20 results about "Poincare sphere" patented technology

The invention discloses a broadband full-Stokes polarization imaging device based on non-orthogonal four-channel polarization beam splitting, which comprises a dielectric substrate, a non-orthogonal polarization beam splitting metasurface and a focal plane detection array, and is characterized in that the non-orthogonal polarization beam splitting metasurface is composed of a plurality of sub-wavelength unit structures; the sub-wavelength unit structure is a single-layer anisotropic medium nano-column array and is configured to perform intrinsic-loss-free non-orthogonal four-channel beam splitting on incident light and form four spatially separated focuses on the same focal plane, and polarization states corresponding to the four channels form a regular tetrahedron measurement base on a Poincare sphere; therefore, full Stokes parameter measurement is completed by four pixels; the sub-wavelength unit structure forms a super-pixel unit, and the super-pixel unit is aligned and integrated with a focal plane detection array in a two-dimensional array form to form a compact polarization camera facing focal plane segmentation. The device is simple in structure, easy to integrate with the CMOS front end, small in device size, low in power consumption and high in mass production feasibility.

The invention provides a cross-domain data classification method and system based on hyperbolic manifold alignment, and the method is executed by a computer, and comprises the steps: respectively mapping the hierarchical data of a source domain and a target domain into hyperbolic vector sets through a Poincare sphere model; training a Mobius transformation mapping function based on the common node pair, and establishing a cross-domain alignment relationship; in response to a to-be-registered node of a target domain, generating an initial mapping vector, and constructing a hyperbolic implication cone taking an upper node of the initial mapping vector as an axis; if the initial vector exceeds the cone range, correcting the initial vector to the cone boundary through geodesic boundary projection to obtain a corrected target vector; and finally, calculating a hyperbolic distance between the to-be-identified sample and the target vector set to output a classification result. According to the method, the exponential dilatation characteristic and the logic constraint mechanism of hyperbolic geometry are utilized, the semantic drift problem in small sample cross-domain migration is effectively solved, and the accuracy of hierarchical data classification is improved.

This invention relates to beam manipulation, providing a method and apparatus for generating randomly polarized beams, and pertains to the field of optical manipulation technology. The method includes generating a highly coherent linearly polarized beam using a continuous excitation light source unit; modulating the linearly polarized beam using a phase modulation unit to obtain a vortex beam with arbitrary topological charge; and modulating the vortex beam into a randomly polarized beam using a polarization modulation unit, wherein the response range of the random polarization can cover any point on the surface of a Poincaré sphere. The beam generation method provided by this invention can modulate a vortex beam carrying arbitrary topological charge into a beam with randomly varying polarization states, exhibiting high modulation efficiency, uniform polarization state distribution, and low cost, thus possessing significant practical value.

The present application relates to the technical field of large model, and more particularly to a data set semantic consistency generation and quality stratification method for large model training, which comprises the following steps: step one: performing semantic coding on candidate samples containing synthetic samples and original samples to obtain unit direction vectors, determining radial coordinates according to the semantic sparsity of the nearest neighbor original samples, and dividing radial shell layers in the Poincare ball hyperbolic space; step two: establishing a support cone for the original sample, determining the synthetic samples falling into the inner support cone and anchored with the shell layer as the semantic consistent samples together with the original sample, and determining the quality level according to the radial shell layer number; step three: constructing a core matrix in each quality level to perform a determinant point process greedy selection, and outputting the selected samples and quality level of the data set. The present application suppresses the semantic drift and redundancy of synthetic samples, and improves the semantic consistency and coverage balance of data.

The invention provides an optical fiber polarization control optimization method and device, computer equipment and a storage medium, and relates to the technical field of optical waveguides. The method comprises the following steps: based on input polarization state distribution under different wavelengths and a Mueller matrix of a polarization maintaining optical fiber, calculating an output polarization state of each wavelength transmitted by the polarization maintaining optical fiber; mapping the output polarization state of each wavelength to a Poincare sphere, and quantitatively evaluating the consistency of the output polarization states according to the spatial distribution characteristics of the output polarization state of each wavelength on the spherical surface of the Poincare sphere to obtain a consistency evaluation function; the optimal optical fiber length is solved through an intelligent optimization algorithm by taking the minimization of a consistency evaluation function as a target, and the minimization of the consistency evaluation function represents the global minimization of the distance of the multi-wavelength output polarization state on the Poincare sphere.

