44results about How to "Feature retention" patented technology

The invention discloses an electric power image data augmentation method based on a generative adversarial network. The method comprises the following steps: obtaining an original image during electric power system inspection; performing image processing on the original image, eliminating interference and influence of noise and the like in the original image, and obtaining a high-quality target image; performing feature enhancement on the target image, and extracting image features; and performing game synthesis of the target image and the background image through a generative adversarial network method to generate new image data. According to the scheme, data augmentation is carried out on the power image through the generative adversarial network method, image annotation does not need tobe carried out on the synthesized image, manpower and material resource cost is saved, and meanwhile the synthesis method is scientific, reasonable, high in augmentation result quality, high in efficiency and high in scheme feasibility.

The embodiment of the invention discloses a high-resolution remote sensing image weak and small target detection method and device based on deep learning. The method comprises the steps of obtaining ato-be-processed remote sensing image; inputting the remote sensing image to be processed into a pre-trained convolutional neural network, carrying out 4-time downsampling, 8-time downsampling and 16-time downsampling respectively on the remote sensing image to be processed through the convolutional neural network; obtaining the priori boxes of different sizes corresponding to the to-be-processedremote sensing image, identifying the target priori boxes of which the target category confidence is greater than a preset threshold, and determining the coordinate information of a target included inthe to-be-processed remote sensing image through a preset clustering algorithm according to the coordinate information of each target priori box, wherein the first layer of the convolutional neural network comprises a residual component, the second layer, the third layer and the fourth layer of the convolutional neural network each comprise four residual components, and each residual component comprises two convolutional layers and a fast link. By applying the scheme provided by the embodiment of the invention, the weak and small target detection precision can be improved.

The invention discloses a phase-coherent accumulation noise elimination-based radio frequency fingerprint feature extraction and recognition method. The method comprises a phase-coherent accumulation noise elimination step S4 which is used for a radio frequency fingerprint identification technology and a step S6 which is used for performing multi-resolution analysis on waveforms in the radio frequency fingerprint identification technology. According to the S4, phase-coherent accumulation is performed on a plurality of power-on instantaneous signal sample point amplitude value functions of the same device, so that signal strength can be multiple times of original signal strength, and phase-coherent accumulation noise elimination signals are obtained. According to the S6, multi-resolution analysis is performed on the phase-coherent accumulation noise elimination signals obtained in the step A. According to the phase-coherent accumulation noise elimination-based radio frequency fingerprint feature extraction and recognition method of the invention, the phase-coherent accumulation method is adopted in the radio frequency fingerprint identification technology; the signal-to-noise ratio of the waveforms is improved, so that the accuracy of radio frequency identification can be improved; and on the basis of improving the accuracy of the radio frequency identification based on the phase-coherent accumulation, the multi-resolution analysis is utilized, and therefore, the computational complexity of the support vector machine-based radio frequency fingerprint identification technology is reduced.

The invention provides a modeling method based on wind power fluctuation multi-scale decomposition, comprising: according to the first wind power historical data collected in advance, analyzing the time characteristic and statistical characteristic of wind power fluctuation, so as to determine the wind power fluctuation decomposition component and the time scale corresponding to the wind power fluctuation decomposition component; According to the wind power fluctuation decomposition component and its corresponding time scale, a two-stage WMMF filter is used to decompose the wind power multi-scale fluctuation of the first wind power historical data to obtain a low-frequency trend component, an intermediate-frequency fluctuation component and a high-frequency fluctuation component. Accordingto the low frequency trend component, the intermediate frequency fluctuation component and the high frequency fluctuation component, a multi-dimensional probability model is established. Through theabove method, the multi-dimensional probability model can be established according to the fluctuation characteristics and correlation of wind power fluctuation of the original wind power time series,thus retaining the characteristics of the complete wind power fluctuation process and simulating the wind power output characteristics to the maximum extent.

The invention provides a processing method for drawing longer data on a low-resolution screen, and belongs to the technical field of data collection. Extract the maximum and minimum feature quantities of the data in sections from a large amount of data, and then draw only these feature quantities on the small screen, so as to achieve the purpose of showing the overall characteristics of the data. The curve displayed after compression by this method can well represent the waveform characteristics of the original data. The principle is simple and clear, and the algorithm is simple and efficient. It is especially suitable for long data representation in embedded state monitoring systems.

The invention discloses an electrocardio and heart sound signal combined analysis based coronary heart disease non-invasive screening system. The system comprises an electrocardiosignal preprocessingmodule used for filtering, denoising and normalizing an original electrocardiosignal, a heart sound signal preprocessing module used for filtering, denoising and normalizing an original heart sound signal, a characteristic extraction and selection module used for extracting and selecting multi-domain characteristics of the processed electrocardiosignal and heart sound signal, a five-channel cascade module used for performing wavelet transformation and reconstruction on the heart sound signal and connecting the reconstructed four-frequency-band heart sound signal and the electrocardiosignal into a five-channel group signal, and a double-input neural network classification module composed of a multi-layer perceptron, a deep learning model and an activation model and used for processing and analyzing electromechanical eigenvectors and the five-channel group signal. Therefore, the coronary heart diseases can be screened more accurately and non-invasively.

The invention discloses a geographic space multi-dimensional data visual analysis method based on parallel coordinate axis arrangement. The method comprises: carrying out the clustering analysis of geographic space objects, carrying out the visual display of the clustering analysis result of the geographic space objects on a map, and obtaining a view of geographic space clustering; displaying themulti-dimensional attribute information by using parallel coordinates, and performing clustering analysis on the multi-dimensional attribute information by using kernel density clustering to obtain anattribute category; the correlation between the clustering analysis result of the geographic space object and the attribute category is measured by using mutual information; embedding a parallel coordinate system into a view of geographic space clustering, determining the arrangement sequence of parallel coordinate axes in the parallel coordinate system by utilizing the correlation, and rearranging the parallel coordinate axes in the parallel coordinate system according to the arrangement sequence. The problems that geographic space objects are difficult to integrate in a parallel coordinatesystem and multi-dimensional attribute information association features are difficult to identify and present can be effectively solved.

The invention provides a text classification feature selecting method capable of reducing the characteristic dimension and the classification complexity and improving the classification accuracy. The method comprises the following steps that a feature set S and a target class C are obtained, the relevancy Rc(x(i)) between each feature x(i) in the feature set S and the target class C is calculated, and sort descending is conducted on the feature set S according to the size of the relevancy Rc(x(i)); the redundancy Rx and the synergy degree Sx between every two features in the feature set S are calculated, and the sensitivity Sen of each feature is calculated with the combination of the relevancy Rc(x(i)) between each feature and the target class, the sensitivities Sen are compared with a preset threshold th, and the feature set S is divided into a candidate set Ssel and an excluding set Sexc with the combination of the sort descending result of the feature set S according to the threshold th; the sensitivities Sen between the features in the candidate set Ssel and the features in the excluding set Sexc are calculated, the sensitivities Sen are compared with the preset threshold th, and the candidate set Ssel and the excluding set Sexc are adjusted according to the threshold th. The text classification feature selecting method is suitable for the field of machine learning text classification.

The present invention relates to a partial discharge denoising method. The method comprises the following steps of: 1) employing an ITD decomposition method to perform adaptive decomposition of original partial discharge signals including noise, and obtaining a plurality of proper rotation components (PRC) including noise; 2) determining the denoising threshold of each PRC, performing processing of each PRC according to the obtained denoising thresholds, and obtaining the preliminary denoising result of each PRC; 3) performing processing of each PRC after the preliminary denoising according tothe energy of each PRC, removing burrs at singular points after preliminary denoising, and obtaining each processed PRC; and 4) performing ITD reconstruction of each PRC obtained in the step 3), andobtaining partial discharge signals after denoising. The method has adaptivity, a good denoising performance and a fast calculation speed, can greatly remove a lot of background noise included in thepartial discharge signals, and can be widely applied in the online monitoring field of insulation of electric devices.

The invention relates to a method and device for carrying out wind driven generator fault diagnosis in a dimension reduction mode. The method comprises the steps of constructing an original data spaceaccording to operation data of the wind driven generator in different operation states; performing feature selection on the sample data in the original data space by using a first mode to obtain a new data space; constructing a corresponding symmetric positive definite matrix manifold for any sample data in the new data space obtained based on the first mode by using a second mode; performing feature extraction on the symmetric positive definite matrix manifold of the sample data in a third mode to obtain a low-dimensional feature set of the sample data in the symmetric positive definite matrix manifold; and inputting the obtained low-dimensional feature set into a support vector machine, and detecting the fault of the wind driven generator according to the output information of the support vector machine. The fault detection precision of the wind driven generator can be improved.

The invention discloses a soybean plant rapid three-dimensional reconstruction method based on phenotype-oriented accurate identification, and the method comprises the steps of enabling a user to useKinect to scan a soybean plant, i.e., a soybean plant, at intervals of a set angle, and obtaining a plurality of frames of point cloud scenes where the soybean plant is placed on an automatic rotatingdisc; after removing a point cloud background through straight-through filtering, calculating a rough registration matrix by using a Rodrigs formula according to a rotation angle provided by a user,performing fine registration by using an ICP algorithm, performing hierarchical clustering on plant point clouds to obtain skeleton points, setting each point to be adjacent to nearest K points for the skeleton points to obtain a plurality of connected components, performing trunk Dijkstra path growth, branch growth point selection and branch communication component Dijkstra path growth on trunk communication components where root nodes are located, and finally performing three-dimensional visualization of a plant skeleton through a pyvista visualization library. According to the invention, the plant point cloud does not need to be processed in advance through software to obtain the parameters, the purchase expenditure of modeling software and the learning and using time cost of the modeling software are saved, and the invention is more efficient.

The invention belongs to the technical field of artificial intelligence graph classification, and provides a graph classification method and system fusing high-order structure embedding and composite pooling, and the method comprises the steps: obtaining a to-be-classified graph; inputting a to-be-classified graph into the graph neural network to obtain a category to which the graph belongs; wherein for each sub-graph set of the graph, each convolutional layer calculates the feature of each sub-graph based on the sub-graph set output by the previous neural network layer, each composite pooling layer updates the sub-graph set based on the feature of each sub-graph output by the convolutional layer, and meanwhile, for each sub-graph in the updated sub-graph set, the feature of each composite pooling layer is calculated based on the feature of each sub-graph output by the convolutional layer. The features of the sub-graphs in the local neighborhood are fused through an attention mechanism, and the features of the sub-graphs are updated; and obtaining a graph representation vector by the reading layer, and inputting the graph representation vector into the classifier to obtain a category to which the graph belongs. A high-order structure is utilized, messages are directly transmitted among the sub-graphs, structural information invisible in node level is captured, and the classification precision of the graphs is improved.

The invention discloses a three-dimensional model retrieval method based on LSTM network multi-modal information fusion, and the method comprises the steps: for a given three-dimensional model, extracting a plurality of views of the three-dimensional model arranged according to a rotation angle sequence; extracting skeleton characteristics of a plurality of views in a multi-task and multi-angle manner, and obtaining structured information of the three-dimensional model according to the skeleton characteristics; extracting view feature vectors of a plurality of views, and inputting the view feature vectors into a layer of LSTM network structure; checking whether other feature vectors need to be extracted continuously or not; connecting the skeleton feature vector with the view feature vector subjected to one layer of LSTM to form a new feature vector, and inputting the new feature vector into a second layer of LSTM network structure for fusion; checking whether other to-be-fused featurevectors exist or not, if yes, forming a new feature vector again and inputting the new feature vector into the next layer of LSTM network structure for fusion; and taking the output of the last fusion as the final feature vector Q of the three-dimensional model, and finishing the final detection process of the three-dimensional model in combination with a similarity measurement method.

The invention aims to provide a real-time monitoring and early warning method for subsynchronous oscillation in a regional power grid, which performs real-time measurement and parallel calculation ona plurality of measuring points in the regional power grid, performs centralized monitoring on subsynchronous oscillation on equipment in the power grid at the same time, and provides risk early warning according to the subsynchronous oscillation mode change trend and severity.

The invention belongs to the technical field of commodity recommendation, and discloses a large-scale multi-objective optimization commodity recommendation method based on image coding. The method comprises the following steps: firstly, obtaining a data set of recommended commodities for a user according to probability, then coding the commodities based on commodity images, then constructing a target function based on recommendation accuracy and commodity diversity, then generating an initial population based on a non-dominated sorting and crowding distance method, and executing a memetic evolution system based on variable statistical division to optimize the population, and finally recommending the user according to the coding similarity. According to the method, a large-scale multi-objective optimization algorithm based on image feature modeling is adopted for optimization, and articles with similar features are selected from recommended commodities according to an optimization result for recommendation, so that the recommendation accuracy of diversified commodities can be improved.

The invention discloses a pneumonia image processing method and system and a storage medium; the method comprises the steps: carrying out the filtering reconstruction and feature enhancement of a to-be-processed pneumonia image, and carrying out the fusion of the to-be-processed pneumonia image with an original pneumonia image, and maintaining the features of the original pneumonia image and the feature-enhanced image, thereby facilitating the subsequent neural network learning. Through verification of an InceptionV3 network, the pneumonia image is processed by adopting the method disclosed by the invention, and compared with an unprocessed pneumonia image and a pneumonia image which is processed only by using a Retinex algorithm, the obtained pneumonia image is improved in both accuracy and specificity.