11 results about "Temporal continuity" patented technology

The invention relates to a few-sample video abstraction method based on cross-video time sequence invariance. The method comprises the following steps that a time sequence window attention module enhances video sequence key time features; constructing a time sequence invariance enhancement module alignment support set and abstract semantic features thereof, and generating similar features of the same category; extracting multi-scale features of a query set, a support set, a support abstract and similar features of the same category through a multi-scale feature extraction module, and fusing corresponding features of each scale; through a multi-scale fusion module, performing learning attention weight adaptive fusion on the features corresponding to each scale to obtain fusion features; and fusing the features, outputting frame importance scores through a regression network, selecting representative frames, and generating a video abstract. According to the invention, by combining the time sequence invariance of the cross-video time decay, the time sequence window attention of the inter-frame time decay is enhanced, and the cross-video semantic alignment capability and the time continuity of the abstract in the video are improved.

The invention discloses a brain-level cross-modal perception autonomous behavior generation method and system for a body-equipped agent, and relates to the technical field of body-equipped agents, and the method comprises the following steps: constructing a dynamic weight distribution chain based on the time continuity of a multi-modal input signal of the body-equipped agent, detecting the time difference of arrival of the language signal and the visual signal and generating a weight adjustment instruction; performing delay compensation on the language signal to form a balanced input frame; setting a semantic suppression window to detect the weight change rate in real time, and triggering reduction suppression when the weight change rate exceeds a threshold; generating a perception enhancement band amplification visual feature according to a suppression result; and establishing a time sequence balance list according to the perception enhancement band output, recording weight change and adaptively correcting weight distribution. According to the method, time sequence synchronization and fusion balance of the language signal and the visual signal are achieved through a dynamic weight distribution and semantic suppression mechanism, long-term weight balance is maintained through self-adaptive correction of a perception enhancement and time sequence balance list, and the fusion precision and behavior stability of the intelligent agent with the body are improved.

The invention provides a college student academic activity recommendation method based on a dynamic knowledge graph, and belongs to the technical field of artificial intelligence and knowledge reasoning. A semantic injection mechanism, a relation modulation time evolution unit, time position coding based on a semester period, sequence-level contrast learning and the like are introduced, structural dependency and time continuity of student interest are captured, historical participation sequences of students are modeled through multi-view sequence coding and attention weighted aggregation based on Transformer, and therefore the interests of the students are obtained. And finally, scoring is performed through a multi-layer perceptron and a Sigmoid function, so that intelligent matching and personalized recommendation between the students and academic competitions, scientific research projects or campus activities are realized. According to the scheme, more accurate personalized item recommendation can be realized, and the accuracy and interpretability of student academic activity recommendation are improved.

This invention presents a lightweight physical layer authentication method based on BiLSTM-Linformer. It addresses the decline in model adaptability caused by the complex time-varying nature of CSI data in dynamic environments by proposing a data augmentation method based on temporal interpolation and contrastive learning, a dimensionality reduction method combining PCA and shallow autoencoders, and a lightweight authentication scheme centered on BiLSTM and multi-head self-attention mechanisms. The scheme first uses temporal interpolation to augment CSI data, enhancing temporal continuity, and combines contrastive learning to optimize the model's discriminative ability, improving adaptability to mobile environments. Subsequently, PCA and shallow autoencoders are used for feature dimensionality reduction, reducing computational burden while preserving key CSI patterns and improving authentication efficiency. Finally, BiLSTM is introduced for temporal modeling, and a multi-head self-attention mechanism is combined to optimize temporal feature weight allocation, enabling the model to focus on key changes in CSI data, further improving authentication accuracy and robustness. Experimental results show that the proposed authentication scheme achieves an authentication stability score of 0.95 in dynamic environments, an improvement of approximately 27% compared to existing methods.

The invention belongs to the technical field of digital media processing, and discloses an intelligent video repairing method and system based on spatio-temporal context reasoning. In order to solve the problems that in a large-area or continuous damage scene, the repairing efficiency is insufficient, dynamic inconsistency is caused due to the fact that single-frame repairing ignores time continuity, logic conflicts are caused by semantic deficiency and the like in a traditional method, a space-time multi-dimensional information fusion and reasoning framework is constructed, and a space-time context reasoning generative adversarial network (STCI-GAN) is provided. According to the method, a reference frame is obtained through multi-scale time window sampling, space, time and semantic features are extracted and fused through a semantic enhancement encoder, an initial repair frame is generated by adopting semantic constraint hierarchical repair, and finally, confrontation training optimization is performed through a multi-dimensional discriminator, so that space authenticity, time continuity, motion rationality and semantic compliance are realized. According to the method, the repair efficiency of large-area damage and continuous frame shielding is improved, the dynamic content is ensured to be real and smooth and consistent with semantics, and the method has high robustness and universality.

