301 results about "Discriminant model" patented technology

The invention discloses a hierarchical network attack identification and unknown attack detection method based on deep learning, which comprises the following steps of: learning behavior modes of normal flow and abnormal flow by using an auto-encoder so as to judge whether the flow to be detected is the normal flow or the abnormal flow by using the learned auto-encoder; using a deep neural networkas a discrimination model to discriminate whether the attack type of the to-be-detected flow is a certain known attack type or an unknown type; and fusing the results of the auto-encoder and the discrimination model to complete network attack classification and unknown attack detection. According to the method, normal flow can be identified, known abnormal flow can be classified, novel unknown network attacks can be detected, and the detection accuracy is improved by combining the advantages of the generation model and the discrimination model.

The invention discloses a regional automatic weather station hourly rainfall data quality control method. The method comprises the steps that a discriminant model is established according to historical data; raindrop spectrometer data, weather radar data and automatic weather station rainfall data are collated in real time; the data of a number of raindrop spectrometers are used to evaluate and revise a radar echo intensity result; the relationship between the radar echo intensity and the rain intensity is fitted according to different precipitation types; the fitted relationship is substituted into a radar quantitative estimation precipitation module to realize radar quantitative rainfall estimation to acquire hourly rainfall; the set of weather radar estimated hourly rainfall and regional automatic weather station hourly rainfall is established; and comparing with the pre-established discriminant model is carried out to judge whether the regional automatic weather station hourly rainfall data are abnormal. The automatic weather station rainfall data are accurate and reliable. The observation data quality is improved. The method lays a good foundation for weather monitoring, early warning, forecasting and other businesses.

The invention discloses a pedestrian re-recognition method based on migration learning. The pedestrian re-recognition method includes the steps of 1) preprocessing pictures in a data set by using a multi-scale Retinex color image enhancement algorithm to remove the influence caused by illumination; 2) concentrically extracting relevant features of pedestrians from the preprocessed data set; 3) establishing an asymmetric multi-task discriminant model; 4) refining the model by using a flow-shaped structure without label data; 5 ) optimizing the objective function by using a conjugation gradientmethod to obtain a final classifier; and 6 ) testing the pedestrian re-recognition rate on the classifier based on the pedestrian re-recognition standard database. Few label data and the label-free data are reasonably applied, and the model obtained through training can well improve the pedestrian re-recognition rate.

The invention discloses an image scene segmentation and layering joint solution method based on component set sampling. The method comprises the following steps of: performing over-segmentation treatment on an input image to obtain a super-pixel set of the image; training on a training dataset to obtain a discriminant model of semantic category and a discriminant model of layer category, and obtain a probability value (of each super-pixel in the input image) belonging to each semantic category and a probability vale belonging to each layer category according to the two models; structuring a candidate graph structure of the input image and calculating a node weighted value, a positive side weighted value and a negative side weighted value; and based on the a candidate graph structure, obtaining an optimal solution by reasoning via a component set sampling algorithm, wherein the optimal solution includes the exact semantic category and the exact layer category of each super-pixel of the input image. The method disclosed by the invention can be widely used for semantic information and layer information labeling of computer visual systems of military, aviation, aerospace, monitoring and manufacturing, and the like.

The invention discloses a brake shoe breaking target detection method, and relates to a rail wagon fault detection method. The objective of the invention is to solve the problems of high cost, low efficiency and low accuracy in train brake shoe image inspection in the prior art. The method comprises the steps of 1, acquiring a linear array image; 2, performing coarse positioning; 3, generating anadversarial network DCGAN, and generating a fault image based on the adversarial network DCGAN; wherein the adversarial network is composed of a discrimination model and a generation model, down-sampling convolution is adopted in a discriminator, and up-sampling convolution is adopted in a generator. The step 3 comprises the further steps of 3.13, constructing an adversarial network DCGAN discrimination model; 3.2, constructing an adversarial network DCGAN generation model. The method also comprises the following steps of 4, establishing a deep learning training data set; 5, segmenting a faulttarget; and 6, performing prediction based on the trained segmentation network model to obtain information of the fault component. The beneficial effect of the invention is that the method is used inthe field of railway freight car fault detection.

The embodiment of the present invention provides a method and a device for discriminating threat information credibility based on a multi-dimensional trusted feature. The method comprises the following steps: acquiring threat information to be detected; obtaining a verification threat information set corresponding thereto according to a category of the threat information to be detected; accordingto a content verification consistency recognition algorithm, calculating a similarity value between the threat information to be detected and the verification threat information; comparing the similarity value with a preset threshold value, and determining the threat information to be detected with the similarity value greater than the threshold as initial credibility threat information; extracting the multi-dimensional trusted feature of the initial trusted threat information, and constructing a multi-dimensional trusted feature vector; inputting the multi-dimensional trusted feature vector into a deep belief network DBN discriminant model, and outputting a credibility discrimination result of the initial trusted threat information. The embodiment of the present invention judges the threat information to be detected twice by using the content verification consistency method and the DBN discriminant model, thereby improving the accuracy of judging the threat information to be detected.

The invention discloses a Jinhua ham grading and identifying method based on an electronic nose technology. The method comprises the following steps of: putting standard samples into a sealed container, and sealing the standard samples in a thermostat at the temperature of 30 degrees for 30 minutes, wherein special grade, first grade and second grade Jinhua hams which are identified by a sensing method according to national recommended standard of GB / T19088-2008 original producing area products of Jinhua hams are taken as standard samples; pumping headspace gas into an electronic nose through a sampling channel, adsorbing a certain amount of volatile substances by a sensor so as to change the conductivity, acquiring signals by a data acquisition system, and storing the signals in a computer; and performing principal component analysis, linear discriminant analysis and partial least square analysis on the acquired data by using Unscrambler software, an WinMuster software and the like, establishing discriminant models of the hams in various grades, and predicting the samples to be tested by using the models. The method provided by the invention is convenient to operate and accurate and objective in result, and the traditional sensory grading can be replaced..

The method according to the present invention includes: a training sample measurement process (S1, S2) in which, for a food product belonging to the same kind as a determination target, a plurality of training samples individually labeled with a known state of quality are subjected to a measurement using a chromatograph mass spectrometer under the same analysis condition; a training sample data collection process (S3, S4) in which an index value related to the magnitude of a peak observed on an extracted ion chromatogram obtained by the measurement is acquired for each training sample, and the index value of the peak at each retention time common to the training samples is extracted; and a discrimination model creation process (S5-S7) in which a supervised training is performed to create a discrimination model, using, as the training data, the index value of the peak at each retention time common to the training samples acquired for each of the labeled training samples. Measurement data for an unknown sample is inputted into a discriminator based on the discrimination model, to obtain a quality discrimination result.

To provide an optimized query generating device capable of generating an optimized query to be given with domain knowledge when generating a discriminant model on which the domain knowledge indicating user's knowledge or analysis intention for a model is reflected.A query candidate storage means 86 stores candidates of a query which is a model to be given with domain knowledge indicating a user's intention. An optimized query extraction means 87 extracts queries having low uncertainty of a discriminant model estimated by queries given with domain knowledge when the domain knowledge is given thereto from query candidates.

The embodiment of the invention provides a generative adversarial network training method and device, and the method comprises the steps: fixing parameters of a discriminator model, and carrying out iterative optimization of the parameters of a generator model through a loss function till the similarity of the generator model reaches a first threshold value; fixing the parameters of the generatormodel, and iteratively optimizing the parameters of the discriminator model through a loss function until the discrimination rate of the discriminator model reaches a second threshold; wherein the value of the first threshold value and the value of the second threshold value are both related to the current alternating frequency; adding 1 to the alternating times; repeating the above steps until the discriminator model and the generator model reach Nash equilibrium; and determining the trained generative adversarial network according to the current discrimination model and the current generation model. According to the embodiment of the invention, the iteration time of the generator model and the discriminator model can be controlled, the training efficiency of the generative adversarial model can be improved, and the situation of model crash is avoided.

