66results about How to "Prove superiority" patented technology

The invention provides a method for detecting an image saliency target, and belongs to the technical field of image saliency detection. The method comprises the following steps: in a first stage, constructing a classification network CNet and a character sequence generation network PNet to obtain saliency maps from classification and character sequence generation tasks respectively; in a second stage, establishing two complementary training data sets, namely a natural image data set with a noise label and a network image data set, by utilizing CNet and PNet, and alternately training SNet by utilizing the two complementary data sets; and in the third stage, updating the natural image data set and the network image data set by using the prediction result of the SNet, and recursively optimizing the model. In the test stage, only the SNet is used to predict the saliency maps. Experiments show that the method is superior to unsupervised and weakly supervised methods, and still has the goodperformance compared with some supervised methods.

The invention belongs to the technical field of urban road traffic flow prediction and analysis and particularly relates to a method for road traffic flow prediction under a suddenly occurred trafficevent. The method comprises the steps that road traffic event alarm information data is preprocessed, and abnormal data is removed and repaired; a place name address space positional information database is established, traffic event position information is obtained and classified, and spatial and temporal distribution characteristics of road traffic flow and spatial and temporal characteristics of road traffic flow under the suddenly occurred event are analyzed; a random forest algorithm, an ARIMA method and a Kalman filtering method are utilized to conduct traffic flow prediction on time series data and space state data under the suddenly occurred event; a weighted least square method is used for conducting fusion processing on prediction results obtained through a time series data prediction method and a spatial series prediction method, and new prediction results are obtained. In the method, three indexes including a comprehensive error percentage absolute value mean value, an error absolute value mean value and a square-error mean value reach an ideal prediction effect.

The invention discloses an electronic chromatic dispersion compensation equalizer for optical communication and a method for adjusting a tap thereof. The electronic chromatic dispersion compensation equalizer comprises two symmetrical linear equalizers arranged at a transmitting terminal and a receiving terminal respectively. The structure of the chromatic electronic dispersion compensation equalizer is simpler than a decision feedback equalizer, and the performance of the chromatic electronic dispersion compensation equalizer is better than the linear equalizer. The invention adopts linear fragmentation algorithm to arrange symmetric linear equalizers at the transmitting and the receiving terminals respectively so as to acquire better anti-interference effect than normal linear equalizers and reduce complexity of the system. The invention further uses adjustment step factors to adjust the tap coefficient of the equalizer, and the method is simple in algorithm and good in convergence effect.

The invention discloses a dialogue behavior identification method and system based on a conditional random field and a structured attention network. The identification method comprises the following steps of (1) in combination with a memory network, performing hierarchical reasoning and semantic modeling on dialogue semantic information according to a word layer, a sentence layer and a dialogue layer; (2) by applying the structured attention network, performing structural section division on dialogue contents according to the correlation between the dialogue contents; and (3) applying the obtained structured information to a linear conditional random field algorithm, and predicting a current dialogue behavior according to a context. Through the method and the system, context information ina dialogue interaction process can be deeply captured, and the sections of the dialogue contents can be dynamically divided; and by combining the structured attention network with the conditional random field algorithm, the dialogue behavior identification accuracy can be further improved.

The invention relates to an estimation method for state of health (SOH) of an electric vehicle lithium battery. The estimation method is characterized by comprising the following steps of 1) buildinga second-order equivalent model of a battery lithium battery; 2) identifying a parameter of the second-order equivalent model, wherein the parameter comprises ohmic internal resistance R0 and a second-order RC link parameter; 3) calculating the ohmic internal resistance R0 by employing an improved UPF method; and 4) calculating an SOH value of the power lithium battery. Compared with the prior art, the estimation method has the advantages of high tracking forecast capability, high estimation accuracy and the like.

