81results about How to "Improve classification accuracy" patented technology

The invention discloses a polarization SAR image classification method based on CNN and SVM, and mainly aims to solve the problem of the existing polarization SAR image classification method that the classification precision is low. The method comprises the steps as follows: (1) inputting a to-be-classified polarization SAR image after filtering; extracting and normalizing the original feature of each pixel point based on a polarization coherence matrix and by taking the neighborhood into consideration; training an AE network, and obtaining the parameter of a CNN convolution layer through softmax fine-tuning; setting a CNN pooling layer as average pooling, and determining the parameter of the CNN pooling layer; and sending the features of CNN learning to an SVM for classification to obtain the classification result of the polarization SAR image. Compared with the existing methods, the spatial correlation of the image is fully considered, a new neighborhood processing method is proposed based on CNN, features more conductive to polarization SAR image classification can be extracted, the classification accuracy is obviously improved, and the method can be used for polarization SAR image surface feature classification and object identification.

The invention discloses a method for realizing classification of scene images. In the method, the scene images are classified by adopting two stages of classifiers, wherein the first-stage classifier obtains candidate classifications by utilizing global structural information characteristics, and judges similar classification pairs according to classification results; and the second-stage classifier distinguishes similar classifications by utilizing local texture information characteristics. By adopting the cascade of the classifiers and comprehensively utilizing the global structural information characteristics and the local texture information characteristics of the scene images, the method can realize robust classification of different scene classifications and effectively distinguish similar scene classifications.

The invention provides an electrocardiogram electrocardiosignal classification method with multi-scale characteristics combined. The method comprises the steps that (1), all electrocardiosignals in a database are read, and a base line and high-frequency noise in the electrocardiosignals are removed; (2), the electrocardiosignals are divided; (3), wavelet packet decomposition of the electrocardiosignals is calculated, and a fourth layer of wavelet packet decomposition coefficient is obtained; (4), electrocardiosignal characteristics extracted in a plurality of periods are arranged to form M-dimensional data, a generalized multidimensional independent component analysis method is applied to the M-dimensional data, and demixing matrixes of all modes are obtained; (5), a heartbeat signal to be tested is input, the fourth layer of wavelet packet decomposition coefficient is obtained through the step 1, the step 2 and the step 3, M-1-dimensional data are formed in an arranged mode, and the fuse characteristics of the tested heartbeat signal are obtained through the step 4; (6) the heartbeat signal fuse characteristics are classified through a classifier, and then the classification result of multiple normal and abnormal electrocardiosignals is obtained.

The invention discloses a method and a system of landslide recognition of a complicated geological background area based on LiDAR (light detection and ranging). The method comprises the following steps: S1, LiDAR data is processed to generate dynamic effect model (DEM) data of a bare earth surface with elimination of vegetation influence; S2, a geomorphic characteristic parameter and a computed texture characteristic parameter are extracted to generate a characteristic parameter file; S3, an optimal characteristic parameter combination is confirmed; S4, a balance coefficient meeting a presetting accuracy condition is obtained; S5, an average user precision, an average producer precision and an overall precision are calculated; and S6, if the precision requirements are met, an edge detector is used for calculating a landslide boundary so as to realize landslide recognition. According to the invention, little data is required, and the utilization potentiality of LiDAR-DEM data in the analysis of the landslide terrain is fully developed. The classification precision of the model is high, so that the automatic recognition of the landslide boundary of the complicated geological background area is realized.

The invention provides a heterogeneous network node representation learning method based on a meta-path. According to the method, various node types and relationship types contained in a heterogeneousnetwork graph are considered, and rich semantic information and structure information in the network graph are stored in a plurality of meta-paths in a meta-path extraction mode. In each meta-path, the feature information of the nodes in the meta-path is saved by learning the vector representation of the nodes, and then the plurality of meta-paths are integrated together for common training, so that the semantic information and the structure information in the whole heterogeneous network are saved in the node vector representation. The method has higher classification accuracy, the feature information of the nodes in the heterogeneous network can be better stored in the vector representation of the nodes, the meta-paths can be freely selected according to specific target tasks, and the method is more flexible.

