Object identification method, device, equipment and medium

Abstract

The present disclosure provides an object recognition method, including: obtaining multiple objects to be recognized, each object including multiple features; using a feature selection matrix to select multiple features of each object respectively, and obtaining multiple objects including at least one feature. Clustering multiple objects including at least one feature to obtain a clustering result. When it is determined that the clustering result meets the preset conditions, according to the clustering result, identifying the information included in the multiple objects; and when it is determined that the clustering result does not meet the preset conditions, updating the feature selection matrix according to the clustering result, and returning to the operation of selecting multiple features of each object respectively using the feature selection matrix. The present disclosure also provides an object recognition device, an electronic device and a computer-readable storage medium.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and more specifically, to an object recognition method, device, equipment and medium. Background Art

[0002] With the rapid growth of high-dimensional data, high-dimensional data clustering plays an increasingly important role in many fields, such as text mining, image search, computer vision, and bioinformatics. For example, in the text mining process, clustering the information included in the text, and identifying the information included in the text based on the clustering results can accurately mine the previously unknown, understandable, and ultimately usable knowledge and information in the text. For another example, in image search, clustering images, identifying the key information included in the image based on the clustering results, and then searching for the image to be searched efficiently and accurately based on the key information. Therefore, how to accurately cluster text or images directly affects the accuracy of text or image information recognition, and thus affects the efficiency and accuracy of text mining or image search. Summary of the invention

[0003] In view of this, the present disclosure provides an object recognition method, apparatus, device and medium.

[0004] One aspect of the present disclosure provides an object recognition method, comprising: acquiring multiple objects to be recognized, each of the objects comprising multiple features; using a feature selection matrix to select the multiple features of each object respectively, to obtain multiple objects including at least one feature; clustering the multiple objects including at least one feature to obtain a clustering result; if it is determined that the clustering result meets a preset condition, identifying information included in the multiple objects according to the clustering result; and if it is determined that the clustering result does not meet the preset condition, updating the feature selection matrix according to the clustering result, and returning to the operation of selecting the multiple features of each object respectively using the feature selection matrix.

[0005] According to an embodiment of the present disclosure, the use of a feature selection matrix to select the multiple features of each object respectively to obtain multiple objects including at least one feature includes: extracting features of different dimensions of each object respectively; combining the features of different dimensions of each object into a feature vector of each object; and using a feature selection matrix to select the feature vector of each object respectively to obtain a feature vector of each object including at least one feature.

[0006] According to an embodiment of the present disclosure, the use of a feature selection matrix to select the multiple features of each object respectively to obtain multiple objects including at least one feature includes: constructing a row sparse matrix; and selecting the multiple features of each object respectively through the row sparse matrix to obtain multiple objects including at least one feature.

[0007] According to an embodiment of the present disclosure, when it is determined that the clustering result does not meet the preset condition, the feature selection matrix is ​​updated according to the clustering result, including: when it is determined that the clustering accuracy included in the clustering result belongs to a first preset range, the feature selection matrix is ​​updated to select different types of features; when it is determined that the clustering accuracy included in the clustering result belongs to a second preset range, the feature selection matrix is ​​updated to select a different number of features.

[0008] According to an embodiment of the present disclosure, the multiple objects including at least one feature are clustered to obtain a clustering result, including: fixing the cluster center, the feature selection matrix and the numerical values ​​of any two variables in the confidence level, calculating the numerical value of another variable, and obtaining a group of numerical values ​​including the above three variables; calculating multiple groups of numerical values ​​including the above three variables in an iterative manner to obtain convergence values ​​of the cluster center, the feature selection matrix and the confidence level.

[0009] According to an embodiment of the present disclosure, the object includes at least one of text, image, voice and video.

[0010] According to an embodiment of the present disclosure, the image set includes a high-dimensional image dataset.

[0011] According to an embodiment of the present disclosure, the high-dimensional image dataset includes the COIL20 dataset or the COIL100 dataset or the ORL face dataset or the YALE face dataset.

[0012] According to an embodiment of the present disclosure, the object is an image; the object includes an image; and extracting features of different dimensions of each object separately includes: extracting at least one of the following features of the image: grayscale value; two-dimensional histogram; scale-invariant feature transformation value; directional gradient histogram.

[0013] On the other hand, the present disclosure provides an object recognition device, including: an acquisition module, used to acquire multiple objects to be recognized, each of the objects including multiple features; a selection module, used to use a feature selection matrix to select the multiple features of each object respectively, and obtain multiple objects including at least one feature; a clustering module, used to cluster the multiple objects including at least one feature to obtain a clustering result; a first determination module, used to identify information included in the multiple objects according to the clustering result when it is determined that the clustering result meets a preset condition; and a second determination module, used to update the feature selection matrix according to the clustering result when it is determined that the clustering result does not meet the preset condition, and return to the operation of using the feature selection matrix to select the multiple features of each object respectively.

[0014] According to an embodiment of the present disclosure, the selection module includes: an extraction unit for extracting features of different dimensions of each object respectively; a composition unit for composing the features of different dimensions of each object into a feature vector of each object. A first selection unit is used to select the feature vector of each object respectively using a feature selection matrix to obtain a feature vector of each object including at least one feature.

[0015] According to an embodiment of the present disclosure, the selection module further includes: a construction unit for constructing a row sparse matrix; and a second selection unit for selecting multiple features of each object respectively through the row sparse matrix to obtain multiple objects including at least one feature.

[0016] According to an embodiment of the present disclosure, the second determination module includes: a first determination unit, used to update the feature selection matrix to select different types of features when it is determined that the clustering accuracy included in the clustering result belongs to a first preset range; a second determination unit, used to update the feature selection matrix to select different numbers of features when it is determined that the clustering accuracy included in the clustering result belongs to a second preset range.

[0017] According to an embodiment of the present disclosure, the clustering module includes: a first calculation unit, which is used to fix the numerical values ​​of any two variables in the cluster center, the feature selection matrix and the confidence level, calculate the numerical value of another variable, and obtain a set of numerical values ​​including the above three variables; a second calculation unit, which is used to calculate multiple groups of numerical values ​​including the above three variables in an iterative manner to obtain the convergence values ​​of the cluster center, the feature selection matrix and the confidence level.

[0018] Another aspect of the present disclosure provides an electronic device, including: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described above.

[0019] Another aspect of the present disclosure provides a computer-readable storage medium storing computer-executable instructions, which are used to implement the above method when executed.

[0020] Another aspect of the present disclosure provides a computer program, which includes computer executable instructions, and the instructions are used to implement the method described above when executed. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The above and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0022] Figure 1 Schematically illustrates an exemplary system architecture 100 that can implement an object recognition method according to an embodiment of the present disclosure;

[0023] Figure 2 A flowchart of an object recognition method according to an embodiment of the present disclosure is schematically shown;

[0024] Figure 3 A flowchart schematically illustrates a selection feature according to an embodiment of the present disclosure;

[0025] Figure 4 A flowchart of selecting features according to another embodiment of the present disclosure is schematically shown;

[0026] Figure 5 The flowchart of the clustering method according to the embodiment of the present disclosure is schematically shown;

[0027] Figure 6 The flowchart of the weighted iterative solution method of the embodiment of the present disclosure is schematically shown.

[0028] Figure 7 A block diagram schematically shows an object recognition device according to an embodiment of the present disclosure;

[0029] Figure 8 Schematically shows a block diagram of a selection module according to an embodiment of the present disclosure;

[0030] Fig. 9 Schematically shows a block diagram of a selection module according to another embodiment of the present disclosure;

[0031] Fig.10A block diagram of an electronic device suitable for implementing the method described above according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0032] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0033] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0034] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0035] In the case of using expressions such as "at least one of A, B, and C, etc.", it should generally be interpreted as the meaning of the expression generally understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.). In the case of using expressions such as "at least one of A, B, or C, etc.", it should generally be interpreted as the meaning of the expression generally understood by those skilled in the art (for example, "a system having at least one of A, B, or C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0036] One method of clustering high-dimensional data is to first perform feature selection on the high-dimensional data to achieve dimensionality reduction, and then use a clustering algorithm to cluster the reduced-dimensional data. However, in this method, the feature selection task is irrelevant to the subsequent clustering task. Feature selection is likely to screen out features that are useful for the clustering task, which directly affects the subsequent clustering performance and results in low clustering accuracy.

[0037] The embodiment of the present disclosure provides an object recognition method and a device capable of applying the method. The method includes obtaining multiple objects to be recognized, each object including multiple features. A feature selection matrix is ​​used to select multiple features of each object respectively, and multiple objects including at least one feature are obtained. The multiple objects including at least one feature are clustered to obtain a clustering result. It is determined whether the clustering result meets the preset conditions. If not, the feature selection matrix is ​​updated according to the clustering result, and the previous operation is returned. If yes, the next operation is performed. According to the clustering result, the information included in the multiple objects is identified.

[0038] Figure 1 The exemplary system architecture 100 for implementing the object recognition method according to the embodiment of the present disclosure is schematically shown. It should be noted that: Figure 1 What is shown is merely an example of a system architecture to which the embodiments of the present disclosure can be applied, in order to help those skilled in the art understand the technical content of the present disclosure, but it does not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios.

[0039] like Figure 1 As shown, the system architecture 100 according to this embodiment may include a database 101, a network 102 and a server 103. The network 102 is used to provide a medium for a communication link between the database 100 and the server 103. The network 102 may include various connection types, such as wired and / or wireless communication links, and the like.

[0040] The database 100 may store objects to be identified, which may include text, images, voice and video. The network 102 may input the objects to be identified into the server 103. The server 103 may be a server that provides various services, such as feature selection and clustering of the objects to be identified, and identifying the information included in the objects based on the clustering results.

[0041] It should be noted that the object recognition method provided in the embodiment of the present disclosure can generally be executed by the server 103. Accordingly, the object recognition device provided in the embodiment of the present disclosure can generally be set in the server 103. The object recognition method provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 103 and can communicate with the database 101 and / or the server 103. Accordingly, the object recognition device provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 103 and can communicate with the database 101 and / or the server 103.

[0042] For example, when identifying an object, the process of clustering the objects and identifying the objects including information based on the clustering results is not directly executed by the server 103, but is executed by a server or server cluster that can communicate with the database 101 and the server 103. After the object identification is completed, the identified information is sent to the server 103.

[0043] It should be understood that Figure 1 The number of databases, networks and servers in the embodiment is only for illustration. Any number of databases, networks and servers may be provided as required.

[0044] Figure 2 The flowchart of the object recognition method according to the embodiment of the present disclosure is schematically shown.

[0045] like Figure 2 As shown, the object recognition method includes operations S201 to S205.

[0046] In operation S201 , a plurality of objects to be identified are obtained, each object including a plurality of features.

[0047] For example, the acquired object to be identified may be at least one of text, image, voice, and video with multi-dimensional features, or may be multi-dimensional data expressed in the form of numerical values, vectors, or matrices.

[0048] In operation S202, a feature selection matrix is ​​used to select multiple features of each object respectively, to obtain multiple objects including at least one feature.

[0049] High-dimensional data has multiple characteristic attributes, that is, its dimensions can reach hundreds or thousands of dimensions, or even higher. In the process of data recognition, some features of high-dimensional data are usually selected in a targeted manner to form new data in order to reduce the amount of calculation for data recognition and reduce the computing cost.

[0050] In operation S203, a plurality of objects including at least one feature are clustered to obtain a clustering result.

[0051] In operation S204, when it is determined that the clustering result satisfies a preset condition, information included in the plurality of objects is identified according to the clustering result.

[0052] In operation S205, when it is determined that the clustering result does not meet the preset condition, the feature selection matrix is ​​updated according to the clustering result, and the operation of selecting multiple features of each object respectively by using the feature selection matrix is ​​returned.

[0053] According to the embodiments of the present disclosure, feature selection and clustering are performed synchronously, the two are interdependent, and the results of feature selection and clustering affect each other. In the clustering process, the number and types of features actually involved in clustering are dynamically controlled to avoid discarding features useful for the clustering process by performing feature selection based on prior experience.

[0054] In order to facilitate understanding of the object recognition method provided by the embodiment of the present disclosure, the following example is provided for description. It should be understood that the example is not intended to limit the present disclosure.

[0055] For example, the object recognition method provided by the embodiment of the present disclosure is applied in commodity recommendation. After a user purchases a commodity through a shopping platform, the corresponding purchase information can be stored in the form of text, for example, user ID, purchase address, type of purchased commodity, etc., that is, the text content includes this information. In order to be able to recommend the commodities of interest to the corresponding user, so as to obtain the correlation between users, it is necessary to mine known valuable information from the text as a reference for commodity recommendation. Such valuable information may include, for example, the cycle or frequency of a user's purchase of a certain commodity, which can accurately reflect the commodities required by the user at different times, so as to recommend the commodities required to users with similar purchase frequencies in batches at the right time, thereby improving the success rate of commodity purchase. At this time, it is necessary to use text mining to mine the cycle or frequency of a user's purchase of a certain commodity from the features including the type of purchased commodity, the number of times the commodity of the type is purchased, and the time interval for purchasing the commodity of the type based on the text data. Using the object recognition method provided by the embodiment of the present disclosure, a large amount of text including user ID, purchase address, and type of purchased commodity is taken as the object, and its massive text is clustered, so as to efficiently and accurately identify the key information included in the text according to the clustering result, and then mine valuable information for commodity recommendation.

[0056] Another example is the application of the object recognition method provided by the embodiment of the present disclosure in image search. When it is necessary to quickly and accurately search for the images it needs from a large amount of data, the object recognition method provided by the embodiment of the present disclosure can be used. With a large amount of images to be searched as objects, the large amount of images are first clustered to obtain the clustering results of the images, and then the information included in different types of images can be quickly identified based on the clustering results, and then the images matching the current needs are searched based on the identified information, so that the search efficiency is higher and the matching is more accurate.

[0057] For another example, the object recognition method provided by the embodiment of the present disclosure is applied to another commodity recommendation. In order to recommend similar commodities to each user, the object to be identified is a massive amount of raw data including user purchase information, and the raw data includes multiple features, such as user age, purchase address, purchase time period, type of purchased commodity, etc. Cluster analysis is performed on the raw data, that is, the users corresponding to the raw data are classified to obtain user sets with certain feature similarities, so as to recommend different commodities to users in different user sets. For the same raw data, clustering can be performed based on different features to obtain different clustering results.

[0058] For example, clustering based on user age and purchased goods type characteristics, the clustering results are a set of users aged 20-30 who purchase clothing products, and a set of users aged 30-40 who purchase daily necessities. For another example, clustering based on purchase address and purchased goods value characteristics, the clustering results are a set of users whose purchase address is in a university area and who purchase goods with a value of less than 500 yuan, and a set of users whose purchase address is in a business center and who purchase goods with a value of more than 500 yuan. For another example, clustering based on user gender, purchase time period and purchased goods type characteristics, the clustering results are a set of female users who purchase clothing between 20:00 and 24:00, a set of male users who purchase electronic products between 22:00 and 24:00, a set of female users who purchase fresh products between 7:00 and 9:00, and a set of male users who purchase fast food between 11:00 and 13:00.

[0059] For the above multiple sets, each set has different clustering qualities. Understandably, the object recognition method provided by the present disclosure needs to identify the maximum similarity of the object, and it is necessary to obtain a user set with the best clustering quality. Therefore, in the clustering process, for each original data, it is impossible to make an accurate judgment based on a priori to determine which features and how many features are based on which clustering can obtain the optimal solution. If feature selection is performed on high-dimensional original data only based on historical feature selection experience, key features may be screened out, and the optimal clustering result cannot be obtained.

