Intelligent evidence credibility evaluation method and system based on neural network

Abstract

The invention relates to an evidence credibility intelligent evaluation method and system based on a neural network, and the method comprises the steps: processing a text evidence to obtain a processed text, extracting time data, place data, character data and event data, constructing a text time node, a text place node, a text character node and a text event node, constructing a text node set, and carrying out the analysis of the text time node and the text place node. The method comprises the following steps: constructing a text relation graph, setting a dynamic acquisition frequency, performing image extraction on video evidence, obtaining an extracted image set, constructing a video time node, a video place node, a video character node, a video event node, a video graph edge set and a video relation graph, setting an evidence state, and calculating evidence similarity. If the evidence similarity is smaller than a similarity threshold value, the evidence state is a false state, and if the evidence similarity is not smaller than the similarity threshold value, the evidence state is a credible state. The evidence credibility evaluation efficiency can be improved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method and system for intelligently evaluating the credibility of evidence based on a neural network. Background Art

[0002] With the rapid development of artificial intelligence and data analysis technology, intelligent assessment of evidence credibility is widely used in judicial, news investigation, social security and other fields. In modern case analysis, text evidence and video evidence play a vital role. Text evidence records the key elements of an event, such as time, place, people and events, through structured language, while video evidence provides real images of the scene, which helps to intuitively show the course of the case.

[0003] Traditional evidence credibility assessment methods usually rely on manual judgment and empirical analysis, and cannot efficiently process massive amounts of evidence information. For text evidence, traditional methods mainly determine the reliability of evidence through keyword matching and manual reasoning, which not only relies on human bias, but also has difficulty in processing complex and diverse evidence. In terms of video evidence analysis, traditional methods mostly rely on simple image processing techniques, such as edge detection and target tracking, which cannot effectively extract and associate important information in the video, resulting in low accuracy and efficiency of evidence analysis. Therefore, how to improve the efficiency of evidence credibility assessment is an urgent problem to be solved. Summary of the invention

[0004] The present invention provides a method and system for intelligently evaluating the credibility of evidence based on a neural network, the main purpose of which is to improve the efficiency of evaluating the credibility of evidence.

[0005] To achieve the above object, the present invention provides a method for intelligently evaluating the credibility of evidence based on a neural network, comprising:

[0006] Acquire text evidence and video evidence, perform text preprocessing based on the text evidence to obtain processed text, perform structural analysis on the processed text, and perform data extraction on the processed text after structural analysis based on a pre-built entity recognition model to obtain time data, location data, person data, and event data;

[0007] Constructing text time nodes, text location nodes, text character nodes and text event nodes according to the time data, location data, character data and event data;

[0008] Constructing a text node set based on the text time node, text location node, text character node and text event node, and constructing a text relationship graph according to the text node set;

[0009] A dynamic acquisition frequency is set, and images are extracted from the video evidence based on the dynamic acquisition frequency and a pre-built convolutional neural network model to obtain an extracted image set, and a video time node, a video location node, a video character node, a video event node, and a video graph edge set are constructed based on the extracted image set, and a video relationship graph is constructed based on the video time node, the video location node, the video character node, the video event node, and the video graph edge set;

[0010] Calculating evidence similarity according to the text relationship graph and the video relationship graph, and judging the evidence status of the text evidence and the video evidence based on the evidence similarity, wherein the evidence status includes: a credible status and a false status;

[0011] Determine whether the evidence similarity is less than a preset similarity threshold;

[0012] If the evidence similarity is less than the similarity threshold, the evidence status is confirmed to be a false status;

[0013] If the evidence similarity is not less than the similarity threshold, the evidence status is confirmed to be credible.