The invention provides a hyper-spectral image classification method and device under a double-branch deep structure so that the image classification accuracy can be improved obviously. On the one hand, the method comprises four steps of image preprocessing, feature extraction, spatial-spectral feature fusion, and classification. At the image preprocessing step, a training set and a testing set needed by two branches are selected and the selected training set and testing set are normalized; at the feature extraction step, a double-branch deep structure is constructed to extract deep semantic information of a hyper-spectral remote-sensing image; at the spatial-spectral feature fusion step, soft-max classifiers are added at tail ends of the two branch network structures to obtain sample probability matrixes respectively, and then a spectral feature and a spatial feature that are obtained by the double-branch deep structure are fused by using a multi-source feature fusion technique; and atthe classification step, the fused features are classified by using a support vector machine. On the other hand, the corresponding device is composed of a pre-processing module, a double-branch deepstructure module, a fusion module, and a classification and evaluation module.

The invention discloses a multispectral image classification method based on surface wave CNN. To-be-classified multispectral images are inputted, normalization processing on the multispectral data is carried out to acquire a matrix, block taking based on a center pixel point is carried out for the matrix after normalization, and a training data set and a test data set are acquired; a classification model based on the surface wave CNN is constructed; the training data set is utilized to train the classification model; the trained classification model is utilized to classify the test data set. The method is advantaged in that a multi-scale depth filter is introduced, multispectral image classification precision is improved, and the method can be applied to target classification.

The invention relates to a polarized SAR image convolutional neural network classification method based on a spatial feature atlas. The method comprises: polarized SAR image data are inputted and preprocessing is carried out to obtain a to-be-classified polarized SAR image; the type of the to-be-classified polarized SAR image is determined; preliminary polarized SAR image feature extraction is carried out on the to-be-classified polarized SAR image to obtain a preliminary image feature combination unit; a feature atlas of a training sample is obtained in the polarized SAR image based on a pixel point and a neighbourhood so as to obtain a training sample space; a convolutional neural network model is erected, a network parameter is adjusted, and the convolutional neural network model is trained by using the feature atlas of the training sample; a combined feature of the to-be-classified polarized SAR image is processed based on a pixel pint and a neighbourhood to obtain a preliminary feature atlas, thereby forming a testing sample space; and then feature extraction is carried out on the testing sample by using the convolutional neural network model and classification is also carried out. Therefore, the classification precision of the polarized SAR image is improved.

The invention discloses a microblog emotion analysis method based on standard dictionaries and semantic rules. The microblog emotion analysis method comprises the following steps: collecting microblog data and manually labeling and marking the emotion value of each microblog; proposing corresponding standard micrblog emotion dictionaries, and establishing an emotion dictionary database; based on the standard emotion dictionaries, adding the semantic rules for assistance, and performing parameter adjustment and optimization on parameters of the semantic rules; based on a real dataset experiment, acquiring the final classification accuracy and precision. The technical scheme provided by the invention is adopted to well analyze the emotion tendency of each microblog user by introducing the standard emotion dictionaries, microblog expression dictionaries and the semantic rules, therefore, higher classification accuracy and precision are achieved.

The invention is applicable to the technical field of internet text categorizing and provides a text categorizing method, device and system. The method includes: extracting characteristics of texts to be categorized; and categorizing the texts to be categorized according to the characteristics of the texts to be categorized to obtain normal texts and junk texts, wherein the characteristics include word properties in the texts to be categorized. By using the property of each word in the texts as the characteristics to conduct characteristic extraction and categorizing, characteristic space is greatly reduced, a relatively complex and precise categorizing model can be selected from a categorizer to categorize the texts to be categorized, and the categorizing accuracy is greatly improved.

The present invention provides a high spectral image classification method of an autoencoder based on entropy rate superpixel segmentation. The problem is mainly solved that a high spectral image classification performance is not good. The realization steps of the method comprise: obtaining a high spectral data training sample set and a test sample set; constructing n layers of autoencoder networks; inputting the training sample set, and employing a loss function to perform training of the autoencoder networks based on the entropy rate superpixel segmentation; and inputting the test data set,and employing the autoencoder networks based on the entropy rate superpixel segmentation after training to perform classification of a high spectral image. The high spectral image classification method of an autoencoder based on the entropy rate superpixel segmentation considers space context neighborhood information, excavates distribution features of data samples to effectively improve classification precision of a high spectral image, and can be applied to distinguishing and identification of surface features in fields of agricultural monitoring, geological prospecting, disaster environmentassessment and the like.

The invention discloses a cross-data-field shift learning sorting method based on differential manifolds. The method comprises the steps of inputting data of each data filed and label data used for training and establishing an adjacent map used for spectrogram geometry regulation for the data; establishing an unified mathematical model by combining the input data, label information and the established adjacent map and an optimal objective; deriving an update formula of variables according to the established mathematical model and updating hidden factors of each dimensionality of each data field, inter-domain sharing relational structures and regression coefficients in a mode of alternant iteration until convergence; and carrying out generic label forecasting on data of a target field and obtaining a forecasting generic label of the data of the target field. The cross-data-field shift learning sorting method based on the differential manifolds is used for obtaining a differential data manifold space, a novel expression factor is of a highly- differential structure which is beneficial for sorting, and an original cluster manifold structure of the data is kept.

The invention relates to the technical field of logistics sorting, in particular to a large-size parcel sorting device used for logistics sorting. A sorting box and a detecting mechanism are included;the detecting mechanism is arranged on the front side of the sorting box; two side plates are fixedly connected to the lower side of the sorting box, and are both fixedly connected to the upper sideof a bottom plate; two driving rollers are arranged between the two side plates, and a plurality of driven rollers are arranged between the two driving rollers; the driving rollers and the driven rollers are rotationally connected with the side plates through bearings; a number one speed reducing motor is fixed to one side plate through a screw; an output shaft of the number one speed reducing motor penetrates through the corresponding side plate and is fixedly connected with one driving roller; and a wide belt is connected between the two driving rollers in a transmission manner. The aerial insulating cable fixing device for electric power supply and distribution has the characteristics of being compact and reasonable in structure and capable of achieving automatic sorting, a reciprocating swinging plate is used for pushing a small-size parcel to enter a slope, a large-volume parcel continues to be transported on a conveying belt, and a vertical plate, a transverse plate and a longitudinal plate form a detecting frame.

The invention relates to a cross-domain video semantic concept detection method based on active learning, comprising the following steps of: (1) taking a Gaussian random field as a standard classifier; (2) selecting and marking unmarked samples by adopting an uncertainty query strategy in an active learning method; (3) updating the standard classifier; and (4) sequentially repeating the steps (2) and (3) until a certain cycle index is completed. In the invention, the Gaussian random field is used for constructing the standard classifier for active learning. Compared with the way that only an original domain sample is taken as a training set of the standard classifier, which is frequently used in other active learning algorithms, the standard classifier in the invention has the advantage that the selected marked samples can reflect distribution of data in a target domain to a greater extent. Weight of a newly selected marked sample from the target domain is increased, thus the standard classifier can rapidly adapt to characteristic distribution of the data in the target domain. The invention also provides an algorithm for rapidly updating a standard model, the complexity of the algorithm is effectively reduced, and the applicability of the algorithm is improved.

The invention discloses a SVM classifier parameter optimization method based on an improved particle swarm algorithm. The method comprises the following steps: (1) performing 10-fold cross verification on collected sample data, selecting the parameter influencing the classifier performance as the to-be-optimized parameter; (2) initializing related parameters of the classifier and the particle swarm algorithm, and updating particle speed and location according to related parameters; (3) setting the to-be-optimized parameter of the classifier as the corresponding dimension value at the current location of the particle, and computing to obtain the fitness value corresponding to the current location of the particle; and (4) obtaining the fitness value evaluation particle according to the fitness value corresponding to the current location of the particle, and updating the individual optimal location and the population optimal location. Through the method disclosed by the invention, the mixed kernel function based on the polynomial kernel function and the Gaussian kernel function is constructed, the traditional particle swarm algorithm is improved, the kernel function parameter is optimized by utilizing the improved PSO-SVM algorithm, and then the comprehensive performance of the classifier is improved, the generalization capacity of the classifier is improved when the high classification precision is guaranteed.

The invention provides an AI based low-confidence sample processing method and system of board sorting. Image data of at least one format of a low-confidence sample is obtained; an image of at least one format of the low-confidence sample is presented in a display device; a new class marked by the low-confidence sample is obtained; and a training method is input to the marked low-confidence sample, and a new classification model is obtained via re-training. According to the method and system, the low-confidence sample can be discovered continuously and utilized, so that the classification precision of a machine learning method is improved gradually.

The invention provides a polarimetric SAR image classification method based on feature attention and a feature improvement network. The polarimetric SAR image classification method mainly solves the problems that an existing polarimetric SAR image classification method based on deep learning is poor in intra-area consistency and inconvenient for end-to-end classification. The implementation schemecomprises the following steps of: 1) inputting a to-be-classified polarized SAR image and filtering the to-be-classified polarized SAR image; 2) synthesizing a pseudo color image and a classificationlabel of the polarized SAR image; 3) extracting initial features of the polarimetric SAR image and preprocessing the features; 4) respectively constructing an input representation layer, a feature attention sub-network, an encoder and a decoder, and sequentially connecting the input representation layer, the feature attention sub-network, the encoder and the decoder to form a feature attention and feature improvement network; 5) training fa eature attention and feature improvement network; 6) Inputting the polarization SAR image into the trained network to obtain the classification result. The polarimetric SAR image classification method is high in intra-area consistency, low in noise and high in classification precision, end-to-end learning and classification are realized, and the polarimetric SAR image classification method can be used for polarimetric SAR image classification.

The embodiment of the invention discloses an electrocardiogram signal classification method and a device. The method comprises the following steps: carrying out heart beat cutting on electrocardiogramsignals to be classified so as to obtain heart beat data, wherein the heart beat data is time-domain heart beat data; then, converting the heart beat data to obtain frequency-domain heart beat data,and extracting statistical characteristics and sampling characteristics of the heart beat data and the frequency-domain heart beat data as the heart beat characteristics of the heart beat data; identifying invalid heart beat data according to the heart beat characteristics of the heart beat data so as to remove the invalid heart beat data from the heart beat data and to obtain updated heart beat data; and inputting the heart beat characteristics of the updated heart beat data into a heart beat classification model to obtain a classification result of the heart beat data. The invention performsclassification by using the heart beat characteristics peculiar to each heart beat data. Compared with the prior art that classification is carried out by sampling a fixed threshold value, the methodimproves the classification accuracy.

The invention discloses an interpretable CNN image classification model-based optical remote sensing image classification method. An explainable CNN ResNet model composed of six groups of 51 basic convolution layers, three full connection layers and one Softmax layer in total is built; the method comprises the following steps of: performing down-sampling by utilizing ResNet; obtaining a feature map containing context information; and then interpretable modification is performed on the ResNet model to obtain a new interpretable CNN network based on the ResNet model, and the ResNet model extracts features through multiple groups of convolution-pooling layers with residual modules and finally inputs the features to a full connection layer to classify the images. The interpretability of an existing deep learning model can be enhanced, and the performance of the model is further improved.

The invention provides a polarimetric SAR image terrain classification method based on self-supervised representation learning. The method mainly solves the problems that an existing polarimetric SARdeep convolutional network needs a large number of tags for classification and is poor in robustness. According to the scheme, the method comprises the following steps: performing polarization coherence matrix modal representation extraction and Pauli color modal representation extraction on polarimetric SAR original data; designing a self-supervised representation learning loss function and a network framework, and training the framework under the condition of not using label data; migrating the trained network framework weight value into a deep convolution integral type network model; carrying out fine tuning training on the deep convolutional integral network by using a small number of labeled samples to obtain a trained classifier; and inputting the test data into the trained classifier to obtain a final classification result. According to the method, the requirement for the label data volume is reduced, the classification precision and robustness of the polarimetric SAR deep convolutional network are improved, and the method can be used for guiding agricultural and marine monitoring.

The invention discloses a power electronic circuit fault diagnosis method based on a longicorn beard optimization deep belief network algorithm, and the method comprises the steps: collecting a DC side bus output current signal of a three-phase PWM rectifier in an open-circuit fault mode of different switching devices, and enabling the DC side bus output current signal to serve as an original dataset; utilizing empirical mode decomposition to extract intrinsic mode function components of output current signals under different switching device open-circuit fault modes, and constructing an original fault feature set; making feature selection based on an extreme tree algorithm to generate a fault state sensitive feature set; optimizing the structure of the deep belief network by adopting a longicorn beard algorithm; and training the optimized deep belief network by using the training set, and performing verification by using the test set to obtain a fault identification result. Accordingto the method, a feature extraction algorithm, an optimization algorithm and a deep learning classification algorithm are combined, so that the feature data volume and classification accuracy of power electronic circuit fault diagnosis are greatly improved.

The invention provides a training method of a food borne pathogenic bacteria Raman spectrum identification method established based on PCA-Stacking. The method aims at a problem of insufficient identification precision for two food borne pathogenic bacterium with similar Raman peaks, namely Escherichia coli 0157:H7 and Brucella S2, and provides a Ramon integrated classification algorithm based onPCA-Stacking, wherein a mathematical statistics model with better robustness and a calculating method are found. For aiming at problems of burr and baseline drift in the Raman spectrum, spectrum pre-processing is realized by means of a Savitzky-Golay filter and an asymmetric least squares. The model parameters are searched through grids. Higher classification accuracy of a Stacking integrated algorithm relative to K neighboring, logic regression, and support vector machine single algorithm model is proved.

The invention relates to a tensor model-based multi-source data classification optimizing method and system. The tensor model-based multi-source data classification optimizing method comprises the following steps: in step a, under a Map-reduce distribution framework, multi-view data is subjected to tensor product operation, high order tensor data is obtained, and an initial support tensor machine classification model is built according to the high order tensor data; in step b, the data of all views is subjected to feature elimination operation via a support vector recursive feature elimination algorithm in an original space, and subscript data where the data of all views retains features is output; in step c, according to the subscript data where the data of all views retains features, parameters of the initial support tensor machine classification model are optimized, and a final support tensor machine classification model is determined; in step d, test samples are input to the support tensor machine classification model and are classified. Via use of the tensor model-based multi-source data classification optimizing method and system, classification accuracy of the classification model can be effectively improved, calculation complexity can be lowered, that redundant information in tensor data can be identified by the classification model is ensured, and classifying speed of the classification model can be further improved.

The invention provides a text training method. The method comprises the following steps: 1) computing a centre vector of each classification of training sample sets; 2) classifying samples in the training sample sets according to the centre vector of the training sample sets; and 3) for the samples which are classified incorrectly, correcting the centre vector of a classification A to which the incorrectly classified samples belong or/and the centre vector of a classification B to which the samples are classified by mistake according to a set drag weight and a set push weight. The text can be classified with high precision and fast speed according to the centre vector obtained by the training method.

The invention discloses a feature fusion transfer learning arrhythmia classification system based on 2D heart beat, and the system comprises: an obtaining module which is configured to obtain a targetelectrocardiosignal; the reconstruction module that is configured to reconstruct the target electrocardiosignal from one dimension to a two-dimensional electrocardiosignal; the feature extraction module that is configured to respectively extract a first feature and a second feature from the two-dimensional electrocardiosignal; the feature fusion module that is configured to perform fusion processing on the first feature and the second feature; and the classification module that is configured to input the fused features into the trained classifier and output an arrhythmia classification resultcorresponding to the current target electrocardiosignal.

The embodiment of the invention discloses a classifying method based on a multi-label flexible support vector machine, which comprises the steps of first, defining a novel distance measuring method in a multi-label space to measure the distance from point to point in the multi-label space under a special classifying target; then, defining a neighborhood for each point in the multi-label space under a special classifying target, wherein the neighborhood of a certain point comprises several points which are closest to the central point in the novel distance measuring method; and finally, combining the neighborhood information in the multi-label space of each sample point and using the novel multi-label flexible support vector machine to carry out multi-label classifying training. The classifying method based on the multi-label flexible support vector machine is used to improve the classifying precision of an identifying classifier in multi-label classification by using information comprised in the multi-label space so as to reduce the influence of noise labels to classification.

The invention discloses a household garbage recycling method, and belongs to the technical field of garbage processing. The household garbage recycling method comprises the following steps of: obtaining an information bar code or image information of the garbage by utilizing image acquisition equipment; scanning the information bar code while the garbage carries the information bar code to obtainnecessary information of garbage; and directly determining how to classify and recycle according to the unnecessary information. If the information bar code is not carried on the garbage or the information bar code is damaged, image information of the garbage is obtained, goods stored in a database of the image identification equipment is related according to the image information to obtain necessary information of the goods, and the way about how to classify is determined according to the necessary information. The method is used for determining how to classify garbage without making mistakeseasily, so that classifying precision of garbage is favorably improved, classifying consciousness of people is conveniently cultivated, later-stage workload of garbage recycling staff is favorably reduced, and mutual pollution caused by garbage classifying errors can be avoided.

The invention relates to the technical field of flow refining, in particular to a method for selecting flow refining marketing parameters. The method is characterized in that data variables are selected for refining user groups according to refining aims, data sample sets of the user groups are extracted from the data set, and parameters having influences on classification accuracy of an AdaBoosting model are determined with the data sample sets of the user groups as training samples of the AdaBoosting model; the classification accuracy of the AdaBoosting model is improved, the consumer demands of users are accurately captured, and the success rate of flow package popularization is improved.

The invention discloses an ELM-based optimized classification method for land use and cover of hyperspectral remote sensing images, which comprises the following steps: firstly, a plurality of ELM-based classifiers are constructed, and a training data set is constructed for each ELM-based classifier; Then, T ELM-based classifiers are trained based on the training data sets of each ELM-based classifier, and the classification and prediction results of training samples in each training data set are obtained. Then, the ELM-based classifier set is pruned based on the classification prediction results. Finally, the hyperspectral remote sensing images to be classified, The spectral features of each pixel point are extracted, and the feature data of the object to be classified are obtained and inputted into the ELM-based classifiers in the set of classifiers retained after pruning, and the hyperspectral remote sensing images to be classified are classified and judged by ensemble, and the classification results of the hyperspectral remote sensing images to be classified are outputted. The invention realizes an optimized classification method for land use and cover of hyperspectral remote sensing images, which improves classification accuracy and classification processing efficiency.

The invention discloses a method for optimizing random forest parameters for machine learning model training. The method comprises the steps that firstly, dividing the whole ant population into a plurality of sub-populations; Then, allowing each sub-population to correspond to one partition in the RDD, and designating independent evolution in one partition; And finally, exchanging information among the sub-populations by utilizing a migration operator. Compared with the traditional grid search, the parallel ant lion algorithm based on the Spark can efficiently find out a better parameter combination to improve the classification precision of the random forest, and under a big data distributed Spark platform, the optimization calculation speed is high, the acceleration effect is obvious, and the parallel ant lion algorithm based on the Spark can be used as a next generation parameter optimizer of a cloud computing platform.