CN113239247B - Multidimensional data searching method, system and storage medium based on brain function partition - Google Patents
Multidimensional data searching method, system and storage medium based on brain function partition Download PDFInfo
- Publication number
- CN113239247B CN113239247B CN202110781954.0A CN202110781954A CN113239247B CN 113239247 B CN113239247 B CN 113239247B CN 202110781954 A CN202110781954 A CN 202110781954A CN 113239247 B CN113239247 B CN 113239247B
- Authority
- CN
- China
- Prior art keywords
- file
- characteristic
- feature
- information
- partitions
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/903—Querying
- G06F16/90335—Query processing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/906—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a multidimensional data searching method, a multidimensional data searching system and a storage medium based on brain function partition. Classifying different files, and acquiring a multi-dimensional characteristic value of each file according to the type of the file; respectively storing the characteristic values into corresponding brain function partitions according to the data types and the characterization items of the characteristic values; fusing the characteristic values of the files of different types by adopting a CMAC neural network to form fused data with a characteristic mapping relation, and directionally storing the fused data to a characterization database; and carrying out multi-dimensional feature depth fusion analysis and feature relation information analysis on input data, forming mapping connection by adopting a CMAC neural network, and carrying out feature matching search in the characterization database based on the mapping connection by adopting an ART network. According to the invention, mass data can be analyzed and processed quickly and accurately, so that the target file can be locked quickly and accurately.
Description
Technical Field
The invention relates to the field of brain-like computing, in particular to a method, a system and a storage medium for searching multidimensional data based on brain function partition.
Background
The brain-like calculation means that the rule of brain information processing is simulated in the aspects of hardware or software algorithm and the like, so that the calculation energy consumption is reduced, and the calculation efficiency and capability are improved. At present, the method for realizing the brain-like parallel computing technology comprises a software algorithm and a neuromorphic device integration method. In the brain-like computing process, different brain function partitions have different functions, thereby handling different things. One neuron in a brain functional partition usually corresponds to thousands of neurons, so that the neuron information among the brain functional partitions has a complex multidimensional mapping relation. In the current stage of big data searching process, most of the big data searching process is an artificial neural network, and the result is locked only by a single-dimensional mapping relation between information from the perspective of space complexity description, but the result is not locked by multi-dimensional searching from multiple features. In the current society with large explosion of information, single-dimension search is not enough to accurately lock result files.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide a method, a system and a storage medium for searching multidimensional data based on brain function partition, which can satisfy multidimensional parallel search and have real-time and accuracy.
The technical scheme adopted by the invention for solving the technical problems is as follows: a multidimensional data searching method based on brain function partition is constructed, and the method comprises the following steps:
s1, classifying different files, and acquiring a multi-dimensional characteristic value of each file according to the type of the file;
s2, storing the characteristic values into corresponding brain function partitions according to the data types and the characterization items of the characteristic values;
s3, fusing the characteristic values of the files of different types by adopting a CMAC neural network to form fused data with a characteristic mapping relation, and directionally storing the fused data into a characterization database;
and S4, performing multidimensional feature depth fusion analysis and feature relation information analysis on the input data by adopting a perception depth network, forming mapping connection by adopting a CMAC neural network, and performing feature matching search in the characterization database based on the mapping connection by adopting an ART network.
In the multidimensional data searching method based on brain function partition according to the present invention, step S1 further includes: and according to the file type, carrying out category splitting on each file to obtain different file information, and respectively adopting different deep neural perception network models to carry out feature extraction on the file information according to the category of the file information.
In the multidimensional data searching method based on brain function partition, the step S1 further includes the following steps:
s11, recognizing the video file into video image information, voice character information and character information;
s12, extracting static characteristics and motion characteristics from the video image information by adopting frame extraction, wherein the static characteristics comprise color characteristics, shape characteristics and texture characteristics, and the motion characteristics comprise object motion characteristics and shooting device motion characteristics;
s13, extracting the voice character features of the voice character information by adopting perceptual linear prediction based on a perceptual neural network;
and S14, extracting the keywords in the text information by using the keywords.
In the multidimensional data searching method based on brain function partition, the step S2 further includes the following steps:
s21, constructing 66 brain function partitions based on a cognitive function structure of a human brain, wherein the 66 brain function partitions comprise 64 attribute feature partitions and 2 relation feature partitions, and the relation feature partitions store association characterization information among the attribute feature partitions;
and S22, according to the data type and the representation item of each characteristic value, respectively and correspondingly storing each characteristic value in each attribute characteristic partition according to the attribute of each brain function partition.
