26results about How to "Excellent accuracy" patented technology

The invention belongs to the technical field of multi-rotor unmanned aerial vehicle control, and particularly relates to a multi-rotor unmanned aerial vehicle self-adaptive feedforward control method based on a mesoscopic wind field model. Comprising the following steps: firstly, based on a relaxation time model of a lattice Boltzmann method, establishing a mapping relation between a Knudsen number and turbulence intensity and thermal noise intensity of a macroscopic wind field, and realizing parameterized representation of a mesoscopic wind field; then, constructing a mesoscopic wind field model by synthesizing a real-time wind field containing average wind, turbulent flow, gust and thermal noise components; then, calculating the wind resistance according to the predicted wind speed, designing a self-adaptive gain mechanism fusing a real-time tracking error, an error wind direction included angle and a historical error trend, and generating a dynamic feedforward control quantity; and finally, combining the feedforward control quantity with a feedback control quantity based on gravity compensation to form a comprehensive wind resistance control law. Starting from the mesoscale, the disturbance suppression capability and trajectory tracking precision of the unmanned aerial vehicle in complex environments such as strong wind and turbulent flow are enhanced.

The invention relates to an inspection algorithm for a composite machine, and the method comprises the steps: obtaining the operation parameters of a function cooperation assembly of the composite machine, eliminating abnormal data through a density clustering model, eliminating false anomalies through the combination of assembly cooperation logic, and generating a standardized parameter set through an interval scaling method; constructing a double-branch feature extraction model, mining single-component local features and cross-component coupling features, and fusing through a feature interaction layer to obtain a multi-dimensional feature set; inputting the feature set into a dual-module detection model, optimizing a random forest to output an initial anomaly probability, performing correction in combination with a working condition and a historical fault rule, performing secondary verification through a component collaboration rule base, and outputting an anomaly parameter type, a time sequence node and a severity degree; abnormal core parameters and grades are positioned, an abnormal conduction path map is generated, a component physical connection relation and a historical maintenance case are matched, and an abnormal basic cause is locked; according to the invention, data reliability and identification accuracy are improved, efficient fault tracing is realized, and maintenance shutdown loss is reduced.

The application provides a vectorization reverse construction method of an existing building indoor structure scene, and relates to the field of data processing. In the method, laser point cloud data and panoramic image sequences are synchronously acquired, and building structure semantics are extracted by image semantic segmentation. The image semantics are robustly migrated to the point cloud to form structure semantic point clouds. On the basis of eliminating non-building structure elements, a structure point cloud skeleton is constructed, and virtual placeholder point clouds are generated for open structures such as doorways and window openings. Then, continuous structure geometry is recovered by combining voxelization blocking completion and a three-dimensional convolution completion network. Finally, a three-dimensional structure grid model binding structure semantics is generated, and a two-dimensional structure vector drawing meeting engineering drawing rules is obtained by projection vectorization. The technical solution provided by the application facilitates improving the accuracy of the vectorization reverse construction of the existing building indoor structure scene.

The invention discloses a novel information system reliability determination method for a multi-input dependent starting scene, and belongs to the technical field of system reliability. Comprising the following steps: drawing a system structure chart containing a multi-input dependency starting module, and dividing a three-stage system structure of a core module-a support signal module-an auxiliary guarantee module; a GO graph is drawn, the multi-input dependent starting module is uniquely identified as a type 29 operator, and the number of input ports of the operator is in one-to-one correspondence with the number of support signals of the corresponding core module; and determining the signal effective probability, the correlation coefficient and the element reliability according to a standardized process, calculating the module reliability by combining a type 29 operator quantitative model, and finally determining the system reliability. According to the method, the problem that an existing GO method operator cannot adapt to a multi-input dependent starting scene is solved, the blank of a special tool for GO method modeling in the scene is filled up, and the application range of the GO method in a novel information system is expanded.

This invention relates to the technical fields of intelligent diagnostic equipment for traditional Chinese medicine (TCM), medical IoT, and equipment and systems for internet-based TCM hospitals. Specifically, it relates to an integrated terminal and system for the four diagnostic methods (inspection, auscultation, inquiry, and palpation), incorporating a built-in TCM health model, wireless networking capabilities, and access to an internet-based TCM hospital platform for online registration, AI-based diagnosis, physician review, and electronic prescription processing. It aims to address the problems of single-function diagnostic equipment, data disconnect, and lack of a closed-loop diagnosis and treatment system. The terminal includes an integrated shell and integrated multimodal sign acquisition, edge computing core, interactive touch display, and secure communication gateway module. The acquisition module consists of inspection, auscultation, inquiry, and palpation units, synchronously acquiring tongue appearance, voice, chief complaint, and pulse wave features under a unified clock. The edge computing core incorporates a built-in TCM health model, deeply fusing multimodal features through a cross-modal attention mechanism.

