91results about How to "Enhance explanatory" patented technology

An apparatus for searching similar sentences that has a translation sentence database includes an input unit to which a sentence is input; first language processing unit configured to perform language processing on sentences input through the input unit; and first language similarity calculating unit configured to refer to previously translated sentences to extract similar sentences for the first language sentence. Further, the apparatus includes translating unit configured to translate a sentence into a second language sentence; second language processing unit configured to perform language processing on a second language sentence; second language similarity calculating unit configured to refer to the previously translated sentences to extract similar sentences for the second language sentence; and a re-ranking unit configured to combine similar sentence extracting results of the first language with those of the second language to re-rank sentence outputs.

The invention discloses a near infrared spectrum information extraction method based on canonical correlation coefficients. The method comprises the steps that 1, original near infrared spectra and chemical values of tobacco leaves are acquired; 2, the abnormal spectra in the original near infrared spectra and the abnormal values in the chemical values are eliminated; 3, the near infrared spectrum is partitioned into a plurality of spectrum sections, and canonical correlation analysis is conducted on the near infrared spectrum and the chemical value in each spectrum section to calculate the canonical correlation coefficients; 4, a threshold value is selected, and PLS modeling is conducted on the wavelength points where the canonical correlation coefficients are larger than the threshold value. According to the near infrared spectrum information extraction method based on the canonical correlation coefficients, the explanation capacity of the near infrared spectrum can be improved, the multicollinearity of near infrared spectrum information can be eliminated, and the stability of predicting a chemical component model through the near infrared spectrum can be improved.

The invention discloses a multi-sensor technology-based grape water stress diagnosis method and a system therefor. The multi-sensor technology-based grape water stress diagnosis method comprises the following steps of 1, acquiring a canopy coverage rate value, a canopy temperature characteristic value and a canopy photosynthetically active radiation value of a grape sample for modeling, 2, building a detection model by the canopy coverage rate value, the canopy temperature characteristic value and the canopy photosynthetically active radiation value as input variables, and a canopy water stress level as an output variable, and 3, by the step 1, acquiring a canopy coverage rate value, a canopy temperature characteristic value and a canopy photosynthetically active radiation value of a grape sample needing to be detected, and substituting the acquired canopy coverage rate value, the canopy temperature characteristic value and the canopy photosynthetically active radiation value into the detection model, and calculating a canopy water stress level of the detected grape sample. Through utilization of the multispectral imaging technology, the thermal infrared imaging technology and the multiple informative data fusion technology, the multi-sensor technology-based grape water stress diagnosis method realizes fast, early-stage and real-time detection of a grape water stress level and improves a detection precision.

The invention relates to a soil type merging and multiple regression-based soil manganese content prediction method. The divide-and-conquer control of different spatial variation characteristics showed by soil trace elements in different soil types is involved; the spatial heterogeneity of the soil trace elements can be detected by analyzing a discrete degree of spatial distribution of effective manganese content of soil; the multicollinearity problem in local regression analysis can be diagnosed through the local regression analysis; and especially in a spatial prediction process, a comprehensive prediction model is built by measuring spatial distribution characteristics of the effective manganese content of the soil under the sampling densities of different soil samples.

The invention discloses a pulmonary nodule recognition and segmentation method based on deep learning, and the method comprises the steps: preprocessing a DICOM file so as to generate a chest CT image, segmenting a lung mask from the chest CT image, and repairing the lung mask; generating a three-channel lung image according to the chest CT image and the repaired lung mask, and inputting the three-channel lung image into a two-dimensional YOLO v3 neural network to detect a suspicious area of pulmonary nodules; standardizing the chest CT image according to the suspicious area to generate a standardized matrix, inputting the standardized matrix into a 3D Dense Net neural network and a C3D neural network for prediction, and generating a target prediction box according to a prediction result;and normalizing the chest CT image according to the target prediction box to generate a normalization matrix, inputting the normalization matrix into the 3D UNet neural network for segmentation, and optimizing a segmentation result. The invention further discloses a pulmonary nodule recognition and segmentation system based on deep learning. According to the invention, the accuracy and speed of identification and segmentation can be effectively improved.

The invention discloses a tobacco chemical value quantifying method based on near-infrared spectrum wave number K-means clustering, comprising the following steps: establishing a training set and a test set, and collecting near-infrared spectrum and target component content of all tobacco samples in the training set; clustering the wave number of the near-infrared spectrum of each tobacco sample in the training set through K-means clustering; after each clustering, using PLS to establish a relationship model between each subclass spectral band and the target component content, and calculating the root mean square error for cross validation of each relationship model; taking the number of clustering with the minimum sum of the root mean square error for cross validation corresponding to the relationship models as the optimal clustering number, and performing weighted summation on the relationship models corresponding to the optimal clustering number to obtain a full-spectrum model; and collecting near-infrared spectrum of each tobacco sample in the test set, and obtaining the target component content of each tobacco sample in the test set on the basis of the full-spectrum model. Compared with the existing PLS method, the method of the invention can significantly reduce the prediction error of a model.

A curved surface actuating mechanism suitable for controlling motions of eyelids for a doll is provided, which includes a main shaft, a first drive unit, a second drive unit, a third drive unit, a sleeve, and a curved surface. The main shaft is a fixed shaft. A first drive wheel of the first drive unit is fitted on the main shaft and driven to rotate by a first driving member. A second drive wheel of the second drive unit is fitted on the main shaft and driven to rotate by a second driving member. The sleeve is fitted on the main shaft and has a vertical split shaft. A third drive wheel of the third drive unit is fitted on the vertical split shaft, and meanwhile engaged with the first and the second drive wheels. The curved surface has a link lever fixed to the third drive wheel.

The invention discloses an optimized and improved fuzzy regression model construction method based on a nondominated sorting genetic algorithm II (NSGA-II). The optimized and improved fuzzy regression model construction method based on the NAGA-II reduces fuzzy sets, fuzzy rule and redundancy of antecedent of the fuzzy rule, and improves explanatory of a fuzzy regression model. The optimized and improved fuzzy regression model construction method based on the NSGA-II comprises the following steps: firstly an initial fuzzy regression model is constructed by a triangle subordinate function and a WM (WangandMendel) algorithm; and then based on a NSGA-II optimized fuzzy regression model, the redundancy of the fuzzy rule is simultaneously deleted by selecting the fuzzy rule and the antecedent of the fuzzy rule, thereby improving accuracy and explanatory of the fuzzy regression model.

The invention provides a potential field anomaly edge enhancement method based on a transition zone and phase congruency. The method is applied to researches of determination geologic body edge positions through potential field (gravity field and magnetic field) anomaly. Judgment of edge positions of underground anomalous bodies is achieved through the steps of improved calculation of co-occurrence matrixes, determination of transition zone anomalous ranges, calculation of thresholding contrast, extraction of phase congruency information, fusion processing of the thresholding contrast and the phase congruency information, judgment of gravity and magnetic data processing results and the like. According to the method, the thresholding contrast and the congruency information are fused, edge positions are guaranteed through enhancement of transition zones, and sensitivity of phase congruency information to weak anomaly can be used simultaneously. The method has the advantages of being capable of highlighting the edge anomaly and revealing subtle differences, and high in accuracy and resolving power.

The invention discloses an optimized and improved fuzzy classification model construction method based on a nondominated sorting genetic algorithm II (NSGA-II). The optimized and improved fuzzy classification model construction method based on the NAGA-II reduces characteristic variable, a fuzzy rule and redundancy of antecedent of the fuzzy rule, and improves explanatory of a fuzzy classification model. The optimized and improved fuzzy classification model construction method based on the NSGA-II comprises the following steps: firstly an initial decision tree is constructed by a C4.5 algorithm, the characteristic variable and fuzzy set numbers are selected; then a triangle subordinate function is utilized to convert the decision tree into the fuzzy classification model; and finally based on a NSGA-II optimized fuzzy classification model, the redundancy of the fuzzy rule is simultaneously deleted by selection of the fuzzy rule and the antecedent of the fuzzy rule, thereby improving accuracy and explanatory of the fuzzy classification model.

The invention provides a novel multi-source data fuzzy clustering algorithm which mainly comprises the following steps: multi-source data is collected, each source in the multi-source data includes a plurality of categories, each category comprises a plurality of dimensions, an object function of a multi-source data fuzzy clustering method for the multi-source data is built, each source in the multi-source data is weighted in the object function, different dimensions in different categories of sources in the multi-source data are weighted, parameters in the object function are initialized, a clustering center and parameters of the object function are subjected to repetitive updating and clustering operation, and multi-source data clustering processes can be finished. According to the novel multi-source data fuzzy clustering algorithm, correlation of multiple sources in the multi-source data and difference in contribution made by different characteristics to different category identification are used, and therefore a novel clustering algorithm which combines different vision angle weighting and different weights of different characteristics is constructed; the novel multi-source data clustering algorithm disclosed in the invention is better than other multi-source data clustering algorithms in explanatory property and reliability of clustering results.

The embodiment of the invention relates to a medical aid decision model interpretation method and device, a storage medium and electronic equipment, and relates to the technical field of big data processing, the method comprises the following steps: obtaining input data of a medical aid decision model and a data prediction result corresponding to the input data, generating a feature subset according to the sub-features included in the input data; combining the target features included in the feature subset to obtain a feature pair, and mining the value domain of the feature value of each target feature included in the feature pair to obtain a candidate value domain of each target feature; fusing the to-be-fused value domain into a current value domain corresponding to the target feature ina preset initial rule base to obtain a target rule base; and matching the decision path from the input data to the prediction result in a target rule base, and explaining the medical aid decision model according to a target rule included in the decision path. According to the embodiment of the invention, the interpretability of the medical aid decision-making model is enhanced.

The invention discloses a model interpretation method and equipment and a readable storage medium. The model interpretation method comprises the following steps of obtaining each model input feature corresponding to a preset black box model, inputting the prediction data set corresponding to each model input feature into a first hash coding model optimized based on each preset sample category; carrying out hash coding on the prediction data set; obtaining a first hash coding result, inputting the prediction data set into a second hash coding model optimized based on each output sample categoryoutput by the preset black box model; and performing hash encoding on the prediction data set to obtain a second hash encoding result, and further determining feature confidence corresponding to eachmodel input feature based on each bit difference degree between the first hash encoding result and the second hash encoding result. The technical problem that the model interpretation effect is pooris solved.

The invention discloses a financial product recommendation method based on a GAMxNN model, relates to the technical field of bank financial product recommendation systems, and solves the problem thatdata prediction in an existing bank financial product recommendation system is either low in precision and high in interpretability or high in precision and low in interpretability. The method comprises: 1, performing data cleaning and preprocessing; 2, feature selection: determining the importance proportion of each variable by adopting a random forest algorithm, and reserving the feature variable with the highest importance proportion for the feature variables with the correlation coefficient of 0.9 or above; 3, training a GAMxNN model, selecting AUC as a model evaluation index, and obtaining an optimal hyper-parameter of the model; and 4, inputting a to-be-recommended customer characteristic value, and obtaining a target variable prediction result. According to the method, high-precision and high-interpretability bank financing product recommendation prediction is achieved, recommendation and non-recommendation results can be given, and meanwhile main influence factors of recommendation and non-recommendation can be given.

The invention discloses a transformer fault diagnosis method and system based on an intelligent integration algorithm, and the method comprises the steps: carrying out the preprocessing of obtained oil chromatography data, and carrying out the feature construction of the preprocessed oil chromatography data, so as to obtain a plurality of features; respectively determining input characteristics corresponding to different types of transformer fault diagnosis algorithms; constructing a transformer fault mode diagnosis model based on an intelligent integration algorithm according to the determined input features corresponding to the different types of transformer fault diagnosis algorithms; determining the fault diagnosis accuracy of each transformer fault mode diagnosis model in each fault mode, and determining an algorithm corresponding to each fault mode and the weight of each algorithm according to the fault diagnosis accuracy; and utilizing the transformer fault mode diagnosis modelbased on the intelligent integration algorithm to diagnose oil chromatography data to be diagnosed, determining a fault mode result, and determining a diagnosis result according to the priority coefficient of each fault mode in the fault mode result.

The invention relates to a keyword-based text marking method, and the method comprises the steps: obtaining training data which comprises a text, a tag and a keyword related to the tag; performing word list representation on the text and the keyword to obtain a text word list and a keyword word list, and connecting the keyword word list to a tag to obtain a first sentence; taking the text word list as a second sentence, and performing fine tuning training of the BERT model based on the first sentence and the second sentence to obtain a trained text marking model; and receiving a to-be-marked text, outputting predicted marking information of the to-be-marked text through the text marking model, and marking the to-be-marked text according to the predicted marking information. According to the method and the system, the problems of complicated steps and lack of interpretability in a related text marking method are solved, and the marking efficiency and the marking effect are improved.

The invention discloses an electrocardiosignal noise reduction method based on an interpretable deep neural network, and the method comprises the steps: building an electrocardiosignal model through a sparse representation algorithm, and converting an optimization problem into optimization of two sub-problems through a semi-secondary splitting algorithm. Then, a neural network is built in the two sub-problems, and an optimal solution is sought through end-to-end training of electrocardiosignal noise data and clean data in the built neural network. By designing the electrocardiosignal noise reduction network, the interpretability of the neural network is improved, and meanwhile, the noise reduction capacity of electrocardiosignals is also improved. An attractive bridge is established between a traditional sparse representation noise reduction algorithm and a deep neural network noise reduction model. In this way, the interpretability of the neural network is improved, and the precision and robustness of a traditional sparse representation noise reduction algorithm are improved.

The embodiment of the invention provides a recommendation method and device and a device for recommendation, and the method specifically comprises the steps: determining a demand word corresponding toexpression behavior data of a user by using a preset word bank; wherein the preset lexicon is used for storing preset words, and the commercial values of the preset words accord with a first condition; determining a recommendation object according to the demand word; and providing the recommendation object to the user, or sending access entry information of the user to a supplier terminal corresponding to the recommendation object, so that the supplier terminal transmits the recommendation object to the user terminal of the user through the access entry information. According to the embodiment of the invention, the interpretability of the recommended object can be improved, and the test cost can be reduced.

The invention discloses a hand model perception-based isolated word sign language identification method, which comprises the following steps that a hand sequence intercepted from a sign language video is converted into a latent semantic representation containing a hand state through a visual encoder; then, a hand model perception decoder works in a model perception mode, the latent semantic representation containing the hand state is mapped into a three-dimensional hand grid, and the position of each hand joint point is obtained; finally, the three-dimensional hand grids are optimized through an inference module, space-time representation of each hand joint point is obtained, classification is conducted, and therefore vocabularies corresponding to the hand sequences are recognized. According to the method, model and data driving can be fused, the hand shape prior is introduced, the recognition accuracy of the system is improved, an intermediate result (i.e., a three-dimensional hand grid) can be visualized, and the interpretability of a framework is enhanced.

The invention relates to a GBDT and LR fusion method and device based on federated learning, equipment and a storage medium, and the method comprises the steps: calculating the gradient of each firstsample, encrypting the gradient, and transmitting the encrypted gradient to a passive side; obtaining the encrypted gradient and group of each group of the passive party; decrypting the gradient sum group, selecting optimal feature division according to the gradient sum, and transmitting a division value corresponding to the optimal feature division to a passive party; obtaining a sample space inwhich the passive party is divided into a left node or a right node; splitting the first sample according to the sample space to obtain a tree structure corresponding to the GBDT model; and constructing a feature matrix according to the tree structure, and performing logistic regression training to obtain an LR model. By means of the GBDT and LR fusion method, device and equipment based on federated learning and the storage medium, financial data can be directly aggregated for fusion model training of GBDT and LR models.