27 results about "Linear complexity" patented technology

The application provides a method and device for graph data processing, and relates to the technical field of artificial intelligence, and the method comprises: a method for graph data processing, which comprises the following steps: acquiring graph data, extracting the basic structure features of the graph data through a static topology path, adaptively screening the key neighbor features in the graph data based on an attention mechanism and in combination with the basic structure features through a dynamic path, establishing long-distance semantic association between nodes through topology-aware coding based on the basic structure features and the key neighbor features through a linear graph transformer with O ( n ) of linear complexity, and establishing long-distance semantic association between nodes through topology-aware coding; performing frequency domain projection on the node features to generate frequency domain features, explicitly extracting and enhancing the periodic discriminant components in the frequency domain features through a Fourier analysis network to generate graph neural network data. Thus, the graph representation learning capability of the graph neural network is effectively improved.

This invention relates to a dual-branch remote sensing image semantic segmentation method and system based on Mamba and adaptive convolution, belonging to the fields of remote sensing image semantic segmentation and artificial intelligence technology. This method aims to solve the problems of existing technologies, such as difficulty in coordinating global dependencies and local texture information, low segmentation accuracy at multiple scales, poor multi-band adaptability, and low training efficiency. The invention includes using a multispectral adaptive processor for spectral grouping and attention enhancement, inputting the data into a dual-path backbone network after downsampling through a convolutional backbone. The global path uses a Mamba module to achieve linear complexity global modeling, while the local path uses adaptive convolution to dynamically adjust weights to capture details. A multi-head cross-attention fusion module then enables bidirectional interaction of multi-scale features, and a feature pyramid network performs cross-scale optimization and resolution restoration. This invention significantly improves segmentation accuracy, reduces computational burden, and adapts to the real-time application requirements of high-resolution remote sensing images.

The RWKV-based novel translation model inference method and system includes the following steps: 1) Collecting novel translations and extracting parallel corpora, and using dynamic MicroBatch concatenation technology for sequence compression; 2) Introducing a lightweight group query attention mechanism on the basis of the RWKV architecture, directly obtaining KV information from the Embedding layer to build the model; 3) Employing a sublinear complexity hybrid parallel training mode, combined with a global scalar scaling FP16 mixed precision strategy for training; 4) Applying a hierarchical distributed heterogeneous architecture, offloading the optimizer to low-performance devices and performing gradient compression transmission; 5) Outputting the translation using a joint decoder and dynamic batch inference technology. This invention, through the above method and system, effectively reduces the computational complexity and memory usage in the long text translation process, improves model training efficiency and inference throughput, and significantly improves the translation efficiency and contextual coherence of ultra-long texts.

The application discloses a kind of physical information neural network driven ocean island typhoon scene generation method and system, comprising: collecting multi-source heterogeneous meteorological data and carrying out space-time alignment preprocessing;Build coarse scale space-time probability prediction model, utilize graph topology learning network to capture the spatial correlation of meteorological elements, integrate the state space model of linear complexity to efficiently process the long-time sequence dependence of typhoon evolution, and generate coarse resolution probabilistic typhoon scene through multivariate joint distribution probability model;Further build physical downscaling model, input coarse scale prediction result as condition, embed atmospheric fluid mechanics equation as physical hard constraint in loss function, and physically consistent downscaling is carried out to coarse scale scene;Finally output high-resolution typhoon scene with probability reliability and physical authenticity.

The application discloses a kind of unstructured grid flow field prediction method and system based on ordered state space model, comprising: the geometric physical feature extraction of input unstructured grid;Massive node features are aggregated and reduced to a small number of macroscopic physical perception slices by learnable physical slicing mechanism;Innovatively introduce the spatial weighting center ordering mechanism based on flow field principal axis, convert unordered slices into ordered sequence strictly conforming to fluid evolution direction;Global feature interaction and evolution are carried out on ordered sequence using bidirectional selective state space model with linear complexity;Finally, the evolved features are restored to the original grid and the prediction result is output.The application reduces the computing cost significantly while achieving high-precision prediction by explicitly modeling the directionality and causality of fluid evolution, and is suitable for rapid flow field analysis and optimization design in the field of vehicle aerodynamics.

This invention relates to the field of computer technology and discloses a lightweight method and system for processing medical text on resource-constrained terminals. This method enables efficient parsing and structured output of medical text under conditions of limited terminal computing power, memory, and power consumption. The method includes: cleaning, abbreviation restoration, and terminology standardization of the original medical text to construct a standard medical text sequence; inputting the sequence into a lightweight feature extraction model, using depthwise separable convolution and linear complexity attention to extract semantic features; retrieving a medical domain ontology knowledge graph based on key semantic nodes and performing cross-source alignment and fusion to obtain a knowledge-enhanced representation; pruning and quantizing the model, combined with operator fusion and computational graph optimization, to generate an inference model adapted to the terminal hardware; and monitoring the terminal load and power consumption status in real time, dynamically adjusting operating parameters, and outputting structured medical information. The system includes modules for data acquisition and preprocessing, feature extraction, knowledge enhancement, hardware collaborative optimization, resource-aware scheduling, and result output. Through this approach, while ensuring the preservation of medical semantics and consistency verification, the inference overhead on the terminal side can be reduced and the real-time response improved.

This invention discloses a three-dimensional breast ABUS image classification method, electronic device, and computer-readable storage medium based on a tokenized bi-branch selective state-space model. The method preprocesses the three-dimensional ABUS volume data and inputs it into a hierarchical pyramidal classification network. The network constructs local convolutional branches and a global state-space branch in its basic modules: the local branch uses lightweight three-dimensional grouped convolution to extract texture and boundary morphology; the global branch aggregates features into a token map through voxel token generation, then performs multi-axis bi-directional selective state-space scanning and adaptive fusion using routing weights. After detoxing, the core features are injected through a gating mechanism. Finally, the classification result is output through three-dimensional global average pooling and a fully connected layer, enhancing the ability to model long-range dependencies across slices with near-linear complexity.

The present invention discloses a high-efficiency and stable method for simulating inextensible ropes in crane and hoist. The piecewise linear inextensible rope is usually represented by the Cartesian coordinates of the vertices and the quaternions on the segments, which uses too many degrees of freedom and needs many constraints, bringing unnecessary numerical difficulties and computational burden to the simulation. The present invention proposes a compact representation which uses the minimum number of degrees of freedom and naturally satisfies all the additional constraints. Specifically, the rope is regarded as a chain of rigid segments, and its shape is encoded as the Cartesian coordinates of its root vertex and the axis-angle representation of the local coordinate system on each segment. Under the representation of the present invention, the implicit time-stepping matrix has a special non-zero pattern. Using the non-zero pattern, the present invention designs a preconditioner which can solve the related linear equations with approximate linear complexity, and its speed is improved by one to two orders of magnitude compared with the widely used block-diagonal linear PCG solver.

The application discloses an intelligent grating microstructure parameter measurement method and system based on a MaLSTM hybrid model, and belongs to the fields of optical precision measurement, micro-nano manufacturing detection and artificial intelligence application. The method comprises the following steps: collecting a diffraction efficiency spectrum of a grating through a front-end data acquisition module; analyzing the spectrum data by using an intelligent processing center with a built-in MaLSTM hybrid model; the MaLSTM hybrid model extracts long-sequence global features of the spectrum through a linear complexity selective state space mechanism of a Mamba sequence modeling module, and finely captures local dynamic time sequence dependence of the spectrum features through a gating mechanism of an LSTM time sequence dynamic module, and then directly maps and outputs parameters such as duty ratio and groove depth of the grating by using a full-connection regression module; and finally, the real-time presentation of the measurement result is realized through a back-end visualization module. The application realizes pure data-driven end-to-end measurement, does not need physical model iterative inversion, and has the characteristics of non-destructiveness, high precision and strong robustness.