The invention provides a method for detecting the content of fluticasone propionate cream. The method comprises the following steps: preparing an internal standard substance solution, preparing a test solution, preparing a reference substance solution, detecting and analyzing. According to the method, the cream is dispersed by heating in a water bath, the problem of uneven dissolution of the cream is solved, and the solution is centrifuged and filtered in a low-temperature environment of 6-10 DEG C to remove matrix impurities, so that the treatment efficiency of a cream sample can be improved, and the influence of personnel operation on a detection result can be effectively reduced; the accuracy of a detection result is improved. Besides, heptyl 4-hydroxybenzoate is selected as an internal standard substance, the retention behavior of heptyl 4-hydroxybenzoate is matched with that of fluticasone propionate, the solubility of heptyl 4-hydroxybenzoate in 65% ethanol is consistent, matrix adsorption and extraction errors can be effectively corrected, and the accuracy and reproducibility are remarkably superior to those of an existing method.

The invention discloses a 3D speaking face generation method of emotion controllable VQ-VAE based on hierarchical decoupling. The method specifically comprises the steps of 1, preprocessing a data set; 2, constructing a first-stage model, setting a loss function and training the model; 3, constructing a second-stage model, setting a loss function and training the model; and step 4, constructing a generative model, and generating a 3D speaking face video. According to the method, through the condition decoupling based on VQ-VAE, facial expressions and actions generated by the 3D speaking face generation model based on condition guidance are more accurate and real, and accurate and stable 3D speaking face generation is realized.

This invention discloses a prefabricated structure and construction method for large-area irregularly shaped ceiling aluminum panels, including a base frame, irregularly shaped aluminum panels, and hook-and-loop components. The irregularly shaped aluminum panels are fixed to the base frame via the hook-and-loop components. All components—the base frame, the irregularly shaped aluminum panels, and the hook-and-loop components—are made of aluminum. The beneficial effects of this invention are: Prefabricated installation allows for pre-correction of the shape and unified leveling, resulting in a more regular overall shape and higher fidelity to irregular curved surfaces; prefabricated installation is suitable for large areas, complex irregular shapes, large spans, and high-clearance locations, offering greater adaptability to complex shapes; prefabricated installation provides stable stress, convenient testing, reliable overall hoisting and positioning, and superior node accuracy and structural stability; prefabricated installation significantly reduces high-altitude work time; labor input is reduced by 60%, and temporary work costs are reduced by 40%; improved accuracy solves the problem of connecting irregularly shaped ceiling designs; and increased efficiency: the on-site construction cycle is shortened by more than 60%, eliminating the need for on-site cutting and welding, and reducing conflicts arising from overlapping operations.

Based on the improved grey wolf optimization algorithm, the PMU optimal configuration method for distribution network is proposed. Under the condition of system full observability, the PMU optimal configuration mathematical model is established. Considering the influence of zero injection node and the system observability under single PMU interruption and single line outage, the constraint conditions of the established PMU optimal configuration mathematical model are modified. On the basis of grey wolf optimization algorithm, the population initialization, nonlinear convergence factor and position update weight coefficient are improved. The improved grey wolf optimization algorithm is used to solve the PMU configuration model under different scenarios. The method builds the optimal configuration model under different scenarios to achieve the goal of system full observability, and uses the improved grey wolf optimization algorithm to solve the PMU optimal configuration scheme. Compared with other algorithms, the method has better optimization effect.

This invention discloses an integrated inversion method for cloud macro- and micro-parameters in the interdisciplinary field of atmospheric remote sensing and deep learning. The method includes: inputting multimodal input data into a dual-branch encoder to extract visual multi-scale features and meteorological physical background features; progressively injecting meteorological physical background features into the visual multi-scale features to achieve cross-scale fusion and obtain physically guided decoding features; based on the decoding features, generating spatial gating weights by constructing a cloud area auxiliary map, followed by element-wise gating modulation and feature refinement to obtain spatially focused features; inputting the spatially focused features into a multi-task prediction head to obtain the integrated inversion results of four cloud macro- and micro-parameters: cloud mask probability map, cloud top height, cloud optical thickness, and effective radius of cloud particles. This invention can solve the problems of difficulty in effectively fusing high-resolution satellite visual features with low-resolution atmospheric background fields and insufficient physical consistency caused by independent inversion of multiple cloud macro- and micro-parameters.