[0060] The object recognition method provided by the embodiment of the present disclosure realizes the continuous change of the number and types of feature selection by dynamically and continuously updating the feature selection matrix during the clustering process. Compared with the technical solution in which the traditional feature selection task is irrelevant to the subsequent clustering task, it can avoid the feature selection from discarding the features that play a key role in the clustering process, and then accurately identify the key information included in the object and the correlation between objects based on the clustering results. And because clustering is performed first and then different classes are identified separately (the same class has commonalities), information recognition is not directly performed from a large number of objects, thereby improving the efficiency of object recognition.

[0061] In order to more clearly illustrate the present invention, the following is a detailed description of the present invention in conjunction with specific embodiments. Figure 2 The object recognition method shown is further explained.

[0062] Figure 3 The flowchart of selecting features according to an embodiment of the present disclosure is schematically shown.

[0063] like Figure 3 As shown, operation S202 may further include operations S301 to S303.

[0064] In operation S301 , features of different dimensions of each object are extracted respectively.

[0065] According to the embodiments of the present disclosure, the object recognition method can be used to recognize text, images, voice, video and other corresponding high-dimensional data. Taking high-dimensional image data as an example, the high-dimensional image data can include COIL20 dataset or COIL100 dataset or ORL face dataset or YALE face dataset. When the object to be recognized is an image dataset, for each data, different features of the image data are extracted, and the extracted features include at least one of gray value, two-dimensional histogram, scale-invariant feature transform value (SIFT), directional gradient histogram (HOG) and other features.

[0066] In operation S302 , features of each object in different dimensions are combined into a feature vector of each object.

[0067] According to an embodiment of the present disclosure, for objects such as text, image, voice, video, etc. with multi-dimensional features, multiple features extracted from each object are combined to form a feature vector of the object, that is, text, image, voice, and video data are converted into feature vector data.

[0068] It should be noted that the feature extraction is performed on each object in the same manner, and the features are combined in the same manner to form a feature vector expressed in the same form.

[0069] For example, when the object to be identified is an image data set, the grayscale value, two-dimensional histogram, SIFT value and HOG corresponding to an image data are extracted, and the grayscale value, two-dimensional histogram, SIFT value and HOG value are serially spliced ​​to form a feature vector representing the image. In the feature vector, each feature can be converted into at least one numerical value, then the feature vector can be a column vector. Among them, the conversion method for the same feature of each object is the same, so that the final column vector corresponding to each object has the same number of rows, and the values ​​of the same number of rows of each column vector represent the same feature.

[0070] In operation S303, a feature selection matrix is ​​used to select feature vectors of each object respectively, so as to obtain a feature vector of each object including at least one feature.

[0071] The method for selecting features provided in the embodiments of the present disclosure converts data in the form of text, images, voice, video, etc. into feature vectors. The obtained feature vectors can more accurately obtain the features that characterize the object, so as to better cluster and identify the objects in the future and improve the accuracy of information recognition.

[0072] Figure 4 A flowchart of selecting features according to another embodiment of the present disclosure is schematically shown.

[0073] like Figure 4 As shown, operation S202 may further include operations S401-S402.

[0074] In operation S401, a row sparse matrix is ​​constructed, where the row sparse matrix is ​​a feature selection matrix.

[0075] In operation S402, multiple features of each object are selected respectively through the row sparse matrix to obtain multiple objects including at least one feature.

[0076] According to an embodiment of the present disclosure, by multiplying a row sparse matrix with a feature vector representing the features of each object, the feature vector can be sparsely processed to control the number of discriminative features actually participating in clustering. The motivation for feature selection is to select at least one feature from multiple original features of high-dimensional data. On the one hand, the original high-dimensional data is reduced in dimension to reduce the amount of computation and computational cost. On the other hand, the features usually selected are discriminative, and the associations between objects can be mined through the discriminative features.

[0077] Figure 5 The flowchart of the clustering method according to the embodiment of the present disclosure is schematically shown.

[0078] like Figure 5 As shown, in operation S205, when it is determined that the clustering result does not meet the preset condition, updating the feature selection matrix according to the clustering result may include operations S501 to S502.

[0079] In operation S501 , when it is determined that the clustering accuracy included in the clustering result belongs to a first preset range, a feature selection matrix is ​​updated to select different types of features.

[0080] In operation S502, when it is determined that the clustering accuracy included in the clustering result belongs to a second preset range, the feature selection matrix is ​​updated to select a different number of features.

[0081] According to an embodiment of the present disclosure, specifically, clustering can be performed using the following objective formula:

[0082]

[0083]

[0084] Where n is the number of objects in the object to be identified, c is the number of cluster centers, i is the number of objects in the object to be identified, j is the number of cluster centers, and x is the number of cluster centers. i is the feature vector corresponding to the i-th object, m j is the vector of the jth cluster center, y ij is the confidence that the i-th object belongs to the j-th cluster, W is the feature selection matrix, that is, the row sparse matrix, W T is the orthogonal matrix of the feature selection matrix, ‖W‖ 2,1 is the row sparse matrix W 2，1 norm, λ is the regularization parameter, 0≤λ≤1, I is the identity matrix, and st represents the constraint.

[0085] It should be noted that in this disclosure, uppercase bold letters are used to represent matrices, such as W, I; lowercase bold italic letters are used to represent vectors, such as x, and lowercase italic letters are used to represent scalars, such as y.

[0086] Looking at the above objective formula as a whole, the first term is the loss term, which represents the similarity between the feature vector corresponding to the projected data and the center vectors of different clusters, where W T x i Indicates the vector after feature selection of the eigenvector. Increasing the orthogonality of the feature selection matrix can avoid falling into a trivial solution in the subsequent solution process. m j Represents the cluster center of the vector after feature selection. The second term λ‖W‖ 2，1 It is used to dynamically control the number of discriminative features that actually participate in clustering. λ is used to control the sparsity of W. The larger the λ value, the sparser the matrix W.

[0087] According to an embodiment of the present disclosure, the preset condition may be an iteration stop condition. When the clustering result satisfies the iteration stop condition, the clustering result is obtained, and an operation of identifying the information included in multiple objects according to the clustering result is performed. When the obtained clustering result does not meet the iteration stop condition, the feature selection matrix is ​​updated, and the updated feature selection matrix is ​​reused to perform feature selection on the object to obtain a new feature vector after feature selection, and the new feature vector is clustered, and the above operation is repeated until the iteration stop condition is met. Usually, the feature vector after the newly obtained feature selection is different from the feature vector after the feature selection obtained last time. It can be understood by those skilled in the art that the iteration stop condition may include the number of iterations, and when the number of iterations reaches the preset number, the iteration is stopped to obtain the clustering result. The iteration stop condition may also be a numerical range that is satisfied by the clustering accuracy. Of course, the embodiments of the present disclosure are not limited to this.

[0088] For example, assuming that the acquired multiple objects have 5 features A, B, C, D, E, the feature selection matrix is ​​used to select the above 5 features. Specifically, in operation S502, if the clustering accuracy does not meet the preset conditions, the feature selection matrix is ​​updated and the previous operation is returned.

[0089] For example, if the value of the clustering accuracy obtained by clustering based on features A and B is less than or equal to 0.3 (the first preset range), it means that the accuracy of clustering based on features A and B is low, so the selected features need to be changed, and the feature selection matrix is ​​updated. The updated feature selection feature can replace one or all of the original features. For example, the updated feature selection matrix can select two features A and C, or select two features C and D, and execute the updated feature selection matrix to select multiple features of the object, obtain multiple feature vectors including two features A and C, or including two feature vectors C and D, and re-cluster.

[0090] If the clustering accuracy value obtained by clustering the two feature vectors including C and D is greater than 0.3 and less than or equal to 0.5 (the second preset range), it means that the clustering result generated based on the features C and D has a certain accuracy, and the similarity between the features can be further explored. Then, the feature selection matrix is ​​updated, and the updated feature selection feature can add a new feature. For example, the updated feature selection matrix can select the three features A, C and D. The updated feature selection matrix is ​​used to select multiple features of the object, and multiple feature vectors including the three features A, C and D are obtained, and clustering is performed again.

[0091] If the clustering accuracy value obtained by clustering the three features A, C and D is greater than 0.5 (the third preset range that meets the preset conditions), it means that the clustering result generated by clustering the three features A, C and D meets the requirements, and then the operation of identifying the information included in multiple objects based on the clustering results can be performed.

[0092] It should be noted that the number of features, feature types, preset conditions, ranges for clustering accuracy, and methods of selecting features listed in the above examples are all exemplary descriptions, and do not limit the application scenarios of the technical solutions provided by the present disclosure, nor do they limit the specific embodiments of the technical solutions provided by the present disclosure.

[0093] Through the embodiments of the present disclosure, the number and types of features are continuously and dynamically adjusted during the clustering process until the clustering result meets the iterative stopping condition, thereby achieving the fusion of the clustering process and the feature selection process. Using an orthogonal matrix of a sparse matrix for feature selection can avoid falling into a trivial solution during the clustering solution process, improve the accuracy of clustering, and then identify object information based on more accurate results, thereby improving the accuracy of information identification. In addition, by dynamically selecting the number of discriminative features that actually participate in clustering through the row sparse matrix, clustering and feature selection can be better integrated to achieve flexible clustering, thereby improving the accuracy and efficiency of clustering, and then improving the accuracy and efficiency of object information identification.

[0094] In order to further improve the accuracy of the object recognition method provided by the embodiment of the present disclosure, the embodiment of the present disclosure also provides a weighted iterative solution method for the above-mentioned target formula. That is, multiple objects including at least one feature are clustered to obtain clustering results, including the values ​​of any two variables in the fixed cluster center, feature selection matrix and confidence, and the value of another variable is calculated to obtain a group of values ​​including the above three variables. Then, multiple groups of values ​​including the above three variables are calculated in an iterative manner to obtain the convergence values ​​of the cluster center, feature selection matrix and confidence.

[0095] Figure 6 The flowchart of the weighted iterative solution method according to the embodiment of the present disclosure is schematically shown.

[0096] like Figure 6 As shown, the weighted iterative solution method may include, for example, operations S601 to S603.

[0097] In operation S601, W and m are fixed. j , update Y to solve the target formula. Matrix Y is the confidence matrix, and the confidence matrix Y is the confidence y that the i-th object belongs to different clusters ij The matrix formed.

[0098] Based on operation S601, W and m are fixed.j The above objective formula can be simplified to:

[0099]

[0100]

[0101] Through the above restrictions, the problem of solving the target formula can be transformed into a constrained linear optimization problem, which can be solved directly using the existing optimization toolbox, such as using the Lagrangian algorithm to solve the minimum value.

[0102] In operation S602, W and Y are fixed, and m is updated. j Solve the target formula.

[0103] Based on operation S602, with W and Y fixed, the above target formula can be simplified to:

[0104]

[0105] wxya T W=I

[0106] For the variable m in the simplified formula j Take the derivative and set it to zero to get m j The solution is

[0107] In operation S603, m is fixed j and Y, update W to solve the target formula.

[0108] Based on operation S603, m is fixed j and Y remain unchanged, the above objective formula can be simplified to:

[0109]

[0110] wxya T W=I

[0111] make in, represents the average direction of the original data of the jth cluster, S w is the intra-cluster divergence matrix, which characterizes the compactness of the clusters, and can be further simplified as:

[0112]

[0113] wxya T W=I

[0114] Where D is a diagonal matrix, and the value of the i-th element on its diagonal is ε is an arbitrarily small constant. Based on the final simplified formula, Y can be directly obtained.

[0115] The above operations S601 to S603 are iterated to solve until W and m j and Y converge to obtain the final clustering result. It should be noted that the execution of the above operations S601 to S603 is not necessarily in the above order, and the logical order of execution can be selected according to actual conditions.

[0116] It should be noted that before clustering objects, the objects to be clustered are projected into a data matrix, and the feature selection matrix can be constructed as Represents, where d represents the dimension of the data matrix, that is, the number of objects, and e is the number of features included in the feature vector corresponding to the object. ‖W T x i -m j ‖ 2 W T x i -m j l 2 norm, ‖W‖ 2，1 is the row sparse matrix W 2.1 Norm.

[0117] The following is a description of the embodiment of the present invention. 2 Norm and l 2.1 The norm is explained in detail.

[0118] For the matrix q ab ,q a Respectively represent the elements at the matrix (a, b) position and the vector composed of the elements in the ath column. Its p The norm is defined as For example, the vector l 0 The norm is expressed as Its 1 The norm is expressed as Where |·| represents the absolute value. 2 The norm is expressed as Expanding from the vector norm to the matrix norm, the F norm of the matrix Q is defined as:

[0119]

[0120] The matrix Q 2.0 The norm is defined as:

[0121]

[0122] The matrix Q 2.1 The norm is defined as:

[0123]

[0124] Based on the above definition, the above target formula can be solved.

[0125] Compared with the clustering operation in the prior art in which feature selection and clustering have nothing to do with each other, the clustering target formula provided in the embodiment of the present disclosure integrates clustering and feature selection well. Increasing the orthogonality of the feature selection matrix can avoid falling into a trivial solution in the subsequent solution process. By dynamically selecting the number of discriminative features that actually participate in clustering through the row sparse matrix and the regularization parameter, clustering and feature selection can be better integrated to achieve flexible clustering, improve the accuracy and efficiency of clustering, and thereby improve the accuracy and efficiency of object information recognition.

[0126] The weighted iterative solution method provided in the embodiment of the present disclosure solves the target formula of clustering by weighted iterative solution until W, m j and Y approach convergence, which can further improve the accuracy of the object recognition method.

[0127] In order to more comprehensively highlight the effectiveness and advantages of the object recognition method provided by the above embodiment of the present disclosure, the data clustering method provided by the embodiment of the present disclosure is used to perform recognition experiments on four standard data sets, namely COIL20, COIL100, ORL face data set and YALE face data set. The information of the object is shown in Table 1:

[0128] Table 1

[0129] Object Number of samples Number of features category COIL20 1440 1024 20 COIL100 7200 1024 100 ORL 400 1024 40 YALE 165 1024 15

[0130] Three methods in the prior art are selected for clustering comparison with the clustering method provided in the embodiment of the present disclosure. Method 1: (Baseline): Use all features for k-means clustering; Method 2 (Max-var): The larger the variance in a certain direction in the data, the stronger the expressive power of the feature. Therefore, the corresponding features are selected according to the variance of the data for k-means clustering; Method 3 (Laplacian Score): Use the classic Laplacian Score feature selection algorithm for feature selection, and then input the selected features into k-menas for clustering. Clustering accuracy is used as the evaluation result of this experiment. Clustering accuracy is obtained by evaluating the accuracy of the final cluster center, feature selection matrix and confidence value solved by the clustering algorithm. The clustering results are shown in Table 2 (clustering results of each method when the number of features is 50) and Table 3 (clustering results of each method when the number of features is 100):

[0131] Table 2

[0132] Dataset Baseline Max-var Laplacian Score The disclosed method COIL20 0.52 0.43 0.49 0.54 COIL100 0.47 0.03 0.29 0.49 ORL 0.46 0.37 0.38 0.53 YALE 0.37 0.34 0.38 0.45

[0133] Table 3

[0134] Dataset Baseline Max-var Laplacian Score The disclosed method COIL20 0.52 0.49 0.51 0.55 COIL100 0.47 0.03 0.31 0.49 ORL 0.46 0.40 0.44 0.54 YALE 0.37 0.35 0.38 0.46

[0135] It can be clearly seen from the result data provided in Table 2 and Table 3 that the data clustering method provided by the embodiment of the present disclosure has the highest clustering accuracy compared to the three methods in the prior art, thereby improving the accuracy of object information recognition.

[0136] Figure 7 The block diagram of the object recognition device according to the embodiment of the present disclosure is schematically shown.

[0137] like Figure 7 As shown, the object recognition device 700 may include, for example, an acquisition module 710 , a selection module 720 , a clustering module 730 , a first determination module 740 and a second determination module 750 .

[0138] The acquisition module 710 is used to acquire multiple objects to be identified, each object including multiple features.

[0139] The selection module 720 is used to select multiple features of each object respectively using a feature selection matrix to obtain multiple objects including at least one feature.

[0140] The clustering module 730 is used to cluster multiple objects including at least one feature to obtain a clustering result.

[0141] A first determination module 740 is configured to identify information included in the plurality of objects according to the clustering result when it is determined that the clustering result satisfies a preset condition; and

[0142] The second determination module 750 is used to update the feature selection matrix according to the clustering result when it is determined that the clustering result does not meet the preset condition, and return to the operation of selecting multiple features of each object respectively using the feature selection matrix.

[0143] The following is a Figure 7 The object recognition device 700 shown is further described.

[0144] Figure 8 The block diagram of a selection module according to an embodiment of the present disclosure is schematically shown.

[0145] According to the embodiments of the present disclosure, Figure 8 As shown, the selection module 720 includes:

[0146] The extraction unit 721 is used to extract features of different dimensions of each object respectively.

[0147] The composition unit 722 is used to combine the features of different dimensions of each object into a feature vector of each object.

[0148] The first selection unit 723 is used to select the feature vector of each object respectively by using the feature selection matrix to obtain the feature vector of each object including at least one feature.

[0149] Fig. 9 A block diagram of a selection module according to another embodiment of the present disclosure is schematically shown.

[0150] According to the embodiments of the present disclosure, Fig. 9 As shown, the selection module 720 also includes:

[0151] A construction unit 724, used to construct a row sparse matrix;

[0152] The second selection unit 725 is used to select multiple features of each object respectively through the row sparse matrix to obtain multiple objects including at least one feature.

[0153] According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits, or at least part of the functions of any one of them can be implemented in one module. According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits can be split into multiple modules for implementation. According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits can be at least partially implemented as hardware circuits, such as field programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems on chips, systems on substrates, systems on packages, application specific integrated circuits (ASICs), or can be implemented by hardware or firmware in any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware, and firmware, or in any appropriate combination of any of them. Alternatively, according to the embodiments of the present invention, one or more of the modules, submodules, units, and subunits can be at least partially implemented as computer program modules, and when the computer program modules are run, the corresponding functions can be performed.

[0154] For example, any multiple of the acquisition module 710, the selection module 720, the clustering module 730, the first determination module 740 and the second determination module 750 can be combined in one module / unit / subunit for implementation, or any one of the modules / units / subunits can be split into multiple modules / units / subunits. Alternatively, at least part of the functions of one or more modules / units / subunits in these modules / units / subunits can be combined with at least part of the functions of other modules / units / subunits and implemented in one module / unit / subunit. According to an embodiment of the present disclosure, at least one of the acquisition module 710, the selection module 720, the clustering module 730, the first determination module 740 and the second determination module 750 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or by any one of the three implementation methods of software, hardware and firmware or by a suitable combination of any of them. Alternatively, at least one of the acquisition module 710 , the selection module 720 , the clustering module 730 , the first determination module 740 , and the second determination module 750 may be at least partially implemented as a computer program module, and when the computer program module is executed, a corresponding function may be performed.

[0155] It should be noted that the object recognition device part in the embodiment of the present disclosure corresponds to the object recognition method part in the embodiment of the present disclosure, and their specific implementation details and technical effects are also the same, which will not be repeated here.

[0156] Fig.10 A block diagram of an electronic device suitable for implementing the method described above according to an embodiment of the present disclosure is schematically shown. Fig.10 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0157] like Fig.10As shown, the electronic device 800 according to an embodiment of the present disclosure includes a processor 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage part 808 into a random access memory (RAM) 803. The processor 801 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 801 may also include an onboard memory for caching purposes. The processor 801 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0158] In RAM 803, various programs and data required for the operation of electronic device 800 are stored. Processor 801, ROM 802 and RAM 803 are connected to each other via bus 804. Processor 801 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 802 and / or RAM 803. It should be noted that the program can also be stored in one or more memories other than ROM 802 and RAM 803. Processor 801 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in the one or more memories.

[0159] According to an embodiment of the present disclosure, the electronic device 800 may further include an input / output (I / O) interface 805, which is also connected to the bus 804. The electronic device 800 may further include one or more of the following components connected to the I / O interface 805: an input portion 806 including a keyboard, a mouse, etc.; an output portion 807 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 808 including a hard disk, etc.; and a communication portion 809 including a network interface card such as a LAN card, a modem, etc. The communication portion 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as needed. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 810 as needed, so that a computer program read therefrom is installed into the storage portion 808 as needed.

[0160] According to an embodiment of the present disclosure, the method flow according to an embodiment of the present disclosure can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 809, and / or installed from the removable medium 811. When the computer program is executed by the processor 801, the above-mentioned functions defined in the system of the embodiment of the present disclosure are executed. According to an embodiment of the present disclosure, the system, equipment, device, module, unit, etc. described above can be implemented by a computer program module.

[0161] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0162] According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium. For example, it may include, but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium that includes or stores a program that may be used by or in combination with an instruction execution system, apparatus, or device.

[0163] For example, according to an embodiment of the present disclosure, the computer-readable storage medium may include the ROM 802 and / or the RAM 803 described above and / or one or more memories other than the ROM 802 and the RAM 803 .

[0164] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code includes one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart, can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0165] It will be appreciated by those skilled in the art that the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways, even if such combinations and / or combinations are not explicitly described in the present disclosure. In particular, the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways without departing from the spirit and teachings of the present disclosure. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0166] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above separately, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. The scope of the present disclosure is defined by the attached claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A method for object recognition, include: Acquire a plurality of objects to be identified, each of the objects comprising a plurality of features, and the objects comprising at least one of text, image, voice and video; Using a feature selection matrix to select the multiple features of each object respectively, to obtain multiple objects including at least one feature; Based on the cluster center, the orthogonal matrix of the feature selection matrix and the confidence that the object belongs to the cluster center, clustering the multiple objects including at least one feature to obtain a clustering result, wherein the feature selection matrix is ​​used to dynamically select the number of discriminative features actually participating in clustering; In the case where it is determined that the clustering result satisfies a preset condition, identifying information included in the plurality of objects in units of different classes according to the clustering result; and In the case where it is determined that the clustering result does not satisfy the preset condition, updating the feature selection matrix according to the clustering result, and returning to the operation of selecting the multiple features of each object respectively using the feature selection matrix; Wherein, clustering the multiple objects including at least one feature based on the cluster center, the orthogonal matrix of the feature selection matrix and the confidence that the object belongs to the cluster center, and obtaining the clustering result includes: Fix the values ​​of any two variables among the cluster center, the feature selection matrix and the confidence, calculate the value of another variable, and obtain a set of values ​​including the above three variables; Multiple groups of values ​​including the above three variables are calculated in an iterative manner to obtain the convergence values ​​of the cluster center, the feature selection matrix and the confidence level.

2. The method according to claim 1, in, The feature selection matrix is ​​used to select the multiple features of each object respectively to obtain multiple objects including at least one feature, including: Extract features of different dimensions of each object separately; Combining the features of each object in different dimensions into a feature vector of each object; The feature selection matrix is ​​used to select the feature vector of each object respectively, so as to obtain the feature vector of each object including at least one feature.

3. The method according to claim 1 or 2, in, The feature selection matrix is ​​used to select the multiple features of each object respectively to obtain multiple objects including at least one feature, including: Construct a row sparse matrix; The multiple features of each object are selected respectively through the row sparse matrix to obtain multiple objects including at least one feature.

4. The method according to claim 1, in, When it is determined that the clustering result does not meet the preset condition, updating the feature selection matrix according to the clustering result includes: When it is determined that the clustering accuracy included in the clustering result belongs to a first preset range, updating the feature selection matrix to select different types of features; When it is determined that the clustering accuracy included in the clustering result belongs to a second preset range, the feature selection matrix is ​​updated to select a different number of features.

5. The method according to claim 1, in, The images include a high-dimensional image dataset.

6. The method according to claim 5, in, The high-dimensional image dataset includes the COIL20 dataset, the COIL100 dataset, the ORL face dataset, or the YALE face dataset.

7. The method according to claim 1, in, The object includes an image; and extracting features of different dimensions of each object respectively includes: Extract at least one of the following features of the image: Gray value; 2D histogram; Scale-invariant feature transformation value; Histogram of Oriented Gradients.

8. An object recognition device, include: An acquisition module, used for acquiring a plurality of objects to be identified, each of the objects comprising a plurality of features, and the objects comprising at least one of text, image, voice and video; A selection module, configured to select the plurality of features of each object respectively using a feature selection matrix to obtain a plurality of objects including at least one feature; A clustering module, configured to cluster the plurality of objects including at least one feature based on cluster centers, an orthogonal matrix of the feature selection matrix, and a confidence that the object belongs to the cluster center, to obtain a clustering result, wherein the feature selection matrix is ​​used to dynamically select the number of discriminative features actually participating in clustering; A first determining module is configured to identify information included in the plurality of objects in units of different classes according to the clustering result when it is determined that the clustering result satisfies a preset condition; and A second determination module is used to update the feature selection matrix according to the clustering result when it is determined that the clustering result does not meet the preset condition, and return to the operation of selecting the multiple features of each object respectively using the feature selection matrix; Wherein, the clustering module includes: A first calculation unit is used to fix the values ​​of any two variables among the cluster center, the orthogonal matrix of the feature selection matrix and the confidence, calculate the value of another variable, and obtain a set of values ​​including the above three variables; The second calculation unit is used to iteratively calculate multiple groups of values ​​including the above three variables to obtain the convergence value of the cluster center, the feature selection matrix and the confidence level.

9. The device according to claim 8, in, The selection module comprises: An extraction unit, used to extract features of different dimensions of each object respectively; A composition unit, used for combining the features of each object in different dimensions into a feature vector of each object; The first selection unit is used to select the feature vector of each object respectively by using the feature selection matrix to obtain the feature vector of each object including at least one feature.

10. The device according to claim 8 or 9, in, The selection module also includes: Construction unit, used to construct row sparse matrix; The second selection unit is used to select multiple features of each object respectively through the row sparse matrix to obtain multiple objects including at least one feature.

11. The device according to claim 8, in, The second determining module comprises: A first determining unit is used to update the feature selection matrix to select different types of features when it is determined that the clustering accuracy included in the clustering result belongs to a first preset range; The second determining unit is used to update the feature selection matrix to select a different number of features when it is determined that the clustering accuracy included in the clustering result belongs to a second preset range.

12. An electronic device, include: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

13. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enables the processor to implement the method according to any one of claims 1 to 7.

14. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images