81 results about "Classification rate" patented technology

An instance-level retrieval method and system are provided. A representation of a query image is embedded in a multi-dimensional space using a learned projection. The projection is learned using category-labeled training data to optimize a classification rate on the training data. The joint learning of the projection and the classifiers improves the computation of similarity / distance between images by embedding them in a subspace where the similarity computation outputs more accurate results. An input query image can thus be used to retrieve similar instances in a database by computing the comparison measure in the embedding space.

Proprietary suite of underlying document image analysis capabilities, including a novel forms enhancement, segmentation and modeling component, forms recognition and optical character recognition. Future version of the system will include form reasoning to detect and classify fields on forms with varying layout. Product provides acquisition, modeling, recognition and processing components, and has the ability to verify recognized data on the image with a line by line comparison. The key enabling technologies center around the recognition and processing of the scanned forms. The system learns the positions of lines and the location of text on the pre-printed form, and associates various regions of the form with specific required fields in the electronic version. Once the form is recognized, the preprinted material is removed and individual regions are passed to an optical character recognition component. The current proprietary OCR engine is trained with a variety of Roman text fonts and has a back end dictionary that can be customized to account for the fact that the system knows which field it is recognizing. The engine performs segmentation to obtain isolated characters and computes a structure based feature vector. The characters are normalized and classified using a cluster centric classifier, which responds well to variations in the symbols contour. An efficient dictionary lookup scheme provides exact and edit distance lookup using a TRIE structure. An edit distance is computed and a collection of near misses can be output in a lattice to enhance the final recognition result. The current classification rate can exceed 99% with context. The ultimate goal of this system is to enable the processing of all tax forms including forms with handwritten material.

The invention relates to a complete system of recycling treatment of city household garbage and a method thereof. The invention is characterized in that an electronic screen control system controls mechanical transmission and treatment condition of a garbage mechanical classification system, a fertilizer-making system, a plastic classification system, a pulping system and a sewage treatment system by a contact screen and a circuit, the recycling treatment of day treatment for day garbage, closed operation, automatic sorting, no secondary pollution and fully treatment of resources is realized finally, the city household garbage is divided into 48 types, which are combined into 15 types and seven sorts, the classification rate reaches 100 percent, the recycling rate reaches over 99 percent and the whole-range treatment needs 60 to 80 minutes.

The invention provides a depression automatic evaluation system and method based on phonetic features and machine learning. On the basis of speech processing, feature extracting and machine learning technologies, the system and method search for the relations between phonetic features and depression, and provide objective references for clinical diagnosis of depression. The system includes 1, a speech acquisition recording module for recording speech information of testees at different emotional and communication stimulation; 2, a speech feature calculating module for calculating acoustic features of the speech, 3, a speech database module containing relative data for speech feature efficiency analysis, classifier training optimization and a system classification rate; 4, a machine learning module for determining an extracting mode of effective features and training a classifier for automatic evaluation; 5; an automatic evaluation module. The system and method conduct depression degree classification on the effective speech features of the testees according to the extracting mode of the effective features and a classification mode of the trained classifier.

The invention provides a wafer surface detect feature analysis method. The method comprises the steps that a plurality of wafer images are collected; wafer interesting areas of a wafer are extracted from the wafer images; all suspicious defects in the interesting areas are obtained; a preset number of training samples are selected from the suspicious defects; feature data of the training samples are extracted; feature analysis is conducted on the feature data through the random forest method, and a random forest formed by a plurality of classification models is obtained. By the adoption of the method, the wafer surface defect error classification rate can be reduced, and classification accuracy is improved. The invention further provides a wafer surface detect feature analysis system and a wafer surface detect feature classification method and system.

The invention discloses an improved adaptive genetic algorithm based neural network image classification method. The method comprises the following steps: extracting texture features of sample images by adopting a gray-level co-occurrence matrix based feature extraction algorithm to obtain texture features of a training sample and a test sample; by taking the texture feature of the training sample as an input of an RBF neural network, training the RBF neural network with a genetic optimization based neural network learning method to generate a trained RBF neural network; and inputting the texture feature of the test sample into the trained RBF neural network, and performing an image classification test. For the deficiency that a k-means clustering algorithm and other algorithms are sensitive to initial value selection, the method can better avoid the ''prematurity'' convergence of a genetic algorithm, simplify the network structure of a neural network classifier and improve the generalization capability of the network and the correct classification efficiency of the images.