The application discloses a kind of space neighborhood reconstruction self-encoding multimodal fusion vector generation method, belong to geographic information science, computer vision and artificial intelligence cross technical field.The present mask self-encoding technique is only stopped at the technical limitation of grid level spatial unit, cannot process the heterogeneous modal vector of geographical entity with independent identity, the application proposes a kind of geographical entity-oriented multimodal mask self-encoding training framework: the multiple heterogeneous modal vectors of geographical entity with independent entity identification are randomly masked, the heterogeneous modal vector information of adjacent entity in the space neighborhood determined by k-ring algorithm is used, the masked modal is reconstructed by multimodal fusion network using the remaining modal vector of the entity not masked;With the weighted combination of reconstruction loss and contrast learning loss, the fusion network parameters are jointly optimized.The application also provides a joint optimization method and a cold start and gradual switching method.

The invention discloses a modulation signal identification method based on convolution attention and multi-dimensional feature fusion, and solves the problems that in the prior art, different signals in the same signal family always show similar features under the condition of a low signal-to-noise ratio, so that a model is difficult to distinguish effectively, the identification accuracy is reduced, and the identification accuracy is low. The modulation signal identification under a low signal-to-noise ratio is realized, the identification precision is higher, and the calculation complexity is lower; the method comprises the following steps: acquiring an original complex signal, normalizing the original complex signal to obtain an IQ sequence, and obtaining an AP sequence through the IQ sequence; then, a pre-trained signal identification network is used for identification; a preprocessing module converts the input IQ and AP sequences into IQ and AP feature maps; the feature extraction module extracts local and global features to form a path feature map; the WAFF module performs weighted fusion on the two paths of features; and finally, the classification head outputs a classification result according to the fusion features.

The invention relates to the field of computer vision and camouflage target segmentation, in particular to a camouflage target segmentation method of a reversible expansion network based on SAM guidance, which comprises the following steps of: firstly, constructing a foreground space priori graph, a background space priori graph and a high-quality SAM pseudo mask by utilizing a segmentation cutting model SAM; the prior redundancy is eliminated through low-dimensional orthogonal subspace projection, and the separability of the foreground and the background is enhanced; pixel-level and gradient-level feature fitting items and SAM subspace priori constraint items are fused to construct an overall objective function, the objective function is expanded into a multi-stage alternating iteration process of a foreground optimization submodule SFOS and a background optimization submodule SBOS, and a foreground feature map and a background feature map are refined step by step; and finally, generating a camouflage target segmentation mask according to the iteratively optimized foreground feature map. Through large model prior guidance, two-stage feature modeling and multi-stage expansion optimization, the integrity and accuracy of camouflage target segmentation are significantly improved.

The invention discloses an air supply device and method for a cathode closed electric pile or a methanol fuel cell. The device comprises a high-speed brushless direct current motor, a centrifugal impeller directly driven by the high-speed brushless direct current motor, a volute for packaging the impeller and an electronic control unit. The impeller and the volute are pneumatically optimized, so that the device has high efficiency and a wide stable working area under the back pressure of 1-10kPa. The electronic control unit takes the load current of the electric pile as a main feed-forward signal, quickly sets the target rotating speed of the motor through a built-in mapping table, and realizes accurate tracking by adopting a rotating speed closed loop. The control unit is integrated with mechanisms such as self-learning compensation, anti-disturbance quick response, safety protection and anti-integral saturation. According to the invention, the harsh requirements of a miniaturized fuel cell system on high back pressure, fast response, small size and long service life of an air supply component are met.

The invention belongs to the technical field of natural gas exploration and development, and discloses a fault control type karst residual hill weathering crust gas reservoir favorable area optimization method and related device.The method comprises the steps that well seismic data in a target area are obtained, marker beds are selected according to the well seismic data for layer position calibration, and the stratum thickness of each layer is counted; according to the horizon calibration and the stratum thickness, an impression method and a trend surface removing method are adopted to recover ancient landform and micro-ancient landform forms, and a target area weathering crust interface top surface structure and a fracture distribution diagram are compiled; and calculating an ancient landform evaluation index, a structure evaluation index and a fault evaluation index at the deployment well point according to the ancient landform, the micro-ancient landform, the weathering crust interface top surface structure and the fracture distribution diagram, solving a comprehensive evaluation index of a favorable region, and screening a favorable target region. According to the method, the relevance and interaction of different control factors are explored, a multi-factor fusion method is provided to solve the favorable area, and the method has better evaluation means and accuracy.