[0014] Optionally, constructing a text relationship graph according to a text node set includes:

[0015] Defining a connection type based on the text evidence, wherein the connection types include: time connection, location connection, character connection and event connection, the time connection is used to connect a text time node with a text event node, the location connection is used to connect a text location node with a text event node, the character connection is used to connect a text character node with a text event node, and the event connection is used to connect different text event nodes;

[0016] Constructing a text graph edge set according to the connection type, time connection, location connection, person connection and event connection;

[0017] Construct a text relationship graph based on the text node set and the text graph edge set:

[0018] Q = (W, E)

[0019] Among them, Q refers to the text relationship graph, W refers to the text node set, and E refers to the text graph edge set.

[0020] Optionally, the setting of the dynamic acquisition frequency includes:

[0021] Set the collection time interval and calculate the initial collection frequency based on the collection time interval:

[0022]

[0023] Among them, R1 refers to the initial acquisition frequency, and t refers to the acquisition time interval;

[0024] Extracting images from the video evidence based on the initial acquisition frequency to obtain an initial image set, and calculating an image difference value based on the initial image set;

[0025] Set the highest difference value and the lowest difference value, and set the dynamic acquisition frequency based on the highest difference value, image difference value, lowest difference value and initial acquisition frequency:

[0026]

[0027] Among them, R2 refers to the dynamic acquisition frequency, T refers to the image difference value, T1 refers to the highest difference value, and T2 refers to the lowest difference value.

[0028] Optionally, calculating the image difference value based on the initial image set includes:

[0029] An initial image sequence is acquired based on the initial image set, wherein the initial image sequence includes a plurality of initial images, a first initial image is sequentially extracted from the initial image sequence, and the following operations are performed on the extracted first initial image:

[0030] Based on the extracted first initial image, identifying a second initial image in the initial image sequence, wherein the second initial image is an initial image adjacent to the first initial image and lagging behind the first initial image;

[0031] Performing pixel detection on the first initial image and the second initial image respectively to obtain a first pixel matrix and a second pixel matrix;

[0032] Set the pixel horizontal coordinate, pixel vertical coordinate and decomposition standard deviation, where the value of the decomposition standard deviation is 1 or 2;

[0033] Construct a decomposition function based on the pixel horizontal coordinate, pixel vertical coordinate and decomposition standard deviation:

[0034]

[0035] Wherein, A(x, y, θ) refers to the decomposition function when the pixel abscissa is x, the pixel ordinate is y, and the decomposition standard deviation is θ, x refers to the pixel abscissa, y refers to the pixel ordinate, θ refers to the decomposition standard deviation, exp refers to the exponential function, and π refers to pi;

[0036] Decomposing the first pixel matrix and the second pixel matrix based on the decomposition function and the decomposition standard deviation respectively to obtain a first decomposition pixel set and a second decomposition pixel set;

[0037] Calculating a first pixel difference set based on the first decomposed pixel set, and calculating a second pixel difference set based on the second decomposed pixel set, wherein the first pixel difference set consists of a first pixel difference and a second pixel difference;

[0038] Summarizing the first pixel difference set to obtain a first grayscale image, and summarizing the second pixel difference set to obtain a second grayscale image;

[0039] An image difference value is calculated according to the first grayscale image and the second grayscale image.

[0040] Optionally, decomposing the first pixel matrix and the second pixel matrix based on the decomposition function and the decomposition standard deviation to obtain a first decomposed pixel set and a second decomposed pixel set includes:

[0041] The decomposition standard deviation is set to 1 to obtain a first decomposition function, and the first pixel matrix and the second pixel matrix are convolved based on the first decomposition function to obtain first decomposition pixels and second decomposition pixels;

[0042] The decomposition standard deviation is set to 2 to obtain a second decomposition function, and the first pixel matrix and the second pixel matrix are respectively convolved based on the second decomposition function to obtain first convolution pixels and second convolution pixels;

[0043] The first decomposed pixels and the first convolution pixels are aggregated to obtain a first decomposed pixel set, and the second decomposed pixels and the second convolution pixels are aggregated to obtain a second decomposed pixel set.