In the multidimensional data searching method based on brain function partition, the step S3 further includes the following steps:
s31, carrying out concept mapping by utilizing a CMAC model to traverse all the characteristic values of each file to obtain preliminary characteristic value association information of the file;
s32, carrying out logic analysis on the file to obtain a logic characteristic value of the file, and correcting the preliminary characteristic value association information by adopting the logic characteristic value to obtain a characteristic vector corresponding relation of the file;
and S33, storing the characteristic values into corresponding attribute characteristic partitions in different compression modes according to the characteristic items of the characteristic values, and storing the corresponding relation of the characteristic vectors into corresponding relation characteristic partitions.
In the multidimensional data searching method based on brain function partition, in step S31, each feature value is input to a spatial neighboring point of the CMAC model, and a CAMC model is used for generalization to obtain the preliminary feature value association information of the file.
In the multidimensional data searching method based on brain function partition, the step S32 further includes the following steps:
s321, putting all file information of the file into the same training deep perception network for training to obtain a logic characteristic value of the file;
s322, correcting the preliminary characteristic value correlation information by adopting the logic characteristic value, and performing global characteristic fusion through global attention;
s333, analyzing the file through global attention multiple depth nerve feature connection training and recording the feature vector corresponding relation of the file;
and S334, establishing a corresponding characterization database and storing the corresponding relation of the depth fusion feature vectors obtained by training into the relation feature partition in a coding mode.
In the method for searching multidimensional data based on brain function partition of the present invention, in step S334, the characterization database is a relational function library.
In order to solve the technical problems, the invention adopts a further technical scheme that a multi-dimensional data search system based on brain function partition is constructed, and comprises
The classification extraction module is used for classifying different files and acquiring a multi-dimensional characteristic value of each file according to the type of the file;
the storage module is used for respectively storing the characteristic values into corresponding brain function partitions according to the data types and the characterization items of the characteristic values;
the fusion module adopts a CMAC neural network to fuse the characteristic values of the files of different types to form fusion data with a characteristic mapping relation and directionally stores the fusion data to a characterization database;
and the retrieval module is used for performing multi-dimensional feature depth fusion analysis and feature relation information analysis on input data by adopting a perception depth network, forming mapping connection by adopting a CMAC neural network, and performing feature matching search in the characterization database based on the mapping connection by adopting an ART network.
The other technical scheme adopted by the invention for solving the technical problem is as follows: a storage medium is constructed on which a computer program is stored which, when being executed by a processor, implements the brain function partition-based multi-dimensional data search method.
By implementing the multi-dimensional data searching method, the system and the storage medium based on the brain function partition, the characteristic values of various types of information are fused on the basis of quantum computation through multi-dimensional characteristic search based on brain-like computation, so that massive data can be analyzed and processed quickly and accurately, and a target file can be locked quickly and accurately. The data features are perceived through the deep neural network, the features can be continuously fused through continuous reinforcement learning, and the purpose of continuous learning is achieved.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a flow chart of a first preferred embodiment of a brain function partition-based multi-dimensional data search method of the present invention;
FIG. 2 shows a schematic representation of information characterization of brain functional partitions;
FIG. 3 shows a schematic diagram of a CMAC in accordance with a preferred embodiment of the present invention;
fig. 4 is a schematic block diagram of a first preferred embodiment of the brain function partition-based multi-dimensional data search system of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The invention relates to a multidimensional data searching method based on brain function partition, which comprises the following steps: classifying different files, and acquiring a multi-dimensional characteristic value of each file according to the type of the file; respectively storing the characteristic values into corresponding brain function partitions according to the data types and the characterization items of the characteristic values; fusing the characteristic values of the files of different types by adopting a CMAC neural network to form fused data with a characteristic mapping relation, and directionally storing the fused data to a characterization database; and performing multidimensional characteristic depth fusion analysis and characteristic relation information analysis on input data by adopting a perception depth network, forming mapping connection by adopting a CMAC neural network, and performing characteristic matching search in the characterization database by adopting an ART network based on the mapping connection. In the invention, the characteristics are stored by brain function partition, the ART (Adaptive Resonance Theory) network is used for characteristic learning, and the CMAC network is used for analysis and operation, so that the multi-dimensional parallel constraint is realized, the search time of multi-characteristic and large-capacity data is effectively reduced, and the search success rate is improved.
