30results about How to "Conform to the distribution characteristics" patented technology

The invention relates to a method for optimizing running parameters of a thermal power unit based on a fuzzy set association rule. The method comprises the following steps of: (1) selecting data; (2) preprocessing the data and dividing a working condition; (3) constructing a fuzzy set; (4) extracting the fuzzy set association rule; and (5) performing regression analysis. According to the method provided by the invention, the optimal parameter and working condition of the thermal power unit which can be achieved under the present running working condition are considered; the main controllable parameters and the historic data influencing the running optimization of the thermal power unit are analyzed; the relevant association rule is excavated by utilizing an excavating technology for automatically constructing the fuzzy set data; a target value for the running parameter optimization when the power supply coal consumption of the thermal power unit is lower is confirmed; and the regression analysis is performed, thereby obtaining a running optimization curve of each parameter under a confirmed working condition. The method provided by the invention has high practical value and wide application prospect in the technical field of energy-saving optimization control during a thermal process of a thermal power plant.

The present invention discloses a children speech emotion recognition method, comprising: creating a training voice database, classifier training and recognizing emotion. The classifier training comprises the step of extracting the speech emotion features, which includes performing pretreatment on the analog speech signal, including sampling, denoising and other basic signal treatments, performingHHT transformation on the pretreated emotion sentences, obtaining the Hilbert spectrum of the emotion speech; performing amplitude feature extraction, intonation feature extraction, tone feature extraction and rhythm feature extraction, and obtaining the emotion feature; reducing dimension of the emotion feature, and obtaining different quaternionic feature of emotion space; performing calculation on the sample set of the training speech database, obtaining the projection of a series of quaternionic features in national emotion state in the emotion space i, thus forming an emotion feature projection set; and carrying out classification training based on the SVM/HMM mixed system by using the data in the emotion feature projection set.

The invention discloses a non-local mean value image denoising method based on a filter window and parameter adaption. According to the invention, firstly, noise is detected, and a noise calibration matrix is established according to a detection result; the size of the noise calibration matrix is consistent with the size of an image, and the matrix value at a corresponding position of each noise point is set to be 1, and the matrix value at a corresponding position of each non-noise point is set to be 0. Then, each pixel of a noise image is successively taken as a reference point, and centric to the point, a predetermined number of non-noise reference points are taken in a counterclockwise direction to be involved in computation. Finally adaptive weighting parameters are determined according to the locations of the reference points, and a weighted result is calculated and a restored pixel value is obtained; the corresponding element in the noise calibration matrix is set to be 0 and the pixel point after denoising can be used as a reference point of other noise points. Compared with traditional image denoising methods, the method provided by the invention is added with the noise detection and noise point screening, thus improving algorithm accuracy, changing a reference point selection window and improving algorithm adaptability.

The invention discloses an iterative eye fundus image blood vessel segmentation method based on the distance modulation loss, which comprises the steps of (0) collecting a color eye fundus image to form an original image; (1) performing normalization processing on the original image; (2) iteratively training a dense convolution neural network based on the distance modulation loss; and (3) carryingout iterative blood vessel segmentation by using the trained dense convolution neural network. The method can process color eye fundus images under different collection conditions, can provide interactive blood vessel segmentation experience for ophthalmologists, is more robust for blood vessel detection and provides a reliable guarantee for subsequent auxiliary diagnosis.

The invention discloses an RGB-D image variable metric super voxel segmentation method. By adopting Poisson disk sampling, seed points are selected in data, and iteration clustering is carried out according to color distances and spatial distances between data points and the various seed points, and then an initial super voxel segmentation result is acquired. The super voxels acquired by the initial segmentation are used as vertexes, and the adjacency relations of the super voxels are used as edges to form an undirected graph, and the super voxel integration is carried out by adopting a method based on a graph theory. Super voxel segmentation results having different dimensions are acquired in the same RGB-D image, and the variable metric super voxel segmentation is realized. The RGB-D image variable metric super voxel segmentation method belongs to data pretreatment, and by using the acquired super voxels by segmentation, the large-scale super voxels are acquired in the area having high data consistency, and the small-scale super voxels are acquired in the area having the poor data consistency, and the method is capable of satisfying a human visual cognition characteristic in a better way.

The invention belongs to the technical field of image processing, and discloses a hyperspectral open set classification method based on Euclidean distance and deep learning. The method comprises the following steps: firstly, constructing a category center of each known category, a category prediction function based on an Euclidean distance and a loss function based on the Euclidean distance, and training and optimizing a deep learning network model; secondly, combining a box line graph method and an extreme value theory Weibull distribution model, analyzing and fitting the distance between input data and the category center of each known category, thereby achieving constraint on each known classification boundary, and then achieving judgment on unknown categories. Specifically, a box linegraph is used for providing the number of abnormal points needed for fitting the Weibull distribution model; and judging whether a Weibull model is used or the upper edge of the box line graph is usedfor judging unknown classes according to the number of the abnormal points. Aiming at the open set classification problem in the hyperspectral field, the method is simple, practical, high in classification precision and high in robustness.

The invention discloses a speech emotion recognition method and system based on semi-supervised adversarial variation self-encoding, and the method comprises the steps: S1, constructing a generative adversarial network, and constructing a speech emotion recognition model through the combination of a semi-supervised variation self-encoding model and the generative adversarial network, wherein data with emotion labels in input data and corresponding emotion labels are used as input, data without emotion labels in the input data are used as emotion label attribute missing types for processing, feature probability distribution of the input data in a hidden layer is learned through the generative adversarial network, and an SSAVAE model is constructed; S2, training the constructed SSAVAE model by using a training set; and S3, inputting to-be-processed speech emotion data, and inputting the to-be-processed speech emotion data into the trained SSAVAE model to obtain an emotion recognition result. The method has the advantages of being simple in implementation method, high in recognition precision, good in generalization ability and data disturbance resistance and the like.

The invention provides a digital rock core reconstruction method and system based on a generative adversarial neural network. The method comprises the following steps: executing iterative processing:processing historical noise data according to a weight matrix of a generator and bias of the generator to obtain a generated sample; processing the generated sample according to the weight of the discriminator and the bias of the discriminator to obtain a generated sample discrimination probability; processing the real sample according to the weight of the discriminator and the bias of the discriminator to obtain a real sample discrimination probability; determining a loss function according to the generated sample discrimination probability and the real sample discrimination probability; whenthe current number of iterations reaches a preset number of iterations; processing the current noise data according to the weight matrix of the generator and the bias of the generator in the currentiteration to obtain a digital rock core model; otherwise, updating the weight matrix of the generator, the bias of the generator, the weight of the discriminator and the bias of the discriminator according to the loss function, so that the reconstruction efficiency can be improved, and the digital rock core conforms to the distribution characteristics of the real rock core.

The invention provides an image restoration method, an image restoration system and a flat panel detector, and the method comprises the steps: obtaining a first image, carrying out the connected domain analysis of the first image, and extracting all defect pixels in the first image; dividing the defect pixels into isolated point defects, cluster defects and broken line defects according to the size and shape of a connected domain; and outputting the second image and repairing the second image according to the type of the defective pixel. According to the method, defective pixels can be effectively repaired, and details of an original image can be well reserved. And repairing the second image output by the detector in real time according to the classification of the defective pixels to realize the rapid repair of the defective pixels of the image. The method is only related to the type and size of the defective pixel, the applicability is high, and the working efficiency is remarkably improved. The image restoration system is integrated in the detector, the hardware design does not need to be changed, the product cost can be reduced, and only software needs to be optimized and upgraded in subsequent upgrading and maintenance.

The invention discloses a method for detecting a hidden object in a millimeter-wave human body image based on block mixture Gaussian low-rank matrix decomposition.The method mainly solves the problemsof low imaging quality caused by weak scatter echo of the hidden object and low detection accuracy of a gray scale value of the hidden object and a human body similarity in the prior art. An implementing scheme of the method comprises the following steps of 1, removingabnormal points in an imaging region background in an original millimeter-wave human body image, and dividing the human body imageinto six parts according to proportions of human body parts; 2, decomposing all regions of thehuman body based on a block mixture Gaussian low-rank matrix decomposition algorithm to obtain a low-rankpart and a sparse part; and 3, binaryzing the sparse part by using a typology method, and removing small noise points to obtain a final detection result graph. The method increases the detection rateof various complicated small targets in the millimeter-wave human body image without a large number of training samples, so that the detected hidden object is more complete; and the method can be used for detecting hidden objects carried by people in public places such as an airport and a bus station.

The invention discloses a rainfall weather phenomenon comprehensive identification method based on multi-source observation data. The method comprises the following steps of firstly, extracting minute rainfall particle spectrum distribution (particle size and number concentration) and falling speed information detected by a raindrop spectrometer, and carrying out quality control to obtain rainfall particle spectrum distribution and falling speed after quality control; and then based on the particle spectrum distribution and falling speed information observed by a raindrop spectrometer and the environment temperature information at a meteorological station position, introducing a fuzzy logic identification method to identify different types of rainfall weather phenomena; and finally, for the identified rainfall type, further according to intensity and continuous changes, subdividing specific rainfall sub-types, and realizing more refined weather phenomenon identification. The method is based on a multi-source observation means and a fuzzy logic algorithm, can solve a problem that a current raindrop spectrometer is too high in weather phenomenon false recognition rate, and achieves accurate recognition of a rainfall phenomenon.

The invention discloses a software defect prediction data processing method. According to the method, an independent feature distribution model is established for each feature of a defect sample; andpart of the features are replaced by adopting a random variation mode so as to obtain a new defect sample, and new samples are supplemented continuously until the ratio of a non-defect sample to thedefect sample is balanced so as to obtain a processed software defect prediction data set for subsequent model training. The invention further provides a software defect prediction data processing device based on the method and a machine storage medium, the problem that in the prior art, due to the fact that the number of defect samples is smaller than that of non-defect samples, the defect samplerecognition capacity is insufficient is solved, and the software defect prediction precision is effectively improved.

The invention discloses a method for analyzing actually measured pressure data in a hydraulic transient process, which relates to the field of electric power and comprises the following steps of: extracting mean pressure to obtain instantaneously changed pressure fluctuation data; performing time-domain grid distribution division on the obtained pressure fluctuation data to divide the pressure fluctuation data into a plurality of time grid regions, analyzing the pressure fluctuation data in each time grid region by using an empirical cumulative distribution function to obtain probability distribution of each data, performing confidence probability processing, and combining processed data groups of each time grid region to obtain the probability distribution of each data; obtaining processed pressure fluctuation data; the average pressure is added back to the processed pressure fluctuation data, and the total pressure data subjected to confidence probability processing in the whole time domain are obtained.According to the method, the test actual measurement extreme value result and the transition process pressure change process can be well evaluated, the draft tube pressure fluctuation process can be well analyzed and evaluated, safe and stable operation of a unit is ensured, and the reliability of the draft tube is improved. And a solid technical guarantee is provided for safe and stable operation of a power station.

The invention discloses a high-fault-tolerance genome complex structure variation detection method based on a filtering strategy, and the method comprises the steps: carrying out preprocessing on an input file in an SAM format, and traversing a CIGAR field in an optimal quality comparison reading segment; according to the compared CIGAR field and variation score calculation criterion, calculatinga variation score corresponding to each site in the current reading segment, and storing the variation score in a variation score set of each site in advance; counting an average value in the variation score set of each site as a final variation score of the site and obtaining a variation score function of the sample; carrying out Kalman or Gaussian filtering on the variation score function to obtain a variation score function after filtering and noise reduction; according to the filtered variation score function, setting a threshold value and separating a structure variation region, and carrying out feature extraction; and training a support vector machine (SVM) model, and classifying the structural variation regions by using the trained SVM model to obtain a complex indel result set. According to the invention, the interference of sequencing errors on the determination of structural variation is solved.

The invention belongs to the field of oil gas and geophysical exploration, and particularly relates to a multichannel bipolar sparse spectrum inversion method based on Hessian matrix constraint. The method comprises the following steps: estimating the wavelength of seismic wavelets, determining the length of a time window, and segmenting multiple channels of seismic data; seismic wavelets in the time window are extracted; constructing a multi-channel bipolar sparse spectrum inversion objective function based on Hessian matrix constraint in the time window, and calculating a reflection coefficient of multi-channel seismic data in the time window; and combining the center reflection coefficient results of the time windows along a time sequence to obtain a spectrum inversion result of the multi-channel seismic data. According to the method, the real contact relation of the stratum can be reflected, the real pinching position of the stratum can be depicted, and high-resolution and richer geological information can be provided. While the seismic data resolution is improved, the method can ensure the transverse continuity of stratums in multi-channel data, and can provide more accurate and rich geological and reservoir information.