797 results about "Cross-validation" patented technology

The invention provides an automatic detection method for a multi-scale polymorphic target in a two-dimensional image. The automatic detection method comprises the following steps: preprocessing the two-dimensional image; performing target labeling on the preprocessed image to complete the production of a data set; adding a spatial mapping layer into a basic feature extraction network of the targetdetection network, and fusing the feature map after spatial mapping and other feature maps together to form a feature pyramid to adapt to the change of a target in the two-dimensional image; completing region recommendation on a feature pyramid formed by the plurality of fused feature maps by using an anchor box with good prior; training the improved target detection network by using the manufactured data set, and carrying out cross validation for multiple times; and using the trained target detection model to detect the picture possibly containing the target, selecting a threshold to screenout a detection box with relatively high possibility of containing the target, carrying out non-maximum suppression on the screened detection box, and removing an overlapping box to obtain a final target detection result with relatively high accuracy.

Subsurface reservoir properties are predicted despite limited availability of well log and multiple seismic attribute data. The prediction is achieved by computer modeling with least square regression based on a support vector machine methodology. The computer modeling includes supervised computerized data training, cross-validation and kernel selection and parameter optimization of the support vector machine. An attributes selection technique based on cross-correlation is adopted to select most appropriate attributes used for the computerized training and prediction in the support vector machine

The invention relates to a tag noise correction based crowd-sourced tagging data quality improvement method. The method comprises the following steps: running a tag integration algorithm in an initial crowd-sourced tagging data set to form a data set after tag integration, and estimating tagger quality and integrated tag quality information of samples in the process; performing multi-round K-fold cross validation by utilizing the data set after tag integration, and constructing a high-quality data set; determining a tag noise set in combination with the tagger quality and the tag quality of the samples by utilizing a prediction probability of a class tag of each sample in the multi-round K-fold cross validation process; and training a classification model by utilizing the high-quality data set generated in the multi-round K-fold cross validation process, and performing prediction and replacement on the class tag of each sample in the tag noise data set by using the model. With the tag noise correction method, the quantity of potential noise tag samples in the data set after original tag integration is reduced, so that the data quality is improved.

The invention discloses a surface roughness prediction method based on GA-GBRT and a method for optimizing process parameters. The method comprises the steps of: collecting data to construct a data set, and dividing the data set to training set data and test set data, and employing the training set data to perform training of key parameters of a GBRT model; b, performing parameter coding and population initialization: randomly generating a chromosomal sequence for increasing the number of iterations, the maximum depth of the individual regression estimator and the learning rate; c, employing the k-folded cross-validation method to train the GBRT model, and employing the genetic algorithm to calculate the fit goodness fitting value of each individual; d, when the number of cycles does not reach the maximum number of iterations, allowing the population to be selected, crossed and mutated to produce a new generation of populations, and continuously performing training of the GBRT model; and e, repeatedly performing the steps c and d until the number of cycles reaches the maximum evolution algebra or exceeds the maximum number of iterations to obtain the optimal model parameters. The surface roughness prediction method based on GA-GBRT and the method for optimizing process parameters are high in test precision and superior in prediction performance and improves the surface processing precision of the workpiece.

The invention provides a retinal vessel segmentation method in a fundus image and a computer readable storage medium. The method comprises the steps of preprocessing the fundus image of a public dataset; cutting and expanding the preprocessed fundus image and dividing the fundus image into a training set and a cross validation set; building a residual non-local attention network, wherein the residual non-local attention network comprises the steps of adopting a coding network decoding network architecture with jump connection as a basic model, and embedding a residual non-local attention module to capture a non-local context dependency relationship between different areas of input features; adding a pyramid pooling module to the encoding network to capture multi-scale features; training aresidual non-local attention network by adopting the training set; determining model parameters of the residual non-local attention network through the cross validation set; and adopting the trainedresidual non-local attention network to segment the fundus image to be processed, so that a retinal vessel segmentation image can be obtained. Segmentation performance is better.

