CN112464977A - Object classification method, computer equipment and storage medium - Google Patents
Object classification method, computer equipment and storage medium Download PDFInfo
- Publication number
- CN112464977A CN112464977A CN202011105733.3A CN202011105733A CN112464977A CN 112464977 A CN112464977 A CN 112464977A CN 202011105733 A CN202011105733 A CN 202011105733A CN 112464977 A CN112464977 A CN 112464977A
- Authority
- CN
- China
- Prior art keywords
- classification
- feature set
- feature
- sub
- prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 65
- 230000004927 fusion Effects 0.000 claims abstract description 30
- 238000004422 calculation algorithm Methods 0.000 claims description 21
- 239000011159 matrix material Substances 0.000 claims description 19
- 238000000605 extraction Methods 0.000 claims description 17
- 238000004590 computer program Methods 0.000 claims description 10
- 230000000007 visual effect Effects 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 4
- 230000007774 longterm Effects 0.000 claims description 2
- 238000012549 training Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 5
- 238000010187 selection method Methods 0.000 description 4
- 238000007621 cluster analysis Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 229910052500 inorganic mineral Inorganic materials 0.000 description 2
- 239000011707 mineral Substances 0.000 description 2
- 238000000638 solvent extraction Methods 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 241001122767 Theaceae Species 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 239000008267 milk Substances 0.000 description 1
- 235000013336 milk Nutrition 0.000 description 1
- 210000004080 milk Anatomy 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000001953 sensory effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 235000013618 yogurt Nutrition 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
Abstract
The application discloses an object classification method, computer equipment and a storage medium, wherein the method comprises the following steps: acquiring a feature set of a target object; performing first selection on the feature set to obtain a first sub-feature set, and performing second selection on the feature set to obtain a second sub-feature set, wherein the first selection mode and the second selection mode are different; carrying out classification prediction on the first sub-feature set to obtain a first classification prediction result, and carrying out classification prediction on the second sub-feature set to obtain a second classification prediction result; and performing decision-level fusion on the first classification prediction result and the second classification prediction result to obtain a final classification result. In this way, the classification accuracy of the object can be improved.
Description
Technical Field
The present application relates to the field of classification technologies, and in particular, to an object classification method, a computer device, and a storage medium.
Background
In the process of classifying objects through traditional machine learning, it is very important to classify objects by using the modal features of salient objects, and especially to classify object objects by using human sensory features, for example, visual or tactile perception is usually required to analyze the features of object objects, so as to further classify the object objects.
Disclosure of Invention
In order to solve the above problems, the present application provides an object classification method, a computer device, and a storage medium, which can improve the object classification accuracy.
In order to solve the above technical problem, one technical solution adopted by the present application is to provide an object classification method, including: acquiring a feature set of a target object; performing first selection on the feature set to obtain a first sub-feature set, and performing second selection on the feature set to obtain a second sub-feature set, wherein the first selection mode and the second selection mode are different; carrying out classification prediction on the first sub-feature set to obtain a first classification prediction result, and carrying out classification prediction on the second sub-feature set to obtain a second classification prediction result; and performing decision-level fusion on the first classification prediction result and the second classification prediction result to obtain a final classification result.
Wherein, obtaining the feature set of the target object comprises: collecting data information of a target object under multiple modes; preprocessing data information; and performing feature extraction on the preprocessed data information to generate a feature set of the target object.
Wherein, the data information in the plurality of modalities comprises visual information, tactile information and auditory information; performing feature extraction on the preprocessed data information to generate a feature set of the target object, wherein the feature set comprises: performing image statistical feature extraction on the preprocessed visual information to obtain a first statistical feature set; performing signal statistical feature extraction on the preprocessed tactile information and auditory information to obtain a second statistical feature set; and performing feature level fusion on the first statistical feature set and the second statistical feature set to generate a feature set of the target object.
The image statistical characteristics comprise at least one of energy, entropy, correlation, inverse difference moment and contrast in a gray level co-occurrence matrix, short-term key characteristics, long-term key characteristics, gray level nonuniformity, directional gradient histograms and roughness, contrast and directionality in Tamura texture characteristics in a gray level run matrix; the signal statistical characteristics include at least one of a maximum, a minimum, a mean, a peak, an absolute mean, a root mean square value, a variance, a standard deviation, a root-mean-square magnitude, a kurtosis, a skewness, a form factor, a peak factor, an impulse factor, and a margin factor.
The method for performing classification prediction on the first sub-feature set to obtain a first classification prediction result and performing classification prediction on the second sub-feature set to obtain a second classification prediction result comprises the following steps: classifying and predicting the first sub-feature set by using a set number of first classifiers to obtain a first classification prediction result of the target object; the first classification prediction result comprises a plurality of groups of first prediction classification labels and corresponding first prediction precision; classifying and predicting the second sub-feature set by using a set number of second classifiers to obtain a second classification prediction result of the target object; and the second classification prediction result comprises a plurality of groups of second prediction classification labels and corresponding second prediction precision.