[0044] Optionally, the aggregating the first pixel difference set to obtain a first grayscale image includes:

[0045] Extracting a first pixel difference and a second pixel difference based on the first pixel difference set, obtaining a maximum function, and calculating a grayscale pixel according to the first pixel difference, the second pixel difference, the maximum function, the pixel abscissa, and the pixel ordinate:

[0046] S(x,y)=ξ(d1(x,y),d2(x,y))

[0047] Wherein, S(x, y) refers to the grayscale pixel with the horizontal coordinate x and the vertical coordinate y, ξ refers to the maximum function, d1(x, y) refers to the first pixel difference with the horizontal coordinate x and the vertical coordinate y, and d2(x, y) refers to the second pixel difference with the horizontal coordinate x and the vertical coordinate y;

[0048] The first pixel difference set is summarized based on the grayscale pixels to obtain a first grayscale image.

[0049] Optionally, calculating the image difference value according to the first grayscale image and the second grayscale image includes:

[0050] Get the total number of pixels and calculate the image error value based on the pixel horizontal coordinate, pixel vertical coordinate, grayscale pixels and the total number of pixels:

[0051]

[0052] Wherein, F refers to the image error value, S2(x, y) refers to the grayscale pixel with the horizontal coordinate x and the vertical coordinate y in the second grayscale image, and N refers to the total number of pixels;

[0053] Calculate a first mean, a second mean, a first variance, a second variance and a comprehensive covariance based on the first grayscale image and the second grayscale image;

[0054] A first stable parameter and a second stable parameter are set, and image similarity is calculated based on the first mean, the second mean, the first variance, the second variance, the comprehensive covariance, the first stable parameter, and the second stable parameter:

[0055]

[0056] Among them, F2 refers to image similarity, δ1 refers to the first mean, δ2 refers to the second mean, σ1 refers to the first stability parameter, ε1 refers to the first variance, ε2 refers to the second variance, σ2 refers to the second stability parameter, and ε3 refers to the comprehensive covariance;

[0057] Set a difference weight parameter, and calculate the image difference value based on the image error value, image similarity and the difference weight parameter:

[0058] F3=α×F+(1-α)×F2

[0059] Among them, F3 refers to the image difference value, and α refers to the difference weight parameter.

[0060] Optionally, calculating the evidence similarity according to the text relationship graph and the video relationship graph includes:

[0061] Calculate the similarity of time nodes, and calculate the similarity of location nodes, person nodes and event nodes;

[0062] Transform the text graph edge set and the video graph edge set respectively to obtain a text edge vector and a video edge vector;

[0063] Set the edge vector index, and calculate the graph edge similarity based on the edge vector index, text edge vector, and video edge vector:

[0064]

[0065] Among them, k refers to the graph edge similarity, i refers to the edge vector index, Refers to the text edge vector when the edge vector index is i, μ i Refers to the video edge vector when the edge vector index is i, ||*|| refers to the modulo symbol, Point to the multiplication symbol;

[0066] Set the time node weight, location node weight, person node weight, event node weight and edge similarity weight, and calculate the evidence similarity based on the time node weight, location node weight, person node weight, event node weight, edge similarity weight, time node similarity, location node similarity, person node similarity, event node similarity and graph edge similarity:

[0067] P=α1×γ1+β2×γ2+β3×γ3+β4×γ4+β5×k

[0068] Among them, P refers to the evidence similarity, =1 refers to the time node weight, γ1 refers to the time node similarity, β2 refers to the location node weight, γ2 refers to the location node similarity, β3 refers to the person node weight, γ3 refers to the person node similarity, β4 refers to the event node weight, γ4 refers to the event node similarity, and β5 refers to the edge similarity weight.

[0069] Optionally, the calculating time node similarity includes:

[0070] Summarizing the text time nodes and the video time nodes to obtain a text time node set and a video time node set;

[0071] Calculate the time node similarity based on the text time node set and the video time node set:

[0072]

[0073] Among them, H1 refers to the text time node set, H2 refers to the video time node set, ∩ refers to the intersection symbol, and ∪ refers to the union symbol.

[0074] To achieve the above object, the present invention also provides an evidence credibility intelligent evaluation system based on a neural network, comprising:

[0075] A text relationship map construction module is used to obtain text evidence and video evidence, perform text preprocessing based on the text evidence to obtain processed text, perform structural analysis on the processed text and perform data extraction on the processed text after structural analysis based on a pre-built entity recognition model to obtain time data, location data, character data and event data; construct text time nodes, text location nodes, text character nodes and text event nodes based on the time data, location data, character data and event data; construct a text node set based on the text time nodes, text location nodes, text character nodes and text event nodes, and construct a text relationship map based on the text node set;

[0076] A video relationship graph construction module is used to set a dynamic acquisition frequency, perform image extraction on the video evidence based on the dynamic acquisition frequency and a pre-built convolutional neural network model to obtain an extracted image set, construct video time nodes, video location nodes, video character nodes, video event nodes and a video graph edge set based on the extracted image set, and construct a video relationship graph based on the video time nodes, video location nodes, video character nodes, video event nodes and the video graph edge set;

[0077] An evidence similarity calculation module is used to calculate the evidence similarity according to the text relationship map and the video relationship map, and judge the evidence status of the text evidence and the video evidence based on the evidence similarity, wherein the evidence status includes: a credible state and a false state;

[0078] The evidence status judgment module is used to determine whether the evidence similarity is less than a preset similarity threshold; if the evidence similarity is less than the similarity threshold, the evidence status is confirmed to be a false status; if the evidence similarity is not less than the similarity threshold, the evidence status is confirmed to be a credible status.

[0079] In order to solve the above problem, the present invention further provides an electronic device, the electronic device comprising:

[0080] A memory storing at least one instruction;

[0081] A processor executes instructions stored in the memory to implement the above-mentioned neural network-based evidence credibility intelligent assessment method.

[0082] In order to solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored. The at least one instruction is executed by a processor in an electronic device to implement the above-mentioned neural network-based evidence credibility intelligent assessment method.

[0083] The present invention solves the problems described in the background technology. First, the text evidence is preprocessed to obtain the processed text, the processed text is structurally analyzed, and data is extracted from the processed text after the structural analysis based on a pre-built entity recognition model. The text preprocessing process can effectively remove irrelevant information and interference, remove redundant spaces, punctuation marks, and stop words, and perform accurate word segmentation. Removing redundant spaces, punctuation marks, and stop words can effectively reduce the noise of the processed text and reduce the calculation amount of the entity recognition model. Accurate word segmentation is performed to convert the text evidence into independent words or phrases, which is conducive to the entity recognition model to recognize and can focus more on specific content, especially in multilingual or professional texts. Accurate word segmentation helps to identify important information without being disturbed by irrelevant symbols or words. Structural analysis can better understand the hierarchical structure and grammatical relationship of the text evidence, provide clear data for the entity recognition model, and improve the recognition accuracy and efficiency of the entity recognition model. Secondly, based on the extracted text time nodes, The text location node, the text character node and the text event node construct a text node set, and the text relationship map is constructed according to the text node set. The text relationship map can clearly represent all the data and the relationship between the data in the text evidence, can effectively organize and store information, and provide a clear node structure for the calculation of evidence similarity; then, the dynamic acquisition frequency is set, and the video evidence is extracted based on the dynamic acquisition frequency and the convolutional neural network model to obtain an extracted image set. The dynamic acquisition frequency can be adjusted according to the content of the video evidence to improve the acquisition efficiency, avoid the extraction of redundant images, and save computing resources; then, the video relationship map is constructed, and the video relationship map provides accurate relationship data for the subsequent calculation of evidence similarity, making the calculation of evidence similarity more efficient and accurate; finally, the evidence similarity is calculated and the evidence status is judged based on the evidence similarity. By automatically calculating the evidence similarity and judging the similarity threshold, the errors of human operation are reduced, and the accuracy and efficiency of evidence evaluation are improved. Therefore, the present invention can improve the efficiency of evidence credibility evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 A schematic diagram of a flow chart of a method for intelligently evaluating the credibility of evidence based on a neural network provided in one embodiment of the present invention;

[0085] Figure 2 A functional module diagram of a neural network-based evidence credibility intelligent evaluation system provided by an embodiment of the present invention;

[0086] Figure 3 A schematic diagram of the structure of an electronic device for implementing the neural network-based evidence credibility intelligent assessment method provided in one embodiment of the present invention.

[0087] Description of reference numerals:

[0088] 1. Electronic device; 10. Processor; 11. Memory; 12. Bus.

[0089] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0090] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0091] The embodiment of the present application provides a method for intelligently evaluating the credibility of evidence based on a neural network. The execution subject of the method for intelligently evaluating the credibility of evidence based on a neural network includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided in the embodiment of the present application. In other words, the method for intelligently evaluating the credibility of evidence based on a neural network can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0092] Embodiment 1:

[0093] Reference Figure 1 As shown, this embodiment provides a flowchart of a method for intelligently evaluating the credibility of evidence based on a neural network, and the method for intelligently evaluating the credibility of evidence based on a neural network includes:

[0094] S1. Obtain text evidence and video evidence, perform text preprocessing based on the text evidence to obtain processed text, perform structural analysis on the processed text, and perform data extraction on the processed text after structural analysis based on a pre-built entity recognition model to obtain time data, location data, person data, and event data.

[0095] Explainable. Text evidence refers to evidence presented in the form of text, for example, a news report mentions "On October 1, 2024, Company A in Shanghai announced the launch of a new product production line. Video evidence refers to evidence presented in the form of video, for example, a video recorded by a bank's surveillance equipment. Text preprocessing of text evidence refers to word segmentation, stop word removal, and syntactic analysis of text evidence. Word segmentation refers to breaking down text evidence into independent words or phrases for easy analysis. Stop word removal refers to removing words that do not contribute to semantics, for example , remove "is", "of", etc. Syntactic parsing refers to identifying the grammatical structure of text evidence and parsing the subject, predicate and object components. For example, the text evidence is: On December 12, 2024, Xiao Zhang planned a meeting in London and discussed the project budget with Xiao Zhao. Perform text preprocessing on the text evidence to obtain the processed text: On December 12, 2024, Xiao Zhang, London, planned a meeting, Xiao Zhao, discussed the budget. Processed text refers to the text obtained after text preprocessing of text evidence. For example, the processed text is: On December 12, 2024, Xiao Zhang, London, planned a meeting , Xiao Zhao, discussing the budget. Structural analysis of the processed text refers to dividing the processed text according to four dimensions: time, place, person, and event. For example, the processed text is: December 12, 2024, Xiao Zhang, London, planning meeting, Xiao Zhao, discussing the budget. Structural analysis of the processed text results in: time: December 12, 2024, place: London, person: Xiao Zhang, Xiao Zhao, event: planning meeting, discussing the budget, and the entity recognition model refers to BERT-Bi-LSTM-CRF, which is an entity recognition model that combines BERT, Bi-LSTM, and CRF. The entity recognition model is used to perform data recognition on the processed text and extract time data, place data, person data, and event data. Time data refers to data used to represent time in text evidence, for example, December 12, 2024. Place data refers to data used to represent place in text evidence, for example, London. Person data refers to data used to represent person in text evidence, for example, Xiao Zhang, Xiao Zhao. Event data refers to data used to represent specific events in text evidence, for example, planning meeting, discussing budget.

[0096] S2. Constructing a text time node, a text location node, a text character node and a text event node according to the time data, location data, character data and event data.

[0097] It can be understood that a text time node refers to a node constructed based on time data, which is used to store time data and represents a specific time in the text evidence; a text place node refers to a node constructed based on place data, which is used to store place data and represents a specific place in the text evidence; a text character node refers to a node constructed based on character data, which is used to store character data and represents a specific character in the text evidence; and a text event node refers to a node constructed based on event data, which is used to store event data and represents a specific event in the text evidence.

[0098] S3. Construct a text node set based on the text time nodes, text location nodes, text character nodes and text event nodes, and construct a text relationship graph based on the text node set.

[0099] Interpretable, a text node set refers to a set obtained by summarizing text time nodes, text place nodes, text character nodes and text event nodes, a text relationship graph refers to a structured graph formed by interconnecting text time nodes, text place nodes, text character nodes and text event nodes through connection types, aiming to show the relationship between time data, place data, character data and event data, and to provide support for further analysis or reasoning, a connection type refers to the type of connection between text time nodes, text place nodes, text character nodes and text event nodes, and the connection types include: time connection, place connection, character connection and event connection, time connection is used to connect text time nodes and text event nodes, place connection is used to connect text place nodes and text event nodes, character connection is used to connect text character nodes and text event nodes, and event connection is used to connect different text event nodes.

[0100] In detail, the construction of a text relationship graph based on a text node set includes:

[0101] Defining a connection type based on the text evidence, wherein the connection types include: time connection, location connection, character connection and event connection, the time connection is used to connect a text time node with a text event node, the location connection is used to connect a text location node with a text event node, the character connection is used to connect a text character node with a text event node, and the event connection is used to connect different text event nodes;

[0102] Constructing a text graph edge set according to the connection type, time connection, location connection, person connection and event connection;

[0103] Construct a text relationship graph based on the text node set and the text graph edge set:

[0104] Q = (W, E)

[0105] Among them, Q refers to the text relationship graph, W refers to the text node set, and E refers to the text graph edge set.

[0106] Explainably, the text graph edge set refers to the set formed by the text graph edges, and the text graph edge refers to the edge connecting two nodes in the text relationship graph. For example, the character data in the text character node is John, and the event data in the text event node is planning meeting, then (John, planning meeting) is the edge connecting the text character node and the text event node.

[0107] S4. Set a dynamic acquisition frequency, perform image extraction on the video evidence based on the dynamic acquisition frequency and a pre-built convolutional neural network model to obtain an extracted image set, construct video time nodes, video location nodes, video character nodes, video event nodes and a video graph edge set based on the extracted image set, and construct a video relationship graph based on the video time nodes, video location nodes, video character nodes, video event nodes and the video graph edge set.

[0108] Explainable, dynamic acquisition frequency refers to the frequency of image extraction of video evidence, the dynamic acquisition frequency will be dynamically adjusted according to the size of the image difference value, extracted image set refers to the set of images obtained after image extraction of video evidence, video time node refers to the node constructed according to the time data in the video evidence, video location node refers to the node constructed according to the location data in the video evidence, video character node refers to the node constructed according to the character data in the video evidence, video event node refers to the node constructed according to the event data in the video evidence, video graph edge set refers to the set formed by the video graph edges, video graph edge refers to the edge connecting two nodes in the video relationship graph, for example, the time data in the video time node is December 12, 2024, and the event data in the video event node is a planning meeting, then December 12, 2024, planning meeting) is the edge connecting the video time node and the video event node, and the video relationship graph refers to a structured graph formed by connecting the video time node, video location node, video character node and video event node through the video graph edge set.