The invention discloses a person and post accurate matching method and system based on multi-modal feature fusion, and the method comprises the following steps: collecting text, behavior and image data of candidates and posts, extracting multi-modal features, and mapping the multi-modal features to a unified embedding space with a learnable curvature; fusion features are embedded into a heterogeneous graph structure containing candidates, posts, skills and behaviors, a Poincare sphere model is adopted to perform hierarchical expression on node embedding, and a system performs weighted integration on associated information such as skills and behaviors based on a hyperedge aggregation mechanism constructed by geodesic distances, so that updated candidate and post representations are generated, and the candidate and post representations are updated. And the matching scoring module calculates the matching degree and outputs a matching result, so that the man-post matching accuracy and the hierarchical expression capability are effectively improved. According to the invention, various information features can be integrated, and the accuracy and adaptability of man-post matching are improved.

This invention relates to a method for preparing vector Hermitian-Laguerre Gaussian beams based on collinear interference. Combining the relationship between the Poincaré sphere and the Jones matrix, and based on the vortex light field generated by the complete solutions of two sets of Helmholtz equations, a method for preparing vector Hermitian-Laguerre Gaussian beams based on collinear interference is proposed. Unlike the commonly used single-mode vector light field preparation, this method, based on a phase-type spatial light modulator, a polarization beam splitter, and a half-wave plate, realizes the preparation of a dual-mode vector Hermitian-Laguerre Gaussian beam, reconstructs the polarization distribution on its intensity cross section, and studies the influence of mode difference on the polarization distribution with transmission distance. Compared with traditional single-mode vector light fields, this method achieves good results in preparing vector light fields with different modes. This method belongs to the category of light field manipulation and can be applied to the design of optical polarization elements.

The present application relates to the field of bioinformatics, and proposes a drug target correlation identification method based on spatial atlas and isometric network. Firstly, the three-dimensional spatial geometric atlas of target protein and the microcosmic chemical topological graph of drug molecule are deduced and constructed respectively, so as to losslessly reserve the physical conformation of receptor and the local pharmacophore of small molecule. Secondly, the three-dimensional isometric graph neural network and the graph message passing mechanism are jointly used for deep feature representation. Finally, the affinity measurement strategy based on hyperbolic space is introduced, the features are losslessly projected into the Poincare ball model with constant negative curvature, and the correlation probability between drugs and targets is predicted by using the non-Euclidean geometric distance. The present application can effectively overcome the feature congestion problem of traditional space, fully excavate the implicit hierarchical information inside complex biological molecules, significantly improve the accuracy and generalization robustness of drug target correlation analysis, and provide effective technical support for new drug research and development, drug repositioning and precise medication.

The invention provides a power plant equipment multi-modal fault causal chain retrieval and self-evolution method and system, and the method comprises the steps: obtaining equipment multi-modal heterogeneous data, extracting features through coding and projection, and outputting potential physical states and static semantic features; constructing a hybrid neurodynamic equation, and describing a physical quantity continuous flow by using a hypergraph convolution differential equation with an incidence matrix; driving hypergraph evolution based on formalized Ricci curvature, calculating a Wasserstein distance of a node pair by adopting Sinkhorn entropy regularization, and updating a hyperedge weight through a curvature flow equation; designing a physical topology double constraint loss function; variational neural symbol reasoning is introduced, and gradient propagation of discrete logic variables is achieved through GumbelSoftmax; aligning the physical state with the query vector by utilizing Poincare sphere index mapping in the hyperbolic retrieval space; an invariant causal prediction framework is adopted to carry out anti-fact intervention verification, knowledge base self-evolution is realized based on feedback, and fault causal chain retrieval which is physically consistent, feasible in calculation and deeply fused with multi-modal information is realized.

This invention discloses a method and system for personalized task recommendation for children's eye care, relating to the field of artificial intelligence recommendation technology. The method includes: acquiring children's behavioral data, preprocessing and splicing it to generate daily behavior vectors; normalizing the daily behavior vectors by setting boundaries to obtain normalized vectors; constructing a task library and generating tag vectors; constructing a Poincaré sphere space; mapping the normalized vectors and tag vectors to the Poincaré sphere space; calculating the vector matching degree; and generating a task recommendation list based on the matching degree for sending. Based on the recommendation list, task feedback is statistically analyzed; a dual loss function is constructed based on the task feedback; and a total loss function is constructed based on the dual loss function for iterative optimization to generate a final recommendation list as the basis for personalized task recommendation. This invention effectively improves the scientific rigor and feasibility of eye care task recommendation.