The method for performing decision-level fusion on the first classification prediction result and the second classification prediction result to obtain a final classification result comprises the following steps: and performing decision-making level fusion on the multiple groups of first prediction classification labels, the first prediction precisions corresponding to the multiple groups of first prediction classification labels and the weight occupied by the first classifier, the multiple groups of second prediction classification labels, the second prediction precisions corresponding to the multiple groups of second prediction classification labels and the weight occupied by the second classifier to obtain a final classification result.
The method for performing decision-level fusion on the first classification prediction result and the second classification prediction result to obtain a final classification result comprises the following steps: the following formula is used for calculation:where n is the number of classification labels, m is the number of classifiers, ωmIs the weight occupied by classifier m, PmnThe classifier m calculates the prediction accuracy of the target object belonging to the classification label n.
The first selection of the feature set to obtain a first sub-feature set and the second selection of the feature set to obtain a second sub-feature set include: performing Laplace feature selection on the feature set to obtain a first sub-feature set; and performing multi-cluster feature selection on the feature set to obtain a second sub-feature set.
Wherein, the laplacian feature selection is performed on the feature set to obtain a first sub-feature set, which includes: calculating Laplacian scores of all characteristic parameters in the characteristic set according to a Laplacian characteristic selection algorithm; sorting the laplacian scores of all the characteristic parameters; and collecting the set number of characteristic parameters with smaller scores in the score sorting to obtain a first sub-characteristic set.
Wherein, performing multi-cluster feature selection on the feature set to obtain a second sub-feature set, comprises: calculating the multi-clustering score of all characteristic parameters in the characteristic set according to a multi-clustering characteristic selection algorithm; sorting the multi-clustering scores of all the characteristic parameters; and collecting the set number of characteristic parameters with larger scores in the score sorting to obtain a second sub-characteristic set.
In order to solve the above technical problem, another technical solution adopted by the present application is: there is provided a computer device comprising a processor and a memory for storing a computer program for implementing the object classification method as described above when executed by the processor.
In order to solve the above technical problem, the present application adopts another technical solution: a computer-readable storage medium is provided for storing a computer program which, when executed by a processor, is adapted to carry out the object classification method as described above.
The beneficial effects of the embodiment of the application are that: different from the prior art, the method for classifying the object is provided, the feature set of the target object is subjected to first selection and second selection to obtain a first sub-feature set and a second sub-feature set under two feature selections, the two sub-feature sets are further subjected to classification prediction to obtain a first classification prediction result and a second classification prediction result, and finally the first classification prediction result and the second classification prediction result are subjected to decision-level fusion calculation to output and obtain a final classification result. By the method, the classification results under the two feature selection methods are fused, and the classification accuracy of the object can be improved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts. Wherein:
FIG. 1 is a schematic flow chart diagram illustrating an embodiment of an object classification method provided herein;
FIG. 2 is a schematic flow chart diagram illustrating another embodiment of an object classification method provided herein;
FIG. 3 is a schematic flow chart showing the detail of step 23 in FIG. 2;
FIG. 4 is a schematic flow chart showing details of step 24 in FIG. 2;
FIG. 5 is a schematic flow chart showing the detail of step 25 in FIG. 2;
FIG. 6 is a schematic flow chart diagram illustrating a further embodiment of an object classification method provided herein;
FIG. 7 is a schematic block diagram of an embodiment of a computer device provided herein;
FIG. 8 is a schematic structural diagram of an embodiment of a computer-readable storage medium provided in the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. It is to be understood that the specific embodiments described herein are merely illustrative of the application and are not limiting of the application. It should be further noted that, for the convenience of description, only some of the structures related to the present application are shown in the drawings, not all of the structures. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is explicitly and implicitly understood by one skilled in the art that the embodiments described herein can be combined with other embodiments.
Referring to fig. 1, fig. 1 is a schematic flow chart of an embodiment of an object classification method provided in the present application, and the specific method of the embodiment is as follows:
step 11: a feature set of a target object is acquired.
The target object is an object to be classified or identified, the feature set of the target object represents a set of multi-modal feature data, the multi-modal data are acquired in different modes, so that multi-modal information of the object to be classified is obtained, the multi-modal information is further preprocessed and feature extraction with different dimensions is carried out, and the feature set of the target object to be classified can be obtained preliminarily.
Step 12: the method includes performing a first selection of the feature set to obtain a first sub-feature set, and performing a second selection of the feature set to obtain a second sub-feature set.
