58 results about "Neural net architecture" patented technology

A hardware accelerator and acceleration method based on a VisionTransformer neural network, the accelerator is to deploy the VisionTransformer neural network on a ZYNQ development platform; the acceleration method is: an ARM processor stores a feature picture into a DDR memory, the read data is dispersed to an input cache and a weight cache, the processed feature picture is input to an on-chip cache unit, the processed data is sent to a PL end, the hardware IP of the PL end is configured, and read-write operation is performed at the same time, the final calculation result is obtained, written into the DDR memory, the data in the DDR memory is taken out and probability operation is completed, the probability operation result is transmitted to a PC, a PetaLinux operating system is transplanted to the hardware accelerator system of the VisionTransformer neural network, and a prediction result of inference is obtained from the PC; the neural network structure is optimized through the parallel method of multiple input and multiple output channels, the calculation speed is fast, the hardware resource occupation is low, the recognition accuracy is high, and the image classification task can be efficiently completed.

The application belongs to the technical field of zinc powder purification, and provides a condensation and sedimentation system for zinc powder purification, which comprises the following steps: collecting a time sequence of original radiation intensity of a condensation chamber, performing frequency spectrum analysis through fast Fourier transform, calculating energy values of high and low frequency band frequency domain signals, solving a signal coherence factor and calculating a transmittance attenuation coefficient, constructing a metal thin film radiation compensation model based on a long short-term memory network, inputting the attenuation coefficient and the time sequence of original radiation intensity to output a corrected real temperature of the condensation chamber, expanding an inference layer of a full connection neural network structure, outputting a calibrated attenuation coefficient through the inference layer, selecting an effective attenuation coefficient input model by comparing a deviation threshold value, dynamically responding to temperature measurement nonlinear attenuation caused by zinc film deposition, guaranteeing production continuity and product quality of zinc powder purification, adapting to industrial multi-working condition operation requirements, being convenient to operate and controllable in cost.

The application discloses an electromagnetic interference identification method and device, equipment and a readable storage medium, and relates to the technical field of signal processing. The in-phase and quadrature data of a to-be-identified radar signal in multiple distance banks is acquired, and the maximum power variation coefficient, the skewness coefficient, the kurtosis coefficient and the maximum sharpness of the to-be-identified radar signal are calculated according to the in-phase and quadrature data. The above characteristic parameters are input into a pre-trained electromagnetic interference identification model, the electromagnetic interference identification model adopts a full connection neural network structure, the identification results of the to-be-identified radar signal in the multiple distance banks are obtained through an output layer, and the identification results are used for distinguishing meteorological echo useful signals from electromagnetic interference signals. The pollution area of the to-be-identified radar signal is determined according to the identification results of the to-be-identified radar signal in the multiple distance banks, the pollution area is recorded, and an alarm prompt is sent to technical personnel, so that the missed judgment of electromagnetic interference and the misjudgment of useful signals can be avoided, and the identification accuracy of various electromagnetic interferences is improved.

This invention discloses an adaptive tree neural network structure and processing method for construction project management. The structure includes: an engineering data input module for collecting structured and time-series data of construction projects; a tree structure encoding module for constructing a multi-level decision tree structure based on project hierarchy and encoding node features; a recurrent neural network module for performing time-series modeling of node feature sequences to predict project status; an adaptive feedback module for feeding back the prediction results to the tree structure encoding module to update node features or weights, forming an adaptive closed loop; and an engineering management output module for generating construction management decision information based on the prediction results. This invention solves the problems of missing engineering structure representation, discontinuous time-series prediction, and lack of dynamic adjustment capability in existing technologies by integrating hierarchical engineering structure modeling and time-series prediction and introducing an adaptive feedback mechanism, thereby improving the intelligence level of construction progress prediction, resource scheduling, and risk warning.

The application discloses a kind of high-performance computing cluster resource scheduling method and system based on TR-DQN, first user submits task request, all requests enter waiting queue and wait for scheduling;Then the priority of submitting task is calculated, and the waiting queue is reordered;Then the node information and task information of cluster are collected and processed, and the processed data is input into TR-DQN model for scheduling;Finally, after task scheduling is completed, it enters corresponding node operation.TR-DQN model combines the characteristics of high-performance computing cluster scheduling into deep reinforcement learning, and introduces two-level neural network structure, the first neural network is used to select tasks for immediate execution or reserved execution, and the second neural network is used to select tasks for backfilling, which can improve the resource utilization of the cluster, reduce the waiting time of the task, and quickly adapt to changes in the cluster load environment. In addition, it can also minimize the problem of work starvation in the cluster.

The application discloses a video super-resolution method based on space-time convolution attention, and is specifically implemented according to the following steps: step 1, converting an input high-resolution video sequence into a low-resolution video sequence, so as to obtain a low-resolution video set; step 2, constructing a deep neural network structure for video super-resolution; step 3, training the network based on paired video data; step 4, video super-resolution reconstruction, so as to realize super-resolution recovery of the input video sequence. The application solves the problem in the prior art that the network is regarded as a black box, performance is improved by increasing the network depth and parameter scale, the rapid growth of the calculation complexity is ignored, and the efficiency and the reconstruction quality are difficult to be considered.

This application provides an image processing method and apparatus based on adversarial neural network (DNN) architecture search. The method includes: for any epoch in the DNN architecture search process, iteratively updating the operating parameters and structural parameters of the DNN network using a gradient descent algorithm, an acquired image training set, and an acquired image validation set until the number of iterations reaches a first iteration number; and iteratively updating the structural parameters of the DNN network using preset network vulnerability constraints and the acquired image validation set until the number of iterations within that epoch reaches a second iteration number; when the number of searched epochs reaches the first epoch number, or when the DNN network model converges, generating a target DNN network for image processing based on the obtained structural parameters, and using the target DNN network to process the image to be processed. This method can improve the accuracy of image processing using DNN networks.

An example device for processing video data includes a memory configured to store video data; and a processing system implemented in circuitry, the processing system being configured to: receive data representing a plurality of neural networks associated with a video bitstream, each of the plurality of neural networks having a different type; receive data representing an update to at least one of the neural networks, the data including a type corresponding to the at least one of the neural networks and a neural network structure for the update; update the neural network according to the data representing the update to generate an updated neural network; and provide video data from the video bitstream to the updated neural network to cause the updated neural network to process the video data.

This invention relates to the field of iron and steel metallurgy technology, and in particular to a method, system, and application method for predicting the viscosity of titanium-containing blast furnace molten iron based on a cascaded neural network. The method includes collecting the temperature and mass fractions of Fe, C, Si, Mn, P, S, and Ti elements in the titanium-containing blast furnace molten iron and inputting them into a first-stage neural network. The first-stage neural network predicts the free volume ratio, cluster volume ratio, number of atomic clusters per unit volume, solid phase particle fraction, and superheat. Intermediate structural parameters are then input into a second-stage neural network, outputting the target physical property values ​​of the titanium-containing blast furnace molten iron viscosity. This invention introduces physically interpretable liquid structural parameters as intermediate variables into the neural network structure, achieving accurate and efficient prediction of the viscosity of multi-component, strongly non-ideal melt systems.

The application provides an electrocardiosignal enhancement and classification method based on a gated recurrent diffusion model, comprising the following steps: step 1, obtaining an electrocardiosignal dataset; step 2, setting parameters and hyperparameters of the diffusion model; step 3, establishing a neural network structure of the diffusion model; step 4, training the diffusion model, and supplementing the electrocardiosignal dataset by using the trained diffusion model; step 5, setting hyperparameters of a classification model; step 6, establishing the classification model; and step 7, training the classification model by using the electrocardiosignal dataset obtained in step 4, and using the trained classification model for electrocardiosignal classification. The method introduces the gated recurrent diffusion model to supplement data, compared with a traditional generative model, can capture time sequence information of the electrocardiosignal, and thus simulate the characteristics of the heart rate.

This invention relates to the fields of radio communication technology and artificial intelligence security technology, specifically to a method and system for adversarial example defense of signal recognition networks based on neural network structure search. It constructs a heterogeneous hypernetwork search space for radio signal recognition tasks, configures network depth, channel number, and connection patterns according to signal processing characteristics, and obtains multiple signal recognition subnetworks with significantly different structures and adversarial robustness in a single search through random path sampling and adversarial example generation mechanisms. Based on this, a dynamic ensemble strategy is adopted to randomly select multiple subnetworks from the heterogeneous robust subnetwork library and perform weighted ensemble. The unpredictability of the ensemble combination obfuscates the attacker's estimation of the model gradient, thereby improving the active defense capability of the signal recognition system in complex electromagnetic environments from the system architecture level. This invention overcomes the performance limitations of single baseline models and provides an efficient and scalable adversarial defense solution for intelligent radio signal processing systems.

The application discloses a kind of based on multi-task neural network and route adapter's freight rate prediction method, comprising: constructing and training multi-task neural network model;Feature data acquisition step, data coding step, data fusion step, route adaptation step, prediction output step, multi-task neural network model multi-task calculation and output, including one or more of calculating and outputting freight rate prediction, freight rate variance prediction, multi-day average price prediction and inflection point prediction.The based on multi-task neural network and route adapter's freight rate prediction method, by using multi-task neural network structure, simultaneously executes freight rate point prediction, future seven-day price prediction, price inflection point prediction and uncertainty estimation task in the same model.Sharing sequence coding layer, feature fusion layer and main network, promote model to learn to the implicit feature representation that is effective to multiple price behaviors, reduce the overfitting to noise characteristics, improve the overall description ability to price change law.

The application provides a power distribution network abnormal working condition identification and self-healing method based on meta-reinforcement learning and related devices. The method comprises: acquiring multi-source heterogeneous data of a power distribution network, wherein the multi-source heterogeneous data comprises electrical quantity data, distributed photovoltaic output data, energy storage system state data and non-electrical quantity environmental data; preprocessing the multi-source heterogeneous data to obtain preprocessed data; mapping the power distribution network topology to a space-time graph neural network structure, and constructing a hierarchical meta-reinforcement learning model based on the space-time graph neural network structure; inputting the preprocessed data into the constructed hierarchical meta-reinforcement learning model, and synchronously outputting an identification result of an abnormal working condition and a self-healing control strategy. The application realizes "identification is decision-making" by integrating fault identification and self-healing decision-making into a unified model, significantly shortens the response time from fault occurrence to power restoration, and effectively improves the self-healing efficiency and operation resilience of the power distribution network.

This invention discloses an interpretable robotic arm bearing fault diagnosis method based on physical prior knowledge, belonging to the field of rolling bearing fault diagnosis technology. This method integrates the physical prior knowledge of the bearing with a neural network structure. It obtains fault frequency harmonic features with clear physical meaning and key frequency band broadband features through parallel physical feature extraction networks and broadband feature extraction networks, respectively. A multi-head attention mechanism with sparsity and diversity constraints is used to adaptively filter the fused features, highlighting key fault characterization information. Finally, a single-layer linear classifier achieves transparent fault category decision-making. This invention solves the "black box" problem of existing deep learning fault diagnosis models, providing end-to-end interpretability from feature contribution and attention allocation to classification decision-making while ensuring high diagnostic accuracy, significantly improving the model's reliability and engineering practicality.

The application discloses the technical field of wind turbine power prediction and intelligent modeling, and particularly relates to a wind power prediction method based on a wake perception condition residual error neural network. The method fully utilizes upstream and downstream unit monitoring data and wind farm geometric arrangement, constructs a condition residual error neural network structure with wake perception capability, and conditionally corrects wake residual error through a gate signal with clear physical meaning. Under the premise of ensuring that the prediction accuracy of the no-wake condition does not decrease, the prediction accuracy and robustness of the wake condition, especially the severe wake condition, are improved, and the method can be applied to the wind turbine power prediction scene of the minute level and shorter time scales.

The application discloses an online intelligent monitoring method for oil content and gas content of a drilling fluid, and the method comprises the following steps: collecting image samples of water-based drilling fluid without being separated by a solid-liquid separator by using an online image acquisition device of the drilling fluid, and establishing an image database; further highlighting the gas content and oil content in a unit area, pre-processing the image data, and automatically generating a trainable data set; designing an online intelligent monitoring neural network structure, training and generating an intelligent identification model; packaging the model, and establishing an online intelligent monitoring system. According to the proportion of bubble area, the proportion of oil droplet area and the shape of water-based drilling fluid, the relationship between these characteristics and the gas content and the oil content of the drilling fluid is fitted through a deep residual shrinkage network, so that the volume fraction of the gas content and the oil content in the water-based drilling fluid can be evaluated in real time, and online inquiry and monitoring can be realized through a user interaction module of the monitoring system.

This invention discloses a cementing quality evaluation system and method for deep shale gas wells, belonging to the technical field of cementing quality evaluation for deep shale gas wells. The cementing quality evaluation system for deep shale gas wells includes a data collection component connected to a data set module for data aggregation. The data set module is connected to a strength evaluation component for assessing the strength of cement sheath bonding and a network training component for training a neural network. The network training component is connected to a structure determination component for determining the neural network structure. The structure determination component is connected to a real-time evaluation component for evaluating the cementing quality of shale gas wells. The network training component is connected to the real-time evaluation component. This invention effectively solves the problems of long evaluation time, high cost, and inaccurate evaluation results in existing shale gas well cementing quality evaluation methods.

The present invention relates to artificial intelligence neural network computation apparatus and method, the apparatus comprising: neural network units connected according to a neural network structure and composed of an encoding unit, a decoding unit or a transformer unit connecting the encoding unit and the decoding unit; line buffers that provide input data for processing images to an input neural network unit from among the neural network units; a first weight buffer that provides first weight data to the encoding unit and the decoding unit; a second weight buffer that provides second weight data to the transformer unit; and an output buffer for storing output data from the processing of the images from an output neural network unit from among the neural network units.

This invention provides an automatic search method and apparatus for neural network structures for few-sample fitting. The method includes: generating multiple candidate neural network structures in a search space using a controller; for each candidate neural network structure, performing a comprehensive performance evaluation based on a few-sample dataset to obtain its comprehensive performance index; selecting candidate neural network structures that meet preset requirements based on the comprehensive performance index of each candidate neural network structure, the comprehensive performance index including fitting accuracy and model complexity, and updating the controller parameters according to the performance index of the candidate neural networks that meet the preset requirements; and repeating the steps of generation, evaluation, selection, and updating based on the updated controller parameters until a preset stopping condition is reached to obtain the optimal neural network structure for few-sample fitting, which can then be used for few-sample fitting tasks. This invention effectively overcomes overfitting in few samples and significantly improves the model's generalization ability.

This invention relates to the fields of automated machine learning and neural network technology, specifically to a method and system for searching neural network structures based on dynamic collaborative graph encoding. This method leverages the collaborative evolution mechanism of maintainer, sterile, and restorer lines in breeding optimization algorithms, achieving inter-population structural migration through dynamic hybridization probability control. A support vector machine surrogate model is constructed to replace time-consuming performance evaluation, rapidly predicting the accuracy of offspring structures and screening high-potential candidate structures. Gradient fine-tuning is applied to candidate structures to output the optimal network structure. Modular recombination and topological evolution of the neural network structure are achieved through graph encoding. Utilizing the collaborative mechanism of maintaining the stability of the maintainer line, exploring the diversity of the sterile line, and accelerating the convergence of the restorer line, combined with a two-layer dynamic regulation of population-level migration control and individual-level variation optimization, this method efficiently obtains high-performance deep neural network structures.

The method for sensing the temperature of urban solid waste incinerators based on a self-attention hybrid ensemble network belongs to the field of industrial process modeling and intelligent sensing. This method addresses the highly nonlinear, multi-condition variations, and complex physical-chemical reaction mechanisms in the MSWI process. Combining an ensemble learning framework and improved neural network structure, it introduces moving block bootstrap (MBB) sampling based on maximum likelihood estimation (MLE), a self-attention mechanism, and a hybrid ensemble strategy fusing AdaBoost and Bagging to achieve accurate modeling and dynamic prediction of the Fourier transform (FT). This results in good modeling accuracy, generalization ability, and engineering adaptability. Experimental verification under multiple benchmark problems and actual MSWI conditions demonstrates that the proposed FT sensing method exhibits superior performance in both prediction accuracy and robustness, possessing significant engineering application potential and widespread value.

A spiking neural network system for single-photon imaging is disclosed. The system comprises: a light source for emitting a series of light pulses, one pulse per repetition period, for repeatedly illuminating one or more objects; a single-photon detector for detecting photons received from the one or more objects; a ring circuit connected to the single-photon detector, the ring circuit comprising a set of delay elements connected to one another thereby forming a ring structure, a respective delay element is configured to store a spike received from the single-photon detector, the ring circuit being configured to repeatedly rotate the spikes through the delay elements, a respective spike making a full rotation of the ring circuit in one rotation period such that the ring circuit is synchronized with the light source; a read-out circuit for reading out the spikes stored in the delay elements while maintaining the differences of arrival times of the spikes in the ring circuit; and a spiking neural network configured to receive the spikes from the read-out circuit to generate an image of the one or more objects.

The application provides a vehicle detector data repair performance analysis method considering multi-factor influence, and belongs to the technical field of intelligent traffic information. The application specifically comprises the following steps: building an expressway road network model based on VISSIM traffic simulation software, and generating expressway multi-source data by using the expressway road network model; obtaining the repair value of the expressway traffic flow parameter by repairing the real value of the expressway traffic flow parameter; constructing a vehicle detector data repair performance analysis model based on a BP neural network structure, training the vehicle detector data repair performance analysis model by using the expressway multi-source data and the repair value of the expressway traffic flow parameter; and quantitatively analyzing the vehicle detector data repair performance by using the trained vehicle detector data repair performance analysis model. The application can make the vehicle detector data repair performance reach an ideal level, has guiding significance for traffic managers to arrange expressway multi-source detection equipment, and further improves the reliability of traffic data.

The present application relates to the field of intelligent prediction, and particularly relates to a weak adhesion hydrogel adhesion performance prediction method and system, comprising the following steps: S1: obtaining hydrogel material parameters and interface state information and preprocessing; S2: identifying state labels through a classification model to form an extended data set with state labels; S3: constructing a segmented physical model system driven by the interface state to obtain state-adaptive physical prediction results; S4: embedding the state-adaptive physical prediction results into a physically-constrained neural network structure to construct a prediction model and obtain model prediction results; S5: using a multi-model collaborative fusion mechanism, comprehensively considering the state-driven physical model output and the model prediction results, and obtaining the final prediction value through a weighted fusion manner; S6: taking the final prediction results and model internal parameters as inputs, performing feature contribution degree analysis, identifying key influencing factors, and outputting parameter optimization suggestions. The present application realizes high-precision and high-generalization-capability prediction.

This invention discloses a method for reconstructing X-ray small-angle scattering (SAXS) patterns based on a physical information neural network, belonging to the field of nanostructure measurement. Through a differentiable physical information optimization framework driven by a physical information neural network, a given electron density template is used as input, and the output is a high-resolution, corrected electron density map. A loss function is calculated by comparing the SAXS pattern with an experimental X-ray small-angle scattering pattern, and the neural network structure is optimized based on the loss function to ensure physical realism and computational feasibility, thereby continuously optimizing the predicted electron density map and making it closer to the real sample. This invention eliminates the dependence on a pre-set geometric model, requiring only a single SAXS scattering pattern containing multi-angle information to faithfully reconstruct the true morphology of nanostructures containing arbitrarily complex and non-ideal features such as rounded corners, sidewall curvature, and chamfered edges. This fundamentally solves the model mismatch problem and significantly improves the accuracy, reliability, and applicability of X-ray small-angle scattering metrology.

The application discloses a motor driving sudden change friction characteristic modeling and feedforward compensation method, which realizes decoupling and independent description of non-sudden change characteristics and sudden change characteristics of friction, and constitutes a dynamic neural network friction model with a width neural network structure which integrates the width neural network structure characteristics and the dynamic recurrent neural network structure characteristics; and the motor control system feedforward control compensation based on the friction model: the motor rotation angle is used as the input signal of the friction model, the friction model obtains corresponding friction torque prediction and estimation, the output of the motor controller is corrected through the feedforward control compensation, and the influence of internal friction on the motor execution precision is offset.

This invention provides a method and system for perception and scene understanding of a flying robot in a microgravity environment, belonging to the field of robot intelligent perception and scene understanding. It addresses the problems of crowded facility layouts, complex spatial lighting, and floating object obstruction in microgravity indoor work scenarios, which lead to difficulties in image feature extraction and obstructed sensor lines of sight, affecting the accuracy of 3D environment modeling and target recognition. A lightweight convolutional neural network structure is adopted to reduce computational burden; in terms of multimodal data fusion, data from RGB images and laser rangefinders are combined, and a multimodal information fusion module is added to target detection and pose estimation, suitable for the identification and localization of target objects required for high-precision operations; in terms of task-adaptive semantic segmentation, the network is trained on a specific dataset for specific task scenarios in a microgravity environment; and transfer learning is used to enhance the model's adaptability.

This invention relates to the interdisciplinary fields of marine gravimetry and artificial intelligence. Specifically, it presents a marine gravity field inversion method based on a regularized multilayer perceptron (NRMLP) that can significantly improve the accuracy of marine gravity field inversion. The method involves building and training a NRMLP model to obtain residual gravity anomalies, which are then incorporated into a loss function as constraints during NRMLP model training. A marine gravity field model, NRMLP_GRA, is then established based on the NRMLP model. This method uses the residual gravity anomalies calculated by the marine gravity field model as part of the loss function, acting as a regularization constraint. NRMLP takes the meridional component of the residual perpendicular deviation, the ramusoidal component of the residual perpendicular deviation, and the spherical angular distance as inputs, and the residual gravity anomalies as outputs. Ultimately, this improves accuracy in regions with scarce effective training data, providing a solution to enhance marine gravity field accuracy that differs from neural network structure optimization.

The application discloses a kind of group intelligence driven bottom quality identification method and related equipment, the present application is configured by introducing genetic population to the global optimization of neural network parameter, can effectively avoid the blindness of traditional method relying on artificial parameter adjustment, and then can improve the fitting ability of model to complex multi-beam data;Among them, through dynamic adjustment mechanism, the number of hidden layer neurons can be adaptively increased and population evolution is re-performed, to ensure that the model accuracy meets the requirements, effectively enhancing the structural adaptability of the model;Specifically, through the collaborative optimization of the above group intelligence and neural network structure, the bottom quality sample characteristics of multiple regions can be efficiently utilized to train a recognition model with strong generalization ability, enabling accurate bottom quality classification based on real-time multi-beam data of different water areas, significantly improving the intelligence level and reliability of bottom quality identification, and can be widely applied in the field of data processing technology.