72results about How to "Achieve training" patented technology

The invention discloses a fine-grained multi-modal prompt learning method based on a visual language pre-training model, and relates to the technical field of visual language pre-training models.The method comprises the steps that a multi-modal prompt learning vector used for learning visual features and text features is constructed, and a fine-grained multi-modal prompt learning model is constructed based on the fine-grained multi-modal prompt learning model; global differences, subtle distinguishing visual features and category difference text features among different visual categories are generated, and a multi-modal prompt learning vector is optimized based on a cosine similarity loss function. According to the invention, through combination of dual-granularity visual prompt and text prompt, the problems that in an existing prompt learning tuning method, cooperative reasoning cannot be carried out in a dual-mode prompt representation space, and fine distinguishing features are difficult to capture are effectively solved; and superior migration generalization ability and few-sample learning ability are verified on a plurality of main image identification reference data sets, and the performance is remarkably improved in downstream tasks.

The invention provides an agent construction method and system based on a large language model, a medium and equipment, and relates to the technical field of decision agent construction. According to the method, an appropriate large language model is selected, a series of prompt project texts are established for the large language model, target codes are generated as required by utilizing the large language model and prompt projects, and class library functions of a network model library, a reinforcement learning algorithm and a simulation training environment can be correspondingly called by running the codes. Finally, auxiliary construction and training of the decision-making agent are realized, and the problems in the existing decision-making agent construction and training process are solved.

The invention relates to a lightweight monocular depth estimation method and device based on self-supervised deep learning, and the method comprises the steps: obtaining a plurality of continuous frames of images, carrying out the reconstruction of a target frame through a trained depth estimation model based on the adjacent frames of the target frame, and obtaining the depth information of the target frame, the training process of the depth estimation model comprises the following steps: obtaining a monocular video sequence, and constructing training data by extracting continuous multi-frame images; predicting relative pose information of two adjacent frames of images through pose coding and decoding; through multi-scale feature extraction and dynamic up-sampling, obtaining a feature map with local detail features reserved, and obtaining a depth map based on the feature map; and performing frame reconstruction based on the relative pose information and the depth map, and calculating a loss function value on the premise of considering triple loss based on semantic perception so as to realize training of a depth estimation model. The method has the advantages that detail features are fully reserved during up-sampling, and the expression effect on gradient-rich areas is good.

The invention discloses a physical field prediction method based on a deep fuzzy physical information neural network, and the method comprises the following steps: S1, collecting physical field data, carrying out the preprocessing of the data with noise, and obtaining a processed training data set Dtrain and a test data set Dtest; s2, constructing a deep fuzzy physical information neural network model; s3, inputting a physical field training data set with Gaussian noise and a corresponding partial differential equation, and training the neural network model to obtain a trained neural network model; s4, physical quantities in different time-space coordinates in the test data set are calculated through the trained neural network model, and physical field prediction is achieved; the method solves the problems that the existing physical information neural network method does not consider the condition that noise exists in a training data set, the training effect of a deep neural network model is poor under the interference of a noise data set, and physical information in a partial differential equation cannot be accurately learned, so that the prediction performance of a physical field is seriously reduced.

The invention provides an unfavorable geological type identification method and system based on a joint deep learning network, and relates to the technical field of geological disaster forecasting. Comprising the following steps: acquiring training information including geological data of an excavation section, geophysical prospecting inversion imaging information and drilling information; inputting the training information into a pre-established joint deep learning network model, performing data preprocessing, feature extraction and feature fusion in sequence to obtain joint feature representation, and based on the joint feature representation, obtaining a discrimination type and a morphological description of the unfavorable geologic body; wherein the joint deep learning network model comprises a DNN model, a CNN model and an RNN model, and the DNN model, the CNN model and the RNN model are respectively utilized to extract the features of the digital data, the image data and the text description data in the training information. The method can quickly and accurately carry out type discrimination and form description on the unfavorable geologic body in front of the tunnel, and reduces the manual forecasting pressure.

The invention belongs to the technical field of data processing. The invention provides a spatio-temporal data prediction method and system for heterogeneous photovoltaic diagram structure data fusion, and the method comprises the steps: a current client obtains a photovoltaic power generation prediction result of the current client at each moment in the future after a current moment based on the historical photovoltaic power generation data of the current client and a trained global model; the global model training process comprises the steps that the central server distributes a global model pre-trained according to a photovoltaic generating capacity data set to all clients, and each client trains the global model through independent local graph data to perform local updating; each client uploads a local update result to a central server for aggregation, and the central server trains a GCN part of a global model according to an aggregation result and distributes the trained global model to each client for use; on the premise of ensuring data privacy of each client, updating and training of a global model are realized, and prediction comprehensiveness is greatly improved.

The invention relates to the technical field of image enhancement three-dimensional reconstruction, in particular to an urban building three-dimensional reconstruction method and system based on virtual modeling, and the method comprises the steps: obtaining a gray-scale map and a depth map of an urban building at different visual angles through employing a camera carried by an unmanned plane; obtaining the depth homogeneity of each pixel point in each depth map based on the similarity degree of the depth value change in the local window of each pixel point in each depth map; based on the spatial difference and gray level distribution difference of point cloud data of each visual angle and the adjacent visual angle thereof, and the gray level difference condition of pixel points at the same position in the gray level images of each visual angle and the adjacent visual angle thereof, the reflection standard difference and the depth homogeneity are combined; and obtaining a luminosity compensation weight of each pixel point in each visual angle grey-scale map to construct a loss function, and training a deep learning model to obtain a three-dimensional model of the urban building. According to the invention, the accuracy of virtual modeling three-dimensional reconstruction is improved.

The invention belongs to the technical field of three-dimensional target detection, and particularly discloses a mechanical assembly three-dimensional target detection method and system based on CAD model domain point cloud data, and the method comprises the steps: obtaining model domain point cloud and corresponding labels based on a CAD three-dimensional model of a mechanical assembly, and enabling the labels to reflect the types and positions of parts; training a detector and a domain discriminator through the point cloud data of the model domain and the real domain, and taking a maximum domain discriminator error and a minimum detector error as loss functions; the detector comprises a feature extractor and a label predictor, the feature extractor extracts point cloud features, the domain discriminator discriminates whether the features are from the model domain or the reality domain, and the label predictor outputs corresponding labels according to the model features. And inputting the real domain point cloud into the trained detector, and outputting a corresponding label. According to the invention, the parts of the assembled assembly can be accurately identified and positioned.

The invention discloses a hand holding structure of fitness equipment and a using method of the hand holding structure. The hand holding structure comprises a connecting base, a first dividing plate is fixedly connected to the left side surface of the connecting base, a second dividing plate is fixedly connected to the right side surface of the connecting base, a thrust roller bearing is fixedly connected to the outer surface of the connecting base, and a rotating shaft is rotationally connected to the outer surface of the connecting base; a first pin shaft is fixedly connected to the end of the right side of the left supporting arm, the first pin shaft is rotatably connected with the connecting base, the left supporting arm is slidably connected with the first index plate, and a first arc-shaped lower support is fixedly connected to the end of the left side of the left supporting arm. The front-back angle of the left supporting arm and the front-back angle of the right supporting arm are changed, the shape of the device can be flexibly adjusted according to using requirements in the using process of the device, and the using requirements of various fitness enthusiasts are fully met.

The adaptive tailings pond remote sensing deep learning detection method provided by the present invention is used for large-scale tailings pond identification and extraction. The present invention proposes a new tailings pond remote sensing extraction framework, namely: first, a feature pyramid network is used to perform deep feature learning of tailings ponds of different sizes based on a tailings pond remote sensing image data set; second, a cascade positioning optimization module is used to locate the tailings pond according to the characteristic value and use a multi-level detector to optimize the positioning coordinates; third, candidate detection objects are obtained according to the positioning optimization value, and multi-dimensional features are extracted and fused based on the deep learning algorithm and the traditional feature extraction algorithm to generate feature vectors; finally, an adaptive feature learning classifier is used to perform integrated learning and classification based on the feature vector, eliminate false alarm targets and extract tailings ponds. The present invention has the advantages of high positioning accuracy, low false alarm rate, and wide application range, and can provide an important basis for dynamic updating of tailings pond distribution information and safety management.

The invention discloses a leaping ability training device which comprises a leg muscle training mechanism, a lower bouncing mechanism, an upper bouncing mechanism and a lower bouncing mechanism. The upper bouncing mechanism is arranged at the top end of the lower bouncing mechanism; the leg muscle training mechanism comprises a base, a guide plate is slidably mounted at one end of the base, a supporting frame is mounted at the top of the guide plate, a foot squatting plate is mounted at the top of the supporting frame, and a spring for driving the foot squatting plate is mounted on the base; chair legs are detachably installed at the end, away from the foot squatting plate, of the base, a seat is installed at the top ends of the chair legs, and a safety rod acting on the chair legs is installed on the base. Leg muscles can be trained through the leg muscle training mechanism, so that legs are trained, and the subsequent bouncing effect is improved; the lower bouncing mechanism can improve the bouncing effect of the upper bouncing mechanism, and therefore bouncing training is conducted.

The embodiment of the invention provides an image storage method and device, electronic equipment and a storage medium, and is applied to the technical field of images. According to the method provided by the embodiment of the invention, the to-be-stored image can be encoded through the encoder of the deep learning model, the classifier classifies the encoded first image feature, and the decoder decodes the sampled first potential spatial feature to obtain the first decoded image. Calculating a second loss value according to the first decoded image and the to-be-stored image, and calculating a first loss value according to the first classification result and the extraction code; and adjusting parameters of the deep learning model through the loss value to realize training of the deep learning model. According to the image storage method and device, storage of the to-be-stored image is realized through training of the deep learning model, the first coding information after coding of the to-be-stored image is stored in the potential space, a storage device cannot directly view the to-be-stored image, and the image can be read only by extracting the code, so that the image storage security is improved.

The embodiment of the present invention discloses a training method, device, equipment and medium for an interface data generation model. The method includes: determining the basic model parameters of a pre-built basic interface data generation model, and determining the hardware parameters of the current server to which the basic interface data generation model is applied; updating the basic model parameters according to the hardware parameters to obtain a candidate interface data generation model after the initial update; inputting the current training data in the model training data set into the candidate interface data generation model, outputting the current interface data, and determining the model evaluation score of the candidate interface data generation model according to the current interface data; iteratively training the candidate interface data generation model according to the model evaluation score and the hardware parameters to obtain a target interface data generation model. The accuracy of training the basic interface data generation model is improved, thereby improving the accuracy of subsequently determining the interface display data based on the target interface data generation model.

The present disclosure relates to a training method, information recommendation method and device for an information recommendation model. The information recommendation model training method includes: obtaining initial sample features and actual operation results of sample accounts for sample recommendation information; the initial sample features are obtained based on sample account features and sample information features; the initial sample features are input into the information recommendation model to be trained to obtain predicted operation results of the sample account for sample recommendation information; the model includes a gating structure and multiple feature extraction layers, the multiple feature extraction layers are used to output multiple target sample features, the output value of the gating structure represents the importance of each target sample feature, and the predicted operation result is obtained based on the multiple target sample features and the output value of the gating structure; based on the difference between the predicted operation result and the actual operation result, the information recommendation model is trained to obtain a trained information recommendation model. The present disclosure can improve the recommendation accuracy of the trained information recommendation model.

The utility model discloses a super-nuclear machine training device, which relates to the technical field of sports equipment and comprises a bed body, a first platform, a second platform, a folding structure and a sliding mechanism, a first platform and a second platform are respectively arranged at two ends of the bed body; the first platform and the second platform are connected through a bed body; a slide way is arranged on the inner side of the bed body, the slide way is matched with the sliding mechanism, and the sliding mechanism is connected with the slide way in a sliding mode; one end of the sliding mechanism is detachably connected with the second platform or detachably connected with the first platform; the bed body comprises a first cross beam and a second cross beam; the first cross beam and the second cross beam are both of split structures, the folding structure is arranged at the joint of the split structures, and the bed body is folded through the folding structure. Compared with a traditional device, the folding design is more convenient to store, meanwhile, the device is suitable for families and gymnasiums, and the average-effect ratio of gymnasium operation can be effectively increased.

The application provides a model training method, a speech conversion method and device, equipment and a storage medium, and belongs to the technical field of artificial intelligence. The method comprises the following steps: obtaining sample audio data of a sample speaker object; inputting the sample audio data into a neural network model comprising an encoding network and a decoding network; performing reconstruction processing on the sample audio data by using the encoding network to obtain initial audio data; performing speech alignment on the initial audio data to obtain a sample audio embedding vector; performing decoupling processing on the sample audio embedding vector, pre-acquired sample pitch parameters and a sample timbre feature vector by using the decoding network to obtain synthesized audio data; performing loss calculation on the synthesized audio data and sample speech data by using a loss function to obtain a model loss value; and performing parameter updating on the neural network model according to the model loss value to train the neural network model and obtain a speech conversion model. The application can improve the speech conversion effect.

The present invention discloses a question-answering method, a question-answering model construction method, an electronic device and a storage medium. In the question-answering method, user question information is obtained; based on the question-answering model, an answer corresponding to the user question information is determined; wherein the question-answering model is constructed based on a question-answering material table, the question-answering material table includes at least one header tag attribute, each of the header attribute tags has at least one corresponding attribute data, and each of the header tag attributes has corresponding question-answering annotation information, the question-answering annotation information includes question annotation information or answer annotation information. Thus, the rapid construction of a question-answering model based on a table document is realized, and the constructed question-answering model can realize accurate responses to aggregate calculation type questions.

The invention discloses a psychological stress machine learning combination model classification method and device, equipment and a medium, and belongs to the technical field of psychological stress recognition. The method comprises the following steps: acquiring original ECG data, and acquiring HRV data according to the original ECG data; constructing a plurality of different classification models, and training each classification model by adopting the obtained HRV data; forming a combined model by the plurality of trained classification models, and obtaining a final classification result through a voting mechanism; wherein in the process of training the classification models, the HRV data and the multiple classification models are distributed to the multiple calculation units in a parallel calculation mode, so that each calculation unit independently trains one classification model. According to the method, various machine learning method models are combined, the advantages of various machine learning methods and deep learning methods can be combined, the generalization ability of model prediction classification is improved, and the prediction accuracy of the model is effectively improved.

The invention discloses a diabetic foot massage treatment training device and a training method thereof. A diabetic foot massage treatment training device comprises a pedal and further comprises a supporting frame, and the pedal is rotationally connected to the supporting frame. The supporting frame comprises an upper supporting plate and a lower supporting plate which are arranged up and down, the pedal comprises a main frame and a floating frame, and the main frame is rotationally connected with the upper supporting plate; the upper supporting plate is detachably connected with an arc-shaped guide column, the main frame is connected with the arc-shaped guide column in a sliding mode, the floating frame is provided with a strip-shaped receding opening used for receding the arc-shaped guide column, the arc-shaped guide column is sleeved with a spring, one end of the spring abuts against the upper supporting plate, and the other end of the spring abuts against the main frame. A longitudinally-arranged guide rail is connected to the main frame, the floating frame is slidably connected with the guide rail, the upper end and the lower end of the guide rail are detachably connected with limiting blocks, and floating springs are clamped between the limiting blocks and the floating frame. Massage stones are installed on the main frame, and gaps used for longitudinal sliding of the massage stones are formed in the floating frame.

The utility model relates to the technical field of rehabilitation training devices, in particular to a post-stroke balance function training device which comprises a seat plate, supporting legs, armrests, a backrest, a pedal, a controller and a supporting rod. Supporting legs are installed on the two sides of the seat plate, handrails are arranged on the two sides of the upper portion of the seat plate, a backrest is installed behind the seat plate, an air bag and a pedal are arranged on the seat plate, vibrators are arranged in the air bag and the pedal, a supporting rod is detachably installed behind the seat plate, and a staylace is arranged through the handrails. A vibrator is also arranged in the staylace, and a safety belt is arranged through the backrest. According to the utility model, the multi-stage intensity training of the balance function of a stroke patient can be realized, the recovery of core muscle groups of the patient is facilitated, meanwhile, the training device can be adjusted to be in an upright state, the training of the standing balance function of the patient is realized, the sitting and standing integrated training is realized, the use is convenient, and the positive effect on the recovery of the patient is realized.

The lower limb exoskeleton rehabilitation training robot comprises a support and a robot, the support comprises a supporting column, a cross beam, a light bar and a sliding rail, the left part and the right part of the support are symmetrically designed, the support is fixedly connected with the ground through the supporting column, and the cross beam is slidably connected with the light bar and the sliding rail; the robot comprises a waist fixing assembly, two hip training assemblies, two leg training assemblies and two ankle joint training assemblies, the waist fixing assembly is fixedly connected with the cross beam, and the hip training assemblies, the leg training assemblies and the ankle joint training assemblies are symmetrically designed. The hip training assembly is connected with the waist fixing assembly, the leg training assembly is connected with the hip training assembly and the ankle joint training assembly through keys, and the ankle joint training assembly is a three-degree-of-freedom association mechanism. The leg rehabilitation training and foot ankle joint three-degree-of-freedom training of a patient can be completed.

The present invention discloses an integrated microfluidic chip for capturing and incubating circulating tumor cells, belonging to the field of microfluidic technology. The integrated microfluidic chip of the present invention comprises an upper chip, a middle chip and a bottom chip connected in sequence from top to bottom. The first microcolumn array in the upper chip realizes the primary capture of circulating tumor cells. The drainage channel arranged in the middle chip is used to introduce the blood sample from the upper chip into the center of the bottom chip, so that the blood sample flows radially into the U-shaped microcolumn unit for secondary capture. The radially distributed U-shaped microcolumn units make the fluid flux very large, the blood sample flow rate is fast, and the capture rate is high. The present invention overcomes the shortcomings of the prior art microfluidic chips with low flux, insufficient number and purity of enriched tumor cells, resulting in inaccurate detection process. The integrated microfluidic chip provided integrates the capture function and the incubation function, and has high throughput and high capture rate.

The present invention discloses a training method, terminal and storage medium for accelerating a pulse neural network, the method comprising: obtaining a pulse sequence and an initial synaptic weight for training, generating an input signal of a pulse neuron according to the pulse sequence and the initial synaptic weight; obtaining an actual output pulse sequence generated based on the input signal; obtaining an expected output pulse sequence, updating the initial synaptic weight according to the actual output pulse sequence and the expected output pulse sequence, so as to realize training of the pulse neural network. The present invention reads image data through a host device and converts it into a pulse sequence and sets initial weights. Then, the most time-consuming parameter update task in the pulse neural network training is handled by a special acceleration device, which improves the speed of parameter update and accelerates the training process of the pulse neural network. Thereby solving the problem that the training process of the pulse neural network in the prior art is time-consuming.

The application provides a personalized brain-controlled mechanical arm rehabilitation system based on individual characteristics, comprising the following steps: S1, evaluating the performance of each MI-BCI training cycle of an individual; S2, researching the correlation between the individual's electroencephalogram features and functional indicators; S3, predicting the motor imagery ability according to the cycle rating and electroencephalogram feature quantitative data of the individual, and then establishing a mapping from the motor imagery ability to the mechanical arm movement rate; S4, constructing a personalized feedback regulation loop and establishing a complete closed-loop MI-BCI mechanical arm system. The application has the beneficial effects that: the personalized brain-controlled mechanical arm system based on individual characteristics predicts the motor imagery ability according to the performance of the training cycle and the electroencephalogram features, thereby predicting the mechanical arm rate, and the MI-BCI ability of the individual is displayed in real time by the mechanical arm rate. The individual's initiative is increased, the degree of assistance of the MI-BCI to the individual is reduced, and the rehabilitation effect is improved.

The invention relates to a platform for realizing intelligent rule management and control of accounting, and belongs to the technical field of accounting. The platform comprises a background support class module, an application service class module, an accounting knowledge base related module, a plurality of MCP servers, an agent main control module and a client module. The method comprises the following steps: based on an mcp protocol, taking a large language model as an intelligent core, and constructing an accounting knowledge base by integrating knowledge loaded by an approval process and knowledge in an accounting rule base in an accounting system; through a natural language interaction interface, accountants are assisted in interpreting policies, enterprise accounting criteria and related management requirements and matching with accounting rules, the accountants are assisted in establishing and adjusting the accounting rules, the working efficiency is improved, and the consistency and integrity of the accounting rules are guaranteed.

The present application relates to the technical field of computer vision and image processing, and discloses a remote sensing weakly supervised fine-grained target detection and recognition method and device, based on a general text prompt corresponding to a small amount of labeled remote sensing data samples and a large amount of unlabeled remote sensing data samples, by extracting the geometric features of the samples and the cross-modal features representing the visual text differences, combining the fine-grained class labels of the labeled remote sensing data samples, prior knowledge prototypes of various fine-grained classes are constructed; then, the prior knowledge prototypes of various fine-grained classes are used to construct fine-grained soft labels of the unlabeled remote sensing data samples, which are used as supervision signals to realize the training of the model, solve the cost consumption problem of fine-grained labeling, improve the label labeling efficiency of the unlabeled remote sensing data samples, and further improve the model training efficiency and target recognition efficiency.

The invention relates to the field of intelligent agriculture, and particularly discloses an intelligent greenhouse with an intelligent monitoring system and a use method thereof, the intelligent greenhouse comprises an environment observation module, a data transmission module, a data storage module, an environment data analysis module, a greenhouse equipment control module, a crop image analysis module and a crop growth analysis module; according to the intelligent greenhouse of the intelligent monitoring system and the use method of the intelligent greenhouse, through collection and analysis of greenhouse environment data and crop data, an optimization prediction algorithm is provided, greenhouse environment data in future time is predicted, work of equipment in the greenhouse is controlled, crop images are segmented, and crop growth conditions are identified and analyzed; and taking corresponding measures to regulate and control the growth of the crops based on the identified growth condition.

The invention discloses a physical informed machine learning high-entropy alloy phase prediction system and method based on semi-empirical parameters, and belongs to the technical field of crossing of high-entropy alloy material design and machine learning. The system takes semi-empirical parameter physical constraint as a core, integrates the physical interpretability of an empirical parameter method and the data driving advantages of machine learning through a three-stage cooperation mechanism of'feature multiplexing-independent prediction-Bayesian fusion ', and solves the problems of narrow phase coverage, low multi-phase prediction precision, 'black box' defect and the like of a traditional method. The system can predict more than 10 high-entropy alloy phase types and multi-phase coexistence systems, the single-phase prediction accuracy rate in 856 groups of multi-component high-entropy alloy test sets reaches 100%, the multi-phase coexistence system prediction accuracy rate reaches 77%, phase formation physical mechanism explanation can be output, B2 phase exclusive accurate criteria are provided, and high-entropy alloy design is promoted to be transformed from a trial and error method to an accurate prediction method.

The application provides a kind of flight landing gear actuator fatigue estimation method and system, comprising: obtaining the service history sample set of flight landing gear actuator;The service history sample set is pretreated, and the service state is marked;Respectively using GRNN, RBF and DNN are trained, obtain multiple actuator fatigue estimation models, and the model output is actuator failure flag;The prediction accuracy value of each model in the preset service time cycle is counted;The prediction accuracy value of each model in the preset service time cycle is weighted and summed to determine the weighted prediction accuracy;The model with the highest weighted prediction accuracy is obtained as the final fatigue estimation prediction model;The feature vector of the aircraft is used as input and input into the fatigue estimation prediction model, and the predicted actuator failure flag is output.The application can be used for the failure warning of the actuator of the aircraft landing gear in the service state, to reduce the defects of artificial judgment and subjective judgment.