Wherein the first selection and the second selection are selected in different manners. In this embodiment, the same feature set is calculated by using two different feature selection manners, and unrelated features in the feature set can be removed according to the two different selection criteria, that is, partial features with low relevance are removed, so as to obtain a first sub-feature set and a second sub-feature set at a low latitude respectively, and the first sub-feature set and the second sub-feature set are used as features required by subsequent actual classification, and the partial features are obtained by performing score sorting through two feature selection algorithms.
Optionally, the first selection in this embodiment may be Laplacian feature selection (Laplacian Score), the second selection may be multi-cluster feature selection (MCFS), the two selections may also be interchanged, the order of using the algorithm is not particularly limited, and the following embodiments will describe this step in detail.
Step 13: and carrying out classification prediction on the first sub-feature set to obtain a first classification prediction result, and carrying out classification prediction on the second sub-feature set to obtain a second classification prediction result.
The first classification prediction result and the second classification prediction result respectively represent two classification prediction results of different sub-feature sets under different classifiers, the same classification prediction result comprises multiple groups of prediction classification labels and prediction precisions corresponding to the labels, and the prediction precisions represent the probability that the target object is classified into a certain prediction classification label under the calculation of the classifier.
That is, the classification prediction result usually includes a plurality of prediction classification labels and a probability corresponding to the classification of the target object into each label, and the calculated probability corresponding to each label is different between the first classification prediction result and the second classification prediction result due to the difference of the sub-feature sets. It can be understood that the prediction classification labels are usually set in advance, and the label content and the number thereof can be set according to actual needs, which are not described herein in detail.
Step 14: and performing decision-level fusion on the first classification prediction result and the second classification prediction result to obtain a final classification result.
The decision-level fusion means that after a first classification prediction result and a second prediction classification result are obtained, the classification prediction results calculated under different classifiers are summed according to weights of different classifiers, namely, the same prediction classification labels and the corresponding prediction precisions in the different classifiers are summed, and finally the summed prediction precision is used as the final classification label of the target object to obtain the final classification result. The embodiment adopts the idea of decision-level fusion and integrates the classification results of two feature selection methods, so that the classification accuracy can reach higher precision.
Different from the prior art, the method for classifying the object is provided, the feature set of the target object is subjected to first selection and second selection to obtain a first sub-feature set and a second sub-feature set under two feature selections, the two sub-feature sets are further subjected to classification prediction to obtain a first classification prediction result and a second classification prediction result, and finally the first classification prediction result and the second classification prediction result are subjected to decision-level fusion calculation to output and obtain a final classification result. By the method, the classification results under the two feature selection methods are fused, and the classification accuracy of the object can be improved.
Referring to fig. 2, fig. 2 is a schematic flow chart of another embodiment of the object classification method provided in the present application, and the specific method of the embodiment is as follows:
step 21: and acquiring data information of the target object under multiple modes.
Wherein the data information in the plurality of modalities includes visual information, tactile information, and auditory information of the target object. For the acquisition of the visual information, a device with a photographing function, such as a mobile device, a camera, a tablet computer, and the like, may be used to acquire the target object so as to acquire the visual information; for the acquisition of the tactile information, a mechanical device provided with a tactile sensor can be used for performing a sliding operation on the surface of a target object so as to obtain the tactile information such as friction force, for example, when the method is applied to a robot, the mechanical device is a manipulator; for the collection of the auditory information, a sound recorder or a microphone can be used for recording the sound generated when the surface of the target object is contacted with the mechanical device during the sliding process of the mechanical device on the surface of the target object, so as to collect the auditory information.
The three modal information of the target object can be obtained through the three modes. Optionally, each object is acquired 10 times by the same person as a training set and 1 time by 10 different persons as a test set, each acquisition taking 10 seconds.
Step 22: and preprocessing the data information.
In this embodiment, the tactile information and the auditory information are converted into time domain and frequency signals, and for the visual information, all the information is cut into pictures with the same resolution by cutting.
In other embodiments, the preprocessing may also solve problems such as missing or duplication, noise, etc. in the data information, for example, by using simple substitution or deletion to obtain new data.
Step 23: and performing feature extraction on the preprocessed data information to generate a feature set of the target object.
Specifically, step 23 may be implemented by the method shown in fig. 3, specifically as follows:
step 231: and performing image statistical feature extraction on the preprocessed visual information to obtain a first statistical feature set.
The image statistical characteristics include 11 statistical characteristics such as energy, entropy, correlation, inverse difference moment, contrast in a Gray Level Co-occurrence Matrix (GLCM), Short Run Emphasis (SRM), LONG Run Emphasis (LRM) and Gray Level Non-Uniformity (GLN) in a Gray Level Run Matrix (GLRLM), direction gradient histograms, roughness, contrast and directionality in Tamura texture features.
In this embodiment, an HOG (Histogram of Oriented gradients) feature extraction algorithm may be used to extract image features, the extracted image features are displayed as a digital sequence in a computer, and each extracted sample is put together according to a sequence order, which is a first statistical feature set.