The application discloses a vehicle control method, device, equipment and storage medium, the method comprises: obtaining the current driving data of the vehicle at the current time and the historical working condition data sequence in the historical period; inputting the two to a short-time prediction model to obtain the first working condition data sequence of the vehicle in the first future period; inputting the two and the first working condition data sequence to a long-time prediction model to obtain the second working condition data sequence of the vehicle in the second future period; determining the target working condition data sequence of the vehicle in the second future period based on the first and second working condition data sequences; determining the target control instruction of the vehicle at the future time based on the target working condition data sequence and a preset target function, and controlling the vehicle according to the instruction. Through multi-time scale working condition prediction and fusion, the target control instruction is solved in combination with the preset target function, the problems of insufficient foresight and precision of single time scale prediction are overcome, and vehicle operation resource allocation is optimized.

The invention discloses a night target tracking method and system guided by a feature space position based on physical attribute modeling, and mainly solves the problem of tracking failure caused by poor target feature perception capability and inaccurate positioning in a low-light environment in the prior art. According to the scheme, the method comprises the following steps: acquiring a day-night mixed video sequence from a public data set, and dividing a training data set and a test set; constructing a low-illumination target tracking network model comprising a double-layer topological correlator, a feedback compensation encoder and a prompt generator; inputting a training data set into the model, and carrying out iterative optimization on parameters by using a gradient descent method until training is finished; inputting the test data set into the trained target tracking network model to obtain a classification response diagram and a regression response diagram of the target features; and carrying out post-processing of maximum value search, coordinate mapping and bounding box regression decoding on the classification response diagram and the regression response diagram so as to track the target. The method effectively improves the adaptability and accuracy of target tracking in the night environment, and can be used for man-machine interaction and automatic driving.

A multimodal brain disease diagnostic system based on phenotypic priors and dual-spectral-domain synergistic enhancement. This system belongs to the interdisciplinary field of artificial intelligence and brain science. It addresses the technical problem that existing methods struggle to meet practical needs in terms of diagnostic accuracy and model robustness when dealing with highly heterogeneous multicenter clinical data. The system described in this invention preserves the spectral heterogeneity of rs-fMRI signals to enhance feature expression; constructs a more reliable population graph through diagnostic conflict suppression; aligns cross-modal representations using contrastive learning and gating mechanisms; and captures multi-scale population structure by combining spectral-domain graph filtering, thereby significantly improving the accuracy, robustness, and interpretability of brain disease diagnosis.

The invention discloses a transformer broadband distributed electromagnetic dual model construction method suitable for broadband disturbance analysis. The transformer broadband distributed electromagnetic dual model construction method comprises the following steps of 1) deducing a topological structure of a transformer electromagnetic dual model; 2) performing axial and radial multiple discretization on the iron core and the winding to construct a complete dual model; 3, model parameter identification is carried out, a parameter identification result is input into the complete dual model, and the transformer broadband distributed electromagnetic dual model.The transformer broadband distributed electromagnetic dual model is superior to a traditional broadband model in the aspects of applicable frequency bandwidth and port frequency variation characteristic representation precision.

The application discloses a method for establishing a short-term building cold load prediction model based on a Transformer model, a prediction method and computer equipment, belongs to the technical field of building cold load prediction, and solves the problem of low prediction accuracy of existing building cold load prediction. The method comprises the following steps: selecting feature variables required for building cold load prediction, wherein the feature variables comprise weather feature variables and time feature variables; obtaining linear correlation of each two feature variables, and selecting feature variables linearly independent with each other as input data of a prediction model; establishing a data set with a sequence information mode according to the input data of the prediction model, and generating a sequence input data set according to the data set with the sequence information mode; dividing the sequence input data set into training data and test data, training a model based on a Transformer network; and obtaining the short-term building cold load prediction model based on the Transformer model according to the model based on the Transformer network after training. The application is suitable for predicting short-term building cold load.