The invention aims at an intermittent process with uncertainty, and proposes a constrained 2D tracking control method for the uncertainty intermittent process. The constrained 2D tracking control method comprises the steps of: firstly, designing an iterative learning control law for a given system dynamic model; secondly, converting the original system dynamic model into a 2D-FM closed-loop systemmodel expressed in the form of a predictive value according to a 2D system theory and the designed iterative learning control law through introducing state errors and output errors; and finally, giving a sufficient condition expressed in the form of a linear matrix inequality (LMI) for ensuring robust asymptotic stability of a closed-loop system and an expression form of the optimal control law according to a designed infinite time domain performance index and a Lyapunov stability theory. According to the constrained 2D tracking control method, numerical values of tracking errors under the control of the constrained 2D tracking control method are smaller, and the convergence is faster; more importantly, the control input does not drastically fluctuate and only require slight adjustment, thereby being conducive to resource conservation and reducing troubles caused by frequent operations.

The invention discloses a zero sample learning method and system based on a semantic attribute attention redistribution mechanism, and the method comprises the steps: (1) building a neural network model based on the semantic attribute attention redistribution mechanism; (2) redistributing the weight between the semantic features by using the attention of the semantic attribute space; (3) traininga neural network model by using the image data set with the label; (4) calculating the similarity between the weighted semantic features of the images and the semantic prototypes of the unknown classes, calculating the similarity between the hidden layer features and the hidden layer feature prototypes of the unknown classes, and adding the two similarities to obtain the similarity between the test image and each unknown class; and (5) sorting according to the similarity with the types, and the type with the maximum similarity is selected as the type prediction of the image. According to the method, in the training process of zero sample learning, the semantic space and the hidden layer space can be more closely linked, and a joint classification result combining the two spaces is more robust.

The invention discloses a dialogue reply generation method and system based on self-commenting sequence learning. The dialogue reply generation method comprises the following steps that: (1) carryingout modeling on the context language environment of a current dialogue to obtain a context semantic vector; (2) according to the context semantic vector, establishing a dialogue language based on theself-commenting sequence learning; (3) training and testing the dialogue model to independently obtain a training award value and a testing award value; (4) calculating a difference between two awardvalues, and optimizing the dialogue model through policy gradient calculation; and (5) finishing dialogue model training, and outputting the dialogue reply. By use of the method, in a training processof a dialogue generation model, the reply with a better substantive meaning can be generated towards an evaluation index optimization direction, and dialogue generation instability can be greatly lowered.

The invention discloses an optimization task unloading method based on network time delay and resource management. The method comprises the steps that an optimization unloading PoW problem is studiedin a fusion blockchain and fog computing system based on node remaining resources and network time delay; in a network scene formed by a base station fog node, a fixed position fog node and a mobile fog node, the total duration required for unloading the PoW problem to various types of fog nodes is analyzed; the expenditure of the MUB is analyzed based on the remaining computing resources, storageresources and power resources of the network node and the social relationship between the nodes; an optimization task unloading scheme is provided based on node residual resources and network time delay, a mathematical optimization model is constructed by taking maximization of MUB income as a target, and the optimization model is solved by using an SA algorithm and a game theory; and finally, the superiority of the scheme is proved through analog simulation analysis. An intelligent task unloading mode of introducing artificial intelligence into the system is considered in future work.

The invention discloses a mobile live broadcast system based on MPEG-2 standard. The mobile live broadcast system comprises a mobile equipment end, a streaming media server and a player end, wherein the mobile equipment end comprises a video data collection sub-system which includes an H264 video compressing and encoding module, an MPEG-2 standard packaging module and a thread pool module; the thread pool module is connected with an RTSP signaling module and an RTP transmission module; the streaming media server comprises a live broadcast data forwarding module, an RTCP code rate control module, a video-on-demand module and an RTSP signaling module; the mobile live broadcast system also comprises an RTSP signaling server for sending an RTSP request open signaling or an RTSP request stop signaling. The mobile live broadcast system based on MPEG-2 standard improves the packaging mode and transmission mode of video streaming, can be applied to a mobile phone terminal to build a new signaling system to perform unified management, and thereby communication stability of every part of the system are ensured to be good.

