31results about How to "Reduce storage requirements" patented technology

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for identifying labels for a data set without disclosing the data set to any individual computing system. A method can include receiving, by a first computing system in a multi-party computation (MPC) system, a query including a first share and a second share of a given user profile. The second share is encrypted with a key that prevents the first computing system from accessing the second share. The second share is transmitted to a second computing system in the MPC system. The first computing system and the second computing system generate a machine learning model and identify respective first and second labels. The first computing system receives the second label from the second computing system as a response. The first computing system responds to the query with a response including the first label and the second label.

The present disclosure provides a video coding method and related apparatus. The video coding method includes receiving input data associated with a current block; determining one or more neighboring motion vectors from one or more non-adjacent affine coded neighborhoods of the current block; determining a control point motion vector based on the one or more neighboring motion vectors, wherein if a target neighboring block associated with a target neighboring motion vector is outside of an available area, a derived control point motion vector is generated to replace the target neighboring motion vector; generating an affine merge list or an affine advanced motion vector prediction list, wherein a non-adjacent affine candidate uses motion information to generate a non-adjacent affine predictor according to the control point motion vector; and encoding or decoding the current block using a motion candidate selected from the affine merge list or the affine advanced motion vector prediction list. The video coding method and related apparatus of the present disclosure reduce storage requirements.

The application provides a sparse-to-dense visual positioning method and system based on feature Gaussian splash, and the method comprises the following steps: initializing a color-decoupled feature Gaussian field based on a training image set, optimizing the color-decoupled feature Gaussian field based on a query feature map set in combination with feature rendering and feature alignment loss cyclic optimization, and outputting a compact feature Gaussian scene model; screening a Gaussian landmark set in the compact feature Gaussian scene model by using a matching-oriented sampling strategy; training a scene-specific detector; extracting sparse local features of a query landmark heat map corresponding to a query image and performing sparse feature matching with the Gaussian landmark set to obtain an initial pose of a query perspective camera; based on 3D Gaussian splash, rendering a dense feature map and a depth map of the query perspective in the compact feature Gaussian scene model by using the initial pose of the query perspective camera, performing cluster-based proxy matching-based sparse-to-dense accelerated pose optimization, and obtaining accurate positioning of the query perspective camera.

The utility model discloses a frame-type intelligent circuit breaker, which comprises a circuit breaker main body, the bottom of the circuit breaker main body is movably connected with a frame-type shell, the top of the frame-type shell is provided with two connecting assemblies used for quickly disassembling the circuit breaker main body, and the two sides of the frame-type shell are fixedly connected with a connecting plate respectively. The frame-type intelligent circuit breaker comprises two connecting plates, one end of each connecting plate is fixedly connected with a mounting plate, two sides of each mounting plate are fixedly connected with a first clamping plate, one end of each mounting plate is movably connected with a bolt, and one end of each bolt is in threaded connection with a nut. The problems existing in the aspects of maintenance, disassembly and heat dissipation of an existing circuit breaker are solved, and the working efficiency of the circuit breaker is greatly improved.

This application discloses a method, apparatus, device, and storage medium for model search. The method includes: acquiring a search space, which includes a convolutional module and a visual transformation module; wherein the visual transformation module includes a self-attention sub-module for executing a self-attention algorithm with a complexity of O(n); preprocessing labeled training images using the convolutional module; performing target task-based image processing on the preprocessed results using the visual transformation module; determining a first gradient of the search space based on a first loss between the image processing results and the labels of the training images; updating the search space based on the first gradient to obtain a target search space; wherein the target search space is used to search for a target model, and the target model is used to perform target task-based image processing on the target image to be processed. This allows the searched model to have linear complexity, effectively solving the problem of difficult model deployment.

The utility model provides a detonation excitation phase change rock breaking device which comprises a liquid storage body arranged in a blast hole, the liquid storage body is expandable, an overburden layer is arranged above the liquid storage body in the blast hole, an absorber and an excitation device are arranged in the liquid storage body, an electronic detonator and an emulsion explosive are arranged in the excitation device, and the electronic detonator is provided with a detonating cord extending out of the blast hole. A liquid conveying pipe and an exhaust pipe are further arranged, the lower end of the liquid conveying pipe is inserted into the lower end of the interior of the liquid storage body, and the lower end of the exhaust pipe is communicated with the upper end of the interior of the liquid storage body. The problem that the blasting effect is poor due to the fact that liquefied air rock breaking is low in detonation time precision and insufficient in liquefied air gasification is solved.

This application provides a real-time hyperspectral local anomaly detection method and device based on multi-line, multi-band recursive updates. The method includes: acquiring new data blocks for a target scene in real time using a filter array spectral imaging system; updating the current global mean vector and the multi-line, multi-band correlation matrix used for background modeling for the target scene using a recursive update formula to obtain the updated global mean vector and the updated multi-line, multi-band correlation matrix; acquiring the local spatial neighborhood statistics of the new data blocks; and performing local anomaly detection on the new data blocks based on the local spatial neighborhood statistics to obtain the current anomaly detection result data for the target scene. This application can realize local anomaly detection for multi-line, multi-band hyperspectral data and can solve the problems of high computational cost, large storage requirements, and insufficient real-time performance of traditional global covariance matrix calculations in complex industrial backgrounds.

This invention relates to the field of flight simulator technology, and discloses a method, system, and medium for digital synthesis of engine noise in an aircraft flight training simulator. The method first collects audio samples of a real aircraft engine under multiple different operating conditions and simultaneously records the corresponding engine operating parameters. After performing spectral analysis on the audio samples, a spectral difference database of engine noise is constructed. Then, in the real-time audio rendering thread of the simulator's host, the current engine operating parameters are received and interpolated and dynamically filtered in conjunction with the spectral difference database to output the synthesized engine noise audio signal. This invention uses spectral difference analysis and dynamic filtering control to achieve engine noise synthesis, and directly extracts spectral features from real aircraft recordings to drive the synthesis process. While ensuring high fidelity in acoustic characteristics, it achieves a flight simulator engine noise synthesis output that combines high fidelity, high real-time performance, and smooth continuity.

The present application relates to the technical field of optical fiber sensing, and discloses an anti-aliasing filtering method and device for a distributed optical fiber acoustic sensing system, which dynamically adjusts the cut-off frequency and type of the anti-aliasing filter by analyzing the frequency band characteristics of environmental noise in real time, and combines frequency band energy detection to close invalid sampling channels, thereby significantly suppressing aliasing noise and reducing data volume. A device fault feature library matching mechanism is further introduced to trigger an emergency anti-aliasing mode when a sudden fault noise occurs, thereby improving fault detection accuracy. The method can be widely applied to oil and gas pipeline monitoring, industrial equipment health diagnosis and other fields, and solves the performance bottleneck of traditional fixed filters under wide frequency dynamic noise.

The application discloses a digital communication signal symbol timing deviation estimation method and device, belonging to the technical field of digital communication, comprising: generating periodic peak points without inter-symbol interference in time domain through frequency domain truncation and equal interval modulus summation mechanism; sinc interpolation up-sampling ensures distortionless reconstruction of band-limited signal and complete reservation of spectral information under low sampling rate. Through specific interval reservation and zero operation on the frequency domain sequence after discrete Fourier transform, periodic reference points without inter-symbol interference are generated in subsequent processing; these reference points only carry current symbol information, eliminating adjacent symbol interference. Through equal interval extraction of points in the discrete sequence after twice frequency domain transformation and summation of modulus values, when the first extraction point position is aligned with the real timing deviation, the sum value reaches the maximum; using the monotone corresponding relationship, high-precision timing deviation estimation can be obtained through simple search, realizing effective estimation of symbol timing deviation.

This disclosure provides a virtual object control method, comprising: displaying a game scene of a game match on a graphical user interface of a terminal device, the game scene including a first virtual object corresponding to the terminal device and a second virtual object belonging to a different faction from the first virtual character; determining a second buff effect in response to the first virtual object performing a first defeat operation on the second virtual object, wherein the second buff effect belongs to a second type of buff effect; applying the second buff effect to the first virtual object; updating the continuous defeat state of the first virtual object; determining a first buff effect corresponding to the updated continuous defeat state based on the updated continuous defeat state, wherein the first buff effect belongs to a first type of buff effect; and applying the first buff effect to the first virtual object. This allows players to obtain dynamic buff effects through defeat operations during game matches, achieving a "fight and grow" gaming experience.

The invention relates to an atrial fibrillation ablation postoperative recurrence continuous learning prediction system based on multi-source electrocardiogram data. The system comprises a data acquisition module, a data processing module, a feature recognition module and a risk prediction module. According to the invention, the electrocardiogram representation network is constructed to process different pieces of follow-up visit electrocardiogram information during follow-up visit, so that corresponding risk prediction results can be automatically output at different stages; meanwhile, the model can still obtain stable and transferable electrocardio characteristic representation under the multi-source and multi-device conditions. The cross-domain and cross-device capability enables the model to stably work in a real multi-center clinical environment, retraining for each medical center is not needed, the deployment difficulty and cost are greatly reduced, and controlled updating can be realized by using a small amount of follow-up visit data without depending on historical original electrocardiogram data and retraining the complete model.

This invention discloses a lightweight star catalog target detection method and system. The method includes: constructing a student model, which employs a lightweight feature extractor and a Transformer decoder with width and depth pruning; constructing a teacher model, which employs an un-lightweight feature extractor and a Transformer decoder without width and depth pruning; adjusting the parameters of the teacher and student models, calculating the value of the loss function, stopping the iteration when the value of the loss function is minimized, and obtaining trained teacher and student models; and using the trained student model to perform lightweight star catalog target detection. The advantages of this invention are: lightweight, low computational complexity, and the ability to achieve 3D position detection of star catalog targets.

The application discloses a device data-free federated incremental learning method and system under resource restriction, relates to the technical field of transfer learning, and comprises the following steps: training a model based on an incremental data stream through a bias correction mechanism to obtain local parameters after training, uploading the local parameters to the cloud, calculating the difference of the parameters to obtain the gradient of each edge computing device, calculating the deviation degree of the update direction of the federated average and the gradient, calculating the aggregation weight, weighting and fusing the updated gradient, and obtaining an updated global model; generating synthetic samples of an old task by minimizing diversity loss; based on the synthetic samples, carrying out knowledge distillation, transferring the knowledge of the old task to the updated global model, obtaining a final global model, and distributing the final global model to the edge end. Through bias correction training, double-sensing aggregation of the deviation degree and information entropy, and attention-guided data-free synthesis and distillation, the application realizes the federated incremental learning of the resource-restricted edge device.

This application discloses a vehicle charging port cover control method, device, vehicle, and system. The vehicle charging port cover control method includes: acquiring a first charging port cover opening request sent by a first charging gun and a pre-written first common identifier; determining whether a second common identifier is pre-written in the vehicle; if a second common identifier is pre-written in the vehicle, determining whether the second common identifier matches the first common identifier; if the second common identifier matches the first common identifier, controlling the charging port cover to open according to the first charging port cover opening request. In the embodiments of this application, it is possible to avoid opening the charging port cover in a vehicle that is not bound to a first charging pile, that is, it is possible to avoid accidentally opening the charging port cover in other vehicles of the same type.

The application belongs to the technical field of nonlinear system control, and discloses a method and system for introducing composite learning into nonlinear system output feedback adaptive control. The application directly reuses the existing K-filter in standard output feedback backstepping control to construct an extended prediction error, and does not need to establish an additional observer or a state estimation model (such as a series-parallel estimation model, a fuzzy observer, etc.) in parallel with the K-filter. This functional reuse design eliminates the calculation overhead of the additional dynamic system, reduces the storage requirement and real-time calculation burden of the controller, and significantly reduces the hardware implementation cost and system debugging complexity. Meanwhile, through the composite driving of the tracking error and the extended prediction error, the cumulative information is continuously introduced by using the historical memory of the regression quantity, so that the adaptive and self-learning ability of the control system is significantly enhanced, and the parameter estimation convergence and the system response performance are improved.

The invention relates to a unified memory architecture supporting a multi-post quantum algorithm and a near memory mapping method thereof. The invention relates to the technical field of post quantum cryptography. The architecture comprises a program controller, an instruction ROM, a Keccak module, a sampler and a polynomial arithmetic unit, the program controller decodes an instruction from the ROM, wherein the instruction specifies an operation type, a source / target operand and a control signal; a Hash function mode is selected through a control signal of the Keccak controller; the sampler is tightly coupled with Keccak output and comprises a parallel logic unit of each mode, and an internal arbiter selects sampling variants according to a current instruction and activates corresponding sub-modules; the polynomial arithmetic unit supports high throughput polynomial operation. According to the method, multiple post quantum cryptography algorithms can be efficiently supported on a single hardware platform, and flexible switching among the algorithms can be realized on the premise of not sacrificing performance.

The invention discloses a radar-oriented order constraint Jacobi eigenvalue self-sorting method and system, solves the problem of resource consumption caused by an additional sorting module in a Jacobi iteration eigenvalue decomposition post-processing flow in the prior art, realizes direct output of ordered eigenvalues and eigenvectors, and omits an independent sorting module. The method comprises the steps that a real symmetric matrix A obtained through radar baseband signal sampling and covariance estimation is processed to obtain a compressed array, and a feature vector matrix V is initialized; in the parallel iteration process, index pairs are generated according to a scheduling sequence, sub-matrixes are extracted, two types of asymmetric rotation operators are generated through a condition driving mechanism, and a feature vector matrix is synchronously updated to track feature vectors; global data replacement is carried out, and non-diagonal element energy judgment convergence is calculated; after iteration is ended, feature values arranged in a descending order can be directly extracted from the main diagonal of the final compressed array, and corresponding feature vectors are output and used for direction of arrival estimation.

This disclosure discloses a server integrated firmware system and task processing method, relating to the field of firmware engineering. The system includes: an integrated control module for receiving and processing task instructions from a target application, calling corresponding BIOS modules and / or BMC modules to execute the target task according to the task instructions, and returning the execution result to the target application; and an embedded system platform integrating multiple BIOS modules and BMC modules. The BIOS modules and BMC modules are obtained by modularizing the BIOS and BMC according to task functions using a RISC-V architecture processor. The integrated firmware design of this disclosure improves the response speed and execution accuracy of instruction processing, enabling the server to adapt more quickly to the needs of cloud computing and big data processing.

The invention relates to a neural network model training method and device, a neural network model calculation method and device, electronic equipment and a computer program product. According to the training method of the neural network model, the neural network model comprises a neural network layer, and the method comprises the following steps: dividing a target tensor in the neural network layer into at least two data blocks; for each data block, determining a corresponding public index according to the amplitude information of the elements in the data block; normalizing all elements in the data block relative to the common index to obtain a corresponding mantissa; for each element, only corresponding mantissa and sign bit are stored; the public indexes are stored according to data blocks, layers or channels; when the neural network model is trained, the public index, the bit width of the mantissa and the weight parameter of the neural network layer are updated through back propagation. According to the technical scheme, the requirements for the storage space and the memory bandwidth of the neural network model can be remarkably reduced on the premise that the precision is controllable.

This application discloses an adaptive multi-granularity text embedding representation learning method based on gradient optimization. The method includes: text encoding, multi-granularity embedding generation, difficulty-aware dynamic sampling, sliding window negative sample management, adaptive loss optimization, and model training and validation. It aims to address the performance degradation caused by existing embedding vector dimensionality compression techniques. Through a difficulty-aware dynamic sampling mechanism, a Beta distribution is introduced to achieve progressive learning, significantly improving low-dimensional performance; sliding window negative sample management improves sample quality and training efficiency; and a gradient-oriented update strategy avoids gradient conflicts and optimizes training stability. Furthermore, this invention supports incremental dimensionality expansion, reducing training costs while maintaining consistency across multi-granularity embeddings. This method significantly improves performance in low-dimensional environments, maintains lossless performance in high-dimensional environments, significantly improves training efficiency, greatly reduces storage costs, and exhibits strong generalization ability, making it suitable for data from various domains.

The invention relates to the field of unmanned aerial vehicle image processing, in particular to a corn phenological period monitoring method based on unmanned aerial vehicle images and deep learning, and adopts the technical scheme that a corn planting test point is tested by taking a year as a time unit, and N planting densities are set for M hybrid varieties of the test point, so that the diversity of a data set is improved; features in different scenes are considered; the training subset is used for training a MaizePhenoNet model, the verification subset is used for monitoring model training, and the phenomenon of overfitting of model training is prevented; compressing the hidden dimension of the classification head from a 1280 channel to a 512 channel so as to reduce parameters of a full connection layer; scaling expansion channels in the FUSDIB module and the UIB module and carrying out 8-times alignment, so that main parameter overhead caused by point convolution is reduced; under the condition that a backbone topological structure and a down-sampling strategy are not changed, the method effectively reduces the model parameter scale and storage requirements, and is suitable for an end-to-end corn phenological period identification task.

The application relates to the technical field of intelligent household appliances, and discloses a cross-domain small sample object image classification method and device for an intelligent terminal, which comprises the following steps: obtaining an object image to be identified; processing the object image to be identified by using a feature extraction module to extract a basic feature vector; processing and fusing the basic feature vector by using a double Riemann manifold processing module to generate a comprehensive feature representation; outputting an object classification recognition result by processing based on the comprehensive feature representation through a classification layer; the feature extraction module, the double Riemann manifold processing module and the classification layer constitute a lightweight self-network model; the lightweight self-network model is obtained by removing a teacher network after joint training through topological relationship knowledge distillation and progressive self-training. The lightweight self-network obtained by the method is suitable for the low computing power and low storage requirement of an intelligent terminal edge end, can realize local low-delay and high-precision classification and recognition of an object image, and solves problems such as uneven illumination and domain offset in the intelligent terminal.

The application discloses a reasoning method and device based on a neural network model, equipment and a medium, comprising: determining a first target network layer with the same structure and a second target network layer with different structures in the neural network model; performing network layer splitting on the neural network model to obtain a first calculation subgraph and a second calculation subgraph corresponding to the first target network layer and the second target network layer respectively; generating a shared engine file corresponding to the first target network layer and weight parameters of each first target network layer according to the first calculation subgraph; generating an independent engine file corresponding to the second target network layer according to the second calculation subgraph; the independent engine file includes the weight parameters of the second target network layer; and obtaining a reasoning result through the shared engine file, the weight parameters of each first target network layer and the independent engine file according to the execution order of each calculation subgraph. Through the shared engine file reasoning of the network layers with the same structure, data storage can be reduced, and single-card reasoning capability can be improved.

The invention discloses an FMCW laser ranging light source spectrum degradation suppression method based on strategy online deployment. Belongs to the laser radar field. The objective of the invention is to solve the technical problems of spectrum degradation and measurement precision reduction caused by frequency modulation nonlinearity of an FMCW laser ranging light source. Comprising the following steps: building an FMCW laser nonlinear dynamic system platform, and constructing a deep learning model to simulate a dynamic environment; designing a nonlinear correction process as a Markov decision process, and training a reinforcement learning agent by using a TD3 algorithm; carrying out lightweight processing on an Actor network in the policy network, wherein the lightweight processing comprises pruning and quantization operations so as to reduce the complexity of the model; a lightweight strategy network is deployed on a hardware platform, strategy network forward reasoning is achieved on an FPGA, the strategy network is cascaded with a state extraction module and an action output module to form a closed-loop correction system, and dynamic modulation current optimization is achieved. According to the invention, frequency spectrum degradation caused by frequency modulation nonlinearity can be inhibited on line, and the measurement precision is improved.

The present application relates to a kind of new type of wall protection mud material for ultra-deep underground continuous wall and its preparation method, comprising the following steps: pour mixing water into the first container, while adding bentonite, the first container is stirred to be uniformly dispersed;Sodium carbonate, sodium carboxymethyl cellulose, xanthan gum and silicone acrylic emulsion are mixed and added to mixing water and poured into the second container, and stirred to be fully hydrated;Mix the solution in the first container and the second container, stir to be fully mixed, complete mud preparation.The beneficial effects of the present application are: it can avoid the hidden troubles such as excessive lateral earth pressure, large amount of mud loss, thin and dense mud skin cannot be formed when the excavation depth of underground continuous wall is large, configure a new type of modified bentonite wall protection mud with suitable physical stability and chemical stability, appropriate relative density, good mud skin forming property, which can adapt to the working condition of ultra-deep underground continuous wall, avoid the occurrence of hole collapse and shrinkage phenomenon, so as to ensure the quality of slotting and construction safety.

The present application relates to the technical field of image matching, and discloses a similar image retrieval method, device and equipment and storage medium, comprising: inputting a to-be-retrieved image and a target image into a pre-trained image segmentation model, obtaining panoramic segmentation results and high-level semantic features of each pixel; performing weighted average on feature vectors of pixels inside each entity to obtain entity description features; constructing a spatial relationship matrix of entity pairs, and weighting and aggregating feature vectors of each overlapping area in the matrix into a spatial topology feature; performing similarity matching on entity description features of entities of the same category in two images to obtain entity description similarity; then performing similarity calculation on spatial topology features of the same category and the same topological relationship to obtain spatial topology similarity; and finally, weighting and fusing the two similarities. The present application can be applied to medical image similar case retrieval in the medical health field and bill and voucher image comparison in the financial technology field, and while retaining global structure information, significantly reduces feature storage capacity.

The present application belongs to the technical field of radar signal processing and target identification, and specifically provides a multi-target array structure inversion method based on two-dimensional power spectrum imaging, to solve the problem that the array structure in the multi-target scattering image is difficult to be stably recovered under the condition of strong aliasing, strong speckle and strong coherent interference background; the present application converts the collected radar scattering data into a power spectrum image, generates a global candidate point set with a level label through log power spectrum construction, sub-aperture decomposition, hierarchical candidate point extraction, consistency voting and density clustering deduplication, and further forms a closed loop through translation and combination prediction and bidirectional matching verification, and the original multi-target array structure is inferred from geometric evidence, which has good robustness, engineering feasibility and identification accuracy; and the present application does not depend on phase compensation, is suitable for observation scenes under multi-target, and meets the application requirements of complex target structure identification, multi-target imaging identification and target classification.

The application discloses a cable surface defect detection method and system based on multi-scale grouping convolution and decoupling detection heads. The method comprises the following steps: firstly, a defect detection network is trained by using a cable surface defect image dataset to obtain an initial defect detection model; then, after model optimization is performed on the initial defect detection model, an optimized defect detection model is obtained; finally, a cable surface image to be detected is input into the optimized defect detection model, and a defect detection result is output. The application can improve detection precision in a scene where complex backgrounds, small targets, low contrast and multiple scales of defects coexist, and can reduce model parameter quantity and calculation amount, thereby meeting real-time detection requirements of resource-limited devices.