Step 232: and performing signal statistical feature extraction on the preprocessed tactile information and auditory information to obtain a second statistical feature set.
The signal statistical characteristics comprise 15 statistical characteristics such as a maximum value, a minimum value, a mean value, a peak value, an absolute mean value, a root mean square value, a variance, a standard deviation, a root mean square amplitude, a kurtosis, a skewness, a form factor, a peak factor, an impulse factor, a margin factor and the like.
In this embodiment, signal feature extraction may be performed by using a short-time Fourier transform (STFT), a time-frequency distribution (Wigner-ville distribution, Choi-Willian distribution), a wavelet transform, a hilbert-yellow transform, and other feature extraction methods, and all the extracted features may be collected as the second statistical feature set.
It can be understood that, in the embodiment, simple statistical features are used for feature extraction in the feature extraction stage, so that the calculation steps can be simplified, the complexity of a subsequent training model is reduced, and the classification efficiency is also improved.
Step 233: and performing feature level fusion on the first statistical feature set and the second statistical feature set to generate a feature set of the target object.
The feature set fusion is to collect image statistical features and signal statistical features extracted from the feature set, and identify the target object by adopting a multi-mode fusion idea, so that the target object can be classified according to data of multiple modes, and compared with a single mode, the feature set fusion has a better classification effect and is more comprehensive in classification.
Step 24: and performing Laplace feature selection on the feature set to obtain a first sub-feature set.
The Laplace feature selection algorithm is an algorithm for scoring the features of a training set sample, a score can be scored for each feature in the training set through the algorithm, and finally a set number of features with the lowest scores are taken as a first finally selected sub-feature set, so that the Laplace feature selection algorithm is a standard Filter type method.
Specifically, step 24 may be implemented by the method shown in fig. 4, specifically as follows:
step 241: and calculating the Laplacian scores of all the characteristic parameters in the characteristic set according to a Laplacian characteristic selection algorithm.
The method comprises the following steps:
1) a weight matrix S is constructed.
For supervised learning, let n samples in the training set, first construct an n × n adjacency matrix G according to the sample typeijWhere i and j denote samples, G when type (i) ═ type (j), i.e. both belong to the same type of sampleij1, otherwise 0, and GijAll of the main diagonal elements of (1).
Then for G in the matrixijPoint 1, orderThe matrix thus obtained is the weight matrix S of the training set,wherein z is a fixed constant.
For unsupervised learning, the method is similar to supervised learning, but a K-nearest neighbor method is used to determine whether two samples are nearest neighbors, if a sample i is one of the nearest neighbors of a sample j, a default type (i) ═ type (j) is determined, that is, the default sample i and the sample j belong to the same type of sample, and then the supervised learning can be performed.
2) A laplacian score is calculated.
Specifically, the following formula can be used for calculation:
wherein L isr(ii) a laplace score for the r-th feature; (f)ri-frj) Is the difference between the r-th features of the ith and jth samples, (f)ri-frj) The smaller the absolute value of (A) is, the smaller the change of the characteristic r among the samples of the same type is, the better the characteristic is; sijFor corresponding values in the weight matrix, SijThe larger the weight value is, the better the characteristic is performed in the same sample, because of S between different samplesij0, so it is not considered; var (f)r) For the variance of the r-th feature over all samples, the greater the variance, the more obvious the variation of the feature over different samples, i.e. the more distinguishable the feature is from the samples.
It will be appreciated that for a good feature, SijThe larger the size, the more (f) isri-frj) The smaller the absolute value of (f) because it is undesirable for a feature to vary too much from class to class, while Var (f)r) The larger the better, so the final calculated LrSmaller indicates better characteristics.
Step 242: the laplacian scores of all feature parameters are ranked.
The arrangement may be from small to large or from large to small, and is not limited herein.
Step 243: and collecting the set number of characteristic parameters with smaller scores in the score sorting to obtain a first sub-characteristic set.
The set number of the selected characteristic parameters may be set according to an actual situation, for example, 10, and is not limited herein.
Step 25: and performing multi-cluster feature selection on the feature set to obtain a second sub-feature set.
In this embodiment, the multi-cluster feature selection mainly introduces spectral features to perform multi-cluster analysis to obtain the feature vector of the generalized feature problem with the minimum feature value.
Specifically, step 25 may be implemented by the method shown in fig. 5, specifically as follows:
step 251: and calculating the multi-clustering score of all the characteristic parameters in the characteristic set according to a multi-clustering characteristic selection algorithm.
The method comprises the following steps:
1) and spectrum embedding clustering analysis.
The multi-clustering feature selection algorithm firstly considers each sample object in the data set as a vertex V of the atlas, quantizes the similarity between the vertexes as the weight of a corresponding vertex connecting edge E, and then obtains an undirected weighted graph G (V, E) based on the similarity, thereby converting the clustering problem into the graph partitioning problem. Due to the nature of the graph partitioning problem, considering the continuous relaxation form of the problem is a good solving method, so that the original problem can be converted into the spectral decomposition for solving the similar matrix W or the laplacian matrix L.
There are many ways to solve the similarity matrix, including, for example, 0-1 weighting, thermonuclear weighting, etc., where ITML metric learning algorithms are employed. The method comprises the following specific steps:
defining a diagonal matrix D whose number on the diagonal is the sum of corresponding rows of the similarity matrix W, i.e.The non-standard laplacian matrix of the similarity graph is defined as L ═ D-W, and the first k eigenvectors of the laplacian matrix are calculated by Ly ═ λ Dy and are denoted as Y ═ Y [1,…yk]And completing the mapping of the original sample set to the feature vector space.
Wherein, ykIs the eigenvector of the generalized eigenproblem described above for the smallest eigenvalue, each row of Y is a reduced-dimension representation of a data point, K is the intrinsic dimension of the data, and each Y iskEmbodying the data distribution of the data in that dimension, each y when using cluster analysiskThe distribution of the data in the cluster can be reflected, so that the K can be set to the number of the data clusters.
2) A multimerization score is calculated.
After Y is obtained, the importance of each intrinsic dimension, i.e., each column of Y, can be measured, while the ability of each feature to distinguish between clusters of data can be measured. Given a ykBy minimizing the fitting error, a relevant subset of features can be found, as follows:
wherein, akIs an m-dimensional vector, represents correlation coefficients of different dimensions,is akIs a weight coefficient, akIncludes a function for approximating ykThe coefficients of each feature, due to the nature of the L1 norm, when β is large enough, akSome of the coefficients may become 0. Therefore, the above formula of feature subset is essentially a regression problem, called LASSO (Least Absolute regression and Selection Operator), and the correlation coefficient of each feature can be obtained by solving the above formula using Least Angle Regression (LARs) algorithm.
Further, for each feature j, an MCFS score for each feature may be defined:
wherein, ak,jIs a vector akThe jth element of (1).
Step 252: the multi-cluster scores for all the feature parameters are ranked.
The arrangement may be from small to large or from large to small, and is not limited herein.
Step 253: and collecting the set number of characteristic parameters with larger scores in the score sorting to obtain a second sub-characteristic set.
The set number of the selected characteristic parameters may be set according to an actual situation, for example, 10, and is not limited herein.
It will be appreciated that the order of step 24 and step 25 is not sequential.
Step 26: and carrying out classification prediction on the first sub-feature set to obtain a first classification prediction result, and carrying out classification prediction on the second sub-feature set to obtain a second classification prediction result.
The first sub-feature set and the second sub-feature set are classified and predicted by different classifiers, the same classification prediction result comprises a plurality of groups of prediction classification labels and prediction precisions corresponding to the labels, and the prediction precision represents the probability that the target object is classified into a certain prediction classification label under the calculation of the classifier.
Step 27: and performing decision-level fusion on the first classification prediction result and the second classification prediction result to obtain a final classification result.
The decision-level fusion means that after a first classification prediction result and a second prediction classification result are obtained, the classification prediction results calculated under different classifiers are summed according to weights of different classifiers, namely, the same prediction classification labels and the corresponding prediction precisions in the different classifiers are summed, and finally the summed prediction precision is used as the final classification label of the target object to obtain the final classification result.
Referring to fig. 6, fig. 6 is a schematic flow chart of another embodiment of the object classification method provided in the present application, and the specific method of the embodiment is as follows:
step 61: a feature set of a target object is acquired.
Step 62: the method includes performing a first selection of the feature set to obtain a first sub-feature set, and performing a second selection of the feature set to obtain a second sub-feature set.
Steps 61-62 are similar to steps 11-12 and steps 21-25 of the above embodiments, and are not described herein.
And step 63: and carrying out classification prediction on the first sub-feature set by using a set number of first classifiers to obtain a first classification prediction result of the target object.
Step 64: and carrying out classification prediction on the second sub-feature set by using a set number of second classifiers to obtain a second classification prediction result of the target object.
The first classification prediction result comprises a plurality of groups of first prediction classification labels and corresponding first prediction precision, the second classification prediction result comprises a plurality of groups of second prediction classification labels and corresponding second prediction precision, and the first prediction classification labels and the second prediction classification labels are the same.
In the embodiment, the classification is mainly performed by adopting an algorithm of a support vector machine, the training is mainly performed by a Svmtran function, and the sample prediction is performed by a Svmpress function.
In the classification prediction, the first classifier and the second classifier belong to different classifiers, and the weights occupied by the two classifiers can be set, and in this embodiment, since only two classifiers are adopted, the weights respectively occupy 0.5.
In the following, taking a robot as an example of a main body of recognition, the recognition and classification of a target object will be exemplified, and the predictive classification tag is preset to include 17 kinds of tags such as two pieces of sponge, two different dolls, toilet paper, milk tea, yogurt paper box, three kinds of mineral water bottles of different brands, three kinds of empty mineral water bottles of different brands, large and small glass bottles, and large and small metal bottles:
as can be seen from the above table, due to the difference between the algorithm and the classifier, the first sub-feature set obtained by using the laplacian feature selection algorithm, the first classifier, which is used to calculate the first prediction accuracies corresponding to the prediction classification label of the target object, is different from the second sub-feature set obtained by using the multi-cluster feature selection algorithm, and the second classifier, which is used to calculate the second prediction accuracies corresponding to the prediction classification label of the same target object, so that the prediction classification label and the prediction accuracy can be directly output by the above method.
It can be understood that the correspondence between the prediction classification labels and the prediction accuracy presented in the above table is only a result obtained by collecting and calculating once for a training set. The support vector machine classifier has a good classification effect on multiple classifications of objects, is simple and understandable in algorithm, and can directly output prediction classification labels and prediction accuracy.
Step 65: and performing decision-making level fusion on the multiple groups of first prediction classification labels, the first prediction precisions corresponding to the multiple groups of first prediction classification labels and the weight occupied by the first classifier, the multiple groups of second prediction classification labels, the second prediction precisions corresponding to the multiple groups of second prediction classification labels and the weight occupied by the second classifier to obtain a final classification result.
The decision-level fusion means that after a first classification prediction result and a second prediction classification result are obtained, the classification prediction results calculated under different classifiers are summed according to weights of different classifiers, namely, the same prediction classification labels and the corresponding prediction precisions in the different classifiers are summed, and finally the summed prediction precision is used as the final classification label of the target object to obtain the final classification result.
Specifically, the method can be realized by the following formula:
where n is the number of classification tags, in this embodiment, n is 17, m is the number of classifiers, ω ismIs the weight occupied by classifier m, PmnThe classifier m calculates the prediction precision of the target object belonging to the classification label n, namely the probability of belonging to the label. In the present embodiment, since only two classifiers are provided, the weights of the classifiers are each 0.5.
For the above formula, the calculation is essentially: multiplying and summing the prediction precision of the same label under different classifiers and the weight of the classifier, which is equivalent to calculating to obtain the actual precision corresponding to each label, namely omegamPmnRepeatedly calculating the actual precision corresponding to each label, and finally obtaining the label corresponding to the highest actual precision by using the argmax function, namely the final classification result n*The whole process is decision-level fusion. By adopting the mode, a decision-level fusion method is adopted, classification results under two feature selection methods are integrated, and the classification accuracy can be effectively improved.
In other embodiments, the number m of the classifiers may also be set to be greater than 2, and corresponding weights thereof may also be changed, for example, two classifiers are respectively used to perform classification prediction on the first sub-feature set and the second sub-feature set, so that the four classification prediction results are subjected to decision-level fusion, and the classification accuracy is further improved.
The application has been tested, and the multi-modal data and the simple statistical method adopted by the application have been well verified on public data and LMT-108, so that compared with the previous classification result, the method is better improved, and the method is proved to be feasible.
Referring to fig. 7, fig. 7 is a schematic structural diagram of an embodiment of a computer device provided in the present application, a computer device 70 of the present embodiment includes a processor 71 and a memory 72, the processor 71 is coupled to the memory 72, where the memory 72 is used for storing a computer program executed by the processor 71, and the processor 21 is used for executing the computer program to implement the following method steps:
acquiring a feature set of a target object; performing first selection on the feature set to obtain a first sub-feature set, and performing second selection on the feature set to obtain a second sub-feature set, wherein the first selection mode and the second selection mode are different; carrying out classification prediction on the first sub-feature set to obtain a first classification prediction result, and carrying out classification prediction on the second sub-feature set to obtain a second classification prediction result; and performing decision-level fusion on the first classification prediction result and the second classification prediction result to obtain a final classification result.
Referring to fig. 8, fig. 8 is a schematic structural diagram of an embodiment of a computer-readable storage medium provided in the present application. The computer-readable storage medium 80 of the present embodiment is used for storing a computer program 81, the computer program 81, when being executed by a processor, is adapted to carry out the method steps of:
acquiring a feature set of a target object; performing first selection on the feature set to obtain a first sub-feature set, and performing second selection on the feature set to obtain a second sub-feature set, wherein the first selection mode and the second selection mode are different; carrying out classification prediction on the first sub-feature set to obtain a first classification prediction result, and carrying out classification prediction on the second sub-feature set to obtain a second classification prediction result; and performing decision-level fusion on the first classification prediction result and the second classification prediction result to obtain a final classification result.
It should be noted that the method steps executed by the computer program 81 of the present embodiment are based on the above-mentioned method embodiments, and the implementation principle and steps are similar. Therefore, when being executed by the processor, the computer program 81 may also implement other method steps in any of the above embodiments, which are not described herein again.
Embodiments of the present application may be implemented in software functional units and may be stored in a computer readable storage medium when sold or used as a stand-alone product. Based on such understanding, the technical solution of the present application may be substantially implemented or contributed by the prior art, or all or part of the technical solution may be embodied in a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, a network device, or the like) or a processor (processor) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
The above description is only for the purpose of illustrating embodiments of the present application and is not intended to limit the scope of the present application, and all modifications of equivalent structures and equivalent processes, which are made according to the content of the present specification and the accompanying drawings, or which are directly or indirectly applied to other related technical fields, are also included in the scope of the present application.
Claims (12)
1. A method of classifying an object, the method comprising:
acquiring a feature set of a target object;
performing first selection on the feature set to obtain a first sub-feature set, and performing second selection on the feature set to obtain a second sub-feature set, wherein the first selection mode and the second selection mode are different;
carrying out classification prediction on the first sub-feature set to obtain a first classification prediction result, and carrying out classification prediction on the second sub-feature set to obtain a second classification prediction result;
and performing decision-level fusion on the first classification prediction result and the second classification prediction result to obtain a final classification result.
2. The method of claim 1,
the acquiring the feature set of the target object comprises the following steps:
collecting data information of a target object under multiple modes;
preprocessing the data information;
and performing feature extraction on the preprocessed data information to generate a feature set of the target object.
3. The method of claim 2,
the data information in the plurality of modalities comprises visual information, tactile information and auditory information;
the performing feature extraction on the preprocessed data information to generate a feature set of the target object includes:
performing image statistical feature extraction on the preprocessed visual information to obtain a first statistical feature set;
performing signal statistical feature extraction on the preprocessed tactile information and auditory information to obtain a second statistical feature set;
and performing feature level fusion on the first statistical feature set and the second statistical feature set to generate a feature set of the target object.
4. The method of claim 3,
the image statistical characteristics comprise at least one of energy, entropy, correlation, inverse difference moment and contrast in a gray level co-occurrence matrix, short-term key characteristics, long-term key characteristics, gray level nonuniformity, directional gradient histograms and roughness, contrast and directionality in Tamura texture characteristics in a gray level run matrix;
the signal statistical characteristics include at least one of a maximum value, a minimum value, a mean value, a peak value, an absolute mean value, a root mean square value, a variance, a standard deviation, a root-mean-square magnitude, a kurtosis, a skewness, a form factor, a peak factor, an impulse factor, and a margin factor.
5. The method of claim 1,
the classifying and predicting the first sub-feature set to obtain a first classifying and predicting result and the classifying and predicting the second sub-feature set to obtain a second classifying and predicting result comprise:
using a set number of first classifiers to perform classification prediction on the first sub-feature set so as to obtain a first classification prediction result of the target object; the first classification prediction result comprises a plurality of groups of first prediction classification labels and corresponding first prediction precision;
classifying and predicting the second sub-feature set by using a set number of second classifiers to obtain a second classification prediction result of the target object; and the second classification prediction result comprises a plurality of groups of second prediction classification labels and corresponding second prediction precision.
6. The method of claim 5,
performing decision-level fusion on the first classification prediction result and the second classification prediction result to obtain a final classification result, wherein the decision-level fusion includes:
and performing decision-level fusion on the multiple groups of first prediction classification labels, the first prediction precisions corresponding to the multiple groups of first prediction classification labels and the weight occupied by the first classifier, the multiple groups of second prediction classification labels, the second prediction precisions corresponding to the multiple groups of second prediction classification labels and the weight occupied by the second classifier to obtain a final classification result.
7. The method of claim 6,
performing decision-level fusion on the first classification prediction result and the second classification prediction result to obtain a final classification result, wherein the decision-level fusion includes:
the following formula is used for calculation:
where n is the number of classification labels, m is the number of classifiers, ωmIs the weight occupied by classifier m, PmnThe classifier m calculates the prediction precision of the target object belonging to the classification label n.
8. The method of claim 1,
the first selection of the feature set to obtain a first sub-feature set and the second selection of the feature set to obtain a second sub-feature set include:
performing Laplace feature selection on the feature set to obtain a first sub-feature set; and
and performing multi-cluster feature selection on the feature set to obtain a second sub-feature set.
9. The method of claim 8,
the performing laplacian feature selection on the feature set to obtain a first sub-feature set includes:
calculating Laplacian scores of all characteristic parameters in the characteristic set according to a Laplacian characteristic selection algorithm;
ranking the laplacian scores for all feature parameters;
and collecting the set number of characteristic parameters with smaller scores in the score sorting to obtain a first sub-characteristic set.
10. The method of claim 8,
performing multi-cluster feature selection on the feature set to obtain a second sub-feature set, including:
calculating the multi-clustering score of all the characteristic parameters in the characteristic set according to a multi-clustering characteristic selection algorithm;
ranking the multimerization scores for all feature parameters;
and collecting the set number of characteristic parameters with larger scores in the score sorting to obtain a second sub-characteristic set.
11. A computer device comprising a processor and a memory, wherein the memory is configured to store a computer program which, when executed by the processor, is configured to carry out the object classification method according to any one of claims 1 to 10.
12. A computer-readable storage medium for storing a computer program which, when executed by a processor, is adapted to carry out the object classification method of any one of claims 1 to 10.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011105733.3A CN112464977A (en) | 2020-10-15 | 2020-10-15 | Object classification method, computer equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011105733.3A CN112464977A (en) | 2020-10-15 | 2020-10-15 | Object classification method, computer equipment and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112464977A true CN112464977A (en) | 2021-03-09 |
Family
ID=74833842
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011105733.3A Pending CN112464977A (en) | 2020-10-15 | 2020-10-15 | Object classification method, computer equipment and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112464977A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115719453A (en) * | 2023-01-09 | 2023-02-28 | 水利部交通运输部国家能源局南京水利科学研究院 | Rice planting structure remote sensing extraction method based on deep learning |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108596154A (en) * | 2018-05-14 | 2018-09-28 | 河海大学 | Classifying Method in Remote Sensing Image based on high dimensional feature selection and multi-level fusion |
CN108629373A (en) * | 2018-05-07 | 2018-10-09 | 苏州大学 | A kind of image classification method, system, equipment and computer readable storage medium |
CN111666275A (en) * | 2020-06-10 | 2020-09-15 | 创新奇智(广州)科技有限公司 | Data processing method and device, electronic equipment and storage medium |
-
2020
- 2020-10-15 CN CN202011105733.3A patent/CN112464977A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108629373A (en) * | 2018-05-07 | 2018-10-09 | 苏州大学 | A kind of image classification method, system, equipment and computer readable storage medium |
CN108596154A (en) * | 2018-05-14 | 2018-09-28 | 河海大学 | Classifying Method in Remote Sensing Image based on high dimensional feature selection and multi-level fusion |
CN111666275A (en) * | 2020-06-10 | 2020-09-15 | 创新奇智(广州)科技有限公司 | Data processing method and device, electronic equipment and storage medium |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115719453A (en) * | 2023-01-09 | 2023-02-28 | 水利部交通运输部国家能源局南京水利科学研究院 | Rice planting structure remote sensing extraction method based on deep learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Salimi et al. | Visual-based trash detection and classification system for smart trash bin robot | |
US11282295B2 (en) | Image feature acquisition | |
JP4719224B2 (en) | Classifier combination based on objective evaluation for multimodal identification | |
US7707132B2 (en) | User preference techniques for support vector machines in content based image retrieval | |
Hoiem et al. | Object-based image retrieval using the statistical structure of images | |
CN110659665B (en) | Model construction method of different-dimension characteristics and image recognition method and device | |
KR20040037180A (en) | System and method of face recognition using portions of learned model | |
WO2000048397A1 (en) | Signal processing method and video/audio processing device | |
JPH1055444A (en) | Recognition of face using feature vector with dct as base | |
CN104281835B (en) | Face recognition method based on local sensitive kernel sparse representation | |
US20080159624A1 (en) | Texture-based pornography detection | |
CN110502989A (en) | A kind of small sample EO-1 hyperion face identification method and system | |
CN115393666A (en) | Small sample expansion method and system based on prototype completion in image classification | |
KR101093107B1 (en) | Image information classification method and apparatus | |
CN112464977A (en) | Object classification method, computer equipment and storage medium | |
CN113673465A (en) | Image detection method, device, equipment and readable storage medium | |
CN111428772B (en) | Photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting | |
Jayasinghe et al. | Feature learning and analysis for cleanliness classification in restrooms | |
Gupta | Feature selection and analysis for standard machine learning classification of audio beehive samples | |
CN112257688A (en) | GWO-OSELM-based non-contact palm in-vivo detection method and device | |
Koskela et al. | Evaluating the performance of content-based image retrieval systems | |
Bhattacharjee et al. | Image retrieval based on structural content | |
Yadav et al. | Image classification using Hybrid MLP method | |
Kitamoto | Typhoon analysis and data mining with kernel methods | |
Park et al. | Image classification using gradient-based fuzzy c-means with divergence measure |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |