101results about How to "Shorten the training period" patented technology

The invention provides a distributed parallel training method and system for a neural network acoustic model. The distributed parallel training method comprises the steps that 101) each group of training data is inputted into a client respectively; 102) each of the client receives the inputted training data, and transmits parameters between a plurality of GPU and a first CPU, which are arranged in the client, by adopting a bus, wherein the parameters comprise the model weight and the gradient; each GPU calculates the gradient based on the inputted model weight parameter and inputs the calculated gradient into the first CPU; the first CPU updates a model copy in the corresponding client by using the gradient uploaded by the GPUs, transmits an updated weight parameter back to each GPU so as to be used for carry out gradient calculation again, and meanwhile, the first CPU accumulates the gradient inputted by each GPU and updates a model in a parameter server according to a accumulation result; and 103) latest gradient information acquired by accumulation carried out by the CPU of each client is inputted into the server by adopting network resources, and then a neural network acoustic model stored in the server is updated.

The invention discloses a polyphone model training method for speech synthesis, and a speech synthesis method and device. The method comprises the following steps of processing a voice data set and a text set so as to generate a training corpus set, wherein the text set corresponds to the voice data set, and the training corpus set comprises texts and Pinyin sequences corresponding to the texts; extracting feature information of the texts; and training polyphone models according to the feature information and the Pinyin sequence. According to the polyphone model training method for speech synthesis, in a polyphone model training process, manual labeling on Pinyin of the texts is not required, a training period of the polyphone models is greatly shortened, meanwhile, the circumstance that the trained polyphone models are inaccurate due to wrong manual labeling is avoided, and accuracy of the trained polyphone models is improved.

The invention discloses a real terrain simulation-based parachuting landing training system and method. A parachuting landing training simulation system closer to actual parachuting conditions is built by adopting the advanced technologies of parachuting landing simulation, virtual reality, real terrain simulation and the like, and the landing training problem is solved; the training system solves the problem that lift-off parachuting training is influenced by the airplane guarantee and the weather, the in-air and landing scenes of parachuting are simulated by using a ground device, thus the training cycle can be effectively shortened, the training consumption and security risks are decreased, and the training effect is obvious; and the training system integrates the landing site judging training, parachute controlling training and landing buffer movement training which are independent in the original ground training way, thereby being capable of closer to actual lift-off parachuting conditions, the training level of parachuters is effectively improved, and the mental tension of parachuters is alleviated.

The invention discloses an improved method for compressing a neural network. The method has the advantages that a pruning technology of the neural network is integrated into the original training process of the neural network, so as to reduce the training iteration frequency, thereby shortening the training cycle of a sparse neural network; the pruning technology can also be integrated into the compression retraining process of a trained model, so that the neural network can be compressed at the premises of ensuring the model accuracy.

The invention intends to provide a cell culture apparatus which is able to realize an adequate culture according to the culture state of cells while alleviating the labor of an operator. The apparatus includes: a culture bag (18) for causing the cells to proliferate; a cell inoculation cassette (19) (or culture bag (242) as an antibody stimulating and proliferation culture vessel) for stimulating the cells by an inducer for the proliferation; a culture medium cassette (20) for storing a culture medium supplied to the culture bag (18) and the cell inoculation cassette (19); a CCD camera (88) for acquiring images of the cells in the cell inoculation cassette; and an image processing computer and an operation control computer for determining the culture state (proliferation capability and proliferation ability of the cells) of the cells from the images of the cells acquired by the CCD camera, and causing a culture operation to be carried out on the basis of the determination.

A portable drilling simulation system includes a main control computer, a graphic processing computer, a choke console and a blowout preventer console. The main control computer, the graphic processing computer, the choke console and the blowout preventer console are interconnected by a network and serial ports. The blowout preventer console includes a blowout preventer control panel. On the left of the blowout preventer control panel is provided a blowout preventer control zone. On the upper right is provided a throttling manifold control zone. On the lower right is provided a high-pressure manifold control zone. The flow plug console includes a flow plug control panel. A main control program runs on the main control computer and a graphic processing program runs on the graphic computer. The regular blowout preventer console, the regular throttling manifold, the regular high-pressure manifold and the regular flow plug console are integrated into the portable drilling simulation system, thus the simulated operation of several apparatuses on the control panel is realized.

The invention provides a method for training an acoustic model based on CTC (Connectionist Temporal Classification). The method comprises the steps of 1, training an initial GMM (Gaussian Mixture Model), wherein time point forced alignment on text annotation of training data by using the GMM to obtain a time region corresponding to each phoneme; 2, inserting a blank symbol associated with the phoneme behind each phoneme, wherein each phoneme has a unique blank symbol; 3, constructing a CTC forward and backward calculated search path diagram for a phoneme annotation sequence with the blank symbols being added by adopting a finite state machine; 4, restricting the appearance time range of each phoneme according to a time alignment result, pruning the search path diagram, and cutting off thepath with the phoneme position exceeding the time restrictions so as to obtain a final search path diagram required by calculating a network error in CTC; and 5, performing acoustic model training byadopting the combination of a time-delay neural network (TDNN) structure and the CTC method to obtain a final TDNN-CTC acoustic model.

The invention discloses a data model dual-drive GFDM receiver and method. The method comprises the steps: respectively obtaining a channel estimation and a signal detection neural network; taking thereal-number result of a matrix comprising transmitted pilot frequency information and a received time domain pilot frequency vector as the input of a channel estimation neural network, and outputtingan estimation of the frequency domain channel state information; obtaining an equivalent channel matrix, taking the real-number result of the equivalent channel matrix and the received time domain signal vector as the input of a signal detection neural network, and outputting the result as the estimation of a GFDM symbol; establishing a demapping neural network, taking an estimation of the GFDM symbol outputted by the signal detection neural network as the input, and outputting the estimation as the estimation of the original bit information; determining the output of the demapping network andthe size of a set threshold, and outputting a detection result of the original bit information according to a determination result. The method has the advantages that the training parameters do not change with the data dimension, the training speed is fast, and the adaptability to different channel environments is strong.

The invention discloses a video clip positioning method and device, computer equipment and a storage medium, and belongs to the technical field of video processing. The method includes: obtaining a plurality of video frame features of a target video and text features of a target text through a video recognition model; and determining a candidate fragment which can be matched with the target text,finely dividing the candidate fragment based on the matching degree of the video frame in the candidate fragment and the target text to obtain a plurality of sub-fragments, and taking the sub-fragmentwhich has the highest matching degree with the target text as the target video fragment. Such feature matching by multiple times, the obtained candidate fragments are divided again, the mode of the target video clip is determined; therefore, the video recognition model does not need to learn the boundary characteristics of the target video clip, the sample video does not need to be accurately labeled during model training, the training period of the video recognition model is shortened, and the problem of low model output result accuracy caused by inaccurate sample video labeling is avoided.

The invention provides a human health characteristic big data intelligent calculation method and a human health characteristic big data intelligent calculation device integrating ZED vision. The humanhealth characteristic big data intelligent calculation method comprises the steps of: step 1, constructing a human body natural sign database and a human body extreme sports database; step 2, utilizing a ZED camera carried on a UAV to obtain heights of human bodies, and screening a human body group to be identified; step 3, performing rapid identification on human body face images to be identified based on a Hadoop platform by utilizing a PID neural network; step 4, tracking the identified human bodies by means of the UAV acquiring health data of human bodies to be identified and transmittingthe health data to a remote server; step 5, inputting the acquired health data and eye and mouth features into a human health assessment model based on a deep belief network DBN, and acquiring healthdetection degrees of the human bodies to be identified. The human health characteristic big data intelligent calculation method and the human health characteristic big data intelligent calculation device are based on a big data platform, utilize wearable equipment, the UAV and the binocular camera to acquire health information of human bodies participating in outdoor sports, and perform real-timehealth monitoring, so as to make a reasonable decision.

The invention relates to the technical field of biological recognition, and provides a body posture recognition method based on LSTM. The body posture recognition method comprises the steps: obtainingan action video of a to-be-recognized main body; extracting action feature information in the acquired action video of the to-be-identified main body through OpenPose, wherein the action feature information at least comprises skeleton key point information; and recognizing an action specification degree corresponding to the action feature information according to the action feature information and a pre-trained and generated body posture recognition model, wherein the body posture recognition model is a target neural network model generated according to a preset standard action, and the target neural network model is generated by training according to standard action feature information arranged according to a time sequence. According to the body posture recognition method, video actionsdo not need to be cut into isolated features to be recognized, and learning recognition is carried out through cooperation with the neural network model, and the body posture recognition process is rapid and accurate, and the user experience is improved.

The invention discloses a high-speed railway onboard signal simulation training system and method. The high-speed railway onboard signal simulation training system comprises an external environment simulation device, an internal processing device and a training effect feedback device, wherein the external environment simulation device is configured to simulate a ground environment and a train operating state so as to provide train operating conditions to the internal processing device, and to receive train protection information outputted by the internal processing device; the internal processing device is configured to compute in accordance with operating signals of a train driver and the train operating conditions so as to generate the train protection information; and the training effect feedback device is configured to conduct quantitative evaluation on the operations of the train driver in accordance with computation data of the internal processing device and the train operating conditions. The high-speed railway onboard signal simulation training system and method provided by the invention can reduce training cost and shorten a practical training cycle.

The invention discloses a recommendation model training method and device, computer equipment and a storage medium, and belongs to the technical field of computers. According to the method, the secondpreference information of the user for the unpurchased object is preliminarily predicted based on the first preference information of the user for the purchased object, and the features of the user dimension and the features of the object dimension are extracted from the obtained first preference information and second preference information respectively; then the currently acquired preference information based on the features of the two dimensions is corrected to obtain third preference information with higher accuracy, and the training data is generated according to the third preference information so that data modeling is not required to be directly performed according to original online data, and the complexity and time consumption of a training data generation process are reduced; therefore, the training period of the model is shortened, the model can be updated in time according to the latest user data, and the output result of the model can better accord with the current user preference.

The invention discloses a human body behavior recognition method and a human body behavior recognition system based on a 3D attention residual error model. The human body behavior recognition method comprises the following steps of: 1) acquiring a human body behavior video data set, namely, collecting YouTube website videos, downloading UCF101 and Kinetics-400 public data sets, and collecting video data through using a monocular camera; 2) performing preprocessing operation on the video data in the step 1), including video frame conversion and key frame extraction, and making a data set; 3) establishing a 3D attention residual error model, and extracting features of the data set obtained in the step 2); 4) utilizing a Softmax classifier to classify and identify the features obtained in thestep 3), so as to realize model training; and 5) migrating the model trained in the step 4) according to an actual scene or an actual demand, finely adjusting the model to improve the generalizationability of the model, and finally applying the finely adjusted model to an actual human body behavior recognition task. The human body behavior recognition method and the human body behavior recognition system improve the real-time human body behavior analysis of multi-class and complex video scene processing, and have wide research and practical application values.

Embodiments of the invention provide a patient inquiry conversation simulation method and apparatus, a terminal device and a storage medium. The patient inquiry conversation simulation method comprises the steps of obtaining an inquiry statement input by a user, searching for a first response statement corresponding to the inquiry statement in an inquiry matching template, and if the first response statement corresponding to the inquiry statement exists in the inquiry matching template, outputting the first response statement to the user; and if the first response statement corresponding to the inquiry statement does not exist, determining a corresponding second response statement according to a semantic attribute type of the inquiry statement, and outputting the second response statementto the user. Therefore, the training cost can be reduced and the training cycle can be shortened.