Fig. 1 is a flowchart of a first preferred embodiment of the brain function partition-based multidimensional data searching method of the present invention. As shown in fig. 1, in step S1, different files are classified, and a multidimensional feature value of each file is obtained according to the type of the file. In a preferred embodiment of the present invention, each file may be subjected to class splitting according to file types to obtain different file information, and different deep neural sensing network models are respectively adopted to perform feature extraction on the file information according to the classes of the file information.
For example, the files to be input into the database to be retrieved can be classified from coarse to fine through points, lines and surfaces. For example, the files may include video files, voice files, and text files. For a video file, the video file can be divided into video image information, voice text information and text information. For a voice file, it can be split into voice text information, and for a text file, it can be split into text information. Frame extraction may be employed to extract static and motion features from the video image information. The static features include color features, shape features, and texture features, and the motion features include object motion features and camera motion features. The speech character features of the speech character information can be extracted by adopting perceptual linear prediction based on a perceptual neural network, and keywords in the character information are extracted by adopting the keywords.
In step S2, the feature values are stored in the corresponding brain function partitions according to the data types and the characterization items of the feature values. In the preferred embodiment of the invention, based on the cognitive function structure of the human brain, the external information is subjected to basic classification of different disciplines according to a discipline classification table, brain function partition is performed according to the attributes of the classification result, and different forms of description methods are adopted to characterize the characteristics of different attributes in the brain function partition. And respectively storing the characteristic values in corresponding brain function partitions according to the characterization items of the characteristic values. And coding the characteristic values in the brain function partitions according to the data types of the characteristic values, wherein different data types correspond to different codes. The encoding result is stored by using a quantum potential energy model, such as a one-dimensional infinite deep potential well storage model, so that exponential fast access can be supported.
In a further preferred embodiment of the present invention, 66 brain function partitions may be constructed based on the cognitive function structure of the human brain, the 66 brain function partitions include 64 attribute feature partitions and 2 relationship feature partitions, and the relationship feature partitions store the association characterization information between the attribute feature partitions. And then, according to the data type and the representation item of each characteristic value, correspondingly storing each characteristic value in each attribute characteristic partition according to the attribute of each brain function partition. And the use process of the static learning machine is strengthened, and the limitation of the previous static learning is changed. The learned new features are continuously updated to strengthen learning by continuously loading the model, so that the features stored in the brain partitions tend to be more reasonable.
Fig. 2 shows a schematic representation of information characterization of brain functional partitions. Under the human brain cognitive architecture, the collected information is classified and characterized, and the result is coded, so that the storage of information with different dimensions is completed. The basic function operation of the brain is performed in different brain areas, the different brain areas manage different kinds of information, and the middle parts are connected through a complex neuron system, so that the neural networks in different areas can perform information interaction rapidly and have serial and parallel calculation modes. In the memory storage process, the association between the information is established, and the information is formed by associating the information with the same dimension (such as different characteristics in picture information) and associating the information with different dimensions (such as picture and character information characteristics). In the present application, in order to further simulate the brain storage mode, in addition to establishing the feature value storage databases of different types of files (i.e. the aforementioned 64 attribute feature partitions), an association characterization library (i.e. the aforementioned 2 relationship feature partitions) between each piece of information is separately established to quickly search and match the corresponding feature association items and lock the related information.
In step S3, the feature values of the different types of files are fused by using a CMAC neural network to form fused data with feature mapping relationships, and the fused data is directionally stored in a characterization database. Preferably, the characterization function library is a relational function library. Figure 3 shows a schematic diagram of a CMAC according to a preferred embodiment of the present invention.
In a preferred embodiment of the invention, this step comprises in particular the following operations. First, a CMAC model is used for conceptual mapping to traverse all the characteristic values of each file to obtain preliminary characteristic value association information of the files. In order to completely cover all feature values, for example, each feature value may be input into a spatially neighboring point of the CMAC model, and a CAMC model is used for generalization to obtain the preliminary feature value association information of the file. As known to those skilled in the art, in the CAMC model, partial overlapping units in spatially adjacent points are excited, and points at a far distance are overlapped lower, so that preliminary characteristic value association information of the file is obtained through local generalization.