The invention belongs to the technical field of network security and communication, and discloses an encrypted traffic analysis feature extraction method and system, a storage medium and security equipment. The method comprises the steps: collecting original traffic data ; preprocessing the collected original data packet, and filtering traffic data of SSL / TLS encryption communication; performing deep packet analysis on the streaming data to generate a traffic analysis log; performing log aggregation according to the connection quadruple information and the index information in each log to forma flow feature call chain, and performing feature extraction according to the call chain to obtain an initial data set; and determining an optimal supervised learning classification algorithm in thecurrent environment, determining an optimal parameter by using a grid parameter optimization method, and evaluating the feature extraction accuracy by using a ten-fold cross validation method. According to the method, the classification effect is optimal by adopting the random forest algorithm, the obtained accuracy is as high as 99.96%, the result shows that SSL / TLS encryption features used by all malicious families are different, and the classification effect is remarkable.

The invention discloses a data-based qualitative analysis method of influence factors of circuit breaker faults. The method solves prior problems of insufficient utilization of circuit breaker fault data, too subjective positioning of fault causes, insufficient stability of a qualitative analysis model and the like. According to the method, the key influence factors of all types of the circuit breaker faults are found through mining and analyzing the circuit breaker fault data. The method includes the realization steps of defining an original fault data set; cleaning the fault data; transforming the fault data; reducing the fault data; constructing a qualitative analysis model of the influence factors of the faults, and carrying out ten times of ten-fold cross-validation; and obtaining association rules between the faults and attributes, and obtaining the influence factors of all the types of the circuit breaker faults by qualitative analysis. According to the method, a supervised learning algorithm of CMAR is utilized for modeling, and with data amount increasing, the accuracy of the model will be continuously improved. At the same time, the robustness of the qualitative analysis model is guaranteed by the ten times of the ten-fold cross-validation, and the influence factors of the circuit breaker faults can be effectively analyzed.

The invention discloses a treated sewage quality prediction method based on combination of support vector classification and a GRU neural network, and belongs to the technical field of sewage treatment. Missing value processing, abnormal value elimination and data standardization are carried out on the collected sewage historical data, a PCA principal component analysis method is adopted to carryout dimension reduction on the data, and the selected auxiliary variable is used as an input variable of a sewage quality prediction model; a sewage effluent key prediction model is established by adopting a GRU neural network suitable for processing time series data, a support vector machine model is firstly introduced to classify sewage quality data, and then the classified data is respectivelymodeled through the GRU neural network algorithm to predict effluent quality. When the SVM model is trained, a grid search method and a cross validation method are used for optimizing model parameters, the prediction precision of the obtained joint prediction model is more accurate, the model effect is better, the network performance can meet the actual application requirements, and accurate prediction of the effluent quality of the sewage treatment system can be realized.

The invention relates to the field of big data processing, in particular to a customs declaration data processing method and device, computer equipment and a storage medium. The method comprises the following steps: receiving customs declaration data sent by a declaration main body, wherein the customs declaration data carries an order identifier, a payment order identifier and a waybill identifier; obtaining encrypted order data, encrypted payment order data and encrypted waybill data from corresponding nodes of the block chain according to the order identifier, the payment order identifier and the waybill identifier; obtaining a directionally authorized decryption key corresponding to the encrypted order data, the encrypted payment bill data and the encrypted waybill data; decrypting thecorresponding data through the obtained decryption key of the directional authorization; performing cross validation on the decrypted order data, payment data and waybill data to obtain a cross validation result; and when the cross validation result shows that the order data, the payment order data and the waybill data are inconsistent, increasing the check rate of customs clearance articles corresponding to the customs clearance data. By adopting the method, the data security can be ensured.

The invention discloses an analysis method and application of EEG signals based on a complex network. The analysis method of the EEG signals based on the complex network comprises the following steps: constructing multi-scale level limited penetrable visibility graph complex networks; calculating characteristic indexes of each multi-scale level limited penetrable visibility graph complex network; combining a support vector machine to classify the EEG signals, namely using a leave-one-out cross-validation and support vector machine classifier to classify all two-dimensional index vectors, and using a ten-fold cross-validation and support vector machine classifier to classify all the two-dimensional index vectors. According to the invention, multi-scale ideas and level limited penetrable visibility graph theories are combined to construct an EEG multi-scale level limited penetrable visibility graph complex network so as to extract complex network indexes, and the support vector machine classifier in machine learning is combined to realize high-accuracy classification for different EEG signals. The analysis method and application of the EEG signals based on the complex network can be applied to smart head-mounted wearable equipment, and sleep EEG signals are measured through analyzing the smart wearable equipment to monitor the brain state of a user, furthermore, necessary early warning can be provided.

