19 results about "Cosine Distance" patented technology

The present application belongs to the technical field of tobacco quality evaluation, and discloses a tobacco quality similarity evaluation method and system based on near-infrared spectroscopy, which comprises the following steps: scanning tobacco samples by using a near-infrared spectrometer to obtain spectral data, reducing the dimension of the spectral data by using principal component analysis to obtain the first D principal components and their scores whose cumulative variance contribution rate is greater than 0.99, calculating the cosine distance and Euclidean distance between samples based on the principal component scores corresponding to the target sample set, converting the two-dimensional vector composed of the above two parameters into a one-dimensional scalar by introducing a kernel function and L2 regularization, and finally obtaining the similarity value, so as to realize the digital evaluation of the quality similarity between target samples. The present application provides a new technical idea for tobacco raw material substitution, leaf group formula auxiliary design, and cigarette product quality stability evaluation.

This invention relates to the field of image processing and discloses a method for tracking multiple moving targets with resistance to occlusion interference. The method includes: identifying target pixel clusters in a video frame sequence and determining their occlusion state; extracting residual visible pixel clusters for target pixel clusters in the occluded state, and constructing a local topological map using the geometric relationships between feature pixels to generate a topological feature vector; extracting optical flow vectors from the occlusion edges and converting them into size adjustment parameters for the spatiotemporal trajectory constraint envelope to determine the search area for subsequent video frames; calculating the cosine distance between candidate pixel clusters and the topological feature vector within the search area to complete trajectory association. This invention utilizes local topological invariants to maintain target identity, eliminating the system's dependence on global feature smoothness; and sensing blind zone movement trends through edge optical flow to narrow the search range, suppress identity switching, and ensure continuous and stable trajectory under complex conditions.

This application discloses a method and system for multimodal content source tracing analysis of social media propagation chains, including: collecting and preprocessing multimodal propagation data; encoding visual and semantic modal data through a deep feature extraction network and projecting it into a cross-modal alignment space to generate a fused semantic representation vector; calculating the weighted cosine distance through a modality adaptive attention mechanism to obtain the semantic fidelity attenuation, and combining it with platform encoding degradation estimation to obtain the net semantic attenuation; constructing a propagation dynamic graph based on propagation metadata with the net semantic attenuation as edge weights; aggregating propagation features through an edge-weight-aware graph convolutional network; tracing and locating the source node by performing reverse semantic gradient tracing through a source likelihood scoring network; and achieving specialized source tracing of adversarial mutation propagation through semantically invariant region detection and deep semantic structure correlation mining. This invention achieves deep coupling source tracing of content-level semantic information and propagation topology information.

This invention discloses a privacy-preserving robust federated learning aggregation method to resist backdoor attacks. Based on a dual-server architecture, clustering algorithm, and threshold filtering algorithm, it filters out malicious clients while ensuring the privacy of local updates by federated learning clients. First, this invention uses a dual-server architecture and, through arithmetic sharing and quadratic association pairs, enables the two servers to calculate the cosine distance between local updates of each client without knowing the plaintext of the client's local updates, thus achieving privacy protection for client data. Furthermore, this invention combines clustering operations and threshold filtering, allowing the servers to filter out updates from malicious clients based on both the direction and magnitude of the client's local updates, thereby improving the robustness of federated learning.

A digital human voice and lip synchronization detection method, system, medium and product, in the method, an image frame group and a corresponding audio segment are extracted from a video; visual features are extracted using a visual coding model, context, content and tone features of the audio are extracted through a self-supervised speech model and a language model respectively; a distance matrix is calculated after fusing the audio features; a mapping relationship between the visual and audio features is established based on the minimum cumulative distance path; finally, the correlation weight is calculated through the cross-modal attention mechanism, and whether it is synchronized is judged by the cosine distance. When the cosine distance is less than the threshold value, it is determined to be in a synchronous state. The present application is used to improve the accuracy of digital human voice and lip synchronization detection in complex contexts.

This invention provides a model update method and apparatus, belonging to the field of machine learning technology. The model update method includes: after training a product recognition model using a first training dataset, calculating the uncertainty and typical representativeness of each sample data in the first training dataset, wherein the typical representativeness is the cosine distance between the sample data and a first vector, and the first vector is the standard vector corresponding to the category to which the sample data belongs; selecting a first dataset from the first training dataset based on the uncertainty and typical representativeness of each sample data; obtaining a second training dataset; and retraining the product recognition model using the first dataset and the second training dataset. The technical solution of this invention can reduce the training time for model updates.

The application relates to the technical field of damage detection, in particular to a cross-structure damage identification method based on domain self-adaption and contrast learning, which solves the technical problem that a large amount of manual cost and a complex mathematical model are required for machine learning to detect building damage in the prior art. The cross-structure damage identification method based on domain self-adaption and contrast learning provided by the application performs adversarial domain self-adaption learning on a feature extractor through a discrimination value of a domain discriminator, uses a probability of feature representation and classifier prediction to construct a conditional representation through a multi-linear mapping, and enhances category alignment in the domain. A comparator compares the cosine distance between the sample features of a source domain sample and the sample features of a sample domain sample, so as to divide the source domain sample and the target domain sample into positive sample pairs and negative sample pairs; the model parameters of the feature extractor are modified according to the positive sample pairs and the negative sample pairs, so that the same samples in the cross-domain are pulled close to each other, the different samples are pushed far away from each other, the category clusters are clear, and the misjudgment probability is reduced.

Various embodiments of the disclosure are directed to an Automated electronic Document Identification and Validation (ADIV) system and method. In the ADIV system, received via an API, words and phrases may be extracted from the document content, and Machine Learning (ML) models may be used to classify the document type. The extracted words / phrases are parsed, and key value pairs of data may be extracted. Word embedding vectors (WEVs) may be generated for the extracted words and phrases and for the keywords and key phrases. The cosine distances between the WEVs of the keywords and key phrases the WEVs of the extracted words and phrases are calculated. The key value pairs are identified based on the minimum cosine distance. The key value pairs may be used with one or more rules based models to determine document validity.

The present application relates to a binary function similarity detection method using hypergraph twin neural network, belonging to the field of computer and information science technology. The present application firstly constructs a directed heterogeneous hypergraph containing data flow and control flow hyperedges based on the abstract syntax tree of binary functions; then uses the improved hypergraph twin network to obtain hypergraph embedding vectors, and calculates the cosine distance as the similarity measure of function pairs, effectively learns the structure and semantic features of long-distance statements through multi-layer convolution, uses the initial residual connection and identity mapping to retain the initial features of the statements and avoid the performance degradation caused by the stacking of network layers; finally, the similarity measure is compared with the threshold to determine whether the function pairs are similar. The present application aims at the problem that the existing methods do not generate sufficient information representation of function semantics and structure, and it is difficult to accurately model the mutual influence between long-distance statements in the function, and proposes a binary function similarity detection method using hypergraph twin neural network, which improves the detection accuracy.

The application discloses a kind of based on single camera pedestrian re-identification's no-sense sign-in method and system, the method includes: through single camera acquisition contains personnel front and back image sequence, construct data set with identity, view angle and camera label;Based on ViT architecture, using "self-supervised pre-training+supervised fine-tuning" two-stage training strategy, introduce Student-Teacher twin network and double-branch feature fusion module in pre-training, collaborative optimization global feature comparison and block-level mask reconstruction task;In feature extraction, fusion spatial information enhances embedding, explicitly modeling lens and view angle information;Through generalized mean pooling fusion global and local features, identity matching is carried out using cosine distance, and combined with view angle discrimination generates entry and exit record, finally form structured access log.The application only needs single camera to realize high-precision, high-robustness two-way no-sense sign-in, effectively solve the problems, such as high cost of traditional system, difficult direction discrimination, poor cross-view performance, significantly improve deployment efficiency and practicality.