The application discloses a traffic flow prediction method and system based on CT-SSM and double-path dynamic graph convolution, comprising: constructing a city road network topology graph and constructing a multi-scale periodic historical traffic flow sequence; inputting a multi-period fusion feature tensor into a CT-SSM-based time sequence feature extraction module, performing state recursion after discretization via a zero-order holder based on data-driven selective parameters, and extracting time sequence features; generating a global implicit semantic graph and a local physical evolution graph in parallel based on a space-time attention mechanism and a static mask constraint; performing spatial depth feature aggregation through double-path collaborative dynamic graph convolution; after weighted fusion via a dynamic gating unit, sending the result into a prediction result output module to give a prediction result, and completing model training based on back propagation. The application extracts long-range time sequence dependence with linear complexity, deeply models space-time dependence with a space-time attention mechanism, and breaks the topology solidification through the construction of a double-path dynamic graph, thereby significantly improving the accuracy and robustness of traffic flow prediction.

This invention discloses a method for predicting the operational status of a network system based on the Mamba architecture. The method includes: acquiring raw time-series data from a data acquisition device, the data containing observations reflecting the operational status of the network system; inputting the data into a front-end feature extraction module to extract local, global, and bidirectional temporal features, generating enhanced temporal features; fusing the raw data and enhanced features to construct a joint representation and inputting it into a backbone network; the backbone network employs a temporal processing unit based on the Mamba state-space model to capture long-short-term temporal dependencies with linear complexity, outputting temporal prediction features; and generating a target time prediction value based on the features to indicate the future operational status of the network system. This invention improves both the accuracy and computational efficiency of network system operational status prediction by combining front-end multi-scale feature extraction and feature fusion with the Mamba backbone to capture long-short-term dependencies with linear complexity.

This invention provides a method and system for proactive service migration in vehicle-mounted edge computing, belonging to the field of vehicle communication technology. The invention uses extended Kalman filtering for spatiotemporal alignment and noise reduction of multi-source heterogeneous vehicle data; based on the Mamba architecture, it achieves refined modeling and switching time prediction of long historical trajectories with linear complexity through a selective state-space model and a parallel correlation scanning algorithm; it aggregates network topology features using graph neural networks and generates the optimal migration strategy under multi-dimensional constraints through a near-end policy optimization algorithm; and it uses RDMA technology to inject the KVCache and intermediate tensors of the deep learning task into the shadow container of the target server before the physical link is disconnected, achieving millisecond-level seamless switching. This invention can accurately capture long-distance dependencies while maintaining linear computational complexity, and combined with an intelligent decision-making mechanism that can perceive the global topology, it achieves seamless proactive migration.

This invention provides a driver gaze direction estimation method based on structural inductive bias, relating to the technical field of driving safety. The method includes: extracting multi-scale features from face images using a backbone network; introducing an adaptive directional compression attention enhancement module, while simultaneously enhancing local information using convolutional auxiliary kernels; introducing an ordered multi-scale aggregator module, achieving directed cross-scale information fusion through sequential recursion and gating mechanisms, followed by mean pooling along the sequence length dimension to obtain a global representation; employing LayerNorm for sample-level normalization at the regression head entry point to ensure stable input distribution under mini-batch and strong domain bias conditions; and using a small perceptron consisting of two fully connected layers and GELU to map yaw and pitch angles, configuring isomorphic auxiliary heads for each scale to achieve gaze estimation. This invention's method has linear complexity and does not rely on global aggregation tokens or explicit positional encoding.

The application discloses a double-flow fault diagnosis method based on state space modeling and dynamic edge context perception, which constructs a time domain state flow and a time-frequency edge flow, and the specific implementation process is as follows: the time domain state flow is modeled in a Transform framework to form a MambaT encoder for efficiently capturing long and short term dependencies with linear complexity and accurately representing time domain sensitive features; the time-frequency edge flow transforms vibration data into a two-dimensional time-frequency image through wavelet transform, designs a differentiable edge sensitive mask to adaptively separate impact features and background noise of the time-frequency image, simultaneously designs a boundary content perception adaptive padding to maintain the topological integrity of the image boundary, and cooperatively constitutes a dynamic edge context perception convolution to effectively extract time-frequency edge features; a channel enhancement and channel gating module is designed to establish a dynamic channel interaction gating mechanism, realize time domain-time frequency domain double-flow cascaded representation, and significantly improve fault discrimination ability. Experiments prove that the method has higher fault diagnosis performance.

This invention discloses a face attribute recognition method based on a state-space model, belonging to the fields of computer vision and pattern recognition technology. This invention aims to solve the problem that existing methods struggle to simultaneously achieve fine-grained local feature extraction and global context modeling, and suffer from high computational complexity. The method includes: acquiring and preprocessing a face image dataset; constructing a multi-task learning network, which includes a ConvNeXt module for extracting fine-grained local features, a feature fusion module, a state-space model-based feature enhancement module, and a classification module; jointly training the network using the dataset; and inputting the face image to be recognized into the trained network to obtain the recognition result. This invention combines the efficient local feature extraction capability of ConvNeXt with the linear complexity modeling capability of the state-space model for long-distance dependencies, reducing model complexity while maintaining recognition accuracy, and achieving accurate and efficient classification of face attributes and expressions.

The application discloses a SLiMs prediction method based on linear complexity sparse attention and cross-modal fusion. First, semantic embedding features, physicochemical features and evolutionary information features of a protein sequence are extracted; then, a SLiMs prediction model is constructed, including a linear complexity sparse feature enhancement module, a cross-modal multi-head attention fusion module and a classifier; the linear complexity sparse feature enhancement module is used for respectively enhancing the physicochemical features and the evolutionary information features to obtain enhanced physicochemical features and enhanced evolutionary information features; the cross-modal multi-head attention fusion module is used for fusing the semantic embedding features, the enhanced physicochemical features and the enhanced evolutionary information features to obtain multi-modal deep fusion features; the multi-modal deep fusion features pass through the classifier to output a prediction result; finally, the SLiMs prediction model is trained. The method makes the features of SLiMs more prominent, improves the feature expression capability of SLiMs, and avoids the features of SLiMs from being submerged by surrounding non-motif regions.

This invention proposes a multi-hunter-single escape robustness reconstruction and control method and system based on Set Transformer-HAPPO. The method and system construct a three-dimensional multi-agent arrival-avoidance differential game model with dynamic attrition and replenishment mechanisms. A two-layer network model consisting of a centralized evaluation network and multiple distributed decision networks is designed. The centralized evaluation network introduces Set Transformer to handle the variable number of agent inputs, successfully reducing the computational complexity of traditional attention mechanisms, which increases quadratically with the number of agents, to linear complexity. This makes centralized training of large-scale agent clusters possible with limited computing resources. The distributed decision networks employ a fixed k-neighbor observation mechanism to maintain a constant input dimension, significantly reducing the deployment cost of traditional methods, which increases linearly with the number of application scenarios, to a constant cost requiring only one training iteration. Simultaneously, a two-layer robustness mechanism is formed through adversarial failure training and active mask filtering and dynamic topology reconstruction during the deployment phase, effectively improving the system's failure robustness. The method and system proposed in this invention effectively solve the shortcomings of existing technologies in terms of dynamic scale adaptation, resistance to disabling attacks, cross-scale migration and computational efficiency. It has the advantages of adaptive training scale, strong failure robustness and efficient training deployment, and can achieve stable collaborative capture of a single escapee by multiple pursuers in adversarial environments.

This invention relates to the field of remote sensing image processing and multi-source image fusion, and discloses a high-fidelity remote sensing image fusion method based on PAN-RWKV. The method takes panchromatic (PAN) and multispectral (MS) images as inputs, and includes: upsampling the MS images and aligning them with the PAN; performing convolutional coding and multi-scale feature extraction in the panchromatic and multispectral branches respectively; introducing a panchromatic sharpening receptive gated weighted key (PRWKV) module to model long-range spatial dependencies with linear complexity and perform dynamic channel-gated modulation; employing a Hybrid Context Space-Spectral Attention Fusion (HCSAF) module in the fusion branch to achieve cross-modal adaptive information interaction and selective fusion; and finally fusing global consistency and local details through a Global-Local Attention Weighted Reconstruction (GLAWR) module to output a high-resolution multispectral (HRMS) image. Compared to global attention structures that require explicit construction of attention matrices, this invention significantly reduces computational and storage overhead while maintaining high spatial detail and spectral consistency, making it suitable for efficient panchromatic sharpening processing of high-resolution remote sensing images.

The application discloses a medical image segmentation method based on GPCS-TransUNet, and belongs to the medical image segmentation field.The method improves the TransUNet model, introduces a global perception gate aggregation attention in the encoder stage, enhances the interaction ability between different features, and effectively improves the shortcomings of the model when processing multi-scale feature interaction.In addition, a simplified linear attention module is introduced before the feature enters the Transformer module, which performs linear complexity context interaction and semantic reorganization on the feature, so that the feature has more stable semantic expression.Finally, in the decoder stage, a lightweight channel and spatial attention module is designed, which jointly models the channel and spatial attention, adaptively recalibrates the feature, strengthens the key semantic information and suppresses the redundant response, thereby improving the accuracy of the feature in semantic expression.In summary, the GPCS-TransUNet improves the accuracy and feature expression ability of medical image segmentation through multi-module collaborative optimization.

The application discloses a light field image compression method based on a multi-view representation and a state space model, and belongs to the field of image processing and multimedia technologies. First, the application decomposes a light field image into four view branches based on the multi-dimensional characteristics of the light field image, independently extracts features of the view branches, and fully mines structural information of the light field image. Meanwhile, a high-dimensional light field image data is decomposed into multiple low-dimensional view data for processing in a view-by-view manner, so that the overall calculation cost of the model is reduced. In order to improve the expression ability and modeling ability of features of each view, a multi-scale Mamba module is designed. The module can extract global information of different branches with linear complexity, and obtain more compact feature representation. Finally, the method comprehensively considers the importance of different view features, establishes adaptive fusion multi-view features, and strengthens the expression ability of depth features to key information of the light field image. Finally, efficient and high-quality compression of the light field image is realized.

This invention discloses a method and system for real-time perception of microscopic traffic conditions using UAV edge computing, belonging to the fields of intelligent transportation and deep learning. The method first cleans and discretizes UAV vehicle trajectory data, constructs a spatiotemporal feature tensor, and extracts vehicle physical component features. Then, it establishes a dual complementary structure of a static physical graph and a component-perceived dynamic semantic graph. It employs an M-STGCN Block encoder with a "temporal-spatial-temporal" architecture, using Mamba instead of a Transformer to achieve linear complexity computation and parallel fusion of physical and semantic information. Finally, the decoder outputs the probability distribution of future multi-step microscopic traffic states, identifying five dynamic phases: free flow, shock waves, and congestion. This invention explicitly models the heterogeneity of vehicle components, resulting in lightweight computation, low memory usage, and fast response. It can achieve millisecond-level real-time perception on UAV-borne edge devices, improving the accuracy of mixed traffic flow state prediction and shock wave early warning capabilities.

The application relates to the technical field of safe disposal, and provides a terrorist attack space-time prediction method and system based on a dynamic graph and a multi-scale excitation effect. Key information is extracted from multi-source threat data to construct a "organization-place-event-time" dynamic space-time graph. A plurality of excitation kernel functions are introduced to explicitly distinguish complex time patterns. Meanwhile, an anti-excitation term is introduced to dynamically simulate the suppression effect of external intervention. In the information aggregation stage, a linear complexity space-time self-attention network is used to realize efficient and interpretable information fusion through joint coding of time difference and spatial distance. The method realizes the leap from "macro probability" to "dynamic path" prediction, solves the problems that a traditional model is difficult to depict chain reactions of events, ignores suppression factors and lacks interpretability, and significantly improves the prediction accuracy and real-time performance, thereby providing scientific decision support for public safety early warning.

The application discloses a solid particle heat storage online dedusting and anti-caking control system and method, which comprises a multi-modal ultra-fast encoder based on a state space model, which realizes O(n) linear complexity ultra-long sequence processing; a physical information causal graph model fused with a convection diffusion partial differential equation constraint, which limits a physically feasible region and eliminates false correlation under extreme working conditions; a diffusion generative control strategy guided by a causal Q function, which realizes zero-sample high-dimensional continuous action generation; and a neural-symbolic reasoning engine fused with a physical rule knowledge base, which supports meta-learning adaptive safe truncation. The application can automatically identify abnormal patterns and generate a diagnosis report, reducing the dependence on experts; surpasses traditional correlation learning, realizes counterfactual reasoning and optimal intervention strategy search; the neural-symbolal reasoning engine generates a readable decision chain, meeting the requirements of industrial safety audit, enhancing user trust and realizing the transformation from "fault response" to "predictive maintenance".

The present application relates to the technical field of remote sensing image processing, and specifically discloses a remote sensing image segmentation method, system and product based on a light-weight UMFormer. The present application is based on the light-weight and moderately accurate UMFormer, adopts a two-stage optimization strategy of first improving segmentation accuracy and then balancing light-weight efficiency, and realizes high-precision and light-weight segmentation of remote sensing images. The LEUMFormer model constructed adopts a DecoupleNet light-weight backbone network, suppresses channel redundancy, and realizes model light-weight on the basis of precision improvement. Through the MSAF module, linear complexity is realized to enhance and extract farmland multi-scale features and direction-sensitive structures. In the decoding stage, the SCIA module is introduced, the spatial structure and spectral channel features are dynamically fused through the double-branch parallel structure and adaptive gating. In addition, the BEH module is constructed to explicitly strengthen the subtle boundary at a very low computational cost, solving the problems of segmentation blur and fracture.

The application discloses an automatic driving perception method and system based on dynamic neural operators and physical evolution, and relates to the technical field of automatic driving. Specifically, the method comprises the following steps: extracting a multi-scale initial feature map of an environment image; mapping the initial feature map to a frequency domain to obtain a frequency domain feature, modulating the frequency domain feature by using a physical attenuation mask and a dynamic modulation parameter to simulate the evolution process of the feature in a continuous physical field, and restoring the feature to a spatial domain to obtain a local physical evolution feature; using a linear complexity attention mechanism to capture cross-region dependencies to generate global semantic features; based on the scene complexity difference, performing spatial-level weighted fusion on the local physical evolution feature and the global semantic feature to obtain fused perception features; and outputting a perception result by using the fused perception features. The application aims to introduce physical equation constraints to improve the perception robustness, and to realize global collaboration at a low computational cost, which is suitable for real-time environment perception of a vehicle-mounted edge computing platform.

This invention relates to a method for predicting ozone concentration in complex environments by integrating selective state-space and multi-scale frequency domain features, belonging to the field of air pollution prediction and intelligent environmental monitoring technology. Addressing the problems of high computational cost of traditional chemical transport models, insufficient capture of long-sequence dependencies by machine learning models, and inadequate multi-scale feature extraction, this invention constructs a multi-branch heterogeneous architecture, integrates a selective state-space model to achieve linear complexity long-time series modeling, combines fast Fourier transform frequency domain analysis to capture periodic patterns, employs a wavelet Kolmogorov-Arnold network to characterize nonlinear threshold effects, and utilizes dynamically fused gated adaptive weighted multi-branch outputs. In a field study in a basin, this method reduced the root mean square error of ozone concentration prediction to 11.1 μg·m³. ‑3 The coefficient of determination was improved to 0.912, significantly optimizing prediction accuracy, computational efficiency and model interpretability.

A method, apparatus, device, and readable storage medium for modeling complex linear structures are disclosed. The modeling method includes: dividing the axis of the complex linear structure into segments based on its linear complexity, making each segment approximately a straight line; establishing a rectangular coordinate system with one end of the axis as the origin, the tangent to the axis as the x-axis, and the normal as the y-axis; unfolding each segment axis along the x-axis direction into a straight line; determining the coordinates of the segment points after unfolding; establishing a straight line solid model based on these coordinates; extracting the coordinates of all nodes in the solid model; determining a coordinate transformation equation based on the coordinates of the segment points before and after unfolding; using this equation to convert all node coordinates of the solid model back to their original coordinates; and constructing the complex linear structure model based on these original coordinates. This significantly simplifies the modeling process and improves modeling efficiency.