The invention discloses an abnormity detection method for enterprise industry classification, which comprises the following steps of: firstly, extracting to-be-mined text and non-text information in taxpayer industry information, and carrying out feature processing and coding processing; Secondly, constructing a deep network structure conforming to the industry classification abnormity detection problem, and determining the number of neurons of an input layer and an output layer of the network according to the characteristic dimension of the coded data; Thirdly, on the basis of the constructeddeep network structure, adopting different training strategies to train the industry large-class network and the industry detail network through cross validation; And finally, carrying out abnormitydetection on the industry large class by using dimension reduction characteristics of the industry large class network in combination with an SOS abnormity detection algorithm, and carrying out abnormity detection on industry details according to reconstruction characteristics of the industry detail network. According to the invention, the TADM model is utilized to carry out abnormal detection onthe original data, and macroscopic management work such as national statistics, tax collection and industrial and commercial management can be analyzed more reasonably and accurately.

The invention discloses an Adaboost software defect unbalanced data classification method based on improvement, and mainly solves the problem that an existing software defect data classification method is poor in classification effect on minority classes. The method comprises the following steps that 1, software data is acquired from a software data set and then preprocessed, software module data is divided into a training set and a testing set for training and testing, and cross validation is performed for ten times; 2, feature selection of the software data is performed by combining a genetic algorithm based on improvement with a BP neural network to obtain an optimal feature subset, and then dimension reduction processing is performed on the software features; 3, the unbalancedness of the software defect data is fully considered, and an Adaboost classifier based on improvement is trained to classify software modules. According to the Adaboost software defect unbalanced data classification method based on improvement, the classification precision of the minority classes can be improved, and the software defect modules can be better detected.

The invention discloses a multi-class arrhythmia detection method based on a lead attention mechanism. The multi-class arrhythmia detection method comprises the steps of: 1, performing down-sampling on an original ECG signal, and processing the original ECG signal until the length of the ECG signal reaches a fixed length; 2, designing a classification model, and integrating four network structures, including a lead attention mechanism, a convolutional neural network, a bidirectional gating circulation unit and a time attention mechanism; 3, training the model on a public data set by adopting four-fold cross validation; and 4, achieving an arrhythmia classification task by utilizing the trained model. According to the invention, high-accuracy automatic arrhythmia detection can be achieved,so that assistance is provided for diagnosis of doctors.

The invention belongs to the field of milk product analysis and particularly relates to a spectral fingerprint identification method for milk, mare milk, camel milk, goat milk and buffalo milk. The invention relates to the intermediate infrared spectroscopy. The method comprises the following steps: 1) taking fresh milk, mare milk, camel milk, goat milk and buffalo milk samples; 2) scanning the samples in a mid-red spectral range to obtain mid-infrared spectral data; 3) preprocessing original mid-infrared spectrum to remove abnormal values; 4) dividing the preprocessed data set into a trainingset and a test set according to a stratified sampling principle; 5) performing dimension reduction processing on the training set to improve the model training speed; 6) establishing a model for distinguishing milk, mare milk, camel milk, goat milk and buffalo milk through 10-fold cross validation by using algorithms such as nearest neighbors on the training set, and evaluating and screening themodel by using the use accuracy, the balance precision and the Kappa coefficient; and 7) estimating the generalization performance of the optimal model.

The invention relates to a high-resolution satellite data-based TRMM rainfall data downscaling method. The method comprises the following steps of: firstly obtaining multiple pieces of satellite meteorological data through an official way, and correcting TRMM remote sensing rainfall data through a Leave-one-out cross validation method; and selecting a variable through step regression according tofactors influencing the rainfall data, considering hysteresis quality, for vegetation, of rainfall and spatial correlation of the vegetation, establishing a multi-spatial scale model, and selecting anoptimum model to downscale the rainfall data. The method is simple in process and correct in rainfall prediction.