This invention relates to the field of AI data processing technology and discloses an iterative optimization method for landscape design schemes based on the linkage of two AI models. The method first obtains the pixel matrix, mask, and text feature vector of the current landscape scheme and retrieves historical datasets. Then, a regularization constant is introduced to map the three primary colors of pixels to a light-independent logarithmic chromaticity value matrix. Subsequently, the cosine distance of the unmodified area is extracted based on the mask to obtain the style coherence decay rate. Combining the local area proportion with the Euclidean distance of the text vectors, a weighted local modification semantic divergence is obtained. Using a time decay coefficient, the decay rate and divergence are fused and summed with historical decay to obtain an iterative fluctuation total score. A baseline stability constant is introduced, and its ratio to the mean of historical coupling interference terms is calculated to obtain a global redraw trigger index. Finally, the trigger index and the total score are substituted into a nonlinear variant function to output the computational power allocation ratio of the two models.

This invention discloses an improved DeepSort algorithm for pose tracking and a system, relating to the field of computer vision technology. The steps are as follows: acquiring the image to be detected; using the YOLOX model to perform target detection on the image to be detected to obtain candidate boxes, and then filtering them using a non-maximum suppression algorithm to obtain detection boxes; based on the results of the previous frame, using Kalman filtering to predict the target region of the current frame; calculating the Mahalanobis distance, apparent cosine distance, and pose cosine distance between the detection boxes and the predicted target regions; using the Hungarian algorithm and cascaded matching to compare the similarity of targets between two consecutive frames to perform target matching and obtain the tracking result; and correcting the tracking result using the target's pose information. This invention optimizes the DeepSORT algorithm, resulting in more accurate human tracking and pose information extraction, and solves the problem of poor target tracking performance caused by similar target appearances and inaccurate target regions.

The application claims a kind of adaptive cloud sampling medical image anomaly detection method and system, belong to medical image analysis and intelligent diagnosis technical field. Including: 1) a novel medical image anomaly detection method, the network model of this method is composed of encoder, cloud sampling feature enhancement module, multi-scale self-attention fusion module and decoder.2) to improve the expression ability of medical image abnormal mode in high-level feature space and depict its distribution uncertainty, design cloud sampling enhancement module, combine channel attention mechanism with cloud model uncertainty modeling, through parameterized modeling and random sampling enhancement to high-level semantic features, while maintaining the consistency of semantic structure, improve the expression of abnormal features.3) to perceive multi-scale information and maintain feature consistency, design multi-scale self-attention fusion module, through attention mechanism to the fusion weight of different scale features is explicitly modeled and normalized constraint, to depict the relative importance and competitive relationship of multi-scale features under different samples.4) propose a kind of adaptive global cosine distance loss function, according to the reconstruction consistency of different levels of features dynamically adjusts its loss weight to optimize the model, so as to suppress the noise interference brought by unstable level.5) design a medical image anomaly detection system based on adaptive cloud sampling.

The application discloses a hallucination detection method and system based on semantic drift distribution perception for a large language model, and belongs to the technical field of artificial intelligence and natural language processing. Without accessing internal parameters of the model, accurate and efficient hallucination detection is realized by inputting semantic disturbance and combining an independent semantic encoding model. The method specifically comprises the following steps: receiving an input pair containing an original query and a reply of the large language model, generating a set of disturbed queries; performing vector encoding on the reply and the set of disturbed queries by using a pre-trained semantic encoding model to obtain a reply embedding vector and a sequence of disturbed embedding vectors; calculating the cosine distance between the reply embedding vector and each disturbed embedding vector one by one to form a distance sequence; extracting multi-dimensional statistical features from the distance sequence to form a feature vector; inputting the feature vector into a pre-trained supervised classifier to output a hallucination detection result, and the supervised classifier is trained based on labeled hallucination data.

This invention provides a doctor recommendation method based on demand keyword parsing and multi-dimensional doctor influence assessment. It extracts semantic information such as user symptom entities and modifiers through sliding window segmentation and deep learning models to generate a confidence-based intent representation. It adaptively detects user intent drift using cosine distance variation and outputs the intent evolution strength. It integrates data such as doctor clinical behavior flow and patient evaluations, and uses time-weighted convolution to measure doctor dynamic influence. It dynamically gates and adjusts the doctor influence weights based on intent drift strength, achieving an organic combination of behavior and historical reputation. Combining real-time doctor influence, expertise vectors, and resource supply, it calculates a comprehensive matching score, and then outputs a personalized list of top ten doctor recommendations, generating corresponding demand-focused explanation tags. This invention improves the sensitivity of the recommendation system to changes in user needs and enhances the timeliness and fairness of doctor ranking.

This invention belongs to the field of structural health monitoring technology and discloses a spatial proximity sensing method and system for identifying and associating duct stent damage. The invention constructs a physical proximity graph, transforming topological proximity relationships into numerical constraints; extracts bi-branch features from vibration signals and introduces relative position sensing self-attention; constructs category prototypes in a low-dimensional space, calculates logical values ​​based on cosine distance and radius, enhances open set identification using feature mixing and energy scoring, and introduces proximity constraints to maintain feature space continuity; employs threshold discrimination for known / unknown cases; uses linear Warmup and cosine annealing for joint optimization; and for unknown samples, re-ranks them based on similarity and proximity weights, associating them with adjacent reference regions. This invention uses a threshold discrimination strategy, setting differentiated rejection boundaries for different damage categories to avoid over-rejection or mis-acceptance caused by a uniform threshold, thus improving the ability to reject unknown damage and the overall diagnostic reliability.

This invention discloses a rapid identification method for unknown radio signals based on a large electromagnetic spectrum model, relating to the field of electromagnetic spectrum monitoring technology. The method involves acquiring clean spectrum samples for self-supervised mask modeling pre-training to construct an electromagnetic spectrum spatiotemporal cognitive model capable of predicting expected spectral patterns based on context. Real-time frequency sweep data is input into the model to obtain the expected spectrum map, and its similarity difference with the actual spectrum map is calculated. Cosine distance is calculated as the feature difference by extracting features from the intermediate layers of the model, and the two are fused to obtain a cognitive difference matrix, which is then mapped to a two-dimensional surprise map. Spatiotemporal correlation tracking is performed on multiple consecutive frames of surprise maps to form surprise trajectories. The type of credible unknown signal is determined based on the frequency of occurrence, the mean surprise level, and frequency stability conditions. Original spectrum segments from the signal occurrence region are collected as positive samples, and the model parameters are incrementally fine-tuned using a low-rank adaptation matrix. This invention enables the discovery of unknown radio signals without relying on prior knowledge.

The application relates to the technical field of speech processing, and particularly discloses a PTE oral language question and answer evaluation method and system based on intelligent speech, which comprises the following steps: acquiring an original speech signal and extracting an acoustic feature sequence; calculating an instantaneous speech speed through phase space reconstruction and a recursive rate matrix, generating a time length regularization factor when the value exceeds a threshold; determining a fractional Fourier transform order according to the factor, calculating an instantaneous frequency deviation weighted cosine distance matrix, and dynamically programming backtracking to obtain a regularized feature sequence; performing multiple fractal detrended fluctuation analysis to locate a singular fluctuation candidate segment, determining an ambiguous phoneme segment in combination with a recursive graph certainty percentage and a laminar flow percentage, and extracting adjacent stable segment phoneme labels as reconstruction conditions; performing phase reconstruction on a clear feature sequence, calculating a phase mutual correlation peak value and a spectral envelope Riemann geodesic distance, obtaining pronunciation confidence, and weighting and summing fluency parameters to obtain an evaluation score; and the application solves the score distortion problem caused by time length mismatch and unclear read-through boundaries under an abnormal speech speed.

A mechanical fault diagnosis method based on multi-scale symbolic dynamic entropy high-density wavelet transform is proposed. This method performs high-density wavelet transforms of different levels on the mechanical fault signal to obtain low-frequency, mid-frequency, and high-frequency components at different levels. At each level, the multi-scale symbolic dynamic entropy values ​​of the low-frequency and mid-frequency components are calculated. The cosine distance between the two multi-scale symbolic dynamic entropy vectors at each level is calculated and used to measure the similarity between the two vectors. The level with the highest similarity between the two entropy vectors is identified, and this level is adaptively determined as the optimal decomposition scale. The mechanical fault signal is then subjected to high-density wavelet transform, and the low-frequency components are reconstructed. The reconstructed signal is analyzed, and the fault type is determined based on fault feature information. This invention effectively solves the problem of adaptively determining the decomposition level based on mechanical fault signal features, has a stronger ability to extract fault feature frequencies, and has broad application prospects in the field of fault diagnosis.