An information fusion method and system

Abstract

The application provides an information fusion method and system, which comprises the following steps: receiving a plurality of target information sent by a plurality of intelligent systems in real time, and generating an adjacency relation matrix corresponding to the plurality of intelligent systems based on preset rules according to the plurality of target information in real time, wherein each intelligent system corresponds to a target information; filtering out a target matrix parameter corresponding to the adjacency relation matrix in real time, and generating a corresponding information matrix according to the target matrix parameter; performing feature extraction processing on the information matrix to generate a corresponding target feature vector in real time, and calculating an information weight corresponding to each target information according to the target feature vector; and performing convolution fusion processing according to the target feature vector and the information weight to generate a target information vector corresponding to the plurality of target information. The application can avoid complex calculation and data, and improve work efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to an information fusion method and system. BACKGROUND

[0002] With the progress of science and technology and the development of the times, computer technology and Internet technology have been popularized in people's daily life, and have been deeply applied in many fields, which has facilitated people's life.

[0003] Among them, the existing technology has developed intelligent systems with different functions according to existing computers and the Internet, and in the existing multi-agent system, the information fusion of heterogeneous agents is a key process, which can combine the information from multiple different agents together to obtain more accurate and comprehensive scene understanding and decision-making.

[0004] Further, in the process of fusing information of different agents, most of the existing technology completes data processing through existing Bayesian theory or deep learning network, however, although this processing method can ultimately realize data fusion, it has high computational complexity and will produce high-dimensional complex data, which has certain use limitation and reduces work efficiency. SUMMARY

[0005] Therefore, the purpose of the present application is to provide an information fusion method and system to solve the problem of high computational complexity and high-dimensional complex data in the process of data fusion in the prior art, which reduces work efficiency.

[0006] The first aspect of the embodiment of the present application provides:

[0007] An information fusion method, wherein the method comprises:

[0008] real-time receiving a plurality of target information sent by a plurality of intelligent systems in real time, and generating an adjacency relationship matrix corresponding to the plurality of intelligent systems in real time based on a plurality of target information according to a preset rule, wherein each intelligent system corresponds to a target information;

[0009] real-time filtering the corresponding target matrix parameters in the adjacency relationship matrix, and generating a corresponding information matrix in real time according to the target matrix parameters;

[0010] performing feature extraction processing on the information matrix to generate a corresponding target feature vector in real time, and calculating an information weight corresponding to each target information respectively in real time according to the target feature vector;

[0011] Convolution fusion processing is performed according to the target feature vector and the information weight, so as to generate a target information vector corresponding to the target information in real time.

[0012] The present application has the advantages that: by receiving the target information sent by the intelligent system in real time, the corresponding processing object can be determined, based on which, the current target information can be quantitatively processed according to the pre-set rules, and the corresponding adjacency relationship matrix can be generated, based on which, the current adjacency relationship matrix can be processed again, and the target feature vector used for subsequent fusion can be extracted in real time, based on which, only the convolution fusion of the target feature vector and the information weight obtained finally is needed, so that the target information vector required can be obtained, thereby the corresponding fusion processing can be completed, and the fusion of data can be quickly and effectively completed, avoiding complex calculation and data, and improving the work efficiency.

[0013] Further, the step of generating an adjacency relationship matrix corresponding to the intelligent system in real time based on the target information according to the pre-set rules comprises:

[0014] When the target information is obtained in real time, the target information is input into a pre-set feature extraction network, so that the feature extraction network outputs an initial correlation vector in real time;

[0015] The initial correlation vector is matrixed to generate an adjacency relationship matrix corresponding to the intelligent system in real time.

[0016] Further, the step of matrixing the initial correlation vector to generate an adjacency relationship matrix corresponding to the intelligent system in real time comprises:

[0017] When the initial correlation vector is obtained in real time, the initial correlation vector is input into a pre-set softmax layer, and a probability matrix is output in real time through the pre-set softmax layer;

[0018] The probability matrix is input into a pre-set onehot encoding layer, and the probability matrix is encoded in real time through the pre-set onehot encoding layer to output a target relationship code in real time;

[0019] The target relationship code is converted in real time to generate an adjacency relationship matrix corresponding to the intelligent system in real time.