The embodiment of the invention provides a hydrogen detection system and method and a storage medium, and belongs to the technical field of gas detection. The system comprises a pump laser, a Stokes laser, a polarization regulation and control module, an optical path coupling module, a distribution module, a detection module and a processing module. The polarization regulation and control module modulates the polarization state of first laser emitted by the pump laser, and the optical path coupling module couples the first laser modulated by the polarization regulation and control module and second laser emitted by the Stokes laser into third laser. The detection module is used for introducing hydrogen to be detected to generate stimulated Raman scattering with the third laser and outputting a current measurement signal; and the processing module is used for determining a current hydrogen concentration value according to the current measurement signal. A polarization regulation and control module is introduced into a pump light path, so that a polarization state quickly traverses a Poincare sphere within integral time, Raman gain equivalent to polarization average is obtained, and polarization state disturbance is eliminated. According to the invention, high-sensitivity and high-stability distributed trace hydrogen monitoring is realized.

The invention relates to a large model data grading method and device. The method comprises the following steps: acquiring to-be-graded source data and a hierarchical tree; embedding the hierarchical tree into a Poincare sphere model in the hyperbolic space to obtain a label embedding vector; based on a retriever, retrieving the to-be-graded source data to obtain a context; utilizing to-be-graded source data, a context and a label embedding vector to finely adjust the large model to obtain a hierarchical perception model; and based on the hierarchical perception model, searching according to the to-be-graded source data and the context, and determining that the hierarchical structure of the nearest label embedding vector is the hierarchical structure of the to-be-graded source data. According to the scheme provided by the invention, the nodes of the hierarchical tree and the relationship between the nodes are mapped to the Poincare sphere model, and the hierarchical structure of the hierarchical tree is represented by using the exponential growth characteristic of the Poincare sphere model and the characteristics of low dimension and high fidelity, so that serious distortion after the hierarchical structure of the hierarchical tree is represented is avoided, and the information loss is reduced.

The invention discloses a three-dimensional Gaussian modeling method and system based on Poincare sphere and three-plane representation. The method comprises the steps that a human body image is acquired and preprocessed; inputting randomly generated Gaussian variables into three-plane geometric feature branches to obtain geometric features; inputting a UV mapping graph obtained by fitting when the 3D human body reconstruction network model is trained into a hyperbolic U-net UV feature extraction branch to obtain UV features; and fusing the geometric features and the UV features, inputting the fused geometric features and UV features into a multi-branch decoder to obtain 3D Gaussian rendering core parameters, and synthesizing a 3D human body avatar image based on the 3D Gaussian rendering core parameters. According to the invention, a high-fidelity and real-time 3D human body model can be generated.

The invention discloses a method and a device for regulating and controlling a high-order Poincare spherical beam, and provides a method for effectively regulating and controlling the high-order Poincare spherical beam with complex vector polarization distribution in multiple dimensions such as angular momentum, light intensity distribution, polarization state and the like by using a nonlinear optical crystal based on a nonlinear effect of second harmonic. And different phase matching conditions in the second harmonic generation process are set, so that the regulation and control results of the second harmonic of the high-order Poincare spherical beam are enriched. According to the method and the device, a good effect is achieved during nonlinear regulation and control of the high-order Poincare spherical beam, the structure is simple, the cost is low, regulation and control of multi-dimensional optical properties are achieved, and the method and the device are suitable for design of optical nonlinear polarization imaging elements.

The invention provides a broadband light skyrmion generator, which belongs to the technical field of light field regulation and control, and comprises a spin-orbit coupling function unit which has optical anisotropy in rotational symmetry distribution and is used for modulating the polarization state of incident light so as to generate a light field covering all polarization states of a Poincare sphere; the focusing effect function unit has lens type transmission phase distribution and is used for performing spatial reconstruction on the polarization state of the modulated light field; under the combined action of the spin-orbit coupling function unit and the focusing effect function unit, optical Stokes vector distribution of a focusing light field on a focus cross section forms an optical Skyrmion topological structure. According to the optical skyrmion generator, the optical skyrmion generator with a large available bandwidth is realized, and the application of the optical skyrmion in the field of information transmission is promoted.