The invention discloses a satellite precipitation data correction method based on multi-source information fusion and downscaling. The method comprises the following steps: S1) acquiring live precipitation data and a satellite inversion precipitation product; S2) carrying out optimal interpolation according to the real-time rainfall data and a satellite inversion rainfall product; S3) obtaining low spatial resolution NDVI data; S4) selecting an environment variable and low spatial resolution NDVI data, and performing independent variable screening; S5) establishing a GWR downscaling model; S6)utilizing the GWR model to obtain OI-GWR corrected rainfall grid point data of the research area; and S7) checking the OI-GWR corrected precipitation grid point data of the research area by adopting10-fold cross validation. According to the method of the invention, optimal interpolation is introduced to improve the quality of original satellite precipitation data, and a better initial field is obtained; on the basis that an initial field is improved, a GWR method is used for downscaling, and the spatial resolution and precision of obtained downscaled data are improved.

The invention belongs to the technical field of soil detection, and discloses a soil moisture prediction method based on a machine learning algorithm. The method comprises the following steps: establishing a soil moisture data and prediction factor database; soil moisture simulation prediction and time-space characteristic detection: establishing a soil moisture prediction model in an R language environment by adopting a support vector machine, a random forest and a BP neural network algorithm; performing comprehensive evaluation and space-time characteristic analysis on the soil moisture by adopting a related statistical analysis method; analyzing the prediction model through cross validation and other methods, and evaluating the precision of the method through precision indexes; and determining the importance of the soil moisture predictor. According to the method, prediction factors of terrain, weather, soil, vegetation and the like are selected in combination with local actual conditions, and a novel algorithm-machine learning algorithm is introduced to construct a model to predict the soil moisture, so that accurate prediction of the local soil moisture is realized.

The present invention provides a video semantic analysis method. The method comprises the following steps: S1, performing preprocessing of a video training set, and constructing a sparse linear decoder; S2, adding topological property constraints to build a topology linear decoder, and performing image blocking processing of the video training set to train the topology linear decoder; S3, taking the parameters of a trained topology linear decoder as initial parameters of a convolution layer in a convolution nerve network; and S4, building a key frame set to perform fin adjustment of the convolution nerve network based on the video training set through adoption of a multifold cross validation mode, and building a general feature extractor on the base of the video data, and inputting the features extracted on the training set and a test set to support the video semantic classification in a vector machine. The model training method provided by the invention has video-type data samples capable of responding to various contents so as to improve the accuracy and the robustness of the model.

The invention discloses a method for determination of the polychlorotrifluoroethylene (PCTFE) content in a polymer-bonded explosive (PBX) by near infrared spectroscopy. The method includes the steps of: preparing and collecting 260 PBX samples, taking 180 samples of them as a calibration set for establishing a calibration model, taking the remaining 80 samples as a validation set for model validation, and acquiring the near infrared spectrum data of all the samples; using a standard method to determine the PCTFE content in the samples; subjecting the spectrum data of the validation set samples in the wave bands of 6102.0cm<-1>-5697.0cm<-1> and 4680.2cm<-1>-4242.9cm<-1> to a first order derivative treatment, correlating the treated spectrum data with the PCTFE content by a partial least squares method, and establishing the calibration model by cross validation; employing the calibration model to predetermine the PCTFE content of the validation set samples, and selecting an optimal model according to a minimum root mean squared error of prediction (RMSEP) of the validation set; and acquiring the near infrared spectrum data of the samples to be determined, and making use of the optimal model to obtain the PCTFE content directly. Being suitable for determination of the PCTFE content in a PBX explosive, the method has the characteristics of convenient operation, and rapid and accurate analysis.

The invention relates to the field of data analysis, and discloses a method for carrying out transformer area user identification based on optimized supervised learning. The method comprises the following steps of determining a user with a known station user topological relation and a station area and a phase to which the user belongs, determining a corresponding tag of user data according to thestation area and the phase to which the user belongs, establishing a training set, a verification set and a test set, determining k parameters in a KNN model by adopting a cross verification mode, andcompleting model training; and identifying and classifying the voltage data to be identified by adopting the trained model and the determined k value, thereby realizing the identification of the users in the transformer area to be identified. According to the invention, conversion from unsupervised learning to supervised learning is realized; a training set, a verification set and a test set arereasonably set, and k parameters are determined by adopting a cross verification mode, so that the transformer area and the phase of a user are accurately and effectively identified, the problem of cross-transformer-area user ownership is thoroughly solved, and a foundation is laid for comprehensively guiding work in the fields of operation, maintenance, first-aid repair, technical improvement, planning and the like of a low-voltage transformer area.

The invention provides a method for solving collaborative filtering recommendation data sparsity based on a neural network. The method for solving collaborative filtering recommendation data sparsity based on the neural network adopts generalized regression of neural network (GRNN) and conducts full filling on sparse data by a train network model and score prediction. The method for solving collaborative filtering recommendation data sparsity based on the neural network comprises the following steps: before conducting the GRNN training, conducting screening on input variables of the neural network by adopting mean impact value (MIV), choosing characteristic values having great impact on output as valid input variables; using the valid input variable to construct the input matrix of the GRNN; adopting Kfold cross validation circulation to find out an optimal spread value of the GRNN; using the optimal spread value and the corresponding input matrix and output matrix to conduct GRNN training; using the trained GRNN to conduct score prediction on a sparse score matrix; and replacing non-scored data of the sparse score matrix with predicted score values. The method for solving collaborative filtering recommendation data sparsity based on the neural network can conduct fully filling on sparse recommendation data, solve the data height sparsity problem most outstanding in existing collaborative technology, and enable recommendation result to be accurate.

The invention discloses a microgrid energy storage SOC estimation method and a system based on a least square support vector machine. Firstly, a microgrid experiment platform comprising an energy storage system is used, and through measurement, voltage, current, temperature and an SOC data sequence of the energy storage system during a charging and discharging process are obtained; according to the measurement data, training set and test set samples for the least square support vector machine are obtained; a cross validation method is then used for selecting the optimal training parameters for the training set of the least square support vector machine, and a training model is further obtained; and the testing set is then used for testing the training model, the result is evaluated, and finally, the training model can be used for SOC estimation determination on the corresponding type of energy storage system in the microgrid. Defects in the traditional SOC calculation method can be overcome, reference is provided for charging and discharging control on the energy storage system, deep charging and discharging of a battery can be prevented, and the service life of the battery and the operation safety of the microgrid can be improved.

Here the inventors describe a tumor classifier based on protein expression. Also disclosed is the use of proteomics to construct a highly accurate artificial neural network (ANN)-based classifier for the detection of an individual tumor type, as well as distinguishing between six common tumor types in an unknown primary diagnosis setting. Discriminating sets of proteins are also identified and are used as biomarkers for six carcinomas. A leave-one-out cross validation (LOOCV) method was used to test the ability of the constructed network to predict the single held out sample from each iteration with a maximum predictive accuracy of 87% and an average predictive accuracy of 82% over the range of proteins chosen for its construction.

The invention discloses a scanning signal feature extraction method based on independent component analysis. The method comprises the steps that six-lead scanning eye electric signals are collected and subjected to band-pass filtering treatment, an airspace filter set corresponding to different scanning task backgrounds is built through an ICA method for filtered data, linear protection is carried out, and airspace feature parameters of scanning signals are obtained. The invention further discloses a recognition method of the scanning signal feature extraction method based on independent component analysis. An ICA airspace filter set is built for each experiment sample in an eye movement database, feature parameters are extracted, cross validation is carried out through a support vector machine, and the optimal ICA filter set and SVM model parameters are determined; the optimal ICA airspace filter set is used for filtering, and then the result is fed into an SVM classifier to be recognized. The scanning signal feature extraction method based on independent component analysis and the recognition method have the advantages of being higher in recognition accuracy rate, higher in expansibility, good in application prospect and the like.

The invention relates to the technical field of animal and plant breeding and human disease prediction, and provides an efficient high-accuracy whole-genome selection method capable of performing parallel operation. The method comprises the following steps: firstly, reading an original genotype file and a phenotype file, constructing a new genotype file and a new phenotype file, and calculating agenetic relationship matrix of all individuals; then, extracting all individuals in the new phenotypic file as a reference group, and extracting all individuals without phenotypic data in the originalgenotypic file as a prediction group; carrying out whole genome association analysis by utilizing the reference group data, and extracting result characteristics of the whole genome association analysis; constructing a model library with specific characters, sequentially optimizing an optimal fixed effect and an optimal random effect by adopting a cross validation strategy, and selecting an optimal prediction model from the model library; and finally, calculating genome estimated breeding values of the prediction group by utilizing the optimal prediction model. The method can quickly, accurately and stably predict individual genome breeding values, and thus the accuracy and efficiency of whole genome selection are improved.

The invention discloses a shopping mall building air conditioner cooling load prediction method based on GBDT, a storage medium and equipment, and the method comprises: collecting cooling load data, and carrying out the normalization processing to serve as the cooling load energy consumption prediction; establishing a load prediction model based on a gradient lifting decision tree algorithm; inputting the preprocessed data into a prediction model for training, selecting a grid search-cross validation mode, and optimizing the three hyper-parameters with the maximum influence on the performanceof the GBDT model; establishing a final cold load prediction model by completing parameter optimization of the prediction model, and obtaining a predicted cold load curve according to the parameters and the structure of the prediction model; and evaluating the prediction performance of the prediction model, adopting the prediction error for evaluation, enabling the deviation between the true valueand the prediction value to form the prediction error, and completing mall building air conditioner cooling load prediction. The method has good prediction precision, universality and applicability,and is especially suitable for large public buildings with periodically changing cold loads.

The present teaching relates to system, method, medium for in-situ perception in an autonomous driving vehicle. A plurality of types of sensor data are received, which are acquired by a plurality of types of sensors deployed on the vehicle to provide information about surrounding of the vehicle. Based on at least one model, one or more surrounding items are tracked from a first of the plurality of types of sensor data acquired by a first type sensors. At least some of the tracked items are automatically labeled via cross validation and are used to locally adapt, on-the-fly, the at least one model. Model update information is received which from a model update center, which is derived based on the labeled at least some items. The at least one model is updated using the model update information.

The invention relates to the field of machine tool cutter remaining life prediction, and discloses a cutter residual life prediction method based on a machine learning regression algorithm. The cutterresidual life prediction method comprises two parts including model training and online life prediction. The model training comprises the steps of collecting original data of a complete life cycle and establishing a corresponding relation with the actual life of a cutter, preprocessing signals, extracting signal features to form feature vectors, performing cross validation to obtain an optimal cutter life model, and performing hyper-parameter adjustment and optimization. The online life prediction comprises real-time data acquisition, signal preprocessing, signal feature extraction to form afeature vector, input of THE optimal cutter life model based on optimal hyper-parameters and output of the residual life of the cutter. The number of eigenvalues extracted from each channel during model training is large, so that the training precision is high, the residual life of the cutter is accurately predicted, a residual life intelligent prediction model of the cutter is established, different regression models can be intelligently selected according to different working condition environments, and the model is good in generalization performance and high in portability.

The invention discloses a protein lysine malonylation site prediction method based on deep learning, and relates to the technical field of biological information. The method comprises the steps: converting the character information of a protein sequence into a numerical vector by adopting an enhanced amino acid composition, a grouped enhanced amino acid composition, a dipeptide deviation expectedaverage value, a K neighbor score and a BLOSUM62 matrix feature extraction algorithm, performing fusion and obtaining a feature space, wherein the influence of each potential feature on a prediction result is fully considered; performing calculating by using a linear convolutional neural network to obtain malonyl site specificity characteristics; selecting related features and reducing feature dimensions through a maximum pooling layer, classifying malonylation sites and non-malonylation sites in combination with a multi-layer deep neural network, constructing a protein malonylation site prediction model DeepMal, and evaluating prediction performance by using 10-fold cross validation and an independent test data set. The model DeepMal is remarkably improved in evaluation indexes, and further promotion of application of deep learning in protein function prediction is facilitated.

The invention discloses an individual sub-health nutrition and health intervention method based on big data and a management system. The method comprises the steps of establishing a sub-health assessment initial model and a nutrition and health intervention scheme library on the basis of evidence based medicine, epidemiology, dietary nutrition survey and nutrition intervention data; revising sub-health assessment model parameters by food, exercise and sign information analytic statistics results of a large sample of people and carrying out cross validation; acquiring individual sub-health score, an interpreted report and a recommended nutrition and health intervention scheme from personal information; in a process of executing the nutrition and health intervention scheme, assessing individual sub-health at regular intervals and adjusting the individual sub-health to obtain an optimal intervention effect; and meanwhile, continuously calculating and optimizing the individual sub-health assessment model by cumulative data of the personal information. The management system which executes the method consists of four modules including a personal information acquiring module, a calculating module, a display module and a supervision module. By the method and the management system, a scientific, individualized, visual and traceable sub-health solution is provided for residents, and long-term health records are established.