[0020] Further, the step of outputting a probability matrix in real time through the pre-set softmax layer comprises:

[0021] When the initial correlation vector is acquired in real time, the corresponding full connection network is called in real time in the preset softmax layer;

[0022] The full connection network is scanned in whole to detect in real time a plurality of original network nodes contained in the full connection network, and each initial correlation vector is filled in the original network node to output the probability matrix in real time.

[0023] Further, the expression of the algorithm for real-time encoding processing of the probability matrix through the preset onehot encoding layer is:

[0024]

[0025] Wherein, g ij represents the target relationship encoding, p i represents the horizontal matrix parameter in the probability matrix, h i represents the vertical matrix parameter in the probability matrix, tau represents the temperature coefficient, and N represents the number of target relationship encodings.

[0026] Further, the step of calculating the information weight corresponding to each target information in real time according to the target feature vector comprises:

[0027] When the target feature vector is acquired in real time, the corresponding preset algorithm is called in real time in the preset database;

[0028] Each target feature vector is input into the preset algorithm in real time, so that the preset algorithm outputs the information weight corresponding to each target information in real time.

[0029] Further, the expression of the preset algorithm is:

[0030]

[0031] Wherein, w j (Q, I) represents the information weight, K represents the number of target feature vectors, QI j T represents the target feature vector, and D1 represents the size of the target feature vector.

[0032] The second aspect of the embodiment of the application provides:

[0033] An information fusion system, wherein the system comprises:

[0034] The receiving module is configured to receive a plurality of target information transmitted by a plurality of intelligent systems in real time, and generate a neighbor relationship matrix corresponding to the plurality of intelligent systems based on preset rules and the plurality of target information in real time, wherein each of the intelligent systems corresponds to one of the target information;

[0035] The screening module is configured to screen a target matrix parameter corresponding to the neighbor relationship matrix in real time, and generate a corresponding information matrix based on the target matrix parameter;

[0036] The computing module is configured to perform feature extraction processing on the information matrix to generate a target feature vector corresponding to the information matrix in real time, and calculate an information weight corresponding to each of the target information based on the target feature vector;

[0037] The fusion module is configured to perform convolution fusion processing based on the target feature vector and the information weight to generate a target information vector corresponding to the plurality of target information in real time.

[0038] Further, the receiving module is specifically configured to:

[0039] When the plurality of target information is acquired in real time, the plurality of target information is input into a preset feature extraction network, so that the feature extraction network outputs an initial correlation vector corresponding to the plurality of target information in real time;

[0040] The initial correlation vector is matrixed to generate a neighbor relationship matrix corresponding to the plurality of intelligent systems in real time.

[0041] Further, the receiving module is specifically configured to:

[0042] When the initial correlation vector is acquired in real time, the initial correlation vector is input into a preset softmax layer, and a probability matrix is output by the preset softmax layer in real time;

[0043] The probability matrix is input into a preset onehot encoding layer, and the probability matrix is encoded by the preset onehot encoding layer in real time to output a target relationship code in real time;

[0044] The target relationship code is converted in real time to generate a neighbor relationship matrix corresponding to the plurality of intelligent systems.

[0045] Further, the receiving module is specifically configured to:

[0046] When the initial correlation vector is acquired in real time, a full connection network corresponding to the preset softmax layer is called in real time;

[0047] The full connection network is scanned in whole to detect a plurality of original network nodes contained in the full connection network in real time, and each initial correlation vector is filled into the inside of each original network node to output the probability matrix in real time.