The embodiment of the invention provides a video processing method and device, a storage medium and a program product. In the embodiment of the invention, the motion amplitude of the to-be-processed part is determined according to the position change of the to-be-processed part needing deformation processing between the deformation areas in the current frame and the historical frame, and the initial deformation area in the current frame is smoothed based on the motion amplitude and the historical frame, so that the time sequence continuity of the deformation areas is enhanced, and the time sequence uniformity of the to-be-processed part is improved. Meanwhile, the initial deformation strength is attenuated, and the deformation action strength is reduced. And although the deformation area is still updated along with the motion, the track of the smoothed deformation area is more stable, and the intensity after attenuation reduces the pixel disturbance amplitude, so that the background distortion is more gentle in change and higher in consistency between adjacent frames, and the background jitter phenomenon in the video can be effectively relieved.

This invention belongs to the field of computer vision and video understanding technology, and discloses a weakly supervised video temporal action localization method based on adaptive temporal continuity. The method includes acquiring an unedited video to be processed, extracting original segment features from the video, and generating an encoded feature sequence using a temporal encoder; performing parallel processing on the encoded feature sequence to generate class-independent action score sequences and class activation sequences, respectively; constructing a temporal continuity label generation module; constructing an instance dependency update module; and supervising the instance dependency update module using temporal continuity pseudo-labels to obtain the action localization result. This invention solves the technical problem in existing technologies where neglecting the temporal continuity of actions leads to severe instance confusion and boundary ambiguity in the model.

The present application relates to a kind of speech content keeping representation learning method based on structural entropy adaptive segmentation and segment level alignment, belong to natural language processing technical field.It includes: in the teacher side based on frame level representation constructs self-similarity graph and carries out clustering by structural entropy minimization, according to time continuity, the clustering result is split into multiple continuous segments, and the prototype of each segment is obtained by pooling;In the student side, the differentiable allocation of frame to segment is realized using attention soft segmentation mechanism, and the student segment representation is obtained;After normalizing the teacher segment prototype and student segment representation, segment level alignment loss is calculated to optimize student model parameters, while updating teacher model parameters using exponential moving average method, to realize stable self-distillation training.By aligning on the segment level unit with stronger linguistic meaning, the present application reduces the sensitivity of frame level alignment to speech rate changes, speaker disturbance and noise disturbance, and improves the language content keeping ability and robustness of speech representation.

This invention relates to the field of image recognition technology, specifically to a short video annotation method and system, comprising the following steps: acquiring frame feature regions, constructing multi-scale mapping, analyzing dynamic levels, matching semantic tags, and generating a temporal annotation sequence. In this invention, by fusing directional difference and color statistics in short video frames, a joint measure of texture and color changes is achieved, improving the accuracy of regional dynamic feature expression. Edge direction histograms and color moment features extracted by multi-scale windows are standardized and screened using cosine similarity to ensure consistency of regional features across multiple scales and reduce interference. The normalized product of centroid displacement and brightness changes constructs a dynamic intensity distribution, enhancing the discriminativeness of motion level division. Semantic matching combines motion parameters and semantic vectors for joint calculation, improving the semantic association and accuracy of tag generation. Temporal reorganization uses trajectory aggregation to achieve temporal continuity of tags, ensuring the uniformity and stability of multi-frame annotation in both spatial and temporal dimensions.

This invention proposes a video-based person behavior recognition method based on temporal feature modeling, comprising: acquiring and preprocessing a continuous video frame sequence, extracting person behavior features from each frame to obtain a single-frame feature vector; constructing a temporal feature representation containing local and global temporal features based on the single-frame features to capture subtle local changes and global evolution trends of behavior; adaptively segmenting behavior into segments based on the magnitude and stability of temporal feature changes to obtain segments with high behavior consistency; extracting segment features and matching them with a pre-built behavior template library to identify behavior types; and finally, performing standardized judgment based on behavior duration, frequency of occurrence, and scene constraints to output normal or abnormal behavior results. This invention effectively solves the problems of missing temporal continuity, high false positive rate, insufficient stability, and limited practicality in existing technologies, significantly improving the accuracy and reliability of behavior recognition.