35results about How to "Simplify the training process" patented technology

The invention provides a multi-degraded image restoration method based on self-adaptive prompt. The method comprises a two-stage hidden space diffusion model, and various degraded images can be flexibly processed. Specifically, the model generates a degradation cue by learning the difference between a degraded image and a clean image, and encodes the cue information in a hidden space. In the first stage, a degradation prompt is obtained through a degradation adaptive prompt learning mechanism as a key feature for guiding the diffusion model in the second stage to carry out image restoration. Finally, prompt information is integrated into a Fourier transform restoration network, and accurate restoration of different types of degraded images can be realized. The method not only improves the accuracy of image restoration, but also significantly reduces the calculation complexity, and is suitable for image restoration tasks in various real world scenes, including denoising, deblurring, defogging, rain removing, snow removing, raindrop removing and low illumination enhancement.

The invention discloses a visual detection optimization method and system based on data set classification. The method comprises the steps of introducing a scene classification label and carrying out standardization processing on scene features in a data set labeling and preprocessing stage, and carrying out scene classification multi-task learning in a model selection and training stage, so that a model dynamically adjusts feature learning in different scenes, and the multi-scene adaptive capacity is improved. And in the model evaluation and verification stage, scene-based hierarchical evaluation is introduced and corresponding optimization is carried out. In the model deployment and reasoning stage, a dynamic scene is introduced to adapt to reasoning, and the reasoning path of the model is dynamically adjusted to adapt to the characteristics of the scene, so that the reasoning accuracy is improved. In the later optimization, scene feature clustering and transfer learning are used, and the multi-scene performance of the model is optimized. According to the method, under the condition that the model is not jointly trained through multiple data sets, the multi-scene adaptive capacity of the visual detection model is improved through data set classification.

The invention belongs to the technical field of voice signal processing, and discloses a method and system for driving a virtual human by a voice signal, and the method comprises the steps: carrying out the resampling processing of original audio signal data, obtaining a voice signal, solving a linear prediction coefficient based on a dynamic framing strategy and a linear prediction analysis technology, and obtaining a linear prediction result; obtaining a voice feature sequence of the voice signal; performing enhancement processing on the voice feature sequence by using a dynamic feature enhancement technology, obtaining an enhanced feature sequence based on a processing result, and constructing an animation sequence prediction model by taking the enhanced feature sequence as input; and obtaining animation sequence parameters based on an output result of the animation sequence prediction model, and combining the animation sequence parameters with the virtual human role model to perform facial action synthesis and rendering operation. According to the invention, the earlier stage processing flow is greatly simplified, and the usability and applicability in the driving process are improved.

The invention provides an ocean state variable forecasting method based on thermocline adaptive loss, and belongs to the technical field of ocean meteorology. The self-adaptive thermocline loss function designed by the invention can dynamically adjust the weight of the loss function according to the intensity and complexity of the temperature change of the thermocline region, and the self-adaptive mechanism enables the model to pay more attention to the temperature gradient change characteristics of the thermocline region in the training process; therefore, the model is guided to learn and simulate the complex evolution process of the thermocline more accurately. Compared with the prior art, the traditional loss function gives the same weight to the temperature change of all regions, and the importance of the thermocline region cannot be highlighted, so that the model has a poor effect when learning and simulating the complex evolution process of the thermocline; according to actual verification, the accuracy and stability of thermocline forecasting are remarkably improved, meanwhile, the model training process is simplified, and a more efficient and reliable thermocline forecasting tool is provided for marine scientific research and practical application.

The invention discloses a zero-sample food image detection method based on visual semantic bidirectional guidance, which belongs to the technical field of image data processing and generation, takes DETR based on Transform as a detection framework, and fuses a food semantic enhancement network and a cross-modal bidirectional guidance module. Firstly, an image is input to an image backbone network branch to extract visual features, a category label is input to a food semantic enhancement network branch to extract semantic features, then the two obtained features are sent to a cross-modal two-way guide fusion module, the module comprises a two-way guide encoder, a two-way guide query vector and a two-way guide decoder, and the cross-modal two-way guide fusion module comprises a two-way guide query vector and a two-way guide decoder. Visual and semantic features are aligned and fused through coding and decoding processes, and finally, regression and classification tasks of zero sample food image detection are completed in a full connection layer. According to the method, detailed food attributes and visual features can be captured in the detection process, effective interaction of visual semantic modals is achieved, and therefore the detection precision of unseen food categories is improved.

The invention discloses an adjusting and optimizing method and system based on a large power dispatching model, and belongs to the technical field of power dispatching, and the method comprises the following steps: obtaining historical power dispatching knowledge, and dividing the historical power dispatching knowledge into regional knowledge and global knowledge based on regional characteristics; constructing a first teacher model based on the difference between the global knowledge and the regional knowledge, and constructing a student model based on the lowest fluctuation cost; obtaining regional scheduling correlation according to historical power scheduling knowledge, and constructing a second teacher model according to the regional scheduling correlation; associating the first teacher model, the second teacher model and the student model based on knowledge distillation, and constructing a large power dispatching model; and outputting a tuning strategy according to the large power dispatching model and the current power data. The beneficial effects of the invention are that the output of the power tuning strategy gives consideration to pertinence and efficiency.

The invention provides a photovoltaic output prediction method considering haze / dust factors. The method comprises the following steps: firstly, preprocessing data, screening out abnormal data and invalid data, and complementing missing data; the method comprises the following steps: firstly, analyzing haze / sand dust data, then classifying weather data by using a decision tree algorithm, then deeply understanding an internal rule and an operation mechanism by analyzing a haze / sand dust formation mechanism and a photovoltaic power generation principle, determining main characteristics, and then firstly establishing a haze / sand dust influence model 1 and a photovoltaic power generation power prediction model 2 for the haze / sand dust data; the model 1 obtains influence parameters of haze / dust factors on the photovoltaic power generation power, and the model 2 outputs a predicted value of the photovoltaic power generation power. The advantages of mechanism analysis and a data driving model are combined, and the accuracy and reliability of the model are improved.

The invention belongs to the field of computer vision, and discloses a multi-target detection method, a multi-target detection model training method, an edge computing device and a medium, and the multi-target detection model training method comprises the steps: extracting a first-class feature map; respectively inputting the first type of feature map into a first target detection module, a multi-layer feature fusion module and a second target detection module; calculating a target loss function value based on the obtained target detection result and the target labeling result; and adjusting network parameters of the to-be-trained model until a preset model training convergence condition is met, and obtaining a multi-target detection model. According to the method, the generalization ability of the model is improved, and the driving safety is enhanced.

The present application discloses a method, device, equipment and medium for predicting the auxiliary jet air consumption of an air jet loom, and belongs to the field of industrial automation technology. The method comprises: obtaining the structural parameters to be measured of the nozzle to be measured in the air jet loom, the structural parameters to be measured include the inlet diameter, the straight tube diameter, the nozzle taper, the elliptical outlet major axis and the elliptical outlet minor axis; inputting the structural parameters to be measured into the air flow prediction model, and obtaining the nozzle air consumption of the nozzle to be measured output by the air flow prediction model, and the air flow prediction model is obtained based on the sample structural parameters and air consumption label training of the sample nozzle. The method can accurately predict the air consumption of the nozzle in a complex industrial environment.

This invention provides a method for the germination and regeneration of mature rubber tree embryos into seedlings outside of test tubes, belonging to the field of seedling cultivation technology. The method uses somatic embryos derived from anther tissue culture as raw materials, and involves three stages of cultivation in a growth medium: dark culture, low-light culture, and strong-light culture. Then, the mature embryos that meet the requirements undergo closed-bottle / open-bottle acclimatization. After this acclimatization, they can be rooted and grown into seedlings in river sand outside the test tube. This method allows mature embryos to germinate and regenerate directly into seedlings outside the test tube, eliminating the need for a sterile culture room and specialized rooting medium, thus requiring less specialized knowledge from the operator. Furthermore, by combining the embryo germination and seedling regeneration with the in-test-tube and sand-bed acclimatization in one step, the cultivation process is simplified, the cultivation cycle is shortened by two months, and cultivation efficiency is improved while production costs are reduced.

The invention provides an oil-water separation effect prediction method and device based on a Lonion model. The method comprises the following steps: defining an input independent variable vector f (x) = [x1, x2,..., x8] T; performing feature engineering extension on the input independent variable vector to generate an extended feature vector f (z); the extended feature vector f (z) comprises an original feature vector, a polynomial feature vector and a physical interaction feature vector; inputting the extended feature vector f (z) into a pre-trained prediction model to obtain a predicted value, output by the pre-trained prediction model, of the oil content of the effluent of the air flotation synergistic device; inputting the predicted value into an integrated prediction framework to obtain a final predicted value and uncertainty estimation; displaying the final predicted value and the uncertainty estimation to a user through a display interface; according to the technical scheme, the accuracy, robustness and physical consistency of oil content prediction of the effluent can be improved.

The utility model discloses a portable bus duct withstand voltage test device, which relates to the bus duct test field and comprises a withstand voltage tester, a quick-change test connector and a bus duct, the quick-change test connector and the withstand voltage tester are electrically connected with each other, and the outer end of the quick-change test connector can be installed in the bus duct in a plug-in fit manner for wiring; a plurality of test claw heads are fixedly mounted at the outer end of the quick-change test connector, contact copper sheets and springs are arranged on the test claw heads, a plurality of bus duct conductors are arranged in the bus duct, the test claw heads are inserted into the bus duct conductors in a one-to-one correspondence manner, and the contact copper sheets are propped against the bus duct conductors through the springs to form firm connection; compared with the prior art, the beneficial effects of the utility model are that the portable design can facilitate the implementation operation of a withstand voltage test; the design of the quick connector can realize plug and play, so that the use steps and related training are simplified; and the fool-proof design does not need to manually butt the conductors one by one, so that the problem of wrong connection of the conductors is avoided.

The present application relates to the technical field of signal sparse representation, and specifically provides a separation type redundant dictionary learning algorithm for 3D signals, which comprises the following steps: S1: initialization process: initializing redundant dictionaries in three dimensions; S2: main iteration process: comprising a sparse coding stage and a dictionary updating stage; the sparse coding stage comprises: S21: reducing the sparse expression process to a two-dimensional matrix; S22: calculating the expression of samples in the sparse domain according to the initialized redundant dictionaries; S23: calculating the sparse expression error of all samples; the dictionary updating stage comprises: S24: arranging the 3D training sample data blocks into two-dimensional matrices after being respectively unfolded according to three dimensions; S25: updating the redundant dictionaries in three dimensions by solving a formula and performing atom normalization on the dictionaries. The learning algorithm in the present application reduces the time consumption of separation type dictionary training and reduces the sparse expression error of target signals.

The present invention discloses a cell state detection method, device and storage medium based on spatiotemporal feature fusion, the method comprising the following steps: collecting multiple cell images within a set acquisition time period, using a target detection network to identify individual cells in each cell image, and extracting cell features of all individual cells; assigning the same time code to all cell features in each cell image according to the acquisition order of the cell images within the acquisition time period; inputting the cell features with the time code into a self-attention network Transformer, using the time code as the position code of the cell features in the Transformer network, and obtaining the predicted probability of the state of each cell image in the multiple cell images. The method of the present invention combines the temporal features and spatial features of cells, provides more comprehensive cell state information, and can more accurately predict the changes in the cell division state.

This invention provides a multi-layer memory-enhanced neural network architecture, training method, device, and medium, including a backbone network and an attractor network module. The backbone network extracts features from the input data and generates feature representations for the intermediate layers. The attractor network is connected to the intermediate layers of the backbone network and adopts a Hopfield network structure. Through iterative operations, it enhances the memory of the intermediate layer features of the backbone network, enabling the intermediate layer features to recover to a stable feature pattern even in noisy environments. By embedding the attractor network module into a traditional neural network architecture, this invention effectively solves the problem of insufficient robustness of existing deep neural networks when facing noise or unknown perturbations. The attractor network can memorize the feature patterns of the intermediate layers of the backbone network and correct features affected by noise during the inference stage, causing them to converge to a stable memory pattern. This characteristic makes the model more stable and reliable in complex environments.

The present invention belongs to the field of fruit disease detection and analysis, and specifically provides a multi-source teacher distillation and entropy reduction distribution platform system for fruit disease detection, including the following modules: an integrated device module uses an integrated device to collect image data of target fruits; a data processing module performs intelligent spectral information enhancement processing on the hyperspectral images and videos respectively; a model deployment module distills the model for the hyperspectral image data to obtain a model that can be deployed on the edge end; a disease detection module splits the teacher model by coordinating the edge ends of different data types through the master control, obtains a sub-model obtained by splitting from multiple edge ends, and uses the sub-model to perform disease quality detection tasks on fruits on a conveyor belt. The present invention collects hyperspectral images and videos of fruits through integrated equipment, and sends them to each edge device for preprocessing, and then coordinates the edge ends of different data types through the master control to split the distillation model, and quantifies and deploys it on the edge device for detection.

The invention discloses a visual content generation method and a training method and device of a visual content generation model. The visual content generation method comprises the following steps: decoding discrete feature representation to obtain first visual data matched with target semantic content; obtaining a first continuous feature representation of the first visual data; and performing reconstruction processing on the first continuous feature representation through a visual content generation model to obtain a visual content generation result of the discrete feature representation. Compared with a method of directly performing reconstruction from discrete feature representation in related technologies, the method performs reconstruction from continuous feature representation, and the continuous feature representation contains richer spatial and semantic information, so that the complexity of executing a reconstruction process by the visual content generation model is remarkably reduced, and the reconstruction efficiency is improved. The efficiency of visual content generation based on discrete feature representation is improved.

The invention relates to a generation method, a processing method, a device, equipment, a storage medium and a program product, and the generation method comprises the steps: carrying out the feature extraction of the image content of a first image, obtaining a first feature map, and carrying out the feature extraction of the style content of a second image, obtaining a second feature map; performing alignment processing on the content features in the first feature map and the style features in the second feature map, and obtaining a first prediction image based on the aligned content features and style features; inputting the first image and the second image into a first model to obtain a second prediction image; wherein the first model is used for image color mapping; and on the basis of the difference information between the first prediction image and the second prediction image, the model parameters of the first model are updated until the target model is obtained, so that data set acquisition is simple, the difficulty of target model development is simplified, meanwhile, memory occupation is reduced, and the use performance of the electronic equipment is improved.

The present application provides a method and system for predicting a coronal mass ejection event, the method comprising: collecting relevant data of CME within a set time period, obtaining characteristic parameter values ​​of 10 coronal mass ejection events after preprocessing; inputting the characteristic parameter values ​​of the 10 coronal mass ejection events into a pre-established and trained CME judgment model to obtain a prediction of whether the CME will reach the earth; based on the parameter weights obtained by the CME judgment model, the characteristic parameter values ​​of the 10 coronal mass ejection events are matched with the historical events in the pre-established CME historical database to find the closest historical event; for the CME predicted to reach the earth, the arrival time of the earth is calculated. The advantage of the present application is that in terms of ensuring the completeness of the input features, it largely solves the accuracy and refinement requirements of the prediction model, and has special reference significance for the current space weather business forecast.

The present disclosure relates to the field of computer technology, and specifically to a behavior recognition method, a behavior recognition device, a storage medium, and an electronic device. The behavior recognition method includes: obtaining multiple skeletal data features of a target object; using a branch displacement convolution subnetwork to perform branch processing on the multiple skeletal data features to obtain multiple feature maps; using a fusion displacement convolution subnetwork to perform mainstream fusion on the multiple feature maps to obtain a feature vector; and using a fully connected layer to perform recognition based on the feature vector to obtain the behavior information of the target object. The behavior recognition method provided by the present disclosure can solve the problem of large amount of calculation in the prior art when performing behavior recognition based on bones.

The invention discloses a training method of a target key point detection model, a detection method, equipment and a medium. The method comprises the following steps: acquiring a detection target, detection information of the detection target, multiple image features and multi-scale image coding features; target image features are determined based on the multiple image features and the multi-scale image coding features, and a learnable query vector group is constructed for the detection target based on the position information of the detection target; outputting predicted key point information through an initial target key point detection model based on the target image features and the learnable query vector group; and training an initial target key point detection model based on the predicted key point information and the marked key point information. According to the method, the training process of open set detection and the training process of attitude detection are decoupled, the target key point detection capability is added on the basis of the open set detection network, the training process is simplified on the basis of keeping the model universality capability of open set detection, and the training cost is saved.

The application provides a WGAN-based FRP sheet and concrete interface bonding slip model generation method, which can automatically, quickly and accurately obtain the bonding slip model and has a wide application prospect. After the automatic and rapid acquisition of the bonding slip model, the bonding performance between the FRP sheet and the concrete can be accurately reflected, and a safe and reliable reinforcement component design calculation method can be established. The WGAN is used to predict the strain, so that the training process can be simplified and the training can be stable. The WGAN can replace the traditional experimental analysis to quickly and accurately establish a bonding strength model. The LSTM is used as the generator of the WGAN model, so that the gradient disappearance and explosion problems of the RNN are solved, and the strain data related to time can be more accurately predicted. The CNN is used as the discriminator of the WGAN model, so that the quality and convergence speed of the generated samples are improved.

The invention discloses a method for in-vitro amplification of antigen-specific T cells based on general type II peptides, belongs to the technical field of biology, and provides the method for in-vitro amplification of the antigen-specific T cells and a kit used for the method. Flow Tetramer detection can obviously improve the proportion of specific antigen-specific T cells, three different cell factor matching modes and concentration ratios are compared, a group of relatively excellent factor combinations is optimized, flow Tetramer detection can obviously improve the proportion of specific antigen-specific T cells, the influence of polypeptide stimulation times in specific time is compared, and the specific antigen-specific T cells can be more accurately detected. Results show that the proportion of the specific T cells amplified through twice polypeptide stimulation is superior to that of the specific T cells amplified through once polypeptide stimulation, and the method is suitable for preparing the virus peptide, TAA and TSA antigen specific T cells.

The invention belongs to the technical field of civil aviation flight safety, particularly relates to a multi-visibility-based visual attention distribution strategy establishment method, system and equipment, and aims to solve the problem that a pilot lacks effective attention resource distribution strategy guidance in a final approaching landing stage. The operation risk of the flight stage is increased; and the flight training quality and the flight performance are possibly influenced. The method comprises the following steps: acquiring eye movement data and height data; dividing a region of interest and a height interval; obtaining the mapping position of each region of interest in different height intervals; screening out effective eye movement data of the expert pilot in different height intervals; acquiring a change trend of attention distribution and a differentiated attention distribution strategy; and judging the consistency of the differentiated attention distribution strategy and the flight control result, and guiding the attention distribution strategy of the pilot. According to the method, the visual attention distribution strategy can be optimized, a support is provided for a pilot to make a correct decision under a complex meteorological condition, and the flight performance is improved.

The invention discloses a honeycomb-free large-scale MIMO unmanned aerial vehicle associated power control method and equipment based on reinforcement learning, and a medium, and relates to the technical field of wireless communication. The method comprises the following steps: acquiring parameters of a communication system containing multiple unmanned aerial vehicles and access points; a cellular-free large-scale MIMO transmission model is constructed; the method comprises the following steps: determining a state space, an action space and a reward function based on a cellular-free large-scale MIMO transmission model, and respectively constructing intelligent agents for an uplink power control coefficient optimization problem of an unmanned aerial vehicle in a system and an access point unmanned aerial vehicle clustering optimization problem; training the intelligent agent by using a depth deterministic strategy gradient reinforcement learning algorithm; and obtaining an uplink power control coefficient of the unmanned aerial vehicle in the system and an unmanned aerial vehicle clustering result of each access point by using the intelligent agent based on an optimized uplink power distribution strategy and an access point unmanned aerial vehicle clustering strategy. According to the method, efficient joint optimization of unmanned aerial vehicle access point association and uplink power control can be realized under a rapidly changing channel condition, and the method has relatively high practical practicability under a low-altitude economic background.

The invention provides a Parkinson's disease patient demand identification method based on a language enhancement technology, and belongs to the technical field of deep learning language processing. Firstly, continuous vowel sounds and repeated syllable sounds of a Parkinson patient are collected; preprocessing the voice data to obtain standard input data; inputting the data into a model obtained by training, and outputting a demand category of the patient; the designed demand identification model comprises an encoder module, a decoder module and a patient demand identification module; inputting the continuous vowel data and the repeated syllable data into a trained encoder module to obtain fusion features; inputting the fusion features into a trained patient demand identification module to output a patient demand category; the decoder module is used for obtaining reconstructed features, and the reconstructed features utilize a loss function to restrain and guide the expression of the encoder output features. According to the language enhancement technology designed by the invention, through a series of innovative methods, the accuracy and efficiency of identifying the requirements of the Parkinson's disease patients are effectively improved.

The application discloses a kind of based on metric and deep learning's radiation source radio frequency feature extraction method, it belongs to radiation source radio frequency feature extraction technical field.The present application solves the problem that it is difficult to extract RF feature with good stability and separability using existing method, and the extracted RF feature is easily disturbed by IM information and fails.The method of the present application is: constructing modeling dataset and ideal training dataset;Each RF signal sample in the constructed dataset is processed to obtain the processing result corresponding to each RF signal sample;The AE network built is trained using the processing result of the RF signal sample in the ideal training dataset;The FRM network is trained using the processing result of the RF signal sample in the modeling dataset and the trained AE network;After processing the RF signal to be detected, the processing result is input into the FRM network constrained by the signal feature encoder output by the trained, and the radiation source radio frequency feature extraction result is obtained.The present application can be applied to radiation source radio frequency feature extraction.

The invention discloses an emotion analysis method based on a convolutional neural network and a YOLO algorithm, and the method comprises the steps: collecting data, and carrying out the preprocessing of the data; constructing a convolutional neural network model; training and adjusting the model; training and optimizing the model; identifying and reasoning the emotion; and analyzing and fusing the emotion. Through the convolutional neural network, different data modes and problems can be automatically learned and adapted, features can be automatically extracted from input data, complex feature engineering does not need to be manually carried out, the YOLO algorithm is relatively simple in structure and easy to understand and implement, meanwhile, the training process is simple, and emotion analysis can be comprehensively and accurately carried out.

This invention provides a method and system for predicting multimodal drug interaction events based on compact representation learning. The method includes: acquiring multimodal features of each drug in a drug pair to be predicted, wherein the multimodal features include at least two of biological features, molecular structure features, and knowledge graph features; performing intramodal compact representation learning on each modal feature of each drug to eliminate redundant information within a single modality and generate compact sub-representations for each modality; using the compact sub-representations corresponding to the molecular structure features as anchors, performing cross-modal alignment of the compact sub-representations corresponding to the biological features and knowledge graph features through mutual information minimization constraints to eliminate intermodal redundant information; constructing a multimodal fusion representation for each drug based on the compact sub-representations of each modality; concatenating the multimodal fusion representations of the two drugs in the drug pair and inputting them into a classifier to output the prediction result of the interaction event type of the drug pair.

The application discloses a kind of train transmission motor, gear box and axle box fault diagnosis method and system, belong to train transmission system fault diagnosis technical field, method includes respectively designing special expert network for the key components of train transmission system;Introduce multi-head self-attention gate network, dynamically fuse the features output by special expert network;Adopt the joint loss function including cross-entropy loss, KL divergence regularization loss, load balancing loss, train special expert network and gate network;Based on teacher-student model architecture and knowledge distillation, realize new fault category adaptation and historical knowledge retention;The multi-component vibration signal to be diagnosed is input into the trained model, and the probability of the highest fault category is output as the diagnosis result through feature extraction, gate network fusion and classifier prediction.