[0048] Further, the expression of the algorithm for real-time encoding processing of the probability matrix through the preset onehot encoding layer is:

[0049]

[0050] Wherein, g ij represents the target relationship encoding, p i represents a horizontal matrix parameter in the probability matrix, h i represents a vertical matrix parameter in the probability matrix, tau represents a temperature coefficient, and N represents the number of target relationship encodings.

[0051] Further, the computing module is specifically used for:

[0052] When the target feature vector is acquired in real time, the corresponding preset algorithm is called in real time in the preset database;

[0053] Each target feature vector is input into the inside of the preset algorithm in real time, so that the preset algorithm outputs information weights corresponding to each target information in real time.

[0054] Further, the expression of the preset algorithm is:

[0055]

[0056] Wherein, w j (Q, I) represents the information weight, K represents the number of target feature vectors, QI j T represents the target feature vector, and D1 represents the size of the target feature vector.

[0057] The third aspect of the embodiment of the application provides:

[0058] A computer comprises a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the information fusion method as described above when executing the computer program.

[0059] The fourth aspect of the embodiment of the application provides:

[0060] A readable storage medium has a computer program stored thereon, wherein the program is executed by a processor to implement the information fusion method as described above.

[0061] Additional aspects and advantages of the present application will be given in part in the following description, become apparent from the following description, or be learned by the practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 a flow chart of the information fusion method provided by the first embodiment of the present application;

[0063] Figure 2 a structural block diagram of the information fusion system provided by the third embodiment of the present application.

[0064] The following detailed description will further describe the present application with reference to the above mentioned drawings. DETAILED DESCRIPTION

[0065] In order to facilitate the understanding of the present application, the following will make a more comprehensive description of the present application with reference to the relevant drawings. The drawings show several embodiments of the present application. However, the present application can be realized in many different forms, and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0066] It should be noted that when an element is referred to as being "fixed to" another element, it can be directly on the other element or there can be a middle element. When an element is referred to as being "connected" to another element, it can be directly connected to the other element or there can be a middle element. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0067] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terminology used in the description of the present application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0068] Referring to Figure 1 , the information fusion method provided by the first embodiment of the present application is shown. The information fusion method provided by the present embodiment can avoid complex calculations and data, thereby quickly and effectively completing the fusion of each target information, and improving work efficiency.

[0069] Specifically, the present embodiment provides:

[0070] An information fusion method, specifically comprising the following steps:

[0071] Step S10, receiving a plurality of target information sent by a plurality of intelligent systems in real time, and generating an adjacency relationship matrix corresponding to the plurality of intelligent systems based on a preset rule according to the plurality of target information in real time, wherein each intelligent system corresponds to one target information;

[0072] Step S20, filtering out a corresponding target matrix parameter in the adjacency relationship matrix in real time, and generating a corresponding information matrix according to the target matrix parameter;

[0073] Step S30, performing feature extraction processing on the information matrix to generate a corresponding target feature vector in real time, and calculating an information weight corresponding to each target information according to the target feature vector;

[0074] Step S40, performing convolution fusion processing according to the target feature vector and the information weight to generate a target information vector corresponding to the plurality of target information.

[0075] Specifically, in this embodiment, first of all, it needs to be pointed out that in order to quickly and effectively complete the fusion processing of data, it is necessary to analyze the real-time received information at this time, wherein it needs to be pointed out that in the actual application process, the information fusion method provided by the present application is implemented based on the server set in the background, and specifically, the server can establish a communication connection with each existing intelligent system, so that the server provided by the present application can receive a plurality of target information sent by each intelligent system in real time, wherein it needs to be pointed out that each existing intelligent system will only send one target information each time. Based on this, in order to facilitate the fusion processing of the current target information, the present application will immediately call out the pre-set information processing rule in the background, and can quantitatively process the current plurality of target information through the preset rule, and can generate a corresponding adjacency relationship matrix, that is, a matrix that can indicate the relationship between the current target information. Based on this, the present application will analyze the current adjacency relationship matrix in real time for subsequent processing.