And then, carrying out logic analysis on the file to obtain a logic characteristic value of the file, and correcting the preliminary characteristic value association information by adopting the logic characteristic value to obtain a characteristic vector corresponding relation of the file. In the preferred embodiment of the invention, all the file information of the file is put into the same training deep perception network for training to obtain the logic characteristic value of the file; correcting the preliminary characteristic value correlation information by adopting the logic characteristic value, and performing global characteristic fusion by global attention; analyzing the file and recording the corresponding relation of the feature vectors of the file through global attention multiple depth nerve feature connection training; and establishing a corresponding characterization database and storing the corresponding relation of the depth fusion feature vectors obtained by training into the relation feature partition in a coding mode, wherein the relation feature partition can support exponential-level quick access. As known to those skilled in the art, CMAC employs the same learning algorithm as adaptive linear neurons.
And finally, storing the characteristic values into corresponding attribute characteristic partitions in different compression modes according to the characterization items of the characteristic values, and storing the corresponding relation of the characteristic vectors into corresponding relation characteristic partitions. Here, the feature values may be stored in the 64 brain function partitions according to the known characterization item classification of the feature values by using the combined action of the multidimensional information correlation mapper of the CMAC model and the brain cognitive architecture information correlation mapper, and the feature vector correspondence between each type of information is established and stored in the corresponding relationship feature partition classified by the brain cognitive structure in a coded form.
Preferably, after the concept mapping is completed, the actual mapping is realized by a compressed storage space technology. In the prior art, a spurious coding technology is commonly used, a residue division method is used in a spurious storage technology, multiple consideration needs to be given in the analysis process due to overlapping and conflict of CMAC models, if the input is an n-dimensional space, each dimension file information contains q quantization levels, and single output is realized
，
However, the stray storage is prone to conflict, i.e. multiple associative units are mapped to the same unit (here, the aforementioned 64 brain function partitions) of the actual mapping at the same time, resulting in information loss. In order to prevent this problem, as described above, 66 brain function partitions are provided in the present application, where the 66 brain function partitions include 64 attribute feature partitions and 2 relationship feature partitions, and the relationship feature partitions store association characterization information between the attribute feature partitions. Thus, the feature vector correspondence may be stored to 2 relational feature partitions, while the feature values may be stored to 64 attribute feature partitions. Different encoding modes and data compression modes can be adopted because different brain function partitions are adopted to store different types of data, and data cannot be lost.
Preferably, different compression algorithms are adopted for the specific data compression modes according to different attribute characteristics of the data of the representation types. The method mainly comprises text data compression, image data compression, audio data compression, video data compression algorithms and the like, wherein a lossless compression algorithm is adopted, and a lossy compression algorithm is not adopted. Optionally, the text compression data may adopt algorithms such as run length coding and arithmetic coding; the Image data adopts algorithms such as TIFF (tagged Image File Format), PNG (Portable Network Graphic Format), GIF (graphics exchange Format) and the like; the Audio data adopts algorithms such as FLAC (Fee Lossless Audio codec) format, TAK (Tom's Audio Kompresor) format and the like; the video data adopts H.264 and other algorithms. And respectively carrying out directional compression storage on the corresponding relation between the characteristic value and the characteristic vector through a specific algorithm. In addition, in the process of storing the characteristic values according to the brain cognitive architecture, in order to ensure that data loss is not caused when the CMAC is actually mapped, the characteristic values in each storage medium can be correlated through logic analysis in the storage process, and initial information is stored.
In step S4, a perceptual depth network is used to perform multidimensional feature depth fusion analysis and feature relation information analysis on the input data, a CMAC neural network is used to form a mapping connection, and an ART network is used to perform feature matching search in the characterization database based on the mapping connection. In the preferred embodiment of the invention, firstly, input data is determined, then, on the basis of quantum computation, multidimensional characteristic deep fusion analysis and characteristic relation information analysis are carried out on the input data through a CMAC network, multidimensional characteristic value association analysis is established, learning and new mapping connection is established, then, the data quantity and the iteration times of each group of characteristic data and fusion data are used as independent variables, and ART3 is utilized to carry out multiple matching search in combination with constraint conditions under different search situations. It should be noted that the ART3 model simulates the application of a neuron system, and the inter-synaptic signals propagate through media in the information propagation process, while the ART3 model simulates the media propagation, and the media in this case are weights. Different situations and constraint conditions correspond to certain weight, and the corresponding search channel is started through the weight so as to be fused with the CMAC model. ART3 is characterized by adding distributed operation and selective (full compression) operation functions in addition to positive feedback, normalization and nonlinear operation) so that different identification codes can be established among the blocks to support fast locking and hierarchical searching in 66 partitions.