The invention provides a lithium ion battery thermal runaway grading early warning method and early warning system. According to the method, fault-free battery characteristic elements are adopted to train the thermal runaway prediction capability of a prediction model composed of a long short-term memory network, a time convolutional network and a GRU neural network; and the battery characteristic elements of the to-be-monitored battery in the time sequence are input into the prediction model to obtain a prediction value at a certain moment in a normal state, the prediction value is compared with a collected true value corresponding to the moment, and the fact that the difference between the prediction value and the true value is larger indicates that the thermal runaway risk of the battery is higher. Through the operation, the respective advantages are integrated, and the problem of overlarge coefficient deviation of respective model training is avoided, so that high-precision battery thermal runaway grading detection is realized, and the speed and precision of lithium ion battery thermal runaway early warning are improved.

The invention discloses a spinal intervertebral foramen surgery model. The model comprises an artificial spine composed of multiple segments of artificial vertebral bones,an artificial intervertebraldisc located between every two adjacent artificial vertebral bones,an intervertebral disc protrusion simulated object bonded to each artificial intervertebral disc and artificial tissue wrapped on theexterior of the artificial spine; the intervertebral disc protrusion simulated objects are made of viscose fibers and bonded to arbitrary positions of the corresponding artificial intervertebral discs. The intervertebral disc protrusion simulated objects made of the viscose fibers can be glued to appropriate positions outside the corresponding intervertebral discs to simulate different types of intervertebral disc protrusions,and the shape and size of the intervertebral disc protrusion simulated objects can be adjusted according to needs to be better in line with the real state of the human body. Compared with a traditional mode of using corpses or non-dedicated models for training,the model has the advantages that the cost of training doctors can be greatly reduced,the repeatability is excellent,surgical modes are diversified,and the training period can be shortened.

The invention discloses a violent sorting behavior recognition method and device. The method comprises the steps of obtaining a to-be-predicted video; graying the video to be predicted; and inputtingthe grayed to-be-predicted video into a predetermined three-dimensional convolutional neural network model to perform violent sorting behavior identification, wherein the predetermined three-dimensional convolutional neural network model has a behavior feature extraction capability. According to the method, a three-dimensional convolutional neural network model is constructed by utilizing a modelwith behavior feature extraction capability, and learning and violent sorting behavior recognition are carried out by utilizing a grayscale video. Therefore, the violent sorting behavior recognition period is shortened, and the judgment accuracy is improved.

The embodiment of the invention discloses a commodity recognition model training method and device, electronic equipment and a storage medium. The method comprises the following steps of collecting various commodity image samples as a training data set; extracting a set number of various commodity image samples from the training data set as a primary training set; and training a primary recognition model based on a Faster-RCNN network, applying the primary recognition model to the training data set to obtain an effective sample, mixing the effective sample into the primary training set, and performing iterative fine tuning training to finally obtain a commodity recognition model. By adopting the technical means, the number of training samples can be reduced, the number of model training steps can be reduced, and the accuracy of the recognition model is ensured through error sample feedback training, so that the training period of the commodity recognition model is shortened, the new process on commodities is further accelerated, and commodity retail operation is optimized.

The invention discloses a human face set matching method and system based on similarity learning. Part of samples are selected as training samples to carry out training, classifiers are selected, and therefore it is avoided that all samples serve as the training samples to be trained, dimensional reduction processing is carried out on training set samples so that dimensional reduction training samples are obtained, the calculation complexity is prevented from being increased from high dimensional data, the training period is shortened, the training process is simplified, the complex process is avoided, and the training speed is improved. In addition, according to the scheme, the geometric mean of each type of samples in the training set samples is selected to construct a plurality of different classifiers, and the effect of the accurate result through the simple operating process is achieved.

The invention discloses skydiving simulation training equipment, which mainly comprises a suspension bracket system, a control system, a dynamic system, a visual system, a software system, a sound equipment system, a command control system and an evaluation system. According to the invention, in a suspension bracket and harness system, when a parachutist jumps out of a cabin, the flying direction is controlled through a control belt. A sensor is operated to transmit the operation information to a computer. A motion system is driven after operation, and the simulated innervation is provided. Meanwhile, the computer outputs visual images to a head-mounted display, and an immersion type virtual environment is provided. A teacher sets a system initial condition through a teacher table and monitors all students. The invention further provides a simulation training method based on the utilization of the above training equipment. By the adoption of the method and the equipment, the training safety of the skydiving person can be guaranteed. The training period is greatly shortened, and the skydiving training cost is reduced. Great military benefits, economic benefits and social benefits can be generated.

The invention discloses a balanced sample set construction method and device, a duplicated image recognition method and device, equipment and a medium. The method comprises the steps of obtaining an unbalanced duplicated sample set and a preset target balanced sample number; inputting the unbalanced copying sample set into a DVAE-GAN network model in the balanced sample set construction model fortraining; recording the converged DVAE-GAN network model as a trained DVAE-GAN network model until the total loss value of the DVAE-GAN network model reaches a preset convergence condition; if the target equalization sample number is greater than the positive sample number and the negative sample number at the same time, randomly generating a positive copy simulation sample and a negative copy simulation sample by the trained DVAE-GAN network model; and obtaining a balanced copy sample set for training a copy recognition model. According to the invention, the sufficient number of balanced copying samples are generated through the balanced sample set construction model to be trained by the copying recognition model, the time for collecting the copying samples can be shortened, and the recognition accuracy and the hit rate of the copying recognition model are improved.

The invention discloses a picture set classification method and device. The method comprises the steps that a first picture set is acquired from a to-be-classified picture set, the picture types of the first picture set are known, and the number of the picture types is a preset number; target models matched with the preset number are selected from preset classification models, and the target models are branch networks divided according to the type number in the preset classification models; model parameters of the target model are updated through the first picture set; and the second picture set is classified by using the target model of which the model parameters are updated to obtain a classification result of the second picture set, the second picture set being a partial or total picture set remaining after the first picture set is removed from the to-be-classified picture set. According to the invention, the technical problem of low picture classification efficiency due to the factthat a computer image processing method for deep learning in the prior art needs a long time to carry out model training is solved.

Provided is a Myrica rubra branch pruning and maintenance method. The method is characterized by comprising the following steps: (1) determining a pruning object and a pruning period, the pruning object: the Myrica rubra tree with a tall tree body and an inner chamber illuminated by dim light, selecting the branches of which the inserted positions in the Myrica rubra tree are unreasonable, the stretching directions are not ideal and the thickness of the bases are 1.5-4 cm; the pruning period: after the end of Myrica rubra harvest, choosing to conduct pruning in a sunny day or a dry cloudy day before the weather 'into the volts'; (2) pruning operation, using a cutting knife to carve more than three lines on the selected branches in the longitudinal direction, with the carved lines deeply reaching the xylem, then holding the carved branches and forcing to twist, and homeopathic changing the stretching directions of the branches and rotating until the desired position; (3) maintenance after pruning. The method has the advantages of simple operation and good pruning effect.

The invention belongs to the technical field of traditional Chinese medicine materials and particularly relates to a method for bletilla striata seedling production based on a liquid medium. The to-be-solved technical problem is how to provide a novel method for quickly acquiring bletilla striata seedlings. The method comprises a step a of sterilization, a step b of germination, a step c of growth into buds, a step d of rooting culture and a step e of seedling hardening of adult seedlings and transplantation. The method is simple and convenient to operate and low in cost, and each bottle of liquid medium can be used for germinating several tens of thousands of seed. The seeds are germinated tidily, the seedlings emerge consistently, and the pollution rate is low. According to the method, the efficient quick production of the bletilla striata tissue culture seedlings is achieved finally, operating procedures are effectively simplified, the production cost is reduced, and large-scale production is facilitated.

The invention discloses an intelligent human body multidimensional natural feature big data transparent learning method and an intelligent human body multidimensional natural feature big data transparent learning system. The intelligent human body multidimensional natural feature big data transparent learning method comprises the steps of: step 1, constructing a human body natural physical sign database and a human body extreme sports database; step 2, utilizing human body heights obtained by means of a ZED camera carried on a UAV to screen a human body group to be identified and acquire corresponding face images; step 3, step 3, performing rapid identification on human body face images to be identified by adopting an Elman neural network based on a Hadoop platform; step 4, tracking identified human bodies by means of the UAV, acquiring health data of human bodies to be identified and transmitting the health data to a remote server; step 5, and inputting the acquired health data into aBP neural network human body health assessment model, so as to obtain health detection degrees of the human bodies to be identified. The intelligent human body multidimensional natural feature big data transparent learning method and the intelligent human body multidimensional natural feature big data transparent learning system are based on a big data platform, employ the wearable equipment, theUAV and the binocular camera to acquire outdoor sports human body health information, and perform real-time health monitoring, so as to make a reasonable decision.

The invention provides a dialogue system training data construction method, a device, electronic equipment and a storage medium, and the method comprises the steps: carrying out the statistical processing of the historical use data of the conversation system, and determining a historical query statement set corresponding to the conversation system, the query frequency corresponding to each historical query statement, and an identification result corresponding to each historical query statement; according to the query frequency and the identification result corresponding to each historical query statement, obtaining a reference query statement from the historical query statement set;judging whether the number of all reference query statements is greater than a first threshold; and if yes, constructing a training data set of the dialogue system by utilizing all the reference query statements and the identification results corresponding to all the reference query statements. Therefore, through the dialogue system training data construction method, the labor cost is saved, the construction efficiency of the training data set is improved, and the accuracy and reliability of the dialoguesystem are further improved.

Disclosed is an active running wheel odor identification exercise box. The active running wheel odor identification exercise box comprises a box body with a ventilation system, and an exercise module arranged inside the box body, wherein the ventilation system is connected with an odor system. The active running wheel odor identification exercise box is technically characterized in that the exercise module is composed of a positive and / or negative stimulation module arranged inside the box body, a locked turning plate limited inside the box body, and more than one rotatable turning plate; the odor system is matched with the positive stimulation module or the negative stimulation module. The active running wheel odor identification exercise box fundamentally solves the problem that existing exercise devices fail to avoid external interference and are long in exercise period, poor in exercise effects and the like. The active running wheel odor identification exercise box has the advantages of being low in manufacture and maintenance cost, stable in performance, simple in exercise manners, rapid and convenient to operate, remarkable in social and economic benefits, wide in application prospect and the like.

The invention discloses a UAV human body health monitoring method and a UAV human body health monitoring system based on intelligent visual reconstruction. The UAV human body health monitoring methodcomprises the steps of: step 1, constructing a human body natural physical sign database and a human body extreme sports database; step 2, utilizing human body heights obtained by means of a ZED camera carried on a UAV to screen a human body group to be identified; step 3, step 3, performing rapid identification on human body face images to be identified by utilizing a generalized regression neural network based on a Hadoop platform; step 4, tracking identified human bodies by means of the UAV, acquiring health data of human bodies to be identified and transmitting the health data to a remoteserver by means of wearable equipment; step 5, and inputting the acquired health data into a human body health assessment model based on a PID neural network, so as to obtain health detection degreesof the human bodies to be identified. The UAV human body health monitoring method and the UAV human body health monitoring system are based on a big data platform, employ the wearable equipment, the UAV and the binocular camera to acquire outdoor sports human body health information, and perform real-time health monitoring, so as to make a reasonable decision.

The invention relates to a multi-axis 360-degree omnidirectional rotating motion flight simulator, which is a multi-axis rotating motion platform combined by multiple groups of rotating bodies with different sizes and volumes in a crossed manner from the outside in, and is characterized in that a simulation cabin is installed on rotating frames in the multi-axis rotating platform, each rotating frame and an outer frame rotating shaft which supports and fixes the corresponding rotating frame are respectively provided with a set of motor-reducer with a driving wheel, a driven wheel is installed at an outer frame rotating bearing which supports and fixes the corresponding rotating frame, the driving wheel and the driven wheel are connected through a belt, a synchronous belt or a chain, the motor-reducer drives the rotating frame to perform rotating motion around the corresponding driven wheel, coordinate motion among the multiple groups of rotating frames is controlled by a control system so as to enable the simulation cabin installed inside the rotating frames to realize 360-degree omnidirectional rotating motion, and three-dimensional attitudes of an aircraft are completely simulated.