[0076] Further, after the required adjacency relationship matrix is acquired in real time through the above steps, it should be noted that the internal adjacency relationship matrix contains a large number of matrix parameters, but not every matrix parameter needs to be fused subsequently. Based on this, the application will perform corresponding screening processing on the internal adjacency relationship matrix, and can real-time screen the corresponding target matrix parameter in the current adjacency relationship matrix. At the same time, the corresponding information matrix can be created in real time according to the target matrix parameter. Based on this, the current information matrix will be extracted again. The feature extraction processing can be performed, and the corresponding target feature vector can be generated synchronously. On this basis, the current target feature vector is finally calculated, and the information weight corresponding to each target information can be finally calculated. Based on this, the current information weight can be used as the fusion basis to perform convolution fusion processing on the current target feature vector and information weight in real time, so that the required target information vector can be finally fused, thereby avoiding complex calculation and data, and improving the work efficiency.

[0077] Second embodiment

[0078] Further, the step of generating an adjacency relationship matrix corresponding to the plurality of intelligent systems in real time based on the plurality of target information according to the preset rule comprises:

[0079] When the plurality of target information is acquired in real time, the plurality of target information is input into a preset feature extraction network, so that the feature extraction network outputs a corresponding initial correlation vector in real time;

[0080] The initial correlation vector is matrixed to generate an adjacency relationship matrix corresponding to the plurality of intelligent systems in real time.

[0081] Further, the step of matrixing the initial correlation vector to generate an adjacency relationship matrix corresponding to the plurality of intelligent systems in real time comprises:

[0082] When the initial correlation vector is acquired in real time, the initial correlation vector is input into a preset softmax layer, and a probability matrix is output in real time through the preset softmax layer;

[0083] The probability matrix is input into a preset onehot encoding layer, and the probability matrix is encoded in real time through the preset onehot encoding layer to output a target relationship code in real time;

[0084] The target relationship code is converted in real time to generate an adjacency relationship matrix corresponding to the plurality of intelligent systems in real time.

[0085] Further, the step of outputting the corresponding probability matrix in real time through the preset softmax layer comprises:

[0086] When the initial correlation vector is acquired in real time, the corresponding fully connected network is called in real time in the preset softmax layer;

[0087] The fully connected network is scanned in whole to detect in real time a plurality of original network nodes contained in the fully connected network, and each initial correlation vector is filled in each original network node to output the probability matrix in real time.

[0088] Further, the algorithm for encoding the probability matrix in real time through the preset onehot encoding layer has an expression as follows:

[0089]

[0090] Wherein, g ij represents the target relation encoding, p i represents a horizontal matrix parameter in the probability matrix, h i represents a vertical matrix parameter in the probability matrix, τ represents a temperature coefficient, and N represents the number of target relation encodings.

[0091] Further, the step of calculating the information weight corresponding to each target information in real time according to the target feature vector comprises:

[0092] When the target feature vector is acquired in real time, a preset algorithm is called in real time in a preset database;

[0093] Each target feature vector is input in real time into the preset algorithm to make the preset algorithm output the information weight corresponding to each target information in real time.

[0094] Further, the preset algorithm has an expression as follows:

[0095]

[0096] Wherein, w j (Q, I) represents the information weight, K represents the number of target feature vectors, QI j T represents the target feature vector, and D1 represents the size of the target feature vector.

[0097] In addition, in the embodiment, it is also necessary to explain that, in order to facilitate understanding, it is assumed that 𝑂 𝑖 is the information observed by the intelligent agent i, and 𝑚 jOj (j = 1, 2,.., N, j≠i) are the observation information of other agents. The Oi and mj are spliced and input into the feature extraction network, and the output is the relevance vector e 𝑖𝑗 of agent i and other agent information j , which represents the closeness of the relationship between the information m i， and agent i itself. Its process can be represented as a function, as shown in the formula:

[0098] (1) f(O j m ij )→e i1 i2 i3 iN Since the number of agents is N, the final output is (e i1 , e i2 , e i3 ,...e iN ). Finally, the output (e i1 , e i2 , e i3 ,...e iN ) of the feature extraction network is input into the softmax, and finally a neighbor relationship matrix G=[g ij , g ij , g i ,...g 𝑗 ] with only 0 and 1 is output through onehot encoding. If g i1 is 0, it means that agent i refuses to receive the information of agent j; if g i2 is 1, it means that agent i receives the information of agent j.

