609results about How to "Short training time" patented technology

The invention discloses a feedback artificial neural network training method and a feedback artificial neural network calculating system and belongs to the field of calculation of neural networks. According to the artificial neural network training method, the synapse weight is adjusted according to a feedforward signal and a feedback signal at the two ends of each neural synapse; when the signals at the two ends of each neural synapse are an excitation feedforward signal and an excitation feedback signal respectively, the synapse weight is adjusted to the maximum value; when the signals at the two ends of each neural synapse are a tranquillization feedforward signal and an excitation feedback signal respectively, the synapse weight is adjusted to the minimum value. According to the feedback artificial neural network calculating system, each node circuit comprises a calculating module, a feedforward module and a feedback module and the node circuits are connected through the neural synapses simulated by memristors, and a series of pulse signals are adopted to achieve the feedback artificial neural network training method. An artificial neural network provided by the system and the method is high in rate of convergence, and the artificial neural network calculating system is few in control element, low in energy consumption and capable of being applied to data mining, pattern recognition, image recognition and other respects.

This invention discloses one forest measurement method without human aviation cameral remote sensor, which adopts aviation device as remote platform without human and integrated digital codes camera, difference GPS and top platform, wherein, it uses digital camera technique to get high earth resolution rate without human remove image with large proportion rules fully in use of high flexible machine without special plane spot and capable of flying under clouds.

The invention discloses an OCR recognition method. The method comprises the steps of obtaining a to-be-recognized image of business party data; Inputting the to-be-identified image into a general OCRtemplate for identification to obtain text information recorded in the to-be-identified image and position information corresponding to the text information, Wherein the universal OCR template comprises a detection model and a universal identification model, and the universal identification model is obtained by training field image samples of various service types of a service party; And synthesizing the text information and the position information corresponding to the text information into structured identification data. The invention further provides an OCR electronic device. According to the OCR identification method and the electronic equipment thereof, the image of the to-be-identified object (such as a contract, an invoice, a bill, a certificate and the like) can be efficiently andrapidly identified through the general OCR template, the structured identification data is generated, and the identification from the optical character to the text information is completed. The universal OCR template adopted in the method is short in training time, high in adaptability, capable of adapting to various different to-be-identified objects, high in identification accuracy and high in overall efficiency in the identification process.

A method and system for generating a decision-tree classifier from a training set of records, independent of the system memory size. The method includes the steps of: generating an attribute list for each attribute of the records, sorting the attribute lists for numeric attributes, and generating a decision tree by repeatedly partitioning the records using the attribute lists. For each node, split points are evaluated to determine the best split test for partitioning the records at the node. Preferably, a gini index and class histograms are used in determining the best splits. The gini index indicates how well a split point separates the records while the class histograms reflect the class distribution of the records at the node. Also, a hash table is built as the attribute list of the split attribute is divided among the child nodes, which is then used for splitting the remaining attribute lists of the node. The method reduces I/O read time by combining the read for partitioning the records at a node with the read required for determining the best split test for the child nodes. Further, it requires writes of the records only at one out of n levels of the decision tree where n>/=2. Finally, a novel data layout on disk minimizes disk seek time. The I/O optimizations work in a general environment for hierarchical data partitioning. They also work in a multi-processor environment. After the generation of the decision tree, any prior art pruning methods may be used for pruning the tree.

The invention discloses a cloud data center load prediction method based on LSTM (Long Short-Term Memory), and aims to solve the problem that optimal utilization can not be obtained by the limited calculation resources of a cloud data center. The method comprises the following steps that: taking the mass historical records of the cloud data center as a basis to manufacture a training sample and atesting sample; in addition, constructing a neural network connected by LSTM units; continuously inputting training samples on batch to obtain an output value. A neural network optimization algorithmadopts a new adaptation moment estimation method, parameters in each unit are continuously updated through iterative training, and global optimum is realized after training is finished, only the testing sample needs to be input into the network to obtain the next prediction value of a sample sequence; and if an input sequence is continuously updated by the prediction value, a prediction value sequence in one future period of time can be obtained.

The invention discloses a remote sensing image semantic segmentation technology based on a VGG network. The method comprises the following steps: 1, randomly cutting the high-resolution remote sensingimage for training and the corresponding label image into small images; dividing the network structure into an encoding part and a decoding part; adopting the depooling path and the deconvolution path to double the resolution of the coded information; carrying out channel connection on a characteristic image and a result of cavity convolution, recovering the characteristic image to an original size through deconvolution upsampling, inputting an output label image into a PPB module for multi-scale aggregation processing, and finally, updating network parameters in a random gradient descent mode by taking cross entropy as a loss function; inputting the small images formed by sequentially cutting the test images into a neural network to predict corresponding label images, and splicing the label images into an original size. According to the technical scheme, the segmentation precision of the model is improved, the complexity of the network is reduced, and the training time is saved.

The present invention provides a handwritten digital recognition method and system based on depth learning. A convolutional neural network is trained with a sample set which is formed by constructing a handwritten digital picture with a label, the trained convolutional neural network is saved, and the image to be identified is taken as the input, and according to the output vector, the recognition result is obtained. The handwritten digital image is identified through the convolution neural network, so that the displayed feature extraction is avoided, the picture is directly taken as the network input, and the recognition accuracy is high. The network can be repeatedly used after being trained, the processing efficiency is high and the training time is short.

The invention belongs to the field of a prediction method of the urban water supply network tube explosion, and provides a novel urban water supply network tube explosion prediction method to solve the problem that the prediction precision is low and establish a tube explosion prediction model so as to provide novel study basis for urban water supply network tube explosion. Therefore, according to the technical scheme, the urban water supply network tube explosion prediction method comprises the following steps of: (1) acquiring the tube explosion rate of a tested water supply network and historical data of other relevant affect factors which affect the tube explosion rate, and taking the tube explosion rate and the historical data as the detected data; (2) establishing a water supply network tube explosion prediction model of a self-adaptive neural network heterozygosis variable precision rough set; (3) carrying out training and test analysis on the water supply network tube explosion prediction model; and (4) predicting the tube explosion rate of the water supply network by using the water supply network tube explosion prediction model which passes the test. The method is mainly applied to prediction on the urban water supply network tube explosion.

The invention provides a blood vessel segmentation method and a system of fundus image. The blood vessel segmentation method of fundus image comprises the following steps: step S1, constructing a neural network training set of fundus image for neural network training; step S2, amplifying the neural network training set to obtain the amplify neural network training set; step S3, constructing a neural network model for blood vessel segmentation of fundus oculi image based on depth learning; step S4, training the neural network model by using the expanded neural network training set; step S5, obtaining a target fundus image, and performing blood vessel segmentation on the target fundus image by using the trained neural network model. The blood vessel segmentation method and the blood vessel segmentation system of the fundus oculi image of the invention can overcome the limitation that the number of the blood vessel segmentation public data sets is low and can be adapted to the fundus oculi image of poor quality. The invention greatly simplifies the network structure, and can effectively improve the segmentation speed without losing the larger accuracy rate.

The invention provides a cycle nerve network text emotion analysis method by embedding logic rules, and the method comprises the steps of grabbing a text corpus for training and conducting emotion class labeling; then dividing the emotion labeled text corpus into a training set corpus and a test set corpus for dividing the words and removing the stopped words; then conducting training to the training set corpus and the test set corpus whose words are divided and stopped words are removed by using word2vec algorithm so as to obtain a corresponding word vector; inputting the training set corpus and the test set corpus in an existing knowledge base and analyzing in combination with a probability graph model; embedding a first order logic rule in the cycle nerve network through a logic cycle nerve network structure (Logic-RNN and Logic-LSTM). According to the invention, the training direction of the cycle nerve network can be controlled, and inclination to human's intuition is realized; besides, the precision of text emotion analysis is improved. Therefore, the method can be used in other fields like natural language processing and machine learning.

The invention discloses a hybrid brain-computer interface method based on steady state motion visual evoked potential and default stimulation response. The method includes the steps that 1, a testee wears an electrode cap, a reference electrode, a ground electrode and a testing electrode on the electrode cap make contact with the head of the testee, and the vision and the computer screen are in the eye level through visual inspection; 2, a steady state motion visual evoked potential and default stimulation response mixed normal form program is compiled through MATLAB in advance, the testee selects a stimulation target to stare according to a target prompt, and electroencephalogram signals acquired by the electrode cap are stored in a computer; 3, steady state motion visual evoked potential features and default stimulation response features are subjected to feature extraction respectively, and then the stimulation target is subjected to classified recognition; 4, the computer screen displays the stimulation target recognition result, and visual feedback is conducted on the testee; 5, the steps are repeated, and the next round is conducted till the program is ended. According to the hybrid brain-computer interface method, two types of feature recognition information is adopted, and the method has the advantages that operation is simple, less training time is needed, and less electrodes are needed.

The invention discloses a human face identification method and apparatus, and belongs to the field of human face identification. The method comprises: performing feature extraction on a to-be-identified human face image by using a plurality of pre-trained convolutional neural networks to obtain a plurality of sub-feature vectors of the to-be-identified human face image, wherein the sub-feature vectors of the to-be-identified human face image are same in number of dimensions; normalizing the sub-feature vectors of the to-be-identified human face image; performing addition on the normalized sub-feature vectors of the to-be-identified human face image, and multiplying the sum of the normalized sub-feature vectors by a coefficient to obtain a union feature vector of the to-be-identified human face image; and performing human face identification by using the union feature vector of the to-be-identified human face image or/and the sub-feature vectors of the to-be-identified human face image. According to the human face identification method and apparatus, the training time of the convolutional neural networks is shortened, the over-fitting of the convolutional neural networks is avoided, and the operation is simple and convenient; and identification modes are more diversified and the accuracy is higher.

The invention mainly relates to a method for regenerating a plant from camellia callus. The method has the following steps: stripping off the inner and outer seed coats of a camellia fruit seed, and inoculating the seed to a 1/2MS culture medium; when a sterile seedling grows above 3 cm, inoculating the soft leaf of the young plant to a callus induction culture medium which is MS + 0.5mg.L 6-BA+1.0mg.L 2,4-D; when the callus grows to get a diameter about 1 cm, shifting the callus to a callus differentiation culture medium which is MS+mg.L 6-BA20+0.1mg.L I BA+ mg.L KT0.1; when an indefinite bud grows to 0.5 cm, carrying out the separation and inoculating to a strong bud culture medium which is MS + 0.2mg.L 6-BA+0.05mg.L NAA; and when a bud stick grows to 4 to 5 cm, cutting off the basal of the bud stick, immersing the basal of the bud stick in 1,000 g.L I BA, and then inoculating to a MW + 0.2mg.L I BA+0.2mg.L NAA culture medium. The method has the advantages that: the method has a simple culture medium recipe, a simple and convenient operating process, short culture time, high regeneration frequency and a high propagation expansion coefficient, and facilitates the large-scale production of rare camellia plants and the realization of the genetic transformation of exogenous genes.

The invention discloses a robot barrier identification method based on a gradient histogram and a support vector machine. The method comprises two parts of a characteristic extraction stage and a target identification stage, for the characteristic extraction stage, a characteristic extraction algorithm of a power transmission line barrier of a principal component gradient histogram is proposed, the characteristic that typical barriers have different structures and space layouts is utilized, statistics characteristics of common online barriers are calculated, characteristic extraction is carried out by utilizing an HOG algorithm, characteristic points irrelevant to illumination and scale change can be acquired, interference can be effectively removed, moreover, dimension reduction operation for acquired characteristic vectors can be realized by utilizing main component analysis to acquire the principal component gradient histogram, irrelevant characteristics can be effectively reduced, operand is reduced, least characteristics are utilized to establish a characteristic set of the corresponding barriers, and excellent support is provided for next target identification; for the target identification stage, the linearity support vector machine is utilized for identification, and the excellent identification effect is acquired.

The invention discloses a wind turbine state prediction model establishing method based on a grey relation-regression SVM (support vector machine). The method comprises input variable determination, regression SVM training modeling and a prediction stage. The invention aims to solve the problems of multiple input vectors, data redundancy, poor prediction accuracy, long model training time and the like of a standard SVM state prediction model and provides the wind turbine state prediction model establishing method based on the grey relation-regression SVM, firm technological support is provided for guarantee of safe running of the wind turbine and reduction of non-planned shutdown times, traditional methods are improved, two methods are combined skillfully, a state prediction model is established, the wind turbine state is predicted with a simple and practical method, grey relational analysis is performed on each monitoring program, main factors are screened out, and unrelated information is rejected, so that the method is high in prediction accuracy, short in model training time and practicable.

The invention discloses a method for predicting fault trends of steam turbines by the aid of adaptive quantum neural networks. The method has the advantages that the traditional three-layer BP (back propagation) neural network models are improved, the quantum neural networks are introduced into the traditional three-layer BP neural network models, trend contribution force of different historical data is analyzed in input layers, influence of the latest data on the trends can be improved, direct connection weights from the input layers to output layers can be increased, excitation functions can be adaptively adjusted by the output layers according to signal characteristics, accordingly, the convergence speeds can be increased, and the prediction precision can be improved; the convergence speeds can be increased by the aid of the method for introducing the adaptive learning efficiency; the method is excellent in reliability and robustness, is key technical research on prediction of the fault trends of the steam turbines and can be widely applied to predicting the fault trends of the steam turbines.

The invention provides a driver fatigue driving detection system based on machine vision and a detection method, and belongs to the technical field of machine vision and machine learning. The system belongs to a non-intrusive type detection system. When detection is performed, needed information is acquired through a camera, normal driving of a driver is not influenced, and equipment is low in price and small in size; and a Bluetooth camera only needs to be installed in a vehicle, and app software is installed in a mobile phone, and then fatigue detection for the drive can be achieved. Information acquisition of the system is convenient and easy; when the system is used, only the camera is externally added, and then the system can adapt to any-type vehicle and any road condition; and the system has a consistent fatigue judgment criteria and the high fatigue judgment accuracy rate. The system integrates fatigue characteristics of eyes, a mouth and a face, the accuracy rate of fatigue judgments in complex driving environment is improved, and system parameters are rapidly updated through combination with the machine learning according to feedback of the driver so that the system adapts to different characteristics of different drivers. The system has the advantages of short training time, the rapid computing speed and high real-time performance.

The invention discloses an XLPE power cable partial discharge type identification method, and the method comprises the following steps: (1) building an XLPE power cable partial discharge experiment platform, and designing a typical insulation fault partial discharge model; (2) collecting PRPD spectrograms and pulse oscillograms of different insulation faults by utilizing a high-frequency current method, dividing the collected data into a training sample and a test sample, respectively extracting a statistical characteristic quantity from the PRPD spectrograms, extracting a time domain characteristic quantity from an original pulse signal oscillogram, and extracting a frequency domain characteristic quantity from the oscillogram after fast Fourier transform; (3) normalizing the characteristic quantity, setting network parameters by using the training sample, and constructing a fusion extreme learning machine network; And (4) sending the normalized characteristic quantity of the test sample into a trained fusion extreme learning machine network to obtain an identification result of the discharge type. The XLPE power cable partial discharge type identification method can improve the accuracy and stability of XLPE power cable partial discharge type identification.

The invention provides a software quality evaluation method and system based on secondary evaluation. The method includes the steps that a software quality evaluation index space is selected, and a software quality evaluation result identification framework is built; sample data and data of software to be evaluated are collected; the number and the topological structures of BP neural networks aredetermined; the BP neural networks are trained in parallel, and the credibility levels are calculated; the quality evaluation index data of the software to be evaluated is input into each trained BP neural network, and preliminary evaluation results are obtained according to output results of the BP neural networks; the preliminary evaluation results are corrected in combination with the credibility levels of the BP neural networks to generate basic probability assignment of each proposition in the identification framework, and all pieces of evidence are fused according to the DS evidence theory to obtain a fusion result; decision-making is conducted on the fusion result based on decision criteria to generate a final evaluation result. By means of the method, software quality evaluation can be effectively achieved.

The invention relates to a Phlebopus portentosus cultivating method. The method comprises the steps of culture medium preparing, bacterium inoculating, earthening, mycelium culture in an earthening layer and fruiting cultivating. The invention further relates to a fungus culture medium and the purpose of the fungus culture medium in Phlebopus portentosus cultivating.

The invention discloses a pedestrian hash retrieval method based on loss measurement in a depth learning network. The pedestrian hash retrieval method realizes hash code learning of a pedestrian imageby constructing a pedestrian hash learning model CFNPHL of a convolution characteristic network. Then the distance between hash codes of pedestrian images is calculated to realize the retrieval of large-scale pedestrian image data. The method includes establishing convolution neural network model and extracting pedestrian feature information; Mapping binary hash codes; Adding quantified loss to measure loss; setting classification los function to learn that distinguishing characteristic of different pedestrians, and obtaining pedestrian categories; Minimizing network losses; Training the network CFNPHL to obtain the pedestrian hash code library for image retrieval; Then inputting the pedestrian image to be retrieved into the trained network to obtain the hash code of the pedestrian to beretrieved; performing pedestrian retrieval by calculating the distance. The pedestrian retrieval method effectively improves the retrieval speed and has high accuracy rate according to the pedestrianretrieval under the complex scene.

The invention discloses a steady state visual evoked potential brain-computer interface signal identification method, which comprises the following steps of: (1) simultaneously displaying a plurality of different pictures with different flicker frequencies through a visual stimulation unit and acquiring electroencephalogram signals of a testee who stares at the visual stimulation unit; (2) carrying out noise estimation and noise suppression on the electroencephalogram signals by a data processing unit, and then carrying out characteristic extraction and judgment on the processed electroencephalogram signals to primarily determine the picture at which the testee stares; and (3) upsetting the flicker frequencies of the displayed pictures, acquiring the electroencephalogram signals, then carrying out noise estimation and noise suppression on the currently acquired electroencephalogram signals, then carrying out characteristic extraction and judgment on the processed electroencephalogram signals to determine the picture at which the testee stares, if the currently determined picture is the same as the picture determined in the step 2, taking the picture as finally determined identification information to output, and if not, judging that the testee does not stare at any picture displayed by the visual stimulation unit. According to the steady state visual evoked potential brain-computer interface signal identification method, the electroencephalogram signal identification accuracy can be effectively improved.

The invention provides a composite bacillus preparation containing three strains, a preparation method of the composite bacillus preparation and application of the composite bacillus preparation to ecological breeding, and belongs to the technical field of microorganism and feed additives. Bacillus coagulans, butyric acid clostridia and bacillus megaterium are used as fermentative strains respectively, and large-scale, low-cost and high-density fermentation production of spores is conducted through step-by-step magnification; three live bacterium preparations are prepared by mixing bacterial sludge with the spores and dry starch; the three live bacterium preparations are matched according to the equal weight ratio, the composite bacillus preparation is obtained, and the concentration of the spores reaches more than 1*108-109 cfu/g; the composite bacillus preparation serves as a microorganism and feed additive to be added to farmed animal feed according to the mass ratio ranging from 0.02% to 0.5%. The composite bacillus preparation is simple and stable in process, low in cost, high in spore content and environment tolerance, easy to store and high in survival rate; the purposes of composition and synergism are achieved through the collaborative supplementary effects of different bacilli in the aspect of micro-ecology adjustment, growth of farmed animals is promoted, the feed utilization rate and feed rewards are increased, diarrhea is prevented and reduced, and the breeding ecological environment is improved.