3927results about How to "Improve training effect" patented technology

The invention discloses a traffic scene multi-target classification method, to be specific, discloses a traffic scene classification method based on a multi-scale convolution neural network. The traffic scene classification method is characterized in that recessive characteristics based on the multi-scale convolution neural network are extracted; and an optimal covering segmentation tree is acquired. During the realizing of the traffic scene classification, the multi-scale convolution neural network is adopted, and the excellent recessive characteristics having the invariance property are effectively extracted from an original image in different scales, and by comparing with the single-scale convolution neural network, the acquisition of the abundant and effective characteristic information of the image is realized. The effective information extracted by the convolution neural network is combined with the original segmentation tree of the image to form an optimal purity price tree, and the covering having the optimal purity is carried out, and therefore a clearer target contour is acquired, and the classification accuracy is increased. The RGB-D is used as the convolution neural network input, and by comparing with the conventional RGB convolution neural network input, the training characteristic is additionally provided with the depth information, and the classification of the input image is more accurate.

The invention discloses an intensive learning based urban intersection passing method for a driverless vehicle. The method includes a step 1 of collecting vehicle continuous running state informationand position information through a photographing method, the vehicle continuous running state information and position information including speed, lateral speed and acceleration value, longitudinal speed and acceleration value, traveling track curvature value, accelerator opening degree and brake pedal pressure; a second step of obtaining characteristic motion track and the velocity quantity of actual data through clustering; a step 3 of processing original data by an exponential weighting moving average method; a step 4 of realizing the interaction passing method by utilizing an NQL algorithm. The NQL algorithm of the invention is obviously superior to a Q learning algorithm in learning ability when handling complex intersection scenes and a better training effect can be achieved in shorter training time with less training data.

The invention discloses a medical simulation training system and method. The system comprises an object model building module, a model database, a sensor tracking module, a data processing module, an image generation module and a display module; the object model building module acquires the three-dimensional model data of a specific part of a human body and builds a corresponding three-dimensional virtual model object based on the acquired three-dimensional model data; the model database stores the three-dimensional virtual model object data; the sensor tracking module acquires a user' operation data of a three-dimensional model; the data processing module processes the acquired data to obtain the shape data of the three-dimensional model and the user' operation data of the three-dimensional model and outputs the shape data and the user' operation data; the image genera module generates a three-dimensional virtual image including simulation operation result information according to the shape data of the three-dimensional model and the user's operation data of the three-dimensional model; and the display module displays and outputs the three-dimensional virtual image for the user to view a simulation operation result. With the system and method provided by the embodiments of the present invention adopted, a strong sense of immersion can be realized, a real medical scene can be simulated to the greatest extent, the costs of clinical teaching and training about surgeries, intravenous injection and the like can be reduced.

The invention discloses a federated modeling method, device and equipment based on a block chain, and a storage medium. The method comprises the following steps: a training initiator issues configuration information to all training clients based on client information corresponding to the training clients when the number of the training clients is monitored to reach a preset number; the training initiator uploads a to-be-trained model to a main chain; and the training initiator determines a target model based on aggregation gradient and the to-be-trained model. Modeling of federated learning isrealized through the block chain; on the premise of protecting the privacy of federated learning data, the accuracy of federated learning is not affected, the training effect and model precision of federated learning are improved, model parameters such as gradients in transmission do not need to be modified, and the balance between privacy protection of the model parameters such as gradients andmodel convergence or model precision is achieved; and information leakage can be completely prevented, so the safety of data samples in federated learning is improved.

Provided are a stroke patient rehabilitation training system and method based on brain myoelectricity and a virtual scene. Control over the virtual rehabilitation scene is achieved through myoelectric signals, and rehabilitation training intensity is adjusted in a self-adaptation mode with a brain myoelectricity fatigue index combined. The design of the virtual rehabilitation scene is completed with the needs of stroke patient rehabilitation training and the advice of a rehabilitation physician combined, the brain fatigue index is provided, and quantitative evaluation on brain region fatigue is achieved. The surface myoelectric signal features under different motion modes of an arm are extracted, the motion intention of a patient is obtained, and control over the virtual rehabilitation scene is achieved. The muscle fatigue and brain fatigue index comprehensive features are extracted, the fatigue state of a rehabilitation patient is obtained, self-adaptation rehabilitation training scene adjusting is achieved, rehabilitation training intensity is relieved or enhanced, and secondary damage caused by improper training is avoided. The system and method have the advantages of being high in safety, high in intelligence and scientific in training, and damage cannot happen easily.

The invention discloses a projection-type billiard training system and an implementation method thereof. The system comprises a billiard table, an image acquisition apparatus, a billiard lamp, and a projection apparatus, wherein the head end of the image acquisition apparatus mounted in the billiard lamp is flush with the lower end of the lamp, and is suspended at a certain high position above the billiard table to be over against the center of the billiard table; and the projection apparatus is hung beside the billiard lamp and is controlled by a computer. The system implementation method comprises the following steps of: establishing a standard database of training courses and the related video files; reading the above data and files by selecting junior training course to senior training course when the system program runs, and displaying the desired images, such as demonstration video, auxiliary lines, ball positions, aim point and simulated motion trajectory of the cue ball and the target ball through the projection apparatus; and capturing shot video when training according to the video, inputting the shot video to the computer for analysis, and giving suggestions for the next shot by the image acquisition apparatus. The projection-type billiard training system is more scientific, accurate, intuitional and convenient in billiard training, thereby improving the shot accuracy and the cue ball control capability.

The invention relates to a multi-perception emergency training system for a petrochemical fire scenario, which mainly solves the problems of insufficient interaction and low scenario perceptibility in the prior art. The multi-perception emergency training system for the petrochemical fire scenario based on a B/S architecture comprises a 3D driving engine, a thermal flux sensor module, an equipment operation module, a basic knowledge database module, a scenario database module, a desktop deduction, examination and evaluation module, a server and a client computer. The system well solves the above problem, and can be used for multi-perception emergency training of petrochemical fire scenarios.

The logistics operation training system includes manipulator assembling and conveying operation unit, automatic cargo distinguishing and buffering operation unit, stacking operation unit, automatic warehouse operation unit, visual detection operation unit, stacking operation unit, cargo flow direction controlling operation unit and container packing operation unit. Each of the operation units is provided with monolithic computer circuit and logic circuit with modular electric interface, and different units may be selected to form stepped configuration. The present invention is suitable for different teaching and training modes.

The embodiments of the invention disclose an operation training system and method of special vehicles. The system comprises a simulation system, a controller, an immersed type display apparatus and a driving operation console of the special vehicles, wherein the driving operation console comprises operation components consistent with operation components in actual driving cabins of the special vehicles in an airport; the controller is connected with the driving operation console through a control bus and is in communicating connection with the simulation system for receiving operation control instructions sent by users through operating the driving operation console and converting the operation control instructions into digital operation control signals for transmission to the simulation system; the simulation system stores setup data of virtual three-dimensional scenes for realizing synchronization and interaction between operation of the driving operation console and changing of the states of the virtual three-dimensional scenes; and the immersed type display apparatus is used for displaying the virtual three-dimensional scenes of current special vehicles. According to the embodiments of the invention, the problems caused by fewer actual operation opportunities during training of special vehicle drivers of a civil aviation airport in the prior art can be solved.

The invention belongs to the field of rehabilitation medical equipment and in particular relates to a multi-position lower limb rehabilitation training robot. The robot comprises a supporting mechanism, a patient fixing system, a lower limber training device and a control system. The lower limber training mechanism utilizes a neutral equilibrium principle, and can be conveniently converted into an outer dynamic-free training mode in a flat-lying state; a patient weight loss and fixation form of connecting the upper body of a patient and a sliding flat plate is adopted; a sole air bag training mechanism is light and flexible, and the patient can achieve a better training effect under the prompt of a lower limber vibration feedback module; exoskeleton lower limber rehabilitation equipment which can realize neutral equilibrium, a flat plate type fixed weight loss system and functional modules such as the vibration feedback module and the like are integrated to form the system, and the neutral equilibrium, the flat plate with an adjustable distance, and the more intuitive vibration feedback characteristic are fully utilized to realize the functions of the multi-position lower limb rehabilitation training robot in an active control mode, a passive control mode, an impedance control model and the like.

The invention discloses a transformer station simulation training system based on the mixed digital/ analogy simulation technology. The transformer station simulation training system comprises a primary simulation system of an electric power system on the side of trainers and a secondary equipment system of a transformer station on the side of trainees. The primary simulation system of the transformer station is used to simulate running states of primary equipment of the electric power system in normal and failed conditions, and transmit data in normal and failed conditions to the secondary equipment system of the transformer station to drive secondary equipment of the transformer station, the transformer station trainees can monitor entire data of the electric power system through the secondary equipment of the transformer station, and the data reflect the real running states of the primary equipment of the electric power system. The transformer station simulation training system provided by the invention can simulate the running states of a power network, forms a complete real-time closed-loop simulation training scene, and can be used for training transformer station operating personnel on the characteristics of protective equipment in a practical running state, enabling the transformer station operating personnel to practice analyzing and handling power network faults, andtraining the transformer station operating personnel for skill tests.

The invention discloses a distributed well drilling simulation system and an operation method. The distributed well drilling simulation system comprises a choke-line manifold (101), a high pressure manifold (102), a blowout preventer console (103), a flow blocking machine console (104), a long-distance console (105), a driller console (106), a teacher console (107) and a figure projecting unit (108), wherein the driller console (106), the long-distance console (105), the blowout preventer console (103), the flow blocking machine console (104), the choke-line manifold (101) and the high pressure manifold (102) are interconnected through a PPI (processor/peripheral interface) protocol, and the teacher console (107) is connected with the PPI protocol through a PPI interface. The operation method comprises the steps of drilling down, pilling out of a hole, drilling, accident and complex condition processing, well shutting, well killing and the like. The invention has the advantages of realizing the simulation of a top driving well drilling simulation with highly simulation, enhancing the liveness of teaching and training, shortening training period and reducing the training cost.

The invention discloses an interactive gait training and evaluation system based on multi-source information fusion for stroke patients. The interactive gait training and evaluation system structurally comprises a pose acquisition module, a plantar pressure acquisition module, a fatigue information acquisition module, a wireless communication module, a data analysis and processing module, an intelligent cloud platform, an AR (augmented reality) helmet and a mobile terminal. The system aims at stroke patients with certain walking ability and can intelligently provide the patients with high-immersion interactive rehabilitation training game on the basis of the AR technology, and the gait rehabilitation training effect on the patients can be improved effectively; multi-dimensional gait information during rehabilitation training of the patients is acquired by a gait acquisition module and processed by the data analysis and processing module, processed information and data are transmitted to the mobile terminal through the wireless communication module, and gait evaluation results can be fed back to doctors and the patients in real time in a highly visualized manner; safety of the patients in the rehabilitation training process is guaranteed on the basis of muscle fatigue monitoring; the intelligent cloud platform can realize automatic tracking, evaluation and push of gait diseases.

The invention relates to a combination-type text classification method based on deep learning. The method includes the following steps: preprocessing to-be-classified text; using the preprocessed textto obtain feature extraction results, which correspond to all deep learning models, through the different depth learning models; combining all the feature extraction results through a fully connectedlayer to obtain a combined feature vector; and using the combined feature vector as input to carry out classification recognition to obtain a class to which the text belongs. Compared with the priorart, the method has the advantages of lower requirements on training-corpus quality, high transplantability and expandability, a still better effect in a case of more classification classes and the like.

The invention relates to a simulation system of loading and unloading operation processes in pressurized water reactor (PWR) nuclear power plants. The system comprises an LED (light emitting diode) display screen, a touch screen, an operation control table consisting of a component part consisting of a button, a switch, an operating rod, an indicator lamp and an alarm, an information communication and processing device consisting of a programmable logic controller PLC (programmable logic controller) and a PC graphic workstation, a display device consisting of a scene display and a camera, and an industrial personal computer. The simulation system of the loading and unloading operation processes is designed in strict accordance with functions of real loading and unloading machines, is featured with rich simulated objects, reliable simulation data and strong reality, simulates various common abnormal states or failure states, allows interactive operation, clears faults and has a component showing function; the touch screen has an interface control function; the system simplifies layout of an operation table, expands subsequent functions, effectively trains loading and unloading operators and is widely applied to nuclear power stations and related analogue simulation systems.

The invention relates to a multi-mode three-dimensional medical image fusion method and system and electronic equipment. The method comprises the steps of constructing a multi-task generative adversarial network, wherein the multi-task generative adversarial network comprises a generator, a discriminator and a classifier; training the multi-task generative adversarial network according to the MRIimage, the PET image and the diagnosis label information of the subject; inputting an MRI image of a to-be-detected person into the trained multi-task generative adversarial network; and enabling thegenerator to synthesize a corresponding PET image according to the MRI image, input the MRI image of the to-be-detected person and the synthesized PET image into the classifier, and output a disease classification prediction label of the to-be-detected person after fusing the MRI image of the to-be-detected person and the synthesized PET image. According to the method, the conflict problem of lossfunction convergence points possibly occurring when the performance of a generator and the performance of a classifier are considered in a traditional generative adversarial network is solved, and the generator and the classifier can be optimal at the same time.

The invention discloses a named entity recognition model training method, a named entity recognition method and a named entity recognition device. The training method comprises the following steps: preprocessing a corpus sample to obtain a character sequence sample, and labeling a named entity label on the character sequence sample to obtain a training character sequence; pre-training the trainingcharacter sequence based on a first bidirectional language model and a first self-attention mechanism model to obtain a character feature vector and a character weight vector, and fusing the character feature vector and the character weight vector to obtain a second context vector; pre-training the training character sequence based on a second bidirectional language model and a second self-attention mechanism model to obtain a word feature vector and a word weight vector, and fusing the word feature vector and the word weight vector to obtain a second context vector; and training the bidirectional neural network and the conditional random field which are connected in sequence by using the first context vector and the second context vector to obtain a named entity recognition model. According to the method, the training effect of the named entity recognition model is effectively improved, and the named entity recognition accuracy is improved.

The invention discloses a self-adaptive speed regulation system of running device based on a flexible array pressure transducer, comprising the flexible array pressure transducer, a data collection transmission module, a tablet computer, a computer control program, and a motor drive module, wherein the motor drive module is connected with a motor for driving the motor to rotate, and the motor drives the running belt of the running device to rotate. In the invention, stride frequency, stride length, ground-touching time, air time, rearward thrust and other parameters of a runner when in running can be obtained by installing the flexible array pressure transducer on the upper side of the running plate of the running device; acceleration can be calculated by an established relational model of difference of acceleration and the step length between two adjacent steps, the step frequency, the component force of the rearward thrust along the moving directing of the running belt, time for completing two steps (the ground-touching time + the air time), and the weight of the runner; a control command is sent to the motor drive module according to the obtained acceleration; and the motor rotation speed is adjusted; therefore, the speed self-adaption of the running device is realized.

The invention belongs to the technical field of data processing, and discloses an encoder training and representation information extraction method and device. The encoder training and representationinformation extraction method comprises: aiming at the original sample data and at least two pieces of loss data of the original sample data, obtaining corresponding coding characteristics by adoptingencoders with the same model parameters respectively, decoding the characteristics by adopting corresponding decoders, and obtaining predicted loss based on the coding characteristics, the original sample data and the decoding characteristics; and if the prediction loss meets a preset convergence condition, initializing a target encoder by adopting the model parameters, and obtaining representation information of the data by adopting the target encoder. Thus, the training efficiency and effect of encoder training are improved, and the effectiveness of extracted representation information extraction is improved.

The invention discloses an attention training system which comprises a preprocessing module, a processing module, a storage module and an interaction module, wherein the preprocessing module, the processing module and the storage module are connect with one another; the interaction module is connected with the preprocessing module and the processing module respectively; the preprocessing module is used for processing the training content designated by a user, and generating training content data, so that the preprocessing module generates the training content suitable for the user; the preprocessing module transmits the training content data to the processing module, the interaction module and the storage module respectively; after the interaction module receives the training content data, the interaction module displays the training content through an output device connected with the interaction module; the interaction module provides the training content designated by the user through an input device connected with the interaction module; and the processing module is used for judging whether training is finished.

The invention discloses a method of using a fully convolutional neural network to segment a human hand area in a stop-motion animation. The method comprises steps: 1) data inputting is carried out; 2)the fully convolutional network is used for feature extraction and preliminary segmentation; 3) a conditional random field (CRF) algorithm is used to optimize the segmentation effects; 4) the networkmodel is trained; and 5) the model which completes the training is used to segment an inputted picture. The method disclosed in the invention mainly aims to solve the problem that through self building a picture data set containing the human hand area, the network model is built, and the data set is used to train the network model. After training, the network model can carry out high-accuracy segmentation on the human hand area. The method of using the fully convolutional neural network to segment the human hand area in the stop-motion animation has the advantages of high accuracy, good anti-noise performance, simple use, high efficiency, quick speed and the like.