This invention relates to the field of bioinformatics and proposes a drug target association identification method based on spatial maps and isovariant networks. First, a three-dimensional spatial geometric map of the target protein and a microscopic chemical topological map of the drug molecule are respectively deduced and constructed to preserve the physical conformation of the receptor and the local pharmacophores of small molecules without loss. Second, a three-dimensional isovariant graph neural network and a graph message passing mechanism are jointly used for deep feature characterization. Finally, an affinity force strategy based on hyperbolic space is introduced to project features without loss onto a Poincaré sphere model with constant negative curvature, and the association probability between the drug and the target is predicted using non-Euclidean geometric distance. This invention effectively overcomes the feature congestion problem of traditional spatial mapping, fully explores the hidden hierarchical information within complex biomolecules, and significantly improves the accuracy and generalization robustness of drug target association analysis, providing effective technical support for new drug development, drug retargeting, and precision medicine.

The invention discloses a PSOCT system based on feedback adjustment circular polarization and an imaging method, and belongs to the field of polarization sensitive optical coherence imaging. A Stokes vector is calculated in real time by analyzing a backscattered light signal on the surface of a sample tissue collected by the system, and the Stokes vector is compared with a target circular polarization state position on a Poincare sphere. And then, the feedback control module dynamically adjusts a fourth polarization controller according to a deviation signal generated by comparison, so that the polarization state change caused by bending or shaking of the sample arm optical fiber is compensated, and the detection light irradiated to the surface of the sample tissue is always kept in a stable circular polarization state. Under the conditions that additional polarization measurement hardware is not added and the system structure is not changed, the problem that the polarization state of a single-state input PSOCT system is unstable is effectively solved, the accuracy and repeatability of birefringence imaging are remarkably improved, and clinical transformation and application are better facilitated.

The application discloses a polarization decoupling vector Mathieu geometric mode beam construction method, and steps are as follows: firstly, a pure amplitude envelope of a scalar Mathieu geometric mode is extracted as a spatial geometric structure, and independent spiral phases are loaded to the spatial geometric structure respectively to generate orthogonal scalar components, so that the coupling limitation of a topological phase and a transverse mode index is completely broken. Then, polarization orthogonal coherent superposition is performed on the components by combining with a computer holographic technology, and physical synthesis of a vector light field is realized. The application realizes perfect combination of high-definition ray-wave geometric trajectories and any low-order polarization state, and generates a beam which can realize accurate mapping on a high-order Poincare sphere. In addition, the beam can maintain high purity of the polarization state under continuous coordinate stretching deformation of an eccentricity adjustment, and the geometric structure remains absolutely unchanged when the polarization is rotated, so that controllability of the vector light field is greatly improved, and the application has important application prospects in the fields of multi-dimensional light manipulation, high-capacity optical communication and the like.

This invention discloses a method and system for full lifecycle evidence preservation of electronic archives based on blockchain and digital twins. The method includes: acquiring multimodal dynamic archive context stream data to construct a full lifecycle spatiotemporal hypergraph; inputting the hypergraph into a bi-track manifold graph convolutional aggregation network to extract semantic manifold features and temporal environment graph features that ignore physical format conversion; performing Möbius fusion in the Poincaré sphere hyperbolic space to generate a digital twin state feature vector; calculating the isomorphic topological difference metric of the hypergraph and adaptively updating the local subtrees of the differential Merkle forest off-chain; and triggering a smart contract to anchor the hash value of the differential subtree root to the blockchain for consensus. This invention achieves dual tamper-proofing and accurate traceability of archive semantics and behavior while ignoring physical format conversion interference, and effectively isolates high-frequency micro-operations from being uploaded to the blockchain through an adaptive anchoring mechanism, significantly reducing blockchain storage costs and performance congestion [Figure 1].

The invention provides a learning behavior pattern mining and trend prediction method based on group big data, and relates to the technical field of artificial intelligence and big data process.The method comprises the steps that firstly, a multi-level heterogeneous graph is constructed, graph convolution feature embedding is conducted through a Poincare sphere model of a hyperbolic geometric space, and a multi-level heterogeneous graph is constructed; the hierarchical structure and long-tail features of knowledge topology are kept losslessly in a low-dimensional space, the problem of Euclidean space feature extrusion is solved, then a multi-factor orthogonal decoupling mechanism is adopted, mutual information minimization constraint is introduced, mixed features are separated into independent capabilities, preferences and social factors, deep attribution of behavior motivations is achieved, and finally, deep attribution of the behavior motivations is achieved. According to the method, anti-fact deduction is performed on the basis of a structured causal model, the influence of endogenous and exogenous variables is distinguished, a trend prediction curve with causal interpretation and a dynamic intervention strategy are output, and the interpretability of a mining result and the scientificity of teaching decision are effectively improved.