The invention discloses an SQL (Structured Query Language) conversion method and system based on language model coding and multi-task decoding. The method comprises the steps: combining a language model in combination with a field where a data set is located to carry out pre-training, and improving the feature extraction capability in the field; sequentially expanding the query database according to table names and column names, converting a two-dimensional table into a one-dimensional text sequence, and splicing the one-dimensional text sequence into an input sequence X in combination with user questions; inputting the sequence X into a pre-training language model, and outputting a coding result; a multi-task decoder composed of nine different neural networks is utilized to decode and restore the SQL fragments, and cross entropy loss is calculated; different weights are set for loss values of different neural networks, the sum is finally calculated as the total loss of the model, a gradient descent algorithm is utilized to optimize an objective function, and model training parameters are updated; after training is completed. Model parameters are stored, and a corresponding SQL sequence is automatically generated according to the user problem and the target database.

The invention discloses an abnormal detection method for a gas circuit of an aero-engine based on deep learning and Gaussian distribution and relates to an abnormal detection method for a gas circuit of an aero-engine. The invention aims to solve the problems that in an existing abnormal detection method for the gas circuit of the aero-engine, QAR data are not widely applied, the false alarm rate of abnormal detection of the engine is high and the accuracy is low. The method comprises the following steps: I, selecting a parameter set in the QAR data, wherein the data set comprises a performance parameter of the gas circuit of the engine and an external environmental parameter; II, calculating the difference value of the performance parameters of two engines on a same plane in the parameter set, and forming a novel parameter set by the difference value and the external environmental parameter; III, extracting data characteristics of the novel parameter set in the step II by using an accumulating and noise-eliminating automatic coder model in a deep learning method; and IV, performing abnormal detection on the data characteristics obtained in the step III by a density estimation algorithm based on Gaussian distribution to obtain a result. The method disclosed by the invention is used in the technical field of fault diagnosis of the aero-engine.

The invention discloses an area-of-interest detection method based on background prior and a foreground node. The method comprises steps of 1) by use of SLIC algorithm, segmenting an original image into super-pixels; 2) by use of K-means clustering algorithm, carrying out clustering on boundary super-pixels, according to a clustering result, constructing a global color difference matrix and a global space distance matrix, fusing the global color difference matrix and the global space distance matrix into a saliency map based on background prior, and finally, by use of a single-layer cellular automaton, primarily optimizing the saliency map based on the background prior; 3) carrying out adaptive threshold segmentation on the saliency map based on the background prior so as to obtain a foreground node, according to a contrast ratio relation, obtaining a saliency map based on the foreground node, and by use of the biased Gauss filtering, carrying out optimization; and 4) fusing the saliency map based on the background prior and the saliency map based on the foreground node, obtaining the final saliency map. According to the invention, the method is used in an image processing processand can be widely applied in visual working field like visual tracking, image segmentation and target re-positioning.

The invention discloses a scale-invariant feature extreme learning machine-based robot scene recognition method. The method comprises the following steps: firstly, expressing a scene of a robot visual image by using a scale-invariant feature; secondly, realizing a robot scene image expression codebook by using a K mean clustering algorithm; finally, establishing mapping relation between the robot visual scene image and a scene label by using the extreme learning machine algorithm. According to the method, the complexity of parameter estimation and optimization of a conventional neural network is reduced, the training time is further shortened and the recognition rate of the robot scene image is improved by using the advantages of the extreme learning machine.

The invention relates to a statistical machine translation method based on fuzzy tree-to-accurate tree rule, in particular to a method fully and rightly using the source language end syntactic structure knowledge to improve the statistical machine translation quality based on a string-to-tree translation model. The method comprises the steps of: conducting word segmentation, automatic word alignment and syntactic analysis on the bilingual sentence; automatically extracting the fuzzy tree-to-accurate tree translation rule from the parse tree of the bilingual sentence with word alignment; conducting probability estimate on the translation rule extracted, and training a language model of the target end; designing the matching criterion of the source language end syntactic structure with the fuzzy tree-to-accurate tree translation rule, and estimating the matching probability thereof; and designing the optimization objective of the translation model, and using the fuzzy tree-to-accurate tree translation rule and the language model of the target end to search the target translation of the test statement. The availability of the statistical machine translation method is verified on the translation task from Chinese to English in the international mechanical translation evaluation.

The invention discloses an evolutionary algorithm for surface fitting in human body modeling, belonging to the technical field of computer application. Based on a principle of being a burgeoning searching optimizing technology, the evolutionary algorithm has the beneficial effect of providing a general framework for solving complicated systems. According to the invention, B spline surface fitting based on the evolutionary algorithm is applied to the human body modeling; a curve boundary line is determined first by the B spline surface fitting method of the evolutionary algorithm so as to obtain a basal plane; surface control points are obtained through reverse engineering according to the basis principle of the surface fitting; and then optimization selection is carried out on the judgment on the fitness of the control points according to the evolutionary algorithm, so that the surface fitting in the human body modeling is realized finally. According to the invention, the calculation speed and efficiency are improved by virtue of the robustness of the evolutionary algorithm, the solving process is simplified, errors generated in the fitting process are reduced, and further the advantages of the evolutionary algorithm is proved.

The invention provides a gas-liquid two-phase flow pattern dynamics representation and identification method based on multi-scale arrangement entropy. The method includes the steps of firstly carrying out a gas-liquid two-phase flow pattern experiment with air and water as media to collect three kinds of gas-liquid two-stage different-flow-pattern electrical conductance fluctuation signals, then carrying out coarse graining processing on flow-pattern signal sequences according to a multi-scale concept to obtain coarse graining time sequences, calculating arrangement entropy of the time sequences in different scales, drawing a distribution map of the multi-scale arrangement entropy, analyzing dynamics evolution characteristics of the multi-scale arrangement entropy according to the gas-liquid two-phase different-flow-pattern characteristics, ultimately calculating the multi-scale arrangement entropy rate according to distribution map of the multi-scale arrangement entropy of different flow patterns to obtain distribution of the multi-scale arrangement entropy of all the flow-pattern signals, and accordingly achieving identification and classification of the flow patterns. According to the method, complexity of gas-liquid two-phase flow-pattern signals is disclosed in terms of time sequence themselves. The method has the advantages of being simple and quick in calculation, good in robustness and the like and is especially suitable for real-time processing of the two-phase flow-pattern signals.

The invention discloses a hot-rolled strip steel surface defect classification method based on a convolutional neural network, and belongs to the field of computer deep learning. The method comprisesthe following steps: firstly, obtaining a hot-rolled strip steel surface defect typical image sample from an NEU database, and preprocessing the sample; building a convolutional neural network VGG16 model, and building a plurality of classification models for the surface defects of the hot-rolled strip steel based on the VGG16 model in combination with an SGD or Adam optimization algorithm; then,based on a plurality of built classification models, identifying and classifying the hot-rolled strip steel surface defect typical image samples obtained in the step; evaluating results obtained by the plurality of classification models to obtain an optimal classification model; and finally, based on the optimal classification model, hot-rolled strip steel surface defect classification is carriedout. The method for identifying the surface defects of the hot-rolled strip steel is high in accuracy and high in classification speed, and can be effectively applied to on-site real-time detection ofthe surface defects of the hot-rolled strip steel.

The invention discloses a DLLE (Linear Local Embedding of Difference) model-based data dimension reduction and characteristic understanding method, and belongs to the field of computer vision. The method comprises the steps of firstly, obtaining an image sequence through a visual sensor, then analyzing an input motion image sequence, extracting a foreground human body contour region through a background subtraction method, performing binarization, researching a periodic characteristic of a motion, performing key frame extraction on each motion sequence, and extracting a complete motion periodic sequence; performing manifold dimension reduction through a DLLE algorithm to obtain a low-dimensional eigenvector, and storing the low-dimensional eigenvector in a motion database; and performing identification through a nearest neighbor classifier by comparing a mean Hausdorff distance between a test sequence and a motion sequence in a training sample library. According to the method, the application of a differential function and category information-based neighborhood preserving embedding algorithm to human body motion identification is proposed; a DLLE model can not only keep a manifold local geometric structure during dimension reduction but also fully utilize category information of original high-dimensional data; and the extension from unsupervised extension to supervised extension is realized.