The invention relates to an unsupervised Markov random field image segmentation method, which is technically characterized in that: the scope of the classification number of an image M is firstly determined, the image M is then carried out the K-means segmentation from m min to m max type, the BICm which is corresponding to the image M under each type is calculated according to the BIC criteria, a potential energy function is applied to calculate the total energy of the image M, and finally the ICM method is selected to complete the Markov random field image segmentation. The method has the advantages that: the new potential energy function is applied to image segmentation, the anti-noise performance thereof is obviously improved, the segmentation result can remove most of the noise of the image with great noise, the error classification phenomenon can be simultaneously effectively reduced, the error classification rate is reduced by more than 60 percent; the classification number does not need to be manually determined, thus basically realizing the unsupervised image segmentation; and the judgment accuracy of the classification number can achieve more than 95 percent by the artificial synthetic image test.

The invention provides a method for extracting application layer protocol key words from an unknown application layer data, the method can implement the data excavation of the key words by annotating the positions of characters in the conversation and the message and defining the general-purpose degree and the supporting degree of the character string. The key words extracted by the invention can be used for unknown applied characteristic analysis and type recognition, the classification controls comprising the service based on content, classification rate control and abnormal flow filtering and so on for the application can be implemented based on the recognition.

The invention discloses a gear fault diagnosis platform and a gear fault diagnosis method. On the platform, gear faults are simulated and vibration signals are collected, the improved local preserving projection algorithm is combined with a Bayes classifier, the mode recognition effect is judged according to the correct classification rate of the Bayes classifier, the vibration signals of the gear faults are measured by a vibration acceleration sensor, the principal component analysis is conducted firstly, and then, kernel transformation, construction of the nearest neighbor graph, solving of mapping space and the like are conducted; and the Bayes classifier is used for recognition in classification according to multi-fault classification. Compared with the principal component analysis, the laplace algorithm and the local preserving projection, the improved local preserving projection fault recognition rate can be greatly increased. According to the improved local preserving projection algorithm and the Bayes classifier combined fault mode recognition method, the fault recognition rate is increased and the accuracy is improved, and the effect of the fault mode recognition of the gear can be improved. The gear fault diagnosis platform has a simple structure, and a high-accuracy diagnosis platform can be provided for the fault recognition of the gear.

The invention discloses an aluminum plate surface defect classification method based on a BP neural network and a support vector machine. The method comprises the steps that feature values of aluminum plate surface defects are extracted as the input quantity of a BP neural network classification model, and oil spots and the first class of defects are adopted as the output quantity to construct the BP neural network classification model; a plurality of support vector machine classification models are constructed through the first class of defects in a one-to-one classification method; learning samples are obtained, and the BP neural network classification model and the support vector machine classification models are trained; the oil spots and the first class of defects are classified through the BP neural network classification model, the BP neural network classification model is regarded as a testing sample of the oil spots to be removed, and the rest of the first class of defects are classified again through the support vector machine classification models; a classification result is obtained finally through statistics. According to the method, the recognition and classification rate of the oil spots on the surface of a cold rolling aluminum plate is improved, meanwhile, the overall recognition rate of the cold rolling aluminum plate surface defects is improved, and the method can be used for recognizing and classifying other metal surface defects and is simple and easy to implement.

A method, medium, and apparatus with a generated object descriptor using a curvature gabor filter, with complementary object recognition being performed by applying a general gabor filter and a curvature gabor filter having various curvatures, such that the dimension and efficiency of a feature vector can be increased by detecting correlations between features of an object searching for a parameter and its position of an optimal curvature gabor filter through boosting learning, and projecting the extracted features onto a base vector having the highest classification rate.

The invention discloses a fast seismic waveform classification method based on a semi-supervised algorithm, which comprises the following steps: S1, getting seismic waveform data along the horizon, and reducing the dimensionality of the seismic waveform data through an SSDR algorithm based on linear transform; S2, using tag data in the seismic waveform data to train a distance measurement matrix; and S3, using a semi-supervised K-means classification algorithm to classify the seismic waveform data, and generating a seismic phase diagram. As original seismic waveform data is processed through a semi-supervised dimensionality reduction method, the similarity between similar waveform data is increased while redundant data is eliminated, the difference between different categories is enhanced, and the classification result is more accurate. A suitable distance measurement matrix is trained using existing logging data, and the distance measurement method is introduced to the subsequent classification method. Then, a weighted semi-supervised K-means classification method is put forward. Logging data is fully utilized. The classification accuracy is improved, and the classification rate is increased.

