21results about How to "Reduce training difficulty" patented technology

The invention provides a diffusion-driven channel adaptive point cloud semantic communication method, which belongs to the technical field of wireless communication, and comprises the following steps: obtaining a training point cloud data set, constructing a point cloud feature extraction network based on a hypergraph convolutional neural network as a semantic encoder, and constructing a channel adaptive enhancement module as a channel decoder, a channel adaptive recovery module is constructed at a receiving end as a channel decoder, a diffusion reconstruction network is constructed as a semantic decoder, end-to-end joint training is performed on the point cloud feature extraction network, the channel adaptive enhancement module, the channel adaptive recovery module and the diffusion reconstruction network, the diffusion reconstruction network predicts original point cloud distribution, and the point cloud feature extraction network and the channel adaptive enhancement module are subjected to end-to-end joint training. And a loss function is constructed based on the chamfering distance between the predicted point cloud distribution and the original point cloud. In the communication process, the reconstructed semantic features are input into the diffusion reconstruction network to reconstruct an original point cloud structure. The high-order semantic features of the point cloud can be effectively extracted so as to improve the adaptive capacity of the method to the incomplete point cloud.

The application relates to the technical field of machine learning, and provides an animal sound classification method, system, device, medium and product. The application trains a classification model through a staged strategy, including pre-training the classification model based on a general audio dataset, end-to-end training the classification model based on an animal sound dataset, and deploying the trained classification model to an intelligent device; the intelligent device collects original audio signals in real time and extracts log-mel spectrum features as feature inputs of the classification model, and the model finally outputs prediction probabilities for each target sound category, so as to determine an animal sound classification result. The staged training method significantly reduces the model training difficulty, the log-mel spectrum features are used as inputs to improve the robustness of the model to background noise, and the generalization ability and overall accuracy of the classification model are effectively improved.

The method belongs to the technical field of panoramic monitoring of ultra-high voltage converter station, and aims to solve the problem that panoramic monitoring image data is directly uploaded to the cloud, occupying a large amount of cloud resources. By adopting multi-scale convolution blocks in the deep multi-scale residual network model to construct low-order and high-order features of images of various scales, the incomplete phenomenon of image detail extraction is avoided, and the residual learning mechanism is adopted to retain low-order rough features, thereby improving the reconstruction ability of the image. Topology optimization constructs the topology structure of the heterogeneous network, and the framework combining deep reinforcement learning and Monte Carlo tree search is used to construct the network according to the pre-defined topology rules. The search result of the Monte Carlo tree strengthens the learning of the deep convolutional neural network, so as to obtain more accurate prediction in the next iteration. After the data is processed in the edge side, it is transmitted to the cloud storage, saving the cloud storage space and transmission bandwidth.

The application provides a method and system for image data recognition model construction. The method comprises: obtaining an image data set comprising true value labels; based on the image data set and a teacher network model, using an FSP method to initialize the weights of each layer of a student network to obtain at least two student network models with the same network structure; based on the image data set comprising true value labels, using a DML distillation method to train each student network model, and when the trained student network models all meet preset conditions, taking the trained student network models as an image data recognition model. Through the method, at least two student network models with the same structure and completed weight initialization of each layer are obtained by transferring learning from a teacher network model, which can avoid problems such as the inability to well initialize a neural network due to insufficient data, and can reduce the training difficulty. The DML distillation method is used to obtain an image data recognition model with guaranteed recognition ability, and the training cycle is short, the speed is fast, the deployment is easy, and the applicability is wide.

This invention discloses a method for predicting the viewport of panoramic video streaming with viewer type estimation. This method, tailored to different viewer types, trains a corresponding neural network prediction model for the viewport position using a set of viewport position vector samples. This makes the neural network prediction model more viewer-type specific, ensuring that a particular viewport position prediction model is only applicable to a specific viewer type, reducing the training difficulty of the model and improving prediction accuracy. This method determines the viewer type based on the viewer's actual viewport position sequence before the current frame and selects the corresponding neural network prediction model to predict the viewport position of the current frame. This method ensures that for different types of viewers, a corresponding trained neural network prediction model is used to predict the viewport position of the current frame.

The application discloses a kind of based on multi-source fusion and reinforcement learning's composite material additive manufacturing adaptive control method, first in printing process Synchronous acquisition space position, infrared temperature, tension and compression force and multi-source sensor data such as visible light image;Through the lightweight perception module constructed to each road data is parallelly processed and feature extraction, to force data adopts second-order difference to carry out state discrimination, to temperature data adopts multistage threshold segmentation and extracts forming area statistical feature, and reward feature is generated in combination with image recognition and position mutation detection;The multi-dimensional state features after fusion are combined with reward function, and the agent is trained using reinforcement learning algorithm to learn the optimal printing parameter decision strategy;The trained agent is used for real-time control, and the best printing speed and layer thickness instruction are dynamically output to the mechanical arm according to the current state. The application realizes accurate description and real-time adaptive control of complex multi-physical field coupling process, effectively improves printing quality and process stability.

The invention relates to a data set decomposition-based large language model training method, which comprises the following steps of: performing token processing on an original document in a training data set to obtain a one-dimensional token sequence and calculating an original length; performing binary decomposition on the original length, splitting the token sequence into sub-token sequences, storing the sub-token sequences in buckets according to the length, and adding an original document traceability identifier; calculating a single sampling number of each effective bucket, and configuring a sampling probability; executing random sampling according to the sampling probability, and performing homologous constraint splicing to generate a training batch; and inputting the training batch into the to-be-trained large language model, and performing fine adjustment on the sampling probability adjustment rule and the training hyper-parameter according to the results of the conventional reference performance evaluation, the long context capability evaluation and the training efficiency evaluation. By means of the large language model training method based on data set decomposition, cross-document attention interference can be reduced, calculation waste can be reduced, and training efficiency can be improved.

This invention discloses a method for displacement field recognition of cement-based materials based on a one-dimensional convolutional model. The method includes: S1: extracting the displacement field from a speckle image; using speckle image pairs as feature data, and using the lateral displacement field, longitudinal displacement field, and combined displacement field as label data, and normalizing the label data to obtain a sample dataset; S2: constructing a displacement field recognition model for cement-based materials based on a one-dimensional convolutional model; S3: constructing a model loss function containing the design parameters to be optimized by introducing physical loss, and training the displacement field recognition model based on the one-dimensional convolutional model for cement-based materials based on the sample dataset to obtain the optimal displacement field recognition model; and realizing the displacement field recognition of cement-based materials based on the optimal displacement field recognition model. This invention solves the problem of insufficient accuracy and efficiency in the displacement field recognition of cement-based materials by existing methods.

The application discloses a multi-agent cooperative navigation optimization method, and the optimization method takes a multi-agent deep deterministic policy gradient algorithm as a basic framework; the following mechanism is introduced in the training stage: in the multi-agent cooperative navigation training stage, a joint priority experience extraction mechanism is introduced, joint experience samples are extracted and spliced according to the time difference error priority through cross-agent synchronization indexing; an information discarding and correction compensation mechanism is introduced at the input end of the centralized Critic network, information from other agents is divided into independent information blocks, the information blocks are randomly discarded with a preset probability, and a correction factor is applied to the retained information blocks to maintain the mathematical expectation of the input data unbiased, and the input of the reduced Critic network is obtained. The application provides a multi-agent cooperative navigation optimization method which can improve the key experience utilization rate, reduce the dependence of the model on high-dimensional global input, and enhance the robustness in a communication limited environment.

This invention proposes a method for denoising vibration signals from structural health monitoring based on a hybrid neural network. This method, implemented according to embodiments of the invention, eliminates the need for prior signal knowledge and manual parameter settings. It effectively removes various noise types from structural health monitoring vibration signals, significantly improves the signal-to-noise ratio, and reduces the root mean square error, achieving efficient and automated noise reduction of vibration signals.

The application provides a trajectory prediction method and device, electronic equipment and a storage medium. The method comprises: acquiring an initial image and determining at least one target object in the initial image; using a skeleton graph convolution network to evaluate the posture of the at least one target object at different time nodes in the initial image to obtain an intermediate data set; determining the spatial relationship between the at least one target object according to the initial image to obtain a correlation matrix; and inputting the intermediate data set and the correlation matrix into a social spatio-temporal graph convolution network to obtain a trajectory prediction result of the at least one target object.

This disclosure provides a method, apparatus, device, and medium for generating adversarial perturbations for voiceprint recognition. The adversarial perturbation generation method includes: acquiring a training voiceprint sample set; initializing an adversarial sample generation network; determining a target recognition object for each original voiceprint sample in the training voiceprint sample set; inputting the original voiceprint samples into the adversarial sample generation network to obtain adversarial voiceprint samples; inputting the adversarial voiceprint samples into a voiceprint recognition network to obtain a first recognition result vector; determining a sample loss function for the adversarial sample generation network based on the original voiceprint samples, the adversarial voiceprint samples, a first probability, and a second probability; training the adversarial sample generation network based on the sample loss function to generate adversarial perturbations. Embodiments of this disclosure can improve the efficiency of generating adversarial perturbations and also enhance the dominance and stealth of the adversarial perturbations. Embodiments of this disclosure can be applied to artificial intelligence, communication security, etc.

The application relates to the technical field of target detection, in particular to a high-altitude electric power operation violation identification method, which discloses the following steps: collecting high-definition images and low-quality images of an electric power operation site by using a drone, and making an electric power site high-definition data set; constructing an image enhancement network based on a local information enhancement module and a global information enhancement module; building a violation identification model based on an improved YOLOv8; selecting a suitable optimizer and a loss function; training and optimizing the constructed model. The optimal model after training and optimization is converted into an RKNN model in the front end, and is deployed to the front end of the drone. Real-time detection is realized by relying on the camera in the front end, and whether the operation personnel violate the rules is judged by the judgment unit according to the detection result. The application provides a high-altitude electric power operation violation identification method, which considers the construction of an operation data set, the enhancement of low-quality images, and simultaneously proposes a violation identification network model based on an improved YOLOv8, so that the accuracy of detection is maximally improved.

This application provides a training method for a classification model, a text classification method, and related equipment to address the problem of low classification flexibility in classification models. The method includes at least the following steps: based on preset multiple feature dimensions, extracting corresponding multidimensional text features, multidimensional associated label features, and at least one multidimensional other label feature for selected sample text, associated classification labels, and at least one other classification label from among multiple classification labels; determining the negative sample similarity between each of the at least one other classification label feature and the multidimensional text feature based on the hierarchical distance between each of the at least one other classification label and the associated classification label; and adjusting model parameters based on the obtained positive sample similarity and at least one negative sample similarity between the multidimensional text feature and the multidimensional associated label feature. This enables a target classification model to have feature extraction capabilities of different granularities, improving classification flexibility.

This invention provides a hierarchical collaborative control method and system for a heat pump thermal management system in electric vehicles, relating to the field of electric vehicle thermal management. The method includes: collecting raw operating data and performing preprocessing on the raw operating data; constructing an upper-level state space and a lower-level state space, and setting upper-level action spaces and lower-level action spaces that match each level of state space; vectorizing the upper-level state space, lower-level state space, upper-level action space, and lower-level action space; constructing and training a high-fidelity heat pump thermal management simulation environment model containing upper-level and lower-level strategy modules; obtaining the current upper-level state vector and the current lower-level state vector; and using the trained high-fidelity heat pump thermal management simulation environment model for online inference to obtain actual actuator control commands. This invention can effectively alleviate the long-term and short-term conflict between passenger cabin comfort, battery thermal safety, drive system thermal reliability, and vehicle energy consumption.

This invention discloses an automated design method, device, computer equipment, and medium for corner bracing and anti-bracing structures within foundation pits, relating to the fields of smart structures and intelligent structural design. The method includes: constructing images for each stage of a three-stage design process for corner bracing and anti-bracing structures based on design images of these structures; semantically processing key structural information in each stage image to obtain semantically encoded images corresponding to each stage; decomposing the semantically encoded images of each stage into multiple single-bracing structure design images, and then combining and stitching these multiple single-bracing structure design images into a multi-bracing structure design image; and building a generative adversarial network (GAN) model, training the GAN model based on each single-bracing structure design image and each multi-bracing structure design image. This invention provides a fast, efficient, and highly visualized automated design scheme for corner bracing and anti-bracing structures within foundation pits.

This invention provides an image-assisted 3D object detection method, comprising the following steps: Step 1, extracting features from a point cloud containing the 3D object to be detected and its corresponding image; Step 2, fusing the features of the point cloud and the image based on a block self-attention mechanism to obtain fused multimodal features; Step 3, generating a foreground point mask based on the fused multimodal features and using it to extract multi-scale features with a focus on the foreground, obtaining feature maps of the point cloud at different resolutions; Step 4, generating an object detection proposal, i.e., interpolating the feature maps at different resolutions based on the three nearest neighbor interpolation method, fusing the interpolated features with the multimodal features using a multilayer perceptron, and feeding them into a two-stage detection module to generate an object detection proposal, thereby completing the image-assisted 3D object detection.

The application provides a liquid muscle model simulation method and device, electronic equipment and storage medium. The method comprises the following steps: obtaining detection parameters of a liquid muscle model to be simulated, the detection parameters comprising a resting length of an effective deformation part of the liquid muscle and a total length of a rope and an ineffective deformation part of the liquid muscle; determining a minimum relative length ratio of the liquid muscle that can be shortened and a maximum relative length ratio of the liquid muscle that can be stretched; calculating a lower limit of a tendon length of the liquid muscle and an upper limit of the tendon length according to the minimum relative length ratio, the maximum relative length ratio, the resting length and the total length; inputting the minimum relative length ratio, the maximum relative length ratio, the lower limit of the tendon length and the upper limit of the tendon length into a physical engine based on multi-body dynamics to obtain a liquid muscle simulation model of the liquid muscle model. The application can realize long muscle biomechanics simulation across multiple joints while reducing the training difficulty of a robot with a liquid muscle model.

This invention relates to the field of fluid dynamics data reconstruction and intelligent computing technology, specifically a method and system for super-resolution reconstruction of vortex flow fields based on residual U-Net. The invention first performs bicubic downsampling on the high-resolution flow field and interpolates it back to its original size to construct a degenerate input. This input is then fed into a network containing an encoder, bottleneck layer, decoder, skip connections, and global residual connections. The output residual field is added to the input to obtain the reconstructed flow field. During training, a combined loss consisting of mean square error and gradient loss is used for supervised optimization. This scheme can effectively recover small-scale vortices, shear layer boundaries, and high-gradient textures in low-resolution flow fields, reduce over-smoothing, and improve reconstruction accuracy and physical structure realism. It is suitable for recovering flow field data after compressed transmission, enhancing low-precision measurements, and reconstructing numerical simulation results.

The application discloses a kind of gas pipeline leakage detection methods based on acoustic signal, comprising the following steps:S1, the leakage acoustic signal on pipeline is acquired and is handled;S2, the feature of the acoustic signal of the leakage acoustic signal after processing with gas pipeline normal operating condition is analyzed, and the characteristic quantity sensitive to gas pipeline leakage condition is established and extracted;S3, the characteristic quantity of other time-frequency domain of leakage acoustic signal is extracted, and the feature set is constructed, and the optimal characteristic quantity is selected based on feature selection algorithm, and the optimal characteristic quantity is used as the input of support vector machine, and the model is trained;S4, for the signal of real-time acquisition, by steps S1-S3 processing, the model of training is used to identify the state of gas pipeline.According to the application, the dependency problem of characteristic quantity to specific pipeline is effectively solved, a kind of robust gas pipeline leakage detection method is established, the accurate identification of gas pipeline leakage is realized, and the safe operation of city gas pipeline network is guaranteed.

The method for super-resolution reconstruction of panoramic monitoring image of protection system of extra-high voltage converter station belongs to the technical field of power equipment detection, and solves the problems of unclearness and low resolution of the existing panoramic monitoring image, which cannot meet the demand of panoramic monitoring of inspectors; by adopting multi-scale convolution blocks to construct low-order and high-order features of images of various scales in the deep multi-scale residual network model, the phenomenon of incomplete extraction of image details is avoided, the residual learning mechanism is adopted in the network model to retain low-order rough features, reduce the training difficulty, promote the reuse of features, and thus improve the reconstruction capability of the image; the reconstructed image has better structural similarity and peak signal-to-noise ratio performance; the standard data set and the panoramic monitoring image data set of the extra-high voltage converter station are used in sequence for image super-resolution reconstruction and target recognition experiment, and the experimental results show that the high-resolution image reconstructed by the method can meet the demand of panoramic monitoring of inspectors.