32results about How to "Help with training" patented technology

The invention discloses an improved Transformer + CRF-based method for recognizing a named entity of Lambda language, and belongs to the field of recognition of small languages in natural language processing. According to the method, an improved Transformer model is adopted to encode the Lao language, and then a CRF (Conditional Random Field) model is adopted to decode the Lao language to obtain an optimal annotation sequence. The method comprises the following steps: firstly, taking Transformer as a character encoder to carry out character encoding on a single character of the Lambda language; splicing the character-level vector obtained by encoding with a pre-trained word vector and carrying out position encoding; wherein the input signal is used as the input of a Transformer encoding part; through training of a multi-head attention layer and a feedforward neural network layer in sequence, finally, an obtained vector set is input into a CRF model fused with old-age naming and geographical name institution name linguistic features for named entity recognition training, and a named entity recognition model fused with old-age language rules is obtained. Compared with a mainstream BLSTM + CRF named entity recognition method, the recognition effect of the method is improved.

The invention discloses a mixed feature-based fine imaginary action online brain-computer interface method. The method comprises the steps of designing four actions of making a fist, raising the wrists, bending the elbows and abducing the shoulders for different joints of the upper limb on the right side; giving electrical stimulation during prompting of starting through electrodes placed on flexor muscles of the palms, median nerves of the wrists, bicipital muscles of the arms and deltoid muscles, paying selective attention to electrical stimulation positions while imagining the correspondingactions by a testee; effectively fusing event-related desynchronization features of different frequency distribution and electrical stimulation induced steady state somatosensory evoked potential features to form a mixed normal form; by adopting a multi-frequency component spatial filtration-based algorithm, extracting the event-related desynchronization and steady state somatosensory evoked potential features from the mixed normal form, thereby increasing the correct identification rate; and performing mode identification by adopting a multi-classification support vector machine, outputtinga decision value, and feeding back a voice prompt to the testee. The body parts controlled by high-correlation adjacent brain regions can be effectively distinguished; and the method has excellent performance.

The invention relates to an unsupervised medical image segmentation method based on an adversarial network, belongs to medical care informatics, and particularly relates to the technical field of medical image segmentation. According to the technical scheme, the method comprises the following steps: firstly, randomly generating or utilizing a third-party data set to obtain a group of auxiliary masks according to shape prior information, and sending the auxiliary masks and an unlabeled training image into a cyclic consistency adversarial network to generate binary masks; and a discriminator based on variational self-encoding and a generator correction module based on discriminator feedback are utilized to improve the quality of the binary mask. And after the binary mask of the training image is obtained, iterative training is performed by using a noise weighted Dice loss function, so that a final high-precision segmentation model can be obtained. According to the method, the problem that the convolutional neural network needs a large number of manual annotations in the training process of medical image segmentation can be solved, the problems of low performance, poor robustness and the like of an unsupervised segmentation method are solved, and the performance of an unsupervised medical image segmentation algorithm is effectively improved.

The invention discloses a wireless charging system fault evaluation method and system. The evaluation method comprises steps: the wireless charging system is subjected to function model division to build a fault tree, wherein the top event of the fault tree is that the wireless charging system can not perform charging, and the bottom event is a specific fault reason for each subsystem; the probability of occurrence of the bottom event is quantified according to preset data; according to the probability of occurrence of the bottom event, the critical importance of the bottom event is calculated, and according to the critical importance of the bottom event, fault of the wireless charging system is comprehensively evaluated, wherein a fault type is divided in the case of comprehensive evaluation, a different priority coefficient is given respectively, the comprehensive importance of the bottom event with the priority coefficient is calculated, and fault of the wireless charging system isevaluated according to the comprehensive importance. Thus, a fault generation reason can be quickly and automatically analyzed and evaluated, the fault information is more intuitive, a fault maintenance person can be trained, and the maintenance cost is reduced.

The invention discloses a rehabilitation training instrument for vision correction. The rehabilitation training instrument comprises a rack, a training cavity is formed in the rack, and a motion cavity is communicated with the lower portion of the training cavity; a vision training device is arranged in the motion cavity, sight distance adjustment devices are arranged at the left and right sides of the motion cavity and used for adjusting the shortest and longest distances of a lamp panel, and an eyepiece switching device is arranged at the upper side of the left wall of the rack. According tothe rehabilitation training instrument, the most proper training state can be adjusted according to the vision intensity of a person, an eyepiece can be adjusted and switched, rehabilitation trainingonly on a single eye or rehabilitation training simultaneously on double eyes can be selected according to the will, and the rehabilitation training instrument is comfortable and convenient to use and more beneficial to eye muscle training and vision correction.

A real-time embedded server system for controlling, in real-time, at least one infusion pump. The embedded server system includes an embedded server including a memory and a processor electrically coupled to the memory and configured to implement a control engine configured to issue control commands to at least one infusion pump, a messaging engine configured to issue messages to and receive messages from at least one infusion pump, an aggregation engine configured to aggregate data related to the operation of the at least one infusion pump, and a networking engine configured to control network access of the embedded server and the at least one infusion pump. A patient care system including at least one embedded server controlling device configured to make patient-specific care decisions and an associated network of devices operably coupled to a patient and configured to be controlled by the at least one controlling device.

Systems and methods for inspecting a device are disclosed. The method includes arranging the device in a known position relative to a plurality of movable cameras (32, 34). The plurality of movable cameras are mounted on a controllable actuator (36, 38). The plurality of cameras are pointed at the device by controlling the controllable actuator to position the camera with a user interface. A computing device (40) renders a virtual image from a CAD model of the device with a computing device. An image of the device generated by the camera is displayed on a display (44). The computing device also causes the rendered virtual image of the device to be displayed on the display. The camera images and the rendered images can be displayed side -by- side, in an overlay fashion or both.

The invention discloses a semi-supervised learning method special for performing cell nucleus segmentation on a histopathological image dyed by hematoxylin eosin. According to the cell nucleus segmentation method provided by the invention, according to the characteristics of the histopathological image and cell nucleus segmentation, the two dyes of hematoxylin and eosin in the histopathological image are separated by adopting non-negative matrix factorization with sparse constraint, and then the eosin dye in the histopathological image is replaced by the eosin dye in other histopathological images, so that the segmentation efficiency of the cell nucleus is improved. Therefore, a group of positive example samples can be prepared, and the positive example samples have the same hematoxylin staining agent, so that the positive example samples have interpretable invariance. And inputting the multiple groups of positive example samples into an encoder, and outputting a corresponding embedded representation vector by the encoder. And constraining the model by adopting a contrast learning loss function, so that the model can learn invariance in a positive example sample, namely the hematoxylin staining agent. The hematoxylin stain can stain the cell nucleus and other nucleic acid-rich parts, such as ribosome, so that the hematoxylin stain and the cell nucleus have relatively high correlation. When the model learns the characteristics of the hematoxylin stain, the characteristics accord with the characteristics of a cell nucleus segmentation task, so that the training of the downstream cell nucleus segmentation task is facilitated. As positive example sample construction and pre-training do not need labels, a large amount of unlabeled data can be utilized for training in the mode. And finally, the pre-trained encoder is added into the segmentation model, and fine adjustment is performed on a very small amount of labeled data, so that an effect better than supervised learning on a small amount of samples can be achieved. Therefore, the demand of annotation data is also reduced, and the labor cost is greatly reduced.

The invention discloses a rolling bearing fault identification method based on the combination of an SPA map and ResNet, and the method comprises the steps: firstly decomposing an original signal intoa trend term and a detrend term which are greatly different through an SPA method, converting the obtained trend term and detrend term into a color map through the combination of the original signalterm, and finally achieving the fault type identification through a ResNet network model, under the condition that bearing fault information is extracted as much as possible, component items being greatly reduced, and meanwhile, extracting deep bearing fault information through deep ResNet, so that the rolling bearing fault recognition efficiency and accuracy are improved.

The invention provides an upper entry type barrel tennis ball ejection device and a method thereof. The controller controls the motor to drive the rotating shaft to rotate and the rotating shaft to drive the baffle to rotate. The tennis balls in the storage bin enter the independent space one by one in the process of the baffle rotating. With the baffle rotating, the tennis balls are sent into theball outlet channel, and the tennis balls pass through the ball outlet channel and connect the channel to the ballistics. The controller controls the spring in the spring compressor to eject. In thespring ejection process, the emitter ejects and pushes the tennis ball to be launched. The tennis learners strike back according to the direction of the tennis ball, and hit back the ball to the baffle. After the tennis ball is fired, the controller controls the spring in the compressor to return to the spring compressor and wait for the next tennis ball to be fired. The stiffening barrel type tennis ejection device provided by the invention has the advantages of simple structure and convenient use, is helpful for tennis learners to carry out training, improves the training effect and tennis skills of tennis learners, fully exercises the reaction ability of athletes, and lightens the work intensity of coaches at the same time.

The invention discloses an intelligent ball picking robot which comprises a detection device, a moving device, a picking device and an electric energy storage device. The detection device, the pickingdevice and the electric energy storage device are all arranged on the moving device. The detection device comprises a camera which is arranged on the moving device through a supporting rod, and the camera is obliquely aligned to the front of the moving device. A direct-current motor and a motor driving plate are arranged on the moving device, PWM speed regulation is conducted through the motor driving plate to control the direct-current motor to rotate, and the moving device is controlled to advance or retreat or turn corner. The picking device comprises a mechanical arm base 7, a mechanicalarm and a steering engine driving plate. The steering engine driving plate is used for driving the mechanical arm to move, the steering engine driving plate is arranged at the position away from the mechanical arm base, and the tail end of the mechanical arm reaches a certain position in space through motion planning. According to the intelligent ball picking robot and the mechanical arm control method thereof, an autonomous intelligent ball picking mode is adopted, object searching, moving, grabbing and framing are achieved, the device is simple, the grabbing range is wide, and small balls inall directions can be grabbed.

The invention relates to the technical field of false teeth, in particular to a zirconium oxide all-ceramic data processing method and system based on big data, and the method comprises the steps: obtaining false tooth modeling data, and obtaining tooth feature data from the false tooth modeling data; obtaining bad tooth feature data from the tooth feature data, and constructing a bad tooth recognition model according to the bad tooth feature data; when a tooth image to be recognized is obtained, inputting the tooth image to be recognized into a preset bad tooth recognition model; and acquiring a bad tooth recognition result corresponding to the tooth image to be recognized from the bad tooth recognition model. The method and the device have the effect of improving the utilization rate of the data obtained when the false tooth is manufactured.

The invention discloses an interaction system based on augmented reality or virtual reality intelligent head-mounted equipment, and the interaction system comprises head-mounted equipment with a display part, wherein the head-mounted equipment is provided with a first control unit which is provided with a display output end for outputting a picture to the display part and a storage module for providing a storage space; and the head-mounted equipment is also provided with a camera unit, and is also provided with a first wireless communication module. The interaction system further comprises a handle, wherein the handle is provided with a system control chip, a second wireless communication module, a six-axis sensor module, a physical key module, a touch control module and a controllable power supply; the second wireless communication module, the six-axis sensor module, the physical key module, the touch control module and the controllable power supply are in electric control connectionwith the system control chip; the physical key module and the touch control module are used for acquiring operation information input by a user; the six-axis sensor module is used for acquiring position information of a preset point on the handle and position change information of the preset point; and the second wireless communication module is used for sending the operation information, the position information and the position change information to the first wireless communication module.

The invention discloses a counting simulation system and method for neutron fluxes of a nuclear power plant. The system comprises a simulator, a simulation computer and a main control room neutron flux loudspeaker, wherein the simulator is used for generating numerical values of actual neutron fluxes of the nuclear power plant in the full range, the simulation computer is connected with the simulator and used for collecting the numerical values of the neutron fluxes needing to be simulated from the numerical values of the actual neutron fluxes of the nuclear power plant in the full range and converting the numerical values of the simulated neutron fluxes into corresponding audio signals according to a relation table of the numerical values of the neutron fluxes and audio signals, and the main control room neutron flux loudspeaker is connected with the simulation computer and used for receiving and playing the audio signals. The system can achieve counting function simulation of the neutron fluxes to the maximum extent and is contributed to training of related technicians.

The invention discloses a pelvic floor muscle contraction exerciser for postpartum recovery of a pregnant woman, which comprises a bottom plate for overall supporting and fixing, two clamping arms for clamping legs of a user are arranged above the bottom plate, and the opposite ends of the two clamping arms are provided with through holes and are slidably connected with a limiting cross rod through the through holes; the vertical section, away from one end of the bottom plate, of a first support penetrates through the limiting cross rod and is fixedly connected with the limiting cross rod. Through the cooperative use of the structures, the problems that in the actual use process, traditional pelvic floor muscle training is mostly passive training, for example, pelvic floor muscle training is achieved by applying pressure through an external instrument, or pelvic floor muscle training is achieved through electrical stimulation, the stimulation feeling of passive training on pelvic floor muscle is not strong enough, the growth and recovery of muscles are slow, and the overall physical quality of the postpartum pregnant woman is difficult to improve due to single pelvic floor muscle training are solved.