[0100] In order to convert the relevance vector into the onehot form of the neighbor matrix G. Suppose the dimension of the local observation O i3 is D1, and the dimension of the other agent information m iN is D2, then the dimension of the relevance vector (e ij , e i , e i ,...e ij ) is N×(D1+D2), and the linear transformation matrix of the full connection network is (D1+D2)×1, so the dimension of the relevance vector after the full connection network is N×1. After the Softmax function, a probability matrix P can be obtained, and the dimension of P is N×1, and then according to onehot encoding as follows:

[0101]

[0102] Where g i represents the target relationship encoding, and p 𝑗denotes a horizontal matrix parameter in the probability matrix, h i denotes a vertical matrix parameter in the probability matrix, τ denotes a temperature coefficient, and N denotes a number of the target relational encodings, g ij is close to onehot encoding, after onehot encoding, a final Nx1 relational encoding G=[g i1 , g i2 , g i3 ,...g iN ] is obtained, so as to facilitate subsequent processing.

[0103] According to the relational encoding matrix G=[g i1 , g i2 , g i3 ,...g iN ] obtained above, information of an agent corresponding to 1 in G will be information that needs to be received by the agent i, assuming that the agent i receives information of K agents in total, an information matrix is denoted as M, and the dimension of M is KxD2.The local observation information of the agent i is obtained through feature extraction to obtain a 1xD1 vector Q, the information matrix M is obtained through feature extraction to obtain a KxD1 vector I, and the information matrix M is obtained through feature extraction to obtain a vector V.

[0104] According to the vector Q and the vector I, the weight of each agent information is calculated, and the formula is as follows:

[0105]

[0106] wherein g ij denotes the target relational encoding, p i denotes a horizontal matrix parameter in the probability matrix, h i denotes a vertical matrix parameter in the probability matrix, τ denotes a temperature coefficient, and N denotes a number of the target relational encodings.

[0107] A vector W=[w1m1,w2m2,...w K m K ] is obtained, and finally, the W and the vector V are convolved to obtain a 1xD3 information vector Mesg.

[0108] Referring to Figure 2 , the third embodiment of the present application provides:

[0109] An information fusion system, wherein the system comprises:

[0110] The receiving module is configured to receive a plurality of target information transmitted by a plurality of intelligent systems in real time, and generate a neighbor relationship matrix corresponding to the plurality of intelligent systems based on preset rules and the plurality of target information in real time, wherein each intelligent system corresponds to one target information;

[0111] The screening module is configured to screen a target matrix parameter corresponding to the neighbor relationship matrix in real time, and generate a corresponding information matrix based on the target matrix parameter;

[0112] The computing module is configured to perform feature extraction processing on the information matrix to generate a target feature vector corresponding to the information matrix in real time, and calculate an information weight corresponding to each target information based on the target feature vector;

[0113] The fusion module is configured to perform convolution fusion processing on the target feature vector and the information weight to generate a target information vector corresponding to the plurality of target information in real time.

[0114] Further, the receiving module is specifically configured to:

[0115] When the plurality of target information is acquired in real time, the plurality of target information is input into a preset feature extraction network, so that the feature extraction network outputs an initial correlation vector corresponding to the plurality of target information in real time;

[0116] The initial correlation vector is matrixed to generate a neighbor relationship matrix corresponding to the plurality of intelligent systems in real time.

[0117] Further, the receiving module is specifically configured to:

[0118] When the initial correlation vector is acquired in real time, the initial correlation vector is input into a preset softmax layer, and a probability matrix is output by the preset softmax layer in real time;

[0119] The probability matrix is input into a preset onehot encoding layer, and the probability matrix is encoded by the preset onehot encoding layer in real time to output a target relationship code in real time;

[0120] The target relationship code is converted in real time to generate a neighbor relationship matrix corresponding to the plurality of intelligent systems.

[0121] Further, the receiving module is specifically configured to:

[0122] When the initial correlation vector is acquired in real time, a full connection network corresponding to the preset softmax layer is called in real time;

[0123] The full connection network is scanned in whole to detect a plurality of original network nodes contained in the full connection network in real time, and each initial correlation vector is filled into the inside of each original network node to output the probability matrix in real time.

[0124] Further, the expression of the algorithm for real-time encoding processing of the probability matrix through the preset onehot encoding layer is:

[0125]

[0126] Wherein, g ij represents the target relationship encoding, p i represents the horizontal matrix parameter in the probability matrix, h i represents the vertical matrix parameter in the probability matrix, tau represents the temperature coefficient, and N represents the number of target relationship encodings.

[0127] Further, the computing module is specifically used for:

[0128] When the target feature vector is acquired in real time, the corresponding preset algorithm is called in real time in the preset database;

[0129] Each target feature vector is input into the inside of the preset algorithm in real time, so that the preset algorithm outputs information weights corresponding to each target information in real time.

[0130] Further, the expression of the preset algorithm is:

[0131]

[0132] Wherein, w j (Q, I) represents the information weight, K represents the number of target feature vectors, QI j T represents the target feature vector, and D1 represents the size of the target feature vector.

[0133] The fourth embodiment of the application provides a computer, including a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the information fusion method as described above when executing the computer program.

[0134] The fifth embodiment of the application provides a readable storage medium, which stores a computer program, wherein the program is executed by a processor to implement the information fusion method as described above.

[0135] In summary, the information fusion method and system provided by the above embodiments of the present application can avoid complex calculation and data, thereby quickly and effectively completing fusion of each target information, and improving work efficiency.

[0136] It should be noted that each of the above modules can be a functional module or a program module, which can be implemented by software or hardware. For the modules implemented by hardware, each of the above modules can be located in the same processor, or each of the above modules can be located in different processors in any combination.

[0137] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a list of executable instructions for implementing logic functions, and can be embodied in any computer-readable medium for use by an instruction execution system, apparatus, or device, such as a computer-based system, a system including a processor, or other systems that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions, or in conjunction with these instruction execution systems, apparatus, or devices. For the present specification, the "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport programs for use by an instruction execution system, apparatus, or device, or in conjunction with these instruction execution systems, apparatus, or devices.

[0138] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection having one or more wires (electrical devices), a portable computer diskette (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium can even be paper or other suitable medium on which the program is printed, because the program can be electronically obtained, for example, by optical scanning of the paper or other medium, followed by electronic conversion into a form that can be edited, interpreted, or otherwise processed as appropriate, and stored in a computer memory.

[0139] It should be understood that parts of the present application can be realized in hardware, software, firmware, or a combination thereof. In the above-described embodiments, a plurality of steps or methods can be realized by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if realized by hardware, and as in another embodiment, any one or a combination of the following technologies known in the art can be used: discrete logic circuit with logic gates for implementing logic functions on data signals, application specific integrated circuit with suitable combination logic gates, programmable gate array (PGA), field programmable gate array (FPGA), etc.

[0140] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Also, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0141] The above-described embodiments only express several implementation manners of the present application, which are described in a more specific and detailed manner, but cannot be understood as a limitation on the scope of the present application. It should be noted that, for those of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, which are all within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. An information fusion method characterized by, The method comprises: Real-time receiving a plurality of target information sent by a plurality of intelligent systems in real time, and generating an adjacency relationship matrix corresponding to the plurality of intelligent systems based on a plurality of target information in real time according to a preset rule, wherein each intelligent system corresponds to a target information; In the adjacency relationship matrix, the corresponding target matrix parameters are screened out in real time, and the corresponding information matrix is generated in real time according to the target matrix parameters; The information matrix is subjected to feature extraction processing to generate a corresponding target feature vector in real time, and the information weight corresponding to each target information is calculated in real time according to the target feature vector; According to the target feature vector and the information weight, convolution fusion processing is performed to generate a target information vector corresponding to a plurality of target information in real time; The information fusion method is implemented based on a server arranged in the background, the server can establish a communication connection with each intelligent system, and the server can receive a plurality of target information sent by each intelligent system in real time; Assume O i The information observed by the agent i, m j The observation information of other agents, j = 1, 2.., N, j ≠ i; will O i And m j Concatenate and input the feature extraction network, and the output is the relevance vector e of the agent i and other agent information ij , which represents the closeness of the information m j to the agent i itself; its process can be represented as a function, as shown in the formula: f(O i， m j )→e ij Since the number of agents is N, the final output is (e i1 ,e i2 ,e i3 ,...e iN ); finally, the output of the feature extraction network (e i1 ,e i2 ,e i3 ,...e iN ) is input to softmax, and finally a 0, 1 adjacency matrix G=[g i1 ,g i2 ,g i3 ,...g iN ] is output through onehot encoding; if g ij is 0, it means that agent i refuses to receive information from agent j; if g ij is 1, it means that agent i receives information from agent j; In order to convert the relevance vector into an adjacent matrix G in onehot form; suppose the local observation O i has a dimension of D1, and the other agent information m j has a dimension of D2, then the relevance vector (e i1 ,e i2 ,e i3 ,...e iN ) has a dimension of N x (D1+D2), the linear transformation matrix of the full connection network has a dimension of (D1+D2) x 1, so the dimension of the relevance vector after passing through the full connection network is N x 1; after passing through the Softmax function, a probability matrix P can be obtained, P has a dimension of N x 1, and after onehot encoding, a final Nx1 relationship encoding G=[g i1 ,g i2 ,g i3 ,...g iN ] is obtained, so as to facilitate subsequent processing; According to the relationship encoding matrix G=[g i1 ,g i2 ,g i3 ,...g iN ] obtained above, the information of the agent corresponding to 1 in G will be needed by the agent i to receive, assuming that the agent i receives K agent information in total, the information matrix is denoted as M, the dimension of M is K*D2; the local observation information of the agent i is obtained through feature extraction to obtain a 1*D1 vector Q, the information matrix M is obtained through feature extraction to obtain a K*D1 vector I, and the information matrix M is obtained through feature extraction to obtain a vector V; According to vector Q and vector I, the weight of each intelligent body information is calculated, and the formula is as follows: wherein w c (Q, I) represents the information weight, K represents the number of target feature vectors, QI c T represents the target feature vector, D1 represents the size of the target feature vector; A vector W = [w1m1, w2m2,... w K m K ] is obtained, and finally W and the vector V are convolved to obtain a 1 x D3 information vector Mesg.

2. An information fusion system characterized by, The system for realizing the information fusion method of claim 1 comprises: A receiving module for real-time receiving a plurality of target information sent by a plurality of intelligent systems in real time, and generating an adjacency relationship matrix corresponding to the plurality of intelligent systems based on a plurality of target information in real time according to a preset rule, wherein each intelligent system corresponds to a target information; A screening module for screening out corresponding target matrix parameters in the adjacency relationship matrix in real time, and generating a corresponding information matrix in real time according to the target matrix parameters; A computing module for feature extraction processing of the information matrix to generate a corresponding target feature vector in real time, and calculating the information weight corresponding to each target information in real time according to the target feature vector; A fusion module for convolution fusion processing according to the target feature vector and the information weight to generate a target information vector corresponding to a plurality of target information in real time.

3. A computer comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to realize the information fusion method of claim 1.

4. A readable storage medium, having stored thereon a computer program, characterized in that, The program is executed by the processor to realize the information fusion method of claim 1.

Images