanism are constructed in combination with a locality sensitive hashing algorithm; according to the method, the multi-target retrieval precision and efficiency in massive videos are remarkably improved, the target relevance and the complex scene semantic understanding ability are enhanced, meanwhile, the calculation overhead is greatly reduced through rapid approximate matching and a hierarchical retrieval strategy, and the requirements of security application for high accuracy, rapid response and real-time intelligent research and judgment are met.

A system includes a processor that accesses a sequence of device events S from IoT network traffic and a plurality of user activity signatures, each user activity signature U in the plurality of user activity signatures being a known sequence of device events. The processor is configured to infer a user activity by way of approximate user activity signature matching.

This invention discloses an AI voice query system for cultural and tourism information in a two-wheeled vehicle rental scenario, belonging to the field of intelligent transportation and voice interaction technology. The system collects user voice data via an in-vehicle voice terminal, obtaining GNSS latitude, longitude, heading, speed, and timestamps, which are then sent to a server. The server recognizes the voice, outputs the optimal transcription and uncertain results in N-best or obfuscated network / word lattice form, constructs a weighted query graph, and recalls a set of candidate POIs based on POI name approximation matching index. A forward cycling corridor is constructed based on heading and speed to spatially prune the candidates. Cycling accessibility costs are calculated on the road network, and combined with rental fences and return points for filtering or correction, resulting in composite costs and ranking to determine target POIs. Prefetching and caching based on corridors and high-frequency intents are used to adapt to weak network environments, thereby improving the stability of recall and ranking under noisy conditions and reducing irrelevant computational load.

A technique of categorization of binary executable files comprising the following steps:a) a starting step of analysis of input and training data for containing semi-organized, partially monotonic sequences;b) a preprocessing step wherein potential sequences are discarded or accepted on the basis of preset statistical criteria;c) an encoding step of said accepted sequences with metadata;d) a storing step of the encoded sequences, wherein said sequences are stored as part of the data sequences digest containing information describing the spatial organization of the sequences, the monotonicity features thereof and other features valuable for approximate matching;e) a computing step of locality-sensitive hashes corresponding to said input and training data files.

The application discloses a three-party efficient private record linkage method and system for large-scale databases, and belongs to the technical field of privacy protection data linkage. The three-party efficient private record linkage method for large-scale databases comprises the following steps: first, second participants respectively perform non-redundant blocking on local records, and send the records to a linkage unit after adding differential privacy disturbance to generate candidate record pairs and randomly and uniformly distribute the candidate record pairs to both parties; a requested party encodes records by using a Bloom filter, encrypts the encoding and an offset list based on quadratic residues, and sends the encoding and the offset list to a requesting party; the requesting party calculates an encrypted intersection cardinality by homomorphic addition, and generates an encrypted similarity result in combination with an encrypted offset; the linkage unit decrypts the result, returns matching record numbers, and both parties output final matching record pairs. The application realizes low computational overhead, low communication volume, high scalability and approximate matching, and improves the efficiency of privacy protection record linkage in large-scale cross-institutional data integration scenarios.

The invention discloses a data resource classification method and device based on rule and semantic approximation and AI (artificial intelligence), and the method comprises the steps: setting a rule matching module, a semantic approximation matching module and an AI model, and sequentially starting the rule matching module, the semantic approximation matching module and the AI model to classify data resources to be classified. According to the data resource classification method and device, on the premise that the classification efficiency is considered, the data can be classified accurately and conveniently, and then the user experience is greatly improved.

The computing system generates a phoneme index for the input lexical elements and performs an approximate match analysis on the content lexical elements of the inverted index database based on the phoneme index for the input lexical elements. The inverted index database includes a first inverted index corresponding to a phoneme index of the content lexical element and a first normal method and a second inverted index corresponding to a phoneme variant of the content lexical element and a second normal method. The computing system returns one or more search results based on the approximate match analysis.

The invention relates to the field of graph processing, and discloses a graph approximate matching method and device, electronic equipment and a storage medium, and the method comprises the steps: carrying out the smoothing of a to-be-matched non-Manhattan two-dimensional graph, and obtaining a smoothed non-Manhattan two-dimensional graph; breaking a target edge in the smoothed non-Manhattan two-dimensional graph, and segmenting the target edge into a plurality of edges to obtain a processed non-Manhattan two-dimensional graph; determining the number of edges in the processed non-Manhattan two-dimensional graph, and determining the preliminary similarity between the non-Manhattan two-dimensional graph and the reference graph based on the number and the corresponding number in the reference graph; and when the non-Manhattan two-dimensional graph is preliminarily similar to the reference graph, matching the edges of the non-Manhattan two-dimensional graph and the reference graph, and determining the similarity between the non-Manhattan two-dimensional graph and the reference graph. The number of times of matching calculation can be reduced and the approximate matching speed of the non-Manhattan graph can be improved by carrying out smoothing processing on the non-Manhattan two-dimensional graph and carrying out preliminary screening on the number of the edges of the non-Manhattan two-dimensional graph.

The application provides a semantic subdivision and modal alignment inference learning cross-modal retrieval method and a retrieval system. The semantic subdivision and modal alignment inference learning cross-modal retrieval method comprises the following steps: performing modal alignment on original modal features obtained after pre-training based on scaled dot-product attention, so as to realize re-aggregated projection modal alignment features for the original features; after the modal alignment data formed in the above step passes through a weight-shared multi-layer perception machine, a semantic approximate matching and correct matching method is used to realize semantic correct matching and approximate matching mining for the same category label cluster; and an Arc4cmr loss function, a mutual supervision contrast loss function, a contrast loss function between a picture-text feature similarity matrix and a similar label matrix are used to constrain the model. The cross-modal retrieval method further improves the accuracy of cross-modal retrieval.

The invention discloses a three-party efficient private record linking method and system for a large-scale database, and belongs to the technical field of privacy protection data linking. The large-scale database-oriented three-party efficient private record linking method comprises the following steps that: a first participant and a second participant respectively carry out non-redundant partitioning on local records, add differential privacy perturbation and then send the local records to a linking unit to generate candidate record pairs, and randomly and uniformly distribute the candidate record pairs to the first participant and the second participant; the requested party encodes the record by using a Bloom filter, encrypts the code and the offset list by using homomorphic encryption based on secondary residual, and sends the code and the offset list to the requesting party; the requester calculates an encryption intersection cardinal number through homomorphic addition, and generates an encryption similarity result in combination with the encryption offset; and the link unit decrypts the result and returns the matching record number, and the two parties output a final matching record pair. According to the method, approximate matching with low calculation overhead, low communication traffic and high expandability is realized, and the privacy protection record linking efficiency in a large-scale cross-mechanism data integration scene is improved.

The invention designs a non-Chinese term real-time voice transcription error correction method and a non-Chinese term real-time voice transcription error correction system for professional speech. The method comprises the following steps: firstly, segmenting a video into continuous time domain intervals based on slide picture change in the speech video; for the speech segments in each time domain interval, using a streaming automatic speech recognition model to obtain an original speech transliteration text; meanwhile, extracting English specialized vocabularies, English abbreviations, variables, Greek letters and mathematical function symbols in formulas in the current page of slides from the video stream, and constructing a non-Chinese term dictionary; then, English segments in the original voice transliteration text are searched for, and a corresponding structural body is constructed; sequentially adopting a spelling similarity detection method and a phoneme feature matching method to correct English professional vocabularies or abbreviation transcription errors, variable transcription errors and Greek letter transcription errors existing in the English fragments; and finally, correcting mathematical function symbol transcription errors and residual Greek letter transcription errors in the original voice transcription text based on Chinese character phonetic form coding by adopting an improved KMP approximate matching method, thereby finally generating an error-corrected voice transcription text. According to the invention, non-Chinese term errors generated by a voice recognition system for professional speeches can be accurately and efficiently corrected.

Techniques are disclosed for querying with semantic code expansion in a clinical data system. In one aspect, a method includes receiving a query containing a predicate specifying a clinical code and a semantic expansion parameter indicating a request for approximate matching. A vector embedding associated with the specified clinical code is retrieved from a pre-computed embedding index. A similarity search is performed in a vector space to identify semantically similar codes. Exact code mappings are retrieved for the specified clinical code from a mapping registry. A rewritten query predicate is generated include the exact code mappings and the semantically similar clinical code mappings. The rewritten query is executed against a clinical data store to retrieve results matching the exact and / or semantically similar codes. The results are annotated to distinguish between exact and semantic matches.