The invention relates to the technical field of cell type annotation, in particular to a cell type annotation method based on multi-feature contrast learning, which comprises the following steps: acquiring a gene expression matrix to extract a gene name, extracting a gene base sequence from a transcriptome sequence database based on the gene name, and matching the gene base sequence with a motif sequence, obtaining a gene motif matching matrix; multiplying the gene expression matrix by the gene motif matching matrix to obtain a cell motif matrix; preprocessing the gene expression matrix and the cell motif matrix, and constructing a data set based on the preprocessed gene expression matrix and cell motif matrix; training a double-feature coding model by using the training set and the verification set, wherein the double-feature coding model comprises a gene expression encoder, a motif feature encoder and a classifier; and inputting the test set into the trained double-feature coding model to obtain the annotated cell type. According to the method, the annotation result has biological significance, and downstream function analysis is facilitated.

This invention discloses a neoadjuvant therapy efficacy prediction system and method based on a multimodal large model and breast ultrasound, belonging to the field of medical artificial intelligence technology. The method includes: acquiring pre-treatment ultrasound images of patients and multidimensional heterogeneous text data including past medical history, ultrasound reports, and pathological indicators; unifying the format using serialization technology and inputting it into a pre-trained multimodal large model; utilizing the model's powerful cross-modal sequence modeling capabilities to mine deep implicit associations between images and medical history, and outputting a predicted probability value for pathological complete remission (pCR). The system simultaneously outputs image heatmaps and textual decision attribution prompts. This invention achieves high-precision end-to-end personalized efficacy prediction and possesses strong clinical interpretability.

The invention discloses a children sagittal bone face type AI screening model based on facial features and a method thereof, and belongs to the technical field of artificial intelligence auxiliary screening. A 90-degree side appearance picture is used as a main view angle image, a 45-degree side appearance picture and a front smile picture are used as auxiliary view angle images to construct a main and auxiliary view angle feature fusion model, and through feature extraction, feature fusion and classification, probability distribution of samples in three categories of bony I, bony II and bony III is output. Compared with a single-view-angle classification model, the main and auxiliary view angle feature fusion model has the advantages that the recall rate, the specificity, the F1 score, the accuracy rate and the area under an ROC curve are improved, and by taking the model 7 as an example, the classification performance comparison with orthodontic doctors with different clinical experiences (primary, intermediate and high-grade) on an independent test set shows that the model 7 has the advantages that the accuracy is high; the model 7 is obviously superior to primary and intermediate orthodontic doctors in the aspects of macro average and overall accuracy of various indexes, and most of the indexes exceed those of advanced orthodontic doctors.

This invention discloses a photothermal-triggered reusable drug delivery microneedle patch based on phase change materials, its preparation method, and its applications, relating to the field of pharmaceutical technology. The microneedle patch comprises several microneedles, each containing a drug-loaded matrix, a therapeutic drug, and photothermal nanoparticles. The drug-loaded matrix comprises a phase change material with reversible solidification-melting properties, capable of melting under photothermal triggering conditions and reverting to solidification after the photothermal triggering conditions disappear. The melting point of the phase change material is between 38°C and 45°C. The phase change material includes higher fatty acids and polyethylene glycol derivatives, and the photothermal nanoparticles include gold nanorods, graphene oxide, and polypyrrole nanoparticles. This invention proposes a microneedle patch using near-infrared light as a trigger signal and a phase change material with reversible solidification-melting properties. Through the melting process of the same phase change material under photothermal triggering, reusable drug delivery is achieved, overcoming the technical limitations of existing microneedles that only allow for "single-use, continuous drug release."

This invention discloses a landslide detection method based on uncertainty perception, belonging to the field of intelligent landslide disaster identification integrating remote sensing image processing and deep learning. The method includes: acquiring multi-source remote sensing image data of the study area and corresponding landslide ground truth labels, and constructing a training sample set; constructing a deep learning model, which includes a backbone convolutional neural network, a hollow spatial pyramid pooling module, a segmentation decoding branch, and an uncertainty decoding branch; iteratively training the deep learning model using the training sample set, updating the model parameters by minimizing the total loss function; inputting the remote sensing image data of the area to be detected into the trained deep learning model, and outputting a segmentation prediction map and an uncertainty weight map of the landslide area after model processing. This application significantly improves the accuracy and robustness of landslide identification in complex environments by integrating multi-source data, automatically focusing on difficult samples using the uncertainty branch, and combining multi-scale context capture, while also achieving higher training efficiency.