As will be appreciated by those skilled in the ART, ART3 is a simple, powerful mechanism for searching learned pattern recognition codes in parallel in order to meet the computational requirements of ART model embedding into a network hierarchy. The search mechanism includes a reset code attribute that has at least three different functions: category selection to correct errors, learning from reinforcement feedback, and response modification. The search mechanism of ART3 avoids this asymmetry to some extent and is therefore more suitable for parallel search computations.
In the invention, file information is analyzed from a multi-dimensional angle, characteristic values are extracted, then a database is divided based on a human brain cognitive structure, and the multi-dimensional characteristic values are classified and stored in the database (brain function partition). When files of different types are stored, characteristic values of the files are extracted by using different models according to file information types of the files, different information is correlated, the correlation information is stored in a specific storage space to serve as search logic, correlation entanglement is generated among the characteristic values, multi-dimensional information search is facilitated, then the CMAC model is used for analyzing and traversing the multi-dimensional characteristic values, and a correlation degree association effect is generated by using local generalization characteristics of the multi-dimensional characteristic values. The generation of the association effect needs to be carried out under the overall logic constraint of the file, so that the association relation among all characteristic values is corrected and strengthened. The generalized information and the feature association information may be encoded and stored in a targeted manner to a specific storage unit, thereby constructing a feature database for retrieval. When input data needs to be retrieved, multidimensional feature depth fusion analysis and feature relation information analysis are carried out on the input data, a CMAC neural network is adopted to form mapping connection, and ART network is adopted to carry out feature matching search in the characterization database based on the mapping connection. The ART3 model is more suitable for parallel search after the asymmetry in the search module is solved. The matching result can strengthen the identification content, and the weight value is updated after the iteration to the dynamic neural network.
The following description will be made of a specific embodiment of the present application, taking a video file search as an example. Firstly, for information extraction of a video file, the video file can be firstly split into video image information, voice character information and character information. The characteristic values of the video image information, the voice character information and the character information are respectively extracted, corresponding attribute characteristic partitions are stored, and the link relation among the video image information, the voice character information and the character information at different moments can be independently proposed and stored in 2 relation characteristic partitions. In the process of extracting the link relation, the CMAC neural network is adopted to fuse the characteristic values of the files of different types to form fused data with characteristic mapping relation, and the fused data is directionally stored in a characterization database. When the CMAC is adopted to obtain the fusion data, for example, for video image information, the key frame can be extracted on the basis of the motion characteristics, and the condition of frame missing can be effectively avoided. In addition, different characteristic values in the picture can be associated except for the motion characteristics, so that picture information can be accurately locked. And finally, storing the fusion characteristic value and the associated information in a coding form. Taking an action sequence as an example, when a drinking action is carried out, the combined characteristic of hand action characteristics and a cup needs to be locked at the same time, when the combined characteristic appears in a specific range (for example, hand movement and the cup must appear in a head specified range), the action can be determined to be finished, a series of actions are cut out and marked based on a time sequence means, the sequence actions are stored in a specified database in a characteristic vector coding mode, and a speech characteristic value is extracted according to an automatic speech recognition technology (ASR) or a caption text information characteristic is extracted by utilizing a keyword extraction technology and stored. Therefore, when searching for a video file, a whole set of video image information, voice character information and character information connected in series can be obtained according to one video image information, voice character information or character information.
By implementing the multidimensional data searching method based on brain function partition, the characteristic values of various types of information are fused on the basis of quantum computation through multidimensional characteristic search based on brain-like computation, so that massive data can be analyzed and processed quickly and accurately, and a target file can be locked quickly and accurately.
Fig. 4 is a schematic block diagram of a first preferred embodiment of the brain function partition-based multi-dimensional data search system of the present invention. As shown in fig. 4, the multidimensional data search system based on brain function partition comprises a classification extraction module 100, a storage module 200, a fusion module 300, and a retrieval module 400. The classification extraction module 100 is configured to classify different files, and obtain a multi-dimensional feature value of each file according to a type of the file. The storage module 200 is configured to store the feature values into corresponding brain function partitions according to the data types and the characterization items of the feature values. The fusion module 300 is configured to fuse the feature values of different types of files by using a CMAC neural network to form fused data with a feature mapping relationship, and store the fused data in a representation database in a targeted manner. The retrieval module 400 is configured to perform multidimensional feature depth fusion analysis and feature relation information analysis on input data by using a perceptual depth network, form a mapping connection by using a CMAC neural network, and perform feature matching search in the characterization database by using an ART network based on the mapping connection.
Those skilled in the art will appreciate that the above-described classification extraction module 100, storage module 200, fusion module 300, and retrieval module 400 may be constructed in accordance with the embodiments shown in fig. 1-3. Based on the construction of the present invention, those skilled in the art can implement the above-mentioned multidimensional data search system based on brain function partition, and will not be described in detail herein.
Accordingly, the present invention can be realized in hardware, software, or a combination of hardware and software. The present invention can be realized in a centralized fashion in at least one computer system, or in a distributed fashion where different elements are spread across several interconnected computer systems. Any kind of computer system or other apparatus adapted for carrying out the methods of the present invention is suited. A typical combination of hardware and software could be a general purpose computer system with a computer program that, when being loaded and executed, controls the computer system such that it carries out the methods described herein.
The invention also relates to a storage medium on which a computer program is stored, which program contains all the features enabling the brain function partition-based multi-dimensional data search method of the invention. When installed in a computer system, may implement the methods of the present invention. The computer program in this document refers to: any expression, in any programming language, code or notation, of a set of instructions intended to cause a system having an information processing capability to perform a particular function either directly or after either or both of the following: a) conversion to other languages, codes or symbols; b) reproduced in a different format.
By implementing the multi-dimensional data searching method, the system and the storage medium based on the brain function partition, the characteristic values of various types of information are fused on the basis of quantum computation through multi-dimensional characteristic search based on brain-like computation, so that massive data can be analyzed and processed quickly and accurately, and a target file can be locked quickly and accurately.
While the invention has been described with reference to specific embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted without departing from the scope of the invention. In addition, many modifications may be made to adapt a particular situation or material to the teachings of the invention without departing from its scope. Therefore, it is intended that the invention not be limited to the particular embodiment disclosed, but that the invention will include all embodiments falling within the scope of the appended claims.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (8)
1. A multidimensional data searching method based on brain function partition is characterized by comprising the following steps:
s1, classifying different files, and acquiring a multi-dimensional characteristic value of each file according to the type of the file;
s2, storing the characteristic values into corresponding brain function partitions according to the data types and the characterization items of the characteristic values;
s3, fusing the characteristic values of the files of different types by adopting a CMAC neural network to form fused data with a characteristic mapping relation, and directionally storing the fused data into a characterization database;
s4, performing multidimensional feature depth fusion analysis and feature relation information analysis on input data by adopting a perception depth network, forming mapping connection by adopting a CMAC neural network, and performing feature matching search in the characterization database by adopting an ART network based on the mapping connection;
the step S2 further includes the steps of:
s21, constructing 66 brain function partitions based on a cognitive function structure of a human brain, wherein the 66 brain function partitions comprise 64 attribute feature partitions and 2 relation feature partitions, and the relation feature partitions store association characterization information among the attribute feature partitions;
s22, respectively and correspondingly storing each characteristic value in each attribute characteristic partition according to the data type and the representation item of each characteristic value and the attribute of each brain function partition;
the step S3 further includes the steps of:
s31, carrying out concept mapping by utilizing a CMAC model to traverse all the characteristic values of each file to obtain preliminary characteristic value association information of the file;
s32, carrying out logic analysis on the file to obtain a logic characteristic value of the file, and correcting the preliminary characteristic value association information by adopting the logic characteristic value to obtain a characteristic vector corresponding relation of the file;
and S33, storing the characteristic values into corresponding attribute characteristic partitions in different compression modes according to the characteristic items of the characteristic values, and storing the corresponding relation of the characteristic vectors into corresponding relation characteristic partitions.
2. The method for multi-dimensional data search based on brain function partition according to claim 1, wherein step S1 further comprises: and according to the file type, carrying out category splitting on each file to obtain different file information, and respectively adopting different deep neural perception network models to carry out feature extraction on the file information according to the category of the file information.
3. The method for multi-dimensional data search based on brain function partition according to claim 2, wherein the step S1 further comprises the steps of:
s11, recognizing the video file into video image information, voice character information and character information;
s12, extracting static characteristics and motion characteristics from the video image information by adopting frame extraction, wherein the static characteristics comprise color characteristics, shape characteristics and texture characteristics, and the motion characteristics comprise object motion characteristics and shooting device motion characteristics;
s13, extracting the voice character features of the voice character information by adopting perceptual linear prediction based on a perceptual neural network;
and S14, extracting the keywords in the text information by using the keywords.
4. The method for multi-dimensional data search based on brain function partition according to claim 1, wherein in step S31, each feature value is inputted into a spatial neighboring point of the CMAC model, and is generalized by using a CAMC model to obtain the preliminary feature value association information of the file.
5. The method for multi-dimensional data search based on brain function partition according to claim 4, wherein the step S32 further comprises the steps of:
s321, putting all file information of the file into the same training deep perception network for training to obtain a logic characteristic value of the file;
s322, correcting the preliminary characteristic value correlation information by adopting the logic characteristic value, and performing global characteristic fusion through global attention;
s333, analyzing the file through global attention multiple depth nerve feature connection training and recording the feature vector corresponding relation of the file;
and S334, establishing a corresponding characterization database and storing the corresponding relation of the depth fusion feature vectors obtained by training into the relation feature partition in a coding mode.
6. The method for multi-dimensional data search based on brain function partition according to claim 5, wherein in step S334, said characterization database is a relational function library.
7. A multi-dimensional data search system based on brain function partition is characterized by comprising
The classification extraction module is used for classifying different files and acquiring a multi-dimensional characteristic value of each file according to the type of the file;
the storage module is used for respectively storing the characteristic values into corresponding brain function partitions according to the data types and the characterization items of the characteristic values;
the fusion module adopts a CMAC neural network to fuse the characteristic values of the files of different types to form fusion data with a characteristic mapping relation and directionally stores the fusion data to a characterization database;
the retrieval module is used for carrying out multi-dimensional feature depth fusion analysis and feature relation information analysis on input data by adopting a perception depth network, forming mapping connection by adopting a CMAC neural network, and carrying out feature matching search in the characterization database by adopting an ART network based on the mapping connection;
the storage module is further used for constructing 66 brain function partitions based on a cognitive function structure of a human brain, wherein the 66 brain function partitions comprise 64 attribute feature partitions and 2 relation feature partitions, and the relation feature partitions store association characterization information among the attribute feature partitions; according to the data type and the representation item of each characteristic value, respectively and correspondingly storing each characteristic value in each attribute characteristic partition according to the attribute of each brain function partition;
the fusion module is further used for carrying out concept mapping by utilizing a CMAC model so as to traverse all the characteristic values of each file to obtain preliminary characteristic value association information of the file; carrying out logic analysis on the file to obtain a logic characteristic value of the file, and correcting the preliminary characteristic value association information by adopting the logic characteristic value to obtain a characteristic vector corresponding relation of the file; and storing the characteristic values to corresponding attribute characteristic partitions in different compression modes according to the characterization items of the characteristic values, and storing the corresponding relation of the characteristic vectors to corresponding relation characteristic partitions.