The invention discloses a method for optimizing a Parsytec on-line surface defect detection system. The method mainly comprises the following steps of: (1) maintaining hardware; (2) calibrating equipment; (3) grouping materials; (4) verifying defect; (5) establishing and optimizing a classifier; and (6) performing post-treatment. By the method, the Parsytec on-line surface defect detection system is optimized, so that the surface defect detection rate of the system can be increased to over 95 percent in a short time, the surface defect classification rate is increased to over 90 percent, and instantaneous, stable and accurate on-line surface defect detection and classification are realized.

The invention discloses a motor imagery electroencephalogram signal classification method for semi-supervised learning optimization SVM, and the method comprises the steps: obtaining motor imagery electroencephalogram signals, and carrying out the preprocessing of the motor imagery electroencephalogram signals; dividing the preprocessed motor imagery electroencephalogram signals into a training set, a verification set and a test set, and performing SVM classifier initialization training of undetermined parameters based on the training set to obtain mutual information of the verification set; updating the training set, and performing SVM classifier iterative training based on the updated training set to obtain mutual information of the verification set after each iteration; constructing anobjective function based on the obtained mutual information and the SVM classifier of the undetermined parameters, and obtaining optimal parameters of the SVM classifier by adopting an empire algorithm; and obtaining an average iterative classification rate by adopting the optimized SVM classifier. According to the method, the imperialist competitive algorithm is used for optimizing the SVM, better parameters can be obtained, and then a better classification effect is obtained.

The invention provides a tobacco material classifying device and classifying method. The equipment comprises a winnowing box machine body, a tangential material discharging device and a classifying filtering mesh belt; the top of the winnowing box machine body is connected with the tangential material discharging device; the classifying filtering mesh belt is located on the upper portion of the winnowing box machine body, penetrates into one side of the winnowing box machine body and penetrates out of the other side of the winnowing box machine body; after the classifying filtering mesh belt penetrates out of the winnowing box machine body, a conveying device is arranged under the classifying filtering mesh belt. According to the tobacco material classifying device and method, materials move upwards through wind power, sundries fall into a sundry falling area due to the fact that the self gravity is larger than the wind power, cut rolled stems float and move upwards to a filter area, pass through the classifying filtering mesh belt and then collected and output, and flaky cut rolled stems are attached to the lower surface of the classifying filtering mesh belt, are taken out of the winnowing box machine body by the classifying filtering mesh belt, fall into the conveying device and then are discharged. Therefore, continuous, steady and uniform material classification is achieved, the defects that a traditional vibrating screen classifier is high in tobacco damage ratio, low in classification rate and the like are overcome, and the classification and usage requirements for different cut rolled stem forms in cigarette processing can be met.

The invention discloses a switching current circuit fault diagnosis method based on wavelet fractal and kernel principal characteristics. The method comprises the training step of a switching current circuit fault diagnosis module and the step of adopting the switching current circuit fault diagnosis module to carry out real-time fault diagnosis on a switching current circuit with faults. Wavelet decomposition and fractal analysis are combined to carry out characteristic extraction on fault signals, then kernel principal analysis is adopted to carry out optimal characteristic extraction on characteristic signals after wavelet fractal analysis, the faults of the switching current circuit are diagnosed through a support vector machine classifier, and when fault diagnosis is carried out on large-scale switching current circuit fault diagnosis and fault category overlapping, the characteristics of a proper number can be selected for the support vector machine to carry out fault diagnosis according to the complexity degree of fault data and the overlapping degree of fault categories, the fault classification rate is high, fault diagnosis is accurate, and the good generalization performance is achieved.

The invention discloses an observation vector difference based method for classifying SAR image textures. The method mainly solves the problem of terrain classification of SAR images. The method comprises the steps of (1) randomly selecting r images from a training set for partitioning processing, converting to obtain a column vector difference matrix P; (2) observing the P with an observation matrix to obtain a texture observation vector difference matrix X, and conducting clustering on the X to obtain a texture dictionary D; (3) calculating images of the training set according to Step (2) to obtain an observation vector difference matrix Xtr; (4) projecting the Xtr onto the texture dictionary D to form a training image texture column diagram h; (5) representing images of a test set by a test image texture column diagram he; (6) calculating the distance between the he and the h, and determining the classification to which the he belongs according to the distance; and (7) calculating all test images according to Step (6) to obtain a final classification rate. According to the method, the latest compressed sensing theory is applied, the process is simple, the classification identification rate is high, and the method is applicable to terrain texture classification of SAR images.

The invention provides a full connection neural network-based classification method of low interception radar signals and mainly solves the problem of a low correctness classification rate of the lowinterception radar signals with low signal-to-noise ratios in the prior art. The classification method includes the following steps of: 1) obtaining the low interception radar signals with different signal-to-noise ratios; 2) calculating bi-spectrum features of the low interception radar signals, and preprocessing and grouping bi-spectrum feature signals to obtain a data set; 3) designing a modelof a full connection neural network, and training the full connection neural network by utilizing the data set to obtain a well-trained full connection neural network; and 4) preprocessing unclassified low interception radar signals, inputting the unclassified low interception radar signals in the well-trained full connection neural network, and obtaining a classification of the unclassified low interception radar signals through network output. Simulation results show that according to the classification method of the invention, the classification correctness rate of the low interception radar signals with the low signal-to-noise ratios is much higher than that of conventional technology, and the classification method can be used to identify different types of radar signal sources.

The invention discloses a fault prediction and diagnosis method adopting double intelligent algorithms and used for a central air conditioner. The fault prediction and diagnosis method adopting the double intelligent algorithms and used for the central air conditioner comprises the following steps: acquiring a real-time data set of running of the central air conditioner in the field; inputting thereal-time data set in the step 1 into a pre-established fault prediction and diagnosis classification neural network model for the central air conditioner and based on a simulated annealing algorithmand an extreme learning machine algorithm, so as to judge whether the real-time data set in the step 1 meets the minimum misclassification rate of fault prediction and diagnosis for the central air conditioner or not; and if the real-time data set in the step 2 meets the minimum misclassification rate of fault prediction and diagnosis for the central air conditioner, outputting a fault predictionand diagnosis classification result for the central air conditioner. The fault prediction and diagnosis method adopting the double intelligent algorithms and used for the central air conditioner hasfield feasibility, and is convenient to apply for engineering. Meanwhile, an extreme learning machine has high generalization ability and high computation efficiency. Therefore, the globally optimal solution for the problem of evaluating the minimum misclassification rate of the central air conditioner through the simulated annealing algorithm is much easily obtained, and the accuracy and precision of online deployment for fault prediction and diagnosis model classification for the central air conditioner are increased.

The invention discloses a coal rock recognition method based on relativity measurement learning. According to the method, a new relativity measurement function is learnt from a training sample set in a monitoring mode to measure the relativity of coal and rock image samples, so that the relativity measurement value of samples in the same type is larger and larger, the relativity measurement value between samples in different types is smaller and smaller, and the classification rate of unknown samples is increased. The method includes an image preprocessing process, a training process and a recognition process. A preprocessing module is used for simple preprocessing of collected coal and rock images to obtain the training sample set. A training module is used for learning the optimal coal rock classification effect relativity measurement function from the training sample set. A recognition module is used for performing measurement classification by means of the optimal relativity measurement function. By means of the method, images of coal and rock under different illuminances and different viewpoints serve as the training samples, the method is little influenced by illuminance and imaging viewpoint changes, the recognition rate is high, and stability is good.

The invention discloses a target recognition method of a hickory nut picking robot based on machine vision. The method comprises the following steps: down-sampling an original image to obtain a down-sampled image; carrying out smooth denoising processing on the down-sampled image to obtain a denoised image; converting the denoised image in different color spaces, selecting and extracting a featureimage component with obvious target features, and acquiring a first binary image after threshold segmentation; processing the denoised image according to a color threshold range to obtain a second binary image; and performing morphological processing on the first binary image and the second binary image, and then taking a public area of the two images, wherein the public area is a hickory nut fruit recognized by the picking robot. According to the method, the classification rate of hickory nut target images is improved, and the method has good classification robustness and accuracy.

The invention discloses a data feature selection method and a system based on an improved particle swarm algorithm, comprising the following steps: a classifier model of an evaluation feature subset is determined; the correct classification rate of the classifier model is used to guide feature selection, and the number of feature subsets is gradually added to guide feature selection, and the quality evaluation function of feature subsets is established; criteria for feature selection in a dataset are determined; the particle swarm optimization (PSO) algorithm is improved. According to the quality evaluation function of the feature subset, the fitness of each particle in the PSO is calculated, and the inertia weight of the PSO algorithm is updated by Logistic chaotic atlas, so that the improved PSO algorithm can process the data set and obtain the selection results.

The invention relates to the technical field of artificial intelligence, and particularly discloses an interview answer text classification method and device, and the method comprises the steps: obtaining an interview answer text of an interviewer, and obtaining the interview answer text according to the reply of the interviewer to an interview question in an interview; constructing a semantic vector of the interview answer text through a feature extraction layer of the constructed classification model; through each full connection layer of the classification model, making full connection according to the semantic vector, acquiring a feature vector correspondingly wherein the feature vector obtained on the full connection layer is used for representing the feature of the sample answer texton the corresponding set capability item of the full connection layer; and carrying out classification prediction on the feature vectors obtained in each full connection layer, and respectively obtaining score levels of the interviewer on each set ability item. Therefore, automatic expansion of the dictionary is realized, and the classification rate of the interview answer text is improved. Therefore, automatic interview evaluation on the interviewer is realized.

The invention discloses an unsupervised clustering method used for large data volume spectral remote sensing image classification, comprising the following steps: dividing the original data into a plurality of data blocks, and obtaining a cluster center of each data subblock by virtue of a peak density searching method; dividing each cluster center into a plurality of data blocks again, and clustering again by virtue of the peak density searching method, so that number of the cluster centers is reduced; and repeating a partitioning-clustering process until similarity of any two cluster centerscan be represented by using one two-dimensional matrix, and then obtaining a final classification result. The unsupervised clustering method disclosed by the invention has the advantages that applicability is good, so that the method not only can be used for hyperspectral remote sensing image classification with more spectrum bands but also can be used for hyperspectral remote sensing image classification with fewer spectrum bands after multispectral remote sensing image or spectrum band selection; and operation efficiency is relatively high, blocked processing reduces computation redundancyof a similarity matrix, and clustering processing of all the data blocks is mutually independent, so that parallel processing can be adopted, and classification rate is increased.

The invention relates to an image classification method based on crowd intelligence ensemble learning. The technical scheme adopted by the invention is a method for improving the accuracy of image classification based on an ensemble learning method. A traditional svm training mode is changed, the traditional svm is combined with the thought of a random forest, and a final prediction result is morerobust and more reliable. For a face data set collected from a system database of the endocrinology department of a certain hospital, the optimal performance of the comprehensive classifier is 88.1%for the correct classification rate of a face image detection task. It is proved that the integrated learning method can achieve classification of the face images, and in other scene image data sets,the method also achieves a very good effect.

The invention relates to a vacuum classifier, suitable for the field of powder separation. Aiming at the problems that the qualified rate of finished products is very difficult to reach 100% by using a sieve-free classifier and the classification rate is lower because coarse particles are easy to attach on the surface of the sieve by using a sieve classifier, the vacuum classifier disclosed by the invention ensures that qualified particles rapidly enter a fine-powder silo through a cylindrical sieve by using negative pressure air-flow and air-flow generated by high-speed rotation of impellers, and the air-flow generated by high-speed rotation of the impellers enables the particles incapable of passing through the cylindrical sieve to break away from the surface of the cylindrical sieve and drop in a coarse-particle silo. The invention is mainly applied to the industries of pharmacy, food, chemical industry and the like.

The invention relates to a non-uniform big data classifying method which is used for classifying data set categories which cannot be classified and non-uniform category data sets of the data set categories which cannot be classified in a computer memory. Firstly, the size of a sample is determined by a downsampling method according to a theory, and the number of classifiers can be determined according to the number of the samples. An integrated classifier is established for each category of the big data. When a test case is tested, the integrated classifiers of all categories are classified, and the category where the integrated classifier with the highest classification rate is arranged is used as the category of the test case. The method is in linear to time complexity of big data category and can reduce polarization of non-uniform big data classification results. Furthermore, the integrated classifiers improve the accuracy. The method is easy to implement and only involves some simple math models in writing codes.

The invention provides a composite algorithm for early warning of excessive volatile fatty acids in an anaerobic process. The composite algorithm comprises the steps of acquiring a sample data set, and establishing a decision table; performing principal component analysis according to the sample data set in the decision table so as to obtain a principal component table and an initial factor load matrix table; calculating a feature vector according to the principal component table and the initial factor load matrix table, and calculating the proportions of all indexes in the overall data; selecting the index values with high proportions in the proportion data, discretizing the breakpoints in the data set corresponding to the index values, selecting the breakpoints and forming a corresponding interval; generating an initial rule set, and the generating a final rule set of reactors according to the initial rule set and the interval; using the data participating in rough set mining as a test set, and performing test evaluation by using the final rule set so as to obtain a correct classification rate of three decision value ranges of the reactors, and carrying out the early warning of the excessive volatile fatty acids by means of a rough set.