The present invention discloses a quantum cryptography implementation method based on a quantum light source. The method comprises the following steps: generating a labeled single photon source through a PDC process, transmitting signal light to a receiving end, and allowing casual light to be used for local detection. The local detection of the transmitting end comprises a beam splitter (BS), a detector D1 and a detector D2. The casual light passes by the beam splitter to trigger the two detectors to generate four different response events. The four different response events are recorded andused for estimation and processing the signal received by the receiving end, so as to extract a secret key. Due to the usage of a novel passive decoy method, the single photon response rate and bit error rate can be estimated accurately. With a low-loss unequal-arm MZ interferometer used, the consumption of the system is greatly reduced. The single-mode standard commercial light enables the QKD system transmission distance to be over 200 km and based on a quantum light source, so that the final secret key extraction rate is two orders higher than that in the previous QKD system using the quantum light source.

The invention relates to a pulse signal cluster sorting method based on class merging. The method aims to overcome the defects that the class number accuracy in the existing cluster result is low and the cluster number is inconsistent with the real signal number after class merging. The pulse signal cluster sorting method based on class merging specifically comprises the following steps: 1, determining initial cluster centers and sorting distances; 2, obtaining new cluster centers; 3, calculating whether the new cluster centers satisfy signal features; and 4, merging the cluster centers satisfying the signal features, thereby accomplishing signal cluster sorting based on class merging. The method is applied in the field of signal processing.

The invention discloses a power price prediction method based on an improved deep belief network, and the method comprises the steps: dividing a data set and determining the input of network data according to the characteristics of electricity price data and the influence factors of electricity price, and carrying out the data preprocessing of an adopted data set; for the preprocessed data set, calculating a network error by using a second-order reconstruction error, and determining the number of layers of the model RBM; optimizing the number of neuron nodes in the network by using a '3 + 2 'search algorithm combining a trisection method and a bisection method; using a BP neural network and an SVR support vector regression machine used as regression layers of a DBN network, and using the number of layers of an RBM and the number of optimized neuron nodes to construct a DBN-BP model with an optimized structure and the DBN-SVR model with an optimized structure; and predicting the real-time electricity price data. According to the invention, the DBN model with an optimized structure is established, and different combination improvements are carried out on the regression layer of the network, so that the prediction precision of the DBN is improved, and the application prospect is very good.

A scheduling method based on green-energy-aware uses solar energy as green energy and conducts optimizing and distributing on works reaching a data center. The scheduling method based on green-energy-aware comprises the following steps: predicting according to illuminated scope and weather conditions to acquire solar energy which can be used by the data center; preprocessing the works reaching to the data center, each work comprising a plurality of tasks, communication constraint relationships existing among all tasks, conducting grading on works needing to schedule according to the communication constraint relationships in works, and gradually distributing tasks without communication constraint relationships to every grade; under requirement of meeting work time limit, conducting scheduling on preprocessed work assembly according to grades, sequentially distributing task of every grade to every server, fully considering loss conditions of the servers and communication network in the distributing process, and aiming at achieving highest solar energy use ratio. The scheduling method based on green-energy-aware can be applied to scheduling of the data center of random dynamic, achieve the effect that the solar energy use ratio is highest and cost of the data center and carbonic emission are lowered.

The invention discloses a domain adaptive pedestrian re-identification method based on mutual divergence learning. The method comprises the following steps: preparing a pedestrian data set; pre-training the source domain data set, and extracting feature vectors of pictures from the target domain data set; performing density-based clustering on the images of the target domain data set, and taking the number of the cluster as a pseudo label; adding the outliers into a training sample by using an adversarial strategy; mixing the clustered samples and the outliers, sending the mixture into a network, correcting noise of a pseudo tag by adopting mutual divergence learning, inputting a pedestrian image to be queried into a trained pedestrian re-identification model to obtain a pedestrian feature vector to be identified, performing similarity comparison on the pedestrian feature vector to be identified and attribute features in a candidate library, and obtaining a pedestrian re-identification result. According to the invention, the distribution difference between the source domain and the target domain is reduced, the knowledge of the source domain is effectively utilized, and finally, the framework can learn the characteristics with robustness and discrimination.

The invention discloses a method for estimating positions of images based on area mining and space encoding. The method comprises an offline part and an online part, wherein the offline part comprises the following steps of extracting whole features of the images in a GPS (global positioning system) image library, and clustering the images; performing the area mining and BOW (bag of words) position description character generation on each image in the GPS image library; finally, according to visual words, establishing a reverse index table for the whole GPS image library; the online part comprises the following steps of obtaining a candidate image set of the inputted images through the extracting and matching of the whole features; performing the operations on the inputted images, wherein the operations are the same as the area mining and BOW position description character generation of the offline part; utilizing the reverse index table in the offline part to match the images based on the area, and finally obtaining the GPS positions of the inputted images.

The invention discloses a design method of an adaptive inversion controller in a flexible active suspension system. The method comprises the following steps of: 1, establishing an active suspension model; and 2, designing an adaptive inversion controller based on a neural network; and aiming at a two-degree-of-freedom flexible active suspension system with unknown nonlinear dynamics, in the control design of the automobile active suspension system, selecting a BL function to construct a control condition meeting system stability, and meanwhile, introducing a second-order auxiliary system to realize asymmetric control saturation compensation. Through the method, the adaptive inversion controller based on the neural network is designed and obtained; and hardware loop experiments show that compared with traditional PD and LQR control, the control method can better attenuate vertical vibration of sprung mass in the automobile active suspension system under sine and impact excitation in thecontrol design of the automobile active suspension system, and meanwhile grounding performance of the automobile active suspension system is improved.

A Chinese medicine for treating gastritis and gastric or duodenal ulcer is prepared from 7 Chinese-medicinal materials including notoginseng, coptis root, bletilla tuber, corydalis tuber, etc through proportional mixing, pulverizing, sterilizing at 110 deg.C for 30 min, and shaping. Its advantages are high curative effect and cure rate, durable act, and no toxic by-effect.

The invention discloses a zero-sample image classification and recognition method based on a multi-part attention mechanism, and the method comprises the following steps: S1, training a multi-part convolution detector; s2, training a semantic feature extractor; s3, acquiring pictures of the training set, and processing the pictures by training an attention detector; s4, carrying out loss calculation; and S5, repeating the calculation of the steps S3 and S4, performing testing when the algorithm loss is lower than a preset value, and selecting the minimum distance as a category value. Accordingto the method, a semantic segmentation mode is adopted, firstly, semantic segmentation is carried out on the whole picture to obtain effective parts, unnecessary redundant information is screened out, and then feature extraction is carried out on the multiple parts; for different parts, an attention mechanism is put forward to act on the different parts for weighting, so that each sample has a different weighting mode, and therefore, for each sample, parts with high weights can be generated, and the parts can better distinguish the parts from other types.

The invention relates to a biomarker combination identification method and system based on biomedical big data. The method includes steps of 1, acquiring a large amount of clinical data of proteins or metabolic small molecules in serum of patients and healthy people; 2, according to the acquired clinical data, adopting a latest central classifying framework to establish an optimal model of biomarker combination identification; 3, analyzing single features one by one to acquire prediction capacity of the single features, and ranking all the features according to the prediction capacity of the single features; 4, focusing on the features with the prediction capacity, arranging the clinic data into determined formats and inputting the clinic data to the optimal model of biomarker combination identification so as to estimate prediction capacities of multiple biomarker combinations, and determining an optimal biomarker combination according to errors of minimizing classification. The method focuses on identifying the biomarker combinations, and can be applied in biomedical big data of gene expression, protein combination and the like.