8. A storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a brain function partition-based multi-dimensional data search method according to any one of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110781954.0A CN113239247B (en) | 2021-07-12 | 2021-07-12 | Multidimensional data searching method, system and storage medium based on brain function partition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110781954.0A CN113239247B (en) | 2021-07-12 | 2021-07-12 | Multidimensional data searching method, system and storage medium based on brain function partition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113239247A CN113239247A (en) | 2021-08-10 |
| CN113239247B true CN113239247B (en) | 2021-10-26 |
Family
ID=77135389
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110781954.0A Active CN113239247B (en) | 2021-07-12 | 2021-07-12 | Multidimensional data searching method, system and storage medium based on brain function partition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113239247B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114048848B (en) * | 2022-01-13 | 2022-05-10 | 深圳市永达电子信息股份有限公司 | Brain-like computing method and system based on memory mechanism |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6532454B1 (en) * | 1998-09-24 | 2003-03-11 | Paul J. Werbos | Stable adaptive control using critic designs |
| CN101670886A (en) * | 2008-09-11 | 2010-03-17 | 上海海事大学 | Method and device for diagnosing information fusion fault of underwater robot propeller |
| CN104809498A (en) * | 2014-01-24 | 2015-07-29 | 清华大学 | Brain-like coprocessor based on neuromorphic circuit |
| CN111783320A (en) * | 2020-08-03 | 2020-10-16 | 中国人民解放军国防科技大学 | Brain processor-like NoC design space exploration method, system and medium for specified applications |
| CN112487263A (en) * | 2020-11-26 | 2021-03-12 | 杭州安恒信息技术股份有限公司 | Information processing method, system, equipment and computer readable storage medium |
| CN112819171A (en) * | 2021-02-04 | 2021-05-18 | 深圳市永达电子信息股份有限公司 | Data searching method and system based on table function and computer storage medium |
-
2021
- 2021-07-12 CN CN202110781954.0A patent/CN113239247B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6532454B1 (en) * | 1998-09-24 | 2003-03-11 | Paul J. Werbos | Stable adaptive control using critic designs |
| CN101670886A (en) * | 2008-09-11 | 2010-03-17 | 上海海事大学 | Method and device for diagnosing information fusion fault of underwater robot propeller |
| CN104809498A (en) * | 2014-01-24 | 2015-07-29 | 清华大学 | Brain-like coprocessor based on neuromorphic circuit |
| CN111783320A (en) * | 2020-08-03 | 2020-10-16 | 中国人民解放军国防科技大学 | Brain processor-like NoC design space exploration method, system and medium for specified applications |
| CN112487263A (en) * | 2020-11-26 | 2021-03-12 | 杭州安恒信息技术股份有限公司 | Information processing method, system, equipment and computer readable storage medium |
| CN112819171A (en) * | 2021-02-04 | 2021-05-18 | 深圳市永达电子信息股份有限公司 | Data searching method and system based on table function and computer storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113239247A (en) | 2021-08-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10783451B2 (en) | Ensemble machine learning for structured and unstructured data | |
| Shi et al. | A novel multi-branch channel expansion network for garbage image classification | |
| KR101488010B1 (en) | Associative matrix methods, systems and computer program products using bit plane representations of selected segments | |
| CN111127385A (en) | Medical information cross-modal Hash coding learning method based on generative countermeasure network | |
| US11645328B2 (en) | 3D-aware image search | |
| Chen et al. | New ideas and trends in deep multimodal content understanding: A review | |
| CN113159067A (en) | Fine-grained image identification method and device based on multi-grained local feature soft association aggregation | |
| CN115222998B (en) | Image classification method | |
| CN113821670A (en) | Image retrieval method, device, equipment and computer readable storage medium | |
| CN112905862A (en) | Data processing method and device based on table function and computer storage medium | |
| CN115909036A (en) | Local-global adaptive guide enhanced vehicle weight identification method and system | |
| CN113239247B (en) | Multidimensional data searching method, system and storage medium based on brain function partition | |
| Jiang et al. | An intelligent recommendation approach for online advertising based on hybrid deep neural network and parallel computing | |
| CN114547307A (en) | Text vector model training method, text matching method, device and equipment | |
| Arul | Deep learning methods for data classification | |
| CN113240033A (en) | Visual relation detection method and device based on scene graph high-order semantic structure | |
| CN115797795B (en) | Remote sensing image question-answer type retrieval system and method based on reinforcement learning | |
| Bibi et al. | Deep features optimization based on a transfer learning, genetic algorithm, and extreme learning machine for robust content-based image retrieval | |
| Guo | Rough set-based approach to data mining | |
| Lin et al. | High-order structure preserving graph neural network for few-shot learning | |
| CN114170460A (en) | Multi-mode fusion-based artwork classification method and system | |
| CN114707483A (en) | Zero sample event extraction system and method based on contrast learning and data enhancement | |
| CN113157892A (en) | User intention processing method and device, computer equipment and storage medium | |
| CN112364193A (en) | Image retrieval-oriented method for fusing multilayer characteristic deep neural network model | |
| Hu et al. | Expressive local feature match for image search |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |