201results about How to "Guaranteed training effect" patented technology

The invention relates to an adaptive image denoising method based on depth learning, comprising the following steps: step 1) establishing an image set; 2) constructing an adaptive depth convolution neural network; 3) training the adaptive depth convolution neural network: setting the learn rate and momentum parameters of the depth convolution neural network, training the adaptive depth convolutionneural network through the depth learning framework until the train reaches the maximum number of iterations, and generating the trained adaptive depth convolution neural network model; 4) image denoising: inputting the image to be denoised into the trained adaptive depth convolution neural network model to obtain the corresponding residual image, and subtracting the image to be denoised from theresidual image to obtain the denoised image. The method further improves and stabilizes the training performance of the convolution neural network, and ensures the denoising performance, saves time,does not need to manually adjust parameters, and can realize blind denoising of the unknown noise level image.

The invention provides a model training method based on knowledge distillation, which is applied to a student model, and comprises the steps of setting a second output layer which is the same as a first output layer at a distillation position according to the distillation position; obtaining a training set, wherein the training set comprises a plurality of training data; obtaining first data output by the first output layer and second data output by the second output layer based on the training data; obtaining supervision data output by the teacher model in a teacher layer corresponding to thedistillation position based on the training data, wherein the teacher model is a complex model which completes training and completes the same task as the student model; based on the difference between the supervision data and the first data and the second data, obtaining a distillation loss value according to a distillation loss function; and updating parameters of the student model based on thedistillation loss value. Through the disclosed embodiment, the teacher model in knowledge distillation is more emphasized to transmit the knowledge of simple data to the student model, the training efficiency of knowledge distillation is improved, and the accuracy of the student model is ensured.

The invention relates to the technical field of vocational education, and particularly discloses a vocational training system and method and a business model, and the vocational training system comprises a login module, a training module, an assessment module and a statistics module. The training module comprises an offline training unit and an online training unit. The offline training unit is initiated by a training responsible person, and is used for providing training contents for students or recording learning contents of the students, monitoring the training process of the students and recording the learning duration, and the offline training unit supports online batch sign-in, sign-off and sampling inspection of a computer or a mobile phone. The assessment module is used for verifying student training results, generating test questions based on the training contents, performing automatic examination paper marking and generating examination scores. The statistics module is used for summarizing and analyzing data based on the statistics information of the training module and the examination module, wherein the statistics information comprises learning duration and examinationresults. By the adoption of the technical scheme, online training and offline training can be integrated.

Disclosed is a rehabilitation training bed for a stroke patient. The rehabilitation training bed comprises a bedstead, a support plate, a control mechanism and a driving motor set; the support plate is composed of a fixed part and a rotating part which are hinged to each other, one end of the support plate is provided with a head and neck support, the head and neck support is of an inflatable structure, and the outer surface of the head and neck support is provided with electrode sheets capable of releasing stimulating pulses; a power mechanism and a push-pull rod are arranged below the rotating part to achieve upward pushing and downward returning of the rotating part. The support plate is provided with forearm fixing plates and calf fixing plates, wherein electrode sheets are distributed on the surfaces of the forearm fixing plates and the surfaces of the calf fixing plates; the driving motor set can drive the upper limbs or the lower limbs of the patient to make left-right swing and forth-back extension movement through the forearm fixing plates and the calf fixing plates. The rehabilitation training bed for the stroke patient can automatically conduct sitting-up and lying-down training, constant changing of the functional positions of the upper limbs and the lower limbs and pulse massage stimulation on the stroke patient, so that muscle spasm and atrophy of the stroke patient who is bedridden for a long time are effectively avoided, and the rapid rehabilitation of the stroke patient is promoted.

The invention discloses a 360-degree omni-directional overload flight simulator. The center of gravity of a trainee is in the front of the center of the gravity of an airplane like in the real flight process. The 360-degree omni-directional overload flight simulator comprises a vertical lifting system, a first movement system, a second movement system and a flight cockpit system, wherein an intelligent control system has control over a front-and-rear pitching motor and a left-and-right rotating motor, the center of gravity of the trainee is in the front of the center of gravity of the airplane like in the real flight process, and the pilot can feel transverse movement and rotation at the same time; the intelligent control system converts various rotation overload motions of the vision into electric signals, and controls a stepping motor to synchronously move in the X direction, the Y direction and the Z direction. When the 360-degree omni-directional overload flight simulator is used, the training cost can be reduced, the training effect can be ensured, and the airplane driving training time of trainees such as trainee pilots can be shortened.

A method and system for automatically training a machine learning model is disclosed. The method performed by a computing device for automatically training a machine learning model includes the stepsof acquiring a data set including a plurality of data records; determining a feature engineering strategy for the data set through meta-learning; utilizing the determined feature engineering strategyto generate features related to each data record in the data set; constructing a machine learning sample based at least on the generated features; and training a machine learning model by using the constructed machine learning sample based on the automatically obtained hyper-parameter value.

The embodiment of the invention discloses a semantic segmentation model training method and device, equipment and a medium. The method comprises the following steps: training a first semantic segmentation model based on a first sample image and an image class label corresponding to the first sample image; inputting a second sample image into the trained first semantic segmentation model, and obtaining a first pixel semantic tag corresponding to the second sample image according to the output of the trained first semantic segmentation model; based on a saliency map corresponding to the second sample image, correcting the first pixel semantic tag, and determining a corrected second pixel semantic tag; and training the second semantic segmentation model based on the second sample image and the second pixel semantic tag, and taking the trained second semantic segmentation model as a target semantic segmentation model. Through the technical scheme of the embodiment of the invention, the dependence of the semantic segmentation model on dense annotation data can be reduced, and the pressure of data annotation is reduced.

The invention provides a human motion capture system, which comprises a plurality of inertial motion sensors for collecting motion data, a temperature sensor for collecting temperature, a heart rate sensor for collecting heart rate, a data coordinator and a data fusion module. The plurality of inertial motion sensors, the temperature sensor, the heart rate sensor and the data fusion module are connected to the data coordinator. The data coordinator receives the data obtained from the plurality of inertial motion sensors, the temperature sensor and the heart rate sensor. After processed by data fusion module, the data is sent to a VR device through the data coordinator. The nodes of the human motion capture system are small, are easy to carry and can be arbitrarily changed, thus human motion capture system is convenient to use. The human motion capture system and the VR device can be used for rehabilitation training, and can remove the requiremet for sites and trainers, so that patients can experience the effect of immersion training. The training effect is easy to be guaranteed, and the cost is relatively low.

The invention discloses a modulation mode recognition method based on spatial-temporal feature extraction deep learning. The method comprises the following steps: collecting a signal of a to-be-identified modulation mode; constructing an automatic modulation identification deep learning model comprising a parameter estimation module, a parameter change module and a spatial-temporal feature extraction module, and training the automatic modulation identification deep learning model; and performing modulation mode identification on the collected signals by adopting the trained automatic modulation identification deep learning model. In order to solve the problems that a modulation identification model in the prior art is relatively high in complexity and high identification accuracy is difficult to realize under the condition of low model complexity, the invention provides a spatial-temporal feature extraction automatic modulation mode identification deep learning model based on parameter estimation and transformation. The parameter quantity of modulation mode identification by using the model is less than that of an existing automatic modulation identification method based on deep learning, and the training overhead is lower than that of other methods with the same identification accuracy level.

The invention discloses a selection method for a marking strategy, and the method comprises the steps: carrying out the simulated marking processing and supervised learning training of a marked data set through employing a plurality of marking strategies, carrying out the training of a regression prediction model according to the index data in the process, and obtaining corresponding regression prediction models; carrying out the prediction calculation according to the index data through all regression prediction models when the index data of a marking task is obtained, and obtaining a prediction result; determining the marking strategy with the minimum manpower cost as the marking strategy to be used according to a prediction result and an expected result. The index data in the simulatedmarking processing is obtained and trained for obtaining the regression prediction models, the method can achieve the prediction of the marking strategy, and the marking strategy with the minimum manpower cost is determined according to a prediction result, thereby reducing the manpower cost while guaranteeing the training effect. The invention also discloses a selection device for the marking strategy, a server, and a computer readable storage medium, which all have the above beneficial effects.

The invention discloses a controllable safety training grenade, which comprises a grenade body, a control unit for locating the falling point of the grenade and controlling the detonation of the grenade is installed in the grenade body, and the control unit is installed in the grenade In the cavity of the body, the control unit is connected with the explosive in the grenade body. By setting the control unit in the grenade body and combining the wireless signals sent by the reference nodes pre-arranged in the grenade training throwing field to calculate the location of the grenade, the location of the grenade will be obtained. position, the grenade will be detonated. If it is detected that the grenade has not fallen to a safe position, the control unit will cut off the power supply to prevent the grenade from detonating. The invention has the advantages of not only ensuring the training effect, but also preventing potential safety hazards caused by the grenade falling too close to the grenade and accidental injury caused by the phenomenon of accidental injury caused by the grenade dropping or throwing by mistake.

The invention relates to an image recognition model generation method and device, a storage medium and electronic equipment, and the method comprises the steps: taking to-be-labeled image samples corresponding to human body parts as the input of a pre-trained sub-model, so as to obtain the recognition probability of whether each to-be-labeled image sample contains the human body parts or not; screening out a target sample from the to-be-labeled image samples according to the recognition probability and the image similarity between each to-be-labeled image sample and a plurality of labeled image samples corresponding to the human body parts; obtaining a labeled image sample corresponding to each target sample; and retraining the sub-model through the plurality of labeled image samples and atest set to generate an image recognition model. According to the method, the recognition probability of the image samples can be determined through the pre-trained recognition model, and the model is retrained through the pattern sample screened according to the recognition probability and the similarity characteristics of the image samples, so that the image annotation cost is reduced while themodel training effect is ensured.

The invention provides a classification model training method and device and a data classification method and device. The classification model training method comprises: acquiring a sample data set, the sample data set comprising at least three category labels and feature data corresponding to the category labels, and counting the proportion of the number of each category label in the sample dataset; according to the proportion of the number of each category label in the sample data set, dividing the category labels in the sample data set into at least two sample groups; and inputting the sample group into a corresponding classification model for training until a training condition is met. According to the method, the classification labels in the sample data set are unbalanced in proportion, the quality of the processed sample data set is greatly improved, the training effect of the classification model can be further ensured, and the classification accuracy of the trained classification model is greatly improved during actual classification prediction of the trained classification model.

The invention discloses a hip joint segmentation model building method using small sample image training and application thereof, and belongs to the medical image processing field. The method comprises the following steps: projecting small sample CT data marked with a hip joint area from different angles through a digital reconstruction radiographic image method to obtain a lot of analog images with marks; forming a pre-training data set by the analog image and the hip joint area marked in the analog image; establishing a segmentation model for performing image segmentation on the input image to extract a hip joint area, and pre-training the segmentation model by using the pre-training data set; and constructing a training data set by using the small sample Xray image marked with the hip joint area, training the pre-trained segmentation model to finely adjust the segmentation model, and taking the segmentation model as the established hip joint segmentation model after the training is finished. The technical problem that an existing hip joint segmentation method is poor in segmentation result due to the fact that the existing hip joint segmentation method depends on a large amount of annotation data can be solved.

The invention discloses a thin-walled workpiece cutter back-off deformation error prediction model building method and application thereof, and belongs to the field of machining error prediction, and the method comprises the steps: building a cutting force model through a finite element analysis method, and taking cutter parameters, machining process parameters and material cutting force parameters of an actual production line under different machining conditions as the input of the cutting force model; outputting cutting force and cutter back-off deformation errors corresponding to each group of parameters to obtain a training data set; obtaining cutting force and corresponding cutter back-off deformation errors of multiple sets of machining sites, and obtaining a test data set; adding a data enhancement module between an input layer and a first hidden layer of the neural network, and establishing a small sample learning model for predicting a cutter relieving deformation error according to the cutting force; and training and testing the small sample learning model by using the training data set and the test data set respectively to obtain a thin-walled workpiece cutter back-off deformation error prediction model. The prediction precision of the cutter back-off deformation error in the thin-walled workpiece machining process can be improved.

The invention discloses a distant place fault setting and monitoring system of a practical training substation. The system comprises a scheduling monitoring practical training room and a practical training substation, wherein the scheduling monitoring practical training room comprises a regulation and control host machine and a first optical transceiver, and the practical training substation comprises a second optical transceiver, a monitoring host machine, a communication workstation and a group of protective measuring and controlling devices controlled by the communication workstation; and a fault setting module is arranged in the regulating and controlling host machine, a fault receiving and distributing module is arranged in the communication workstation, and an embedded programming module of a protective device is arranged in each protective measuring and controlling device. The system disclosed by the invention has the beneficial effects that an accident handling safe operation training environment which can not be provided on site is provided, trainees and grid equipment can not be damaged in the practical training process, and the system is safe and reliable; fault simulation is more real, the fault information is uploaded to a scheduling end through the existing telecommand channel, handling within the scope of jurisdiction can also be carried out by scheduling personnel and a substation watch keeper, and a synchronous accident drilling chance is obtained; and the drilling can be repeated, and the training effect is guaranteed.

A PPI knowledge graph representation learning method based on protein family clustering is realized through the following steps: a) entity classification in a PPI knowledge graph; b) pre-representation learning; c) construction of a child-parent class loss function; C1) calculation of the sum of squares of distance errors; C2) normalization of the sum of squares of the distance errors; d) overalltraining; d1), training of a universal model; d2) training based on a child-parent class loss function; and e) multiple times of training through the step d). According to the PPI knowledge graph representation learning method, a representation learning model based on a child-parent class loss function is applied to a protein interaction knowledge graph, and representation learning training is performed through a child-parent class relationship between homologous proteins and a protein family. The invention has better accuracy, reliability and interpretability in protein function and interaction reasoning.