The invention discloses an adversarial generative network for defending a text malicious sample and a training method thereof. A Generator and a Discriminator in an adversarial generative network framework are used for defending and generating the malicious sample. The generator part is composed of an auto-encoder, discrete text data is mapped into a continuous high-dimensional hidden space, and therefore the generator can generate malicious text through a hidden vector. The discriminator is a discrimination model and is used for identifying data. A malicious text generated by the generator ismarked with a real label and input into the discrimination model together with a real sample, so as to train the discrimination model. By adding the discrimination model trained by malicious samples,text data can be identified accurately and efficiently. The generator trains the evaluation score of the malicious sample and the difference between the text data and the malicious sample by using adiscrimination model to generate a malicious sample with stronger attack force. Due to the addition of malicious samples in the training process and the network training process of resistance, the text data network recognition capability, the anti-interference capability and the defense capability are greatly improved.

The invention discloses a target pedestrian trajectory prediction model training method and device, electronic equipment and a storage medium, and the target pedestrian trajectory prediction model training method comprises the steps: inputting a first historical position vector set and a second historical position vector set into a target pedestrian trajectory prediction model, and obtaining a prediction position vector set; inputting the real position vector set and the prediction position vector set into a discrimination model to obtain a loss value corresponding to the distance set; and ifthe loss value does not meet the preset condition, training parameters in the target pedestrian trajectory prediction model based on the loss value until the loss value meets the preset condition, wherein the first historical position vector set is composed of first historical position vectors of the target pedestrian at a plurality of historical preset moments, and the second historical positionvector set is composed of a second historical position vector sub-set corresponding to each surrounding pedestrian, so that the target pedestrian trajectory prediction model can be trained based on the data features in the time dimension and the space dimension, and the prediction capability of the target pedestrian trajectory prediction model can be improved.

The invention discloses an adversarial sample defense method based on feature remapping and application. The method comprises the steps of: constructing a feature remapping model, wherein the featureremapping model comprises a significant feature generation model used for generating significant features, a non-significant feature generation model used for generating non-significant features and ashared discrimination model used for discriminating the authenticity of the significant features and the non-significant features; constructing a detector according to the significant feature generation model and the non-significant feature generation model, wherein the detector is used for detecting an adversarial sample and a benign sample; constructing a re-identifier according to the significant feature generation model, wherein the re-identifier is used for identifying the category of the adversarial sample; when adversarial sample detection is carried out, connecting a detector to the output of the target model, and carrying out adversarial sample detection by utilizing the detector; and during adversarial sample identification, connecting the re-identifier to the output of the target model, and performing adversarial sample identification by using the re-identifier. The dual defense effect of detection and re-identification of the adversarial sample can be realized.

According to the crowd counting method based on the multi-scale generative adversarial network, crowd density prediction is carried out by adopting an adversarial training mode. And the maximum and minimum problems of the generation model and the discrimination model are optimized by adopting a combined alternate iteration training mode, wherein the training generation network is used for generating an accurate crowd density map so as to cheat the discriminator, and on the contrary, the discriminator is trained to be used for discriminating the generated density map and a real density map label. Meanwhile, the output of the discriminator can provide feedback of the density map position and prediction precision for the generator. The two networks compete for training at the same time so asto improve the generation effect until the sample generated by the generator cannot be correctly judged by the discriminator. According to the crowd density detection algorithm provided by the invention, after adversarial loss is introduced, an adversarial training mode is adopted to enable the convolutional neural network to generate a density map with higher quality, so that the accuracy of network crowd counting is improved.

The invention provides a method for discriminating types of propolis raw materials. The method comprises the following steps: (1) crushing a propolis sample to be detected, dissolving the sample in absolute ethyl alcohol, and preparing a mixture, wherein the mass and volume ratio of the sample to the absolute ethyl alcohol is 1:25; (2) shaking the mixture for 1.5-3 hours, performing ultrasonic treatment for 1-1.5 hours, performing centrifugal separation at low temperature, and taking the separated liquid; and (3) measuring a chromatic value of the separated liquid by using a colorimeter, substituting in a linear discriminant model to calculate a Y value, and discriminating the sample to be detected as a propolis source type with the maximum type. The provided method is easy to operate, has low requirements on instruments and levels of detectors and is easily popularized in actual production; the accurate discrimination rate is high; the discrimination range of propolis source plants is wide, and common propolis source types in the propolis raw material market in China are almost covered; the measurement result is measurable and is not influenced by subjective judgment of the detectors; moreover, the discrimination result is clear and is easily standardized.

The invention provides a method for predicating a water inrush source by a Fisher discriminant model based on PCA analysis. The method comprises the steps of collecting water quality type data of eachmain aquifer in a research area in recent years; building the Fisher discriminant model of a PCA method based on actual water quality data; and comparing a model result with an actual condition, thereby predicting the water inrush source of the research area. According to the method, the water quality of each main aquifer is analyzed based on the PCA method; overlapping interference factors amongions are eliminated; main influence factors are selected; a water inrush source discriminant model is built in combination with a Fisher discriminant analysis method; and the prediction of the waterinrush source of the research area is carried out. The method has a certain novelty, is simple in operation and high in practicability, and provides a new idea and method for predicting the water inrush source.

An image description generation method based on a conditional generative adversarial network comprises the following steps of 1, constructing a network, wherein a conditional generative adversarial network framework is composed of a generation model and a judgment model, the generation model and the judgment model are similar in structure, but the parameters are independently trained and updated; 2, preprocessing a data set; 3, performing network training, wherein the process comprises 3.1, initializing the parameters of a genearation model and a discrimination model by using the random weights; 3.2, training the generation model; 3.3, training a discrimination model; 3.4, minimizing a loss function by using an RMSprop descent algorithm; and 4, testing the precision, generating the description of the test picture through the operation of the above steps. The image description generation method based on the conditional generative adversarial training is better in robustness and lower in requirement for training data.

The invention provides a target tracking method based on depth feature and average peak correlation energy. The method comprises the following steps: extracting color histogram feature of the target,depth feature and three-layer depth feature of upper, lower, left and right image blocks of the target, and calculating a color histogram discrimination model and a depth feature model; calculating the color histogram characteristic response and the depth characteristic response of the target of the current frame, and predicting the target position of the next frame; calculating an average peak correlation energy of the target response of the current frame; if the average peak correlation energy of the current frame target response is greater than the average peak correlation energy of all frames before the current frame, determining that the confidence level of the frame response is high, and updating the color histogram discriminant model and the depth feature model by using a layered model update scheme, otherwise, the color histogram discriminant model and the depth feature model being not updated; repeating the above steps until the video sequence ends. The invention effectively fuses the depth feature and the average peak correlation energy, and adopts a layered model updating scheme to further effectively improve the tracking performance.

The invention discloses an opportunity network topology prediction method and device based on a cyclic generative adversarial network, and the method comprises the steps: carrying out the slicing of the communication data of an opportunity network through a time sequence analysis method, and obtaining a series of discrete network snapshot data; randomly extracting and grouping all nodes in the opportunity network to obtain a plurality of observation node groups; intercepting the network snapshot by using a sliding window with a preset size, and extracting a network topology structure change sequence of each observation node group under each sliding window; constructing a generative adversarial network model by utilizing the generative model and the discrimination model, training the generative adversarial network model by taking the network topology structure change sequence as a training sample, and performing topology structure prediction on the opportunity network by taking the generative model in the trained generative adversarial network model as a prediction model. According to the method, the prediction model is used to extract the high-order spatial-temporal characteristicsof the topological structure evolution law of the opportunity network, and topology prediction of the opportunity network at the future moment is realized.

The invention provides a method for discriminating unfertilized eggs before incubation on the basis of a visible near infrared spectrum. The method comprises the following steps: (1) preparing a to-be-detected sample and acquiring spectral information of the to-be-detected sample; (2) selecting a representative modeling sample by a mathematical analysis method; (3) determining a discrimination model of the unfertilized eggs before optimal incubation; and (4) discriminating the to-be-detected eggs by the discrimination model of the unfertilized eggs before optimal incubation. The invention alsoprovides a device for detecting the unfertilized eggs before incubation on the basis of the visible near infrared spectrum so as to implement the above method. The experiment proves that the accuraterate of discriminating an unknown sample by the optimal discrimination model which is determined by the method for discriminating the unfertilized eggs before incubation based on the visible near infrared spectrum, which is provided by the invention, is 94.77 percent.

The invention discloses a model interpretation method, apparatus and device, and a readable storage medium. The method comprises the steps of performing interpretation processing on a model predictionresult of a target user by adopting a preset interpretation algorithm to obtain an interpretation result, generating a plurality of interpretation text contents according to the interpretation result, and analyzing each interpretation text content by adopting an acceptability discrimination model to obtain an acceptability corresponding to each interpretation text content, the acceptability discrimination model being obtained by training feedback data of historical interpretation text contents based on historical users, and selecting a target interpretation text content from the interpretation text contents based on the acceptability and outputting the target interpretation text content. According to the method, the actual application effect of model interpretation is improved.

The invention discloses transformer target detection and appearance defect recognition method based on VGG-net style migration, and relates to the field of image recognition. At present, the appearance of the equipment in the substation is mainly based on the image acquisition of the robot. Due to the image acquisition error, the appearance detection may be offset, resulting in inaccurate appearance detection. In addition, the collected equipment images are mostly positive samples, including rust, oil leakage and the like. There are fewer negative samples of appearance defects, which will leadto model training over-fitting, poor generalization ability, and easy to cause false detection. The method first collects samples and constructs a sample set; then uses the SSD target detection algorithm to accurately intercept the target device for detecting appearance defects; and then uses the VGG-net-based style migration algorithm to generate defect samples for the problem of insufficient negative samples. The sample set is extended to improve the generalization ability of the discriminant model; finally, the appearance is detected according to Le-net's discriminant network. Transformertarget detection and appearance defect identification are accurately realized.

The invention provides a rapid detection method for grain fungal pollution cases and applications thereof and relates to the microbiological detection field. The method comprises the following steps: grains polluted with control fungi and grains without fungi are mixed uniformly, and grain samples polluted with different control fungi or grain samples polluted with control fungi with different concentrations are obtained; the Fourier transform attenuation total reflection infrared spectrometer is employed to collect the infrared spectrogram and data of each polluted sample in characteristic bands; the infrared spectrometer data of each sample is subjected to principal component analysis, the principal component value is extracted, fungal discriminant models with different types or different concentrations are established by utilization of linear discriminant analysis or partial least square discriminant analysis, and the reliability of the discriminant models is verified; an infrared spectrogram and data of a grain sample to be detected in the characteristic bands are collected, and the model is employed to determine the type or the concentration of the polluted fungi in the grain sample to be detected. The method can carry out qualitative analysis and rough quantification of polluted mould in grains rapidly.

A method of extracting a combination of a drug and an adverse event related to the drug includes: for each of positive example combinations, negative example combinations and combinations that are neither positive examples nor negative examples, which are combinations of drug and disease, extracting medical events from medical information data about a patient and generating attribute data based on time-series information about the medical events; and learning a discriminant model based on attribute data of the positive and negative examples; and inputting attribute data corresponding to the combinations that are neither positive examples nor negative examples to the discriminant model to determine scores.

The invention provides an aircraft fuel system fault prediction method based on flight parameter data. The method comprises the following steps that a time series data set composed of N aircraft parameters sensitive to faults in an aircraft fuel system is obtained; constructing a long-short-term neural network model according to the time series data set, and obtaining an actual prediction result of the time series data set according to the long-short-term neural network model; determining a fault discrimination model f (x) according to the time series data set; acquiring a real data set of fault conditions of a part of known aircraft fuel systems, and determining a fault threshold according to the fault discrimination model f (x) and the real data set; and determining the probability density of an actual prediction result according to the fault discrimination model, and judging the health condition of the aircraft fuel system by comparing the probability density of the actual prediction result with a fault threshold. According to the method, the aircraft fuel system is subjected to fault prediction in a mode of combining the long-term and short-term neural network model and the fault discrimination model f (x), so that the health condition of the aircraft fuel system can be effectively predicted.

The present invention discloses a muscat production place identifying method, which comprises: (1) collecting grape samples from muscat main production area production places such as Tianjin Chadian, Hebei Changli and Shandong Dazeshan; (2) homogenizing the grape sample, and carrying out digestion to achieve a clear state; (3) detecting the concentrations of K, Na, Ca Fe, Mn and Cu in the digestion solution, wherein the concentrations are sequentially marked as XK, XNa, XCa, XFe, XMn and XCu, and the unit is mg / kg; (4) carrying out statistical analysis on the detected data by using a SPSS statistical software, and establishing discriminant models (1)-(3); and (5) substituting the detected data of the six grape sample elements into the discriminant models (1)-(3) so as to obtain the YChadian, the YChangli and the YDazeshan, wherein the muscat production place is the production area having the maximum value in the YChadian, the YChangli and the YDazeshan. According to the present invention, the discriminant models are established through the six elements, the types of the elements are less, the elements are the conventional elements, the method is simple, the detection cost is reduced, and the determining correct rate is more than or equal to 90%.

The invention discloses a self-adaptive radar marine clutter prediction system which comprises a radar, a database and an upper computer, wherein the radar, the database and the upper computer are sequentially connected; the radar is used for irradiating a marine area to be detected and storing radar marine clutter data to the database; and the upper computer comprises a data preprocessing module, a robustness prediction module modeling module, a self-adaptive gunz optimizing module, a marine clutter prediction module, a discriminant model updating module and a result displaying module. The invention also provides a self-adaptive radar marine clutter prediction method. The self-adaptive radar marine clutter prediction system and method provided by the invention have the advantages of parameter self-adaption, good optimizing effect and high prediction efficiency.

The invention discloses a machine learning-based falling discrimination method of an intelligent carpet. On the intelligent carpet designed according to reasonable human body walking rules, a machinelearning method is used to use data change situations of carpet stepping points on time and space in walking processes as input features of machine learning, information is stored into a database through repeated walking, matching is carried out on stored feature information and actual walking situations of users, and thus a feature environment based on machine learning is realized. A hidden Markov probability transfer model is established before a classification algorithm is carried out, the carpet is enabled to obtain a self-adaptive object discrimination function, invalid stepping point information on the carpet is removed, and target user information is retained. In a training process, an SVM (Support Vector Machine) is used for training, different kernel function feature data are selected for repeated training, two processes of walking and falling are repeated on the intelligent carpet according to different training results, a highly reliable falling discriminant model is established.

The invention provides an optimized model for rapid identification of transgenic soybeans based on morphological analysis. According to the invention, firstly, an establishment method of the optimizedmodel for rapid identification of the transgenic soybean based on morphological analysis is provided; for whole soybeans, spectral information under a characteristic wave band of 9403-5438cm<-1> is selected, the spectrum is preprocessed by adopting a second derivative, and a PLS-DA model is established by adopting a partial least squares-discrimination method; and for powdery soybeans, spectral information under a characteristic waveband of 7505-4597cm<-1> is selected, the spectrum is preprocessed by adopting vector normalization and a first derivative, and a PLS-DA model is established by adopting a partial least squares-discrimination method. According to the invention, the transgenic soybeans are identified by combining the near-infrared spectrum with a discriminant analysis method, and the identification accuracy of the discrimination model can be improved by selecting the sample form, the wavelength range and the spectrum pretreatment method, so that the optimal model is selectedto be applied to actual production.

The invention relates to the field of computer vision and pattern recognition, in particular to a target tracking method based on deep learning and discriminant model training and a memory, and aims to improve the positioning precision of target tracking. The tracking method comprises the following steps of: in an offline training stage, extracting sample frame features from a training image and atest image by using a depth feature extraction network, and calculating a sample frame label of the training image and a sample frame first label of the test image; utilizing a discriminant model solver to train to obtain a discriminant model according to the sample frame feature and the label of the training image; according to the sample frame feature of the test image, using the discriminant model to perform prediction to obtain a second label; calculating network prediction loss according to the second label and the first label so as to drive optimization learning of the deep feature extraction network; in an online tracking stage, using the trained depth feature extraction network in a target tracking algorithm based on online discriminant model training. According to the method of the invention, the positioning precision of target tracking is effectively improved.