35results about How to "Lessen the training load" patented technology

A training apparatus is disclosed which allows a trainee to attain a sense of accomplishment by continuing an exercise until a target is reached, without undue strain. A torque motor applies a load to a handle bar, which is driven by the exercise of a trainee. If the movement of the trainee who moves the handle bar is about to stop, then the load is gradually reduced. If the handle bar once again begins to move due to a load reduction, then it is inferred that the trainee has resumed the exercise, and the load at that time is maintained until, for example, the direction of motion of the handle bar changes. By gradually reducing the load at the fatigue limit of the trainee, it is possible to promote the resumption and the continuance of the exercise.

The invention discloses a short-term load predicting method based on a quick fuzzy rough set. The method comprises the following steps that firstly, electrical load data recorded by an electricity meter installed in a power grid are collected, and an initial attribute decision table is constructed; secondly, a fuzzy subordinate function of the condition attribute and the decision attribute is determined; thirdly, the attribute reduction is carried out according to the quick fuzzy rough set method, and the reduction condition attribute is obtained; fourthly, the reduction condition attribute serves as input data of a neural network to train normalized historical load data; fifthly, the neural network obtained through training is utilized for carrying out the short-term load prediction on an electric power system; sixthly, reverse normalization processing is carried out on the obtained normalization value of the maximum load of the prediction day, and a short-term electric power load prediction result is obtained and is the maximum load of the prediction day. The computing amount of the fuzzy rough set attribute reduction is small, the computing time is short, and the computing efficiency is improved.

The invention discloses a passenger detecting method based on a Haar-PCA characteristic and a probability neural network. The passenger detecting method comprises the steps of manually selecting a large number of passenger pictures and non-passenger pictures, and respectively marking the passenger pictures and the non-passenger pictures as positive samples and negative samples; representing a random positive sample or negative sample Si by means of an Haar characteristic, generating an Haar characteristic vector Hi, wherein i=1, 2, ..., n; selecting a main element subset vector Hi<PCA> which comprises majority part information on the Haar characteristic vector Hi of all samples through main component analysis; determining a classifier CPNN structure and determining to-be-classified samples, similarly with the step 2 and step 3, representing a to-be-determined sample according to the Harr-PCA characteristic vector Hi<PCA>, and inputting the vector into the CPNN which is obtained in the step 4, wherein the to-be-determined sample belongs to a class of which the output neuron is 1. The passenger detecting method has advantages of greatly reducing number of dimensions of a training characteristic vector through the Haar-PCA characteristic, reducing training load of the classifier, and greatly improving detection accuracy through replacing traditional BPANN by means of the probability neural network (PNN).

The invention discloses an ultra-short-term wind power prediction method based on a hybrid intelligent technology, to address the unpredictable challenges of ultra-short-term wind power generation. The proposed method employs a series of data processing techniques on the basis of available raw data, including input variable selection based on statistical analysis, attribute reduction based on principal component analysis (PCA), and attribute reduction based on K-Means clustering algorithm to obtain more relevant and efficient concentrated data as the input information of the prediction. The proposed method uses adaptive neuro-fuzzy inference system (ANFIS) to train and learn the input information in order to obtain the output prediction results. Particle swarm optimization (PSO) algorithmis used to optimize the parameters of ANFIS in order to reduce the prediction error. The hybrid intelligent method is evaluated by the forecasting results of actual wind farms. Experiments show that the method can achieve effective forecasting accuracy.

The invention provides a virtual reality rehabilitation training method and system based on surface myoelectricity information and depth image information for forearm amputation patients. The method comprises: collecting residual limb myoelectricity information of an amputation patient, and obtaining a palm action category corresponding to the myoelectricity information through a trained gesture recognition model; meanwhile, acquiring the stump pose information of the amputation patient through a depth camera, and identifying the arm action category following the palm action based on the poseinformation; and finally, based on the palm action category and the arm action category, generating a control instruction for controlling a virtual hand action in a virtual environment so as to assistrehabilitation training. According to the method and the corresponding system, the surface myoelectricity information and the depth image information can be fused, the rehabilitation training effectand efficiency of the forearm amputation patient are improved, and the patient can be helped to adapt to the artificial limb as soon as possible.

The embodiment of the invention discloses a text summarization method and system based on deep learning, and the method comprises the steps: carrying out Chinese word segmentation and sentence segmentation processing on a target file, and dividing an original text of the target file into independent sentences; converting the original text divided into the independent sentences into text vectors; performing self-attention calculation on the text vector to obtain sentence features containing semantic information; performing weighted integration on each sentence feature obtained through out-of-order self-attention calculation, and performing normalization processing on the sentence features after weighted integration to obtain chapter-level sentence features; inputting the chapter-level sentence features into a pre-constructed classification model, and outputting sentence categories; dividing the original text of the target file into key information and non-key information according to the sentence category; and arranging and organizing the key information according to a reasonable sequence, and synthesizing the arranged and organized key information into an abstract text with smooth semantics and compliant grammar.

The invention provides an image feature extraction and analysis method, system and device, and the method comprises the steps: carrying out the filtering preprocessing of an image collected through anoptical fiber microendoscope; carrying out contrast enhancement processing on the image after filtering preprocessing; carrying out segmentation processing on the image subjected to contrast enhancement processing, and extracting a segmented target feature structure; and carrying out superposition processing on the extracted target feature structure, and carrying out reinforcement processing on the boundary of the image after superposition processing. According to the invention, the morphological characteristics of squamous cells in the optical fiber microendoscope image are enhanced, the work of endoscopic doctors is assisted, the workload and training burden of the doctors are reduced, and the clinical efficiency is improved.

The invention provides a method and device for real-time prediction of the reaction depth of a hydrocracker. The method comprises the steps of establishing steady-state datasets for different working conditions, wherein operation datasets corresponding to different steady-state working conditions are screened out from historical datasets with the similarity measurement method, and the operation datasets are stored in a classified mode according to device operation phases and mixed feeding oil properties; conducting neural network optimal training sample set optimization, wherein based on the feeding oil properties under the current working condition, an optimal training sample is selected from the steady-state working condition operation datasets of the hydrocracker according to the similarity of mixing indexes to achieve fast training of a neural network; conducting real-time prediction of the reaction depth. By the adoption of the method and device, the reaction depth of the hydrocracker is predicted under the current operation phase, the current feed properties and the current operation condition, real-time state information is provided for optimal control of the injection volume of quenching hydrogen in the process, and a basis is provided for protecting the activity of a catalyst, prolonging the on-stream time of the hydrocracker and realizing flexible adjustment of processing amount of a target product.

The embodiment of the invention discloses an eye key point labeling method and device and a training method and device for an eye key point detection model. The eye key point labeling method comprisesthe steps of obtaining a face image and a marked eyelid curve corresponding to the face image; for each marked eyelid curve, based on a mathematical integration principle, integrating the marked eyelid curves, and determining a first curve length of a labelled upper eyelid curve and a second curve length of a labelled lower eyelid curve in the labelled eyelid curves; determining a plurality of eyelid points to be utilized from the labeled upper eyelid curve and the labelled lower eyelid curve; based on the first curve length of the labelled upper eyelid curve, the second curve length of the labelled lower eyelid curve, the eyelid point to be utilized and a preset equal division point number, determining the equally-divided upper eyelid points and the equally-divided lower eyelid points from the labelled upper eyelid curve and the labelled lower eyelid curve respectively, and then determining the eye key points corresponding to each face image by combining the labelled eyelid points. The eye key points with obvious semantic features are labelled, and the marking efficiency is improved.

The invention relates to a machine translation method and device and a storage medium. The method comprises the steps of inputting M source language sentences with the same semanteme and different source languages into corresponding coding layers in a translation model respectively to obtain M original coding results, wherein M is a positive integer; associating words of different source languageswith the same semanteme in the M source language sentences by utilizing a word alignment model to obtain an alignment result; inputting the M original coding results into an interaction layer of a translation model, and obtaining M target coding results according to the alignment result and the M original coding results; and based on the M target coding results, obtaining a target sentence basedon target language representation. According to the method, the information transmission among all source language sentences is greatly enhanced, so that the source end input information is more fullyutilized, and the accuracy of a translation result is improved; and semantic interaction between sentences is guided by utilizing an alignment result, so that the information interaction efficiency is improved, and meanwhile, the training burden of the model is also reduced.

An electric power system voltage stability evaluation misclassification constraint method based on an umbrella-type algorithm comprises: a step 1, constructing an initial data set, and constructing avoltage stability safety classification label based on an electric power system voltage stability evaluation rule; a step 2, selecting a key operation variable to construct an efficient sample set; astep 3, performing offline training on the voltage stability evaluation model; a step 4, sending a new sample set generated under a new operation working condition into the voltage stability evaluation model to update the model; and a step 5, completing online voltage stability evaluation by using the trained voltage stability evaluation model. The invention aims to solve the limitations of a traditional VSA model constructed based on a data driving tool in the aspects of misclassification constraints and a model updating mechanism. An electric power system voltage stability evaluation misclassification constraint method based on an umbrella-type algorithm is provided, so that the VSA model can provide a VSA result which balances the overall classification precision and the class of classification error constraints.

An electric power system dynamic safety assessment method based on a hybrid integration model comprises the following steps: 1, constructing corresponding dynamic safety indexes, and establishing an initial sample set containing a plurality of electric power system operation variables and the corresponding safety indexes; 2, constructing an efficient sample set; 3, obtaining a dynamic safety evaluation model capable of accurately evaluating the operation condition of the power system; and step 4, based on real-time monitoring data of a synchronous vector measurement device, selecting key characteristic variable data and inputting the key characteristic variable data into a DSA model to realize online DSA of the power system. The invention is to provide an electric power system dynamic safety assessment method which is good in generalization ability, and the method has the advantages of being rapid and accurate in assessment, is beneficial for power grid workers to take effective prevention and control measures in time, and can effectively avoid loss caused by faults of the power system.

The invention discloses a cross-platform malicious software confrontation sample generation method and system, belongs to the technical field of software security, and aims to map various types of malicious software samples of different platforms into a unified binary file, improve the generalization ability of a malicious software confrontation sample generation model and the diversification of confrontation samples, and improve the generation efficiency of the confrontation samples. And the malicious software analysis efficiency is improved. Besides, by modifying the action space, a character string confusion method is added, the robustness of the adversarial sample can be better improved, a decision network algorithm and an integrity verification method are applied to malicious software adversarial sample generation, the training calculation overhead can be reduced, and it is guaranteed that the sample function is complete.

The invention relates to a method for dynamically adjusting and migrating a service function chain in a network slice scene, and belongs to the technical field of mobile communication. The method comprises the following steps: S1, in a network slice scene, considering an SFC migration problem caused by traffic change due to service request change and a migration chain reaction brought by SFC migration due to lack of prediction of SFC resource demands, and adopting an algorithm based on an integrated deep neural network to predict a traffic change condition; s2, establishing an SFC migration penalty minimization model under the constraint of calculation, memory and bandwidth resources; s3, converting a traffic prediction result into a future resource request condition of the SFC, sensing a physical node and link resource occupation condition, and minimizing an optimal strategy through migration of the SFC; and S4, obtaining an optimal strategy of SFC migration. By adopting the method, the punishment of operators can be reduced, the probability of migration chain reaction is reduced, and the service reliability is ensured.

The invention discloses a cross-user gesture recognition method for robust electromyographic control. The method comprises the following steps: 1, collecting surface electromyographic signals and extracting features to construct a source domain data set; 2, constructing a student teacher deep network model; and 3, training a student model by using the source domain data set, obtaining network parameters, and obtaining teacher model parameters through index moving averaging. 4, new user electromyographic signals are collected, and features are extracted to construct target domain data; and classifying and generating pseudo labels through the student model and the teacher model. And 5, optimizing a pseudo tag generated by the teacher model through an optimal transmission algorithm. And 6, continuously giving classification results for newly input target domain data by using the student model, carrying out parameter updating, and then updating the teacher model. And 7, executing the step 4 for subsequent new users. According to the method, model migration from the source domain to the target domain can be realized, so that high-precision cross-user gesture recognition is realized.

The invention discloses an intelligent mechanical rack used for being coordinated with a motion assisting lower limb exoskeleton to adjust human body positions and a control method of the intelligentmechanical rack. The intelligent mechanical rack comprises a three-axis motion platform, a posture detection sensor module and a single-chip microcomputer module and is characterized in that the three-axis motion platform comprises an X axis, a Y axis and a Z axis; the posture detection sensor module comprises left foot sole pressure detecting module and a right foot sole pressure detecting modulewhich are used for detecting the treading force of a wearer; a mechanical rack sensor module mainly comprises an ultrasonic sensor, a first displacement sensor and a second displacement sensor, the ultrasonic sensor is used for acquiring the height change data, along the Z axis direction, of the gravity center of the wearer, the first displacement sensor is used for measuring the displacement, along the X axis direction, of the gravity center of the wearer, and the second displacement sensor is used for measuring the displacement, along the Y axis direction, of the gravity center of the wearer; the data of the sensors is transmitted to the single-chip microcomputer module, and the three-axis motion platform is controlled by the single-chip microcomputer module. The intelligent mechanicalrack has the advantages that the burden of a patient when the patient wears the exoskeleton is relieved greatly, and the patient can have enough space to wear the exoskeleton.

The invention provides an image contrast enhancement method, system and device, and the image contrast enhancement method comprises the following steps: carrying out the filtering preprocessing of an image collected through an optical fiber microendoscope; carrying out enhancement processing on the filtered image; carrying out nonlinear gray scale transformation on the enhanced image, and calculating a gray scale value of each pixel point on the image after nonlinear gray scale transformation to obtain a new image; and based on the obtained new image, cutting off the value of the lowest pixel of the image, standardizing the highest pixel value, and standardizing the pixel values between 0 and 255 by taking the minimum value and the maximum value of the pixels of the image as standards to obtain an image with enhanced contrast. According to the method, the morphological characteristics of squamous cells in the optical fiber microendoscopic image can be enhanced, the work of endoscopic doctors can be assisted, the work and training burden of doctors can be reduced, and the clinical efficiency can be improved.

The invention discloses a light field image spatial super-resolution reconstruction method, which constructs a spatial super-resolution network and comprises an encoder, an aperture level feature registration module, a light field feature enhancement module, a decoder and the like. Extracting multi-scale features of the up-sampled low-spatial-resolution light field image, the 2D high-resolution image and the blurred image by using an encoder; the correspondence between the 2D high-resolution features and the low-resolution light field features is learned through an aperture-level feature registration module, so that the 2D high-resolution features are registered to each sub-aperture image, and registered high-resolution light field features are formed; enhancing the extracted shallow light field features by using the registered high-resolution light field features through a light field feature enhancement module to obtain enhanced high-resolution light field features; reconstructing the enhanced high-resolution light field features into a high-spatial-resolution light field image by using a decoder; the method has the advantages that the high-spatial-resolution light field image can be reconstructed with high quality, and texture and detail information can be recovered.

The invention discloses a CT (Computed Tomography) image blind denoising method based on multiple scales and an attention mechanism. The method specifically comprises the following steps of 1, performing feature fusion; step 2, weight distribution; step 3, de-noising processing is carried out; step 4, comparing and screening; the invention relates to the technical field of image processing. According to the CT image blind denoising method based on the multiple scales and the attention mechanism, the attention mechanism is used for weight redistribution, feature maps with more sufficient key detail feature expression are output, assistance is provided for noise level authentication, effective collection of feature maps of different levels of noise is achieved through sampling processing of the different levels of noise, and the accuracy of denoising of the CT image blind denoising method based on the multiple scales and the attention mechanism is improved. When the CT image to be denoised is denoised, the noise of the corresponding level can be accurately and effectively retrieved, then the denoising model matched with the noise can be selected, the definition of the denoised CT image is ensured, the training burden of the image feature set is effectively reduced in cooperation with the setting of the residual network, and a good guarantee is provided for the denoising robustness of the CT image.

The invention discloses a method for improving deep learning channel estimation performance based on optimal data augmentation of an auto-encoder in a wireless communication scene, and the method comprises the following steps: building two basic convolutional neural network models for channel estimation; acquiring a training set of wireless communication channel estimation, and performing data augmentation based on an auto-encoder to obtain augmented data; estimating a basic convolutional neural network model by the augmented data through a channel to obtain a relationship between a mean square error value of the data augmentation method based on the auto-encoder on a test set and the size of a data set; based on a small amount of experimental data in the step 3, providing a simple straight line intersection point detection method, and obtaining the threshold value of the data set size when the self-encoder improves the performance of the wireless communication channel estimation model. According to the method, the data set threshold value when the self-encoder improves the performance of the wireless communication system to the greatest extent can be obtained, and the method has practical value for data augmentation of the model by using the self-encoder.

The invention belongs to the technical field of teaching training equipment, and discloses a simulated training device for extracorporeal membrane oxygenation and continuous kidney replacement therapy. The system comprises a human body simulator which is internally provided with an elastic liquid storage bag and a damping device capable of reducing the flow rate of liquid; the connecting pipe suite comprises a plurality of connecting pipes arranged corresponding to ECMO and CRRT system pipelines, adapters are arranged at the joints of the connecting pipes and the ECMO or CRRT system equipment and the human body simulator, one ends of the adapters are detachably connected with the connecting pipes, and the other ends of the adapters are used for being connected with the corresponding ECMO or CRRT system equipment or the human body simulator; each adapter is provided with an indicating lamp, the connecting pipe is provided with a wire, and the connecting pipe is connected with the corresponding adapter so that the corresponding indicating lamps can be connected in series or in parallel through the wire. The power supply is used for supplying power to the indicator lamp; the training burden can be reduced, and the training effect is improved.

The invention relates to a multi-moving-target visual tracking method and device, and the method comprises the two steps: model offline pre-training and online visual tracking. An intra-class loss function with intra-class constraint is fused on the basis that an original loss function has inter-class constraint to form a total loss function, and the total loss function replaces the original loss function to carry out offline training on a model. Therefore, the inter-class feature vector spacing between different classes of targets and the intra-class feature vector spacing between different entities of the same class are increased at the same time. Through the technical scheme disclosed by the invention, on the basis that a visual tracking task does not increase the training burden of a network model and the network model has relatively good feasibility and real-time performance, the defect that the distinguishing of different entities of the same category is insensitive in the prior art is overcome, and the visual tracking accuracy of multiple moving targets is improved.

The present invention provides a method and system for acquiring voice data, including: when a user makes a voice call, save the voice data stream transmitted in real time in the intelligent terminal system, save the input voice data stream of the microphone as the first voice data, and turn the handset The output voice data stream of the output voice data stream is saved as the second voice data; detect whether the first voice data and the second voice data meet the training requirements of the voice recognition model, if so, continue to judge whether the first voice data is from the application object of the voice recognition model, if so, pass Mark the first voice data as voice data for application, mark the second voice data as voice data for non-application; if not, mark the first voice data and the second voice data as voice data for non-application. Based on the method of the present invention, by improving the voice acquisition method, the burden of training the voice recognition model for the user is reduced, and the user experience is improved.

The invention discloses a short-term load predicting method based on a quick fuzzy rough set. The method comprises the following steps that firstly, electrical load data recorded by an electricity meter installed in a power grid are collected, and an initial attribute decision table is constructed; secondly, a fuzzy subordinate function of the condition attribute and the decision attribute is determined; thirdly, the attribute reduction is carried out according to the quick fuzzy rough set method, and the reduction condition attribute is obtained; fourthly, the reduction condition attribute serves as input data of a neural network to train normalized historical load data; fifthly, the neural network obtained through training is utilized for carrying out the short-term load prediction on an electric power system; sixthly, reverse normalization processing is carried out on the obtained normalization value of the maximum load of the prediction day, and a short-term electric power load prediction result is obtained and is the maximum load of the prediction day. The computing amount of the fuzzy rough set attribute reduction is small, the computing time is short, and the computing efficiency is improved.

The embodiment of the present invention discloses a method and device for marking eye key points and training a detection model thereof. The method includes: obtaining a human face image and an annotated eyelid curve corresponding to the human face image; for each annotated eyelid curve, based on the principle of mathematical integration, integrating the annotated eyelid curve, and determining the first curve length of the annotated upper eyelid curve in the annotated eyelid curve and the second curve length of the marked lower eyelid curve; from the marked upper eyelid curve and the marked lower eyelid curve, a plurality of eyelid points to be used are determined; based on the first curve length of the marked upper eyelid curve, the second marked lower eyelid curve The length of the curve, the eyelid points to be used, and the preset number of equally divided points are determined from the marked upper eyelid curve and the marked lower eyelid curve to determine the equally divided upper eyelid point and the equally divided lower eyelid point, and then combined with the marked eye corner points to determine each person The eye key points corresponding to the face image are used to mark the eye key points with obvious semantic features and improve the labeling efficiency.

The embodiment of the invention discloses a hand function rehabilitation training and evaluation device based on voice control. The hand function rehabilitation training and evaluation device based on voice control comprises an upper computer, a main control unit, a driving unit, a transmission unit, a hand execution unit, a voice input unit and an evaluation unit. The voice input unit is connected with the upper computer, the voice input unit is used for receiving voice information sent by a user, and the upper computer recognizes mode information to be selected according to the voice information sent by the user. According to the hand function rehabilitation training and evaluation device based on voice control, the defects that a hand rehabilitation robot is single in function, large in size and inconvenient to operate can be overcome, a convenient rehabilitation training and evaluation mode can be provided for a user, and the training rehabilitation interest of the user is aroused.

The invention provides a wind speed prediction method based on a hybrid deep learning mechanism. The method comprises the following steps: S1, collecting historical wind power data for preprocessing; S2, inputting the preprocessed historical wind power data into a hybrid deep learning mechanism for training; and S3, performing wind speed prediction on the trained prediction model. The invention further provides a wind speed prediction system and equipment based on the hybrid deep learning mechanism. According to the method, the future wind power is predicted only by using the historical wind power data, the neural network is rapidly trained, and the characteristics of the gating circulation unit and the long-short term memory neural network are combined, so that the contradiction between the prediction time and the prediction accuracy is effectively balanced, and the obtained result can promote the power grid to more fully utilize wind resources.

The invention provides a virtual reality rehabilitation training method and system based on surface electromyography information and depth image information for forearm amputee patients. The method includes: collecting the residual limb electromyographic information of the amputee patient, and using a trained gesture recognition model to obtain the palm action category corresponding to the electromyographic information; meanwhile, obtaining the residual limb position and posture information of the amputee patient through a depth camera, and based on the The posture information identifies the arm motion category following the palm motion; finally, based on the palm motion category and the arm motion category, a control instruction for controlling the virtual hand motion in the virtual environment is generated to assist rehabilitation training. The method and the corresponding system of the present invention can integrate surface electromyography information and depth image information, improve the effect and efficiency of rehabilitation training for forearm amputee patients, and help patients adapt to prosthetic use as soon as possible.

The invention discloses a short track speed-skating sliding guider, and belongs to the technical field of short track speed-skating training devices. For the short track speed-skating sliding guider,an electric track is mounted on the inner side wall of the top of a sliding rail, the sliding rail is fixed above a short track speed-skating runway through fixing racks, a sliding guider body capableof sliding is arranged on the sliding rail, the sliding guider body fetches electricity from the electric track through a sliding contact line current collector, an advancing drive motor driving thesliding guider body to slide on the sliding rail is arranged inside the sliding guider body, a cantilever swing device is arranged on the sliding guider body, a sliding guide plate is mounted below the cantilever swing device through a cantilever, a cantilever drive motor driving the cantilever swing device is arranged in the sliding guider body, a PLC is mounted in the sliding guider body, the PLC is connected with a PC upper computer on the ground through an AB433A wireless 485 passthrough module, through the PC upper computer, the data of the sliding guiding path is transmitted into the PLC, the PLC drives the advancing drive motor through a motor driver so that the sliding guider body advances, during the advancing process, the PLC drives the cantilever drive motor through the motor driver so that the sliding guiding plate swings, and thus the sliding guiding path is simulated.