[0109] In detail, the setting of the dynamic acquisition frequency includes:

[0110] Set the collection time interval and calculate the initial collection frequency based on the collection time interval:

[0111]

[0112] Among them, R1 refers to the initial acquisition frequency, and t refers to the acquisition time interval;

[0113] Extracting images from the video evidence based on the initial acquisition frequency to obtain an initial image set, and calculating an image difference value based on the initial image set;

[0114] Set the highest difference value and the lowest difference value, and set the dynamic acquisition frequency based on the highest difference value, image difference value, lowest difference value and initial acquisition frequency:

[0115]

[0116] Among them, R2 refers to the dynamic acquisition frequency, T refers to the image difference value, T1 refers to the highest difference value, and T2 refers to the lowest difference value.

[0117] It can be explained that the acquisition time interval refers to the time interval between two frames of images extracted from the video. For example, if the acquisition time interval is set to 0.03s, the time interval between two adjacent frames of images extracted from the video is 0.03s. The initial acquisition frequency refers to the frequency of preliminary image extraction of video evidence. The initial image set refers to the set formed by the initial images. The initial image refers to the image extracted from the video evidence according to the initial acquisition frequency. The image difference value refers to the value that quantifies the difference between two consecutive images. The highest difference value refers to the maximum value of the difference between two images. The lowest difference value refers to the minimum value of the difference between two images. When the image difference value is greater than the highest difference value, it means that there is a large difference between the two images, and the image acquisition frequency needs to be increased. When the image difference value is less than the lowest difference value, it means that there is no large difference between the two images, and the similarity is large. It is necessary to reduce the image acquisition frequency and reduce image duplication.

[0118] In detail, the calculating of the image difference value based on the initial image set includes:

[0119] An initial image sequence is acquired based on the initial image set, wherein the initial image sequence includes a plurality of initial images, a first initial image is sequentially extracted from the initial image sequence, and the following operations are performed on the extracted first initial image:

[0120] Based on the extracted first initial image, identifying a second initial image in the initial image sequence, wherein the second initial image is an initial image adjacent to the first initial image and lagging behind the first initial image;

[0121] Performing pixel detection on the first initial image and the second initial image respectively to obtain a first pixel matrix and a second pixel matrix;

[0122] Set the pixel horizontal coordinate, pixel vertical coordinate and decomposition standard deviation, where the value of the decomposition standard deviation is 1 or 2;

[0123] Construct a decomposition function based on the pixel horizontal coordinate, pixel vertical coordinate and decomposition standard deviation:

[0124]

[0125] Wherein, A(x, y, θ) refers to the decomposition function when the pixel abscissa is x, the pixel ordinate is y, and the decomposition standard deviation is θ, x refers to the pixel abscissa, y refers to the pixel ordinate, θ refers to the decomposition standard deviation, exp refers to the exponential function, and π refers to pi;

[0126] Decomposing the first pixel matrix and the second pixel matrix based on the decomposition function and the decomposition standard deviation respectively to obtain a first decomposition pixel set and a second decomposition pixel set;

[0127] Calculating a first pixel difference set based on the first decomposed pixel set, and calculating a second pixel difference set based on the second decomposed pixel set, wherein the first pixel difference set consists of a first pixel difference and a second pixel difference;

[0128] Summarizing the first pixel difference set to obtain a first grayscale image, and summarizing the second pixel difference set to obtain a second grayscale image;

[0129] An image difference value is calculated according to the first grayscale image and the second grayscale image.

[0130] It can be explained that the initial image sequence refers to the sequence obtained by sorting the initial images in the initial image set in the order of the extraction time corresponding to the initial images from the earliest to the latest. For example, 1, the first initial image, 2, the second initial image, 1 and 2 are the initial image sequence, the first initial image and the second initial image both refer to the images extracted in sequence from the initial image set, and the extraction order of the first initial image is before the second initial image, the first pixel matrix refers to the pixel matrix of the first initial image, the second pixel matrix refers to the pixel matrix of the second initial image, the pixel matrix refers to a two-dimensional array, each element in the two-dimensional array represents a pixel point in the image, the pixel horizontal coordinate refers to the horizontal coordinate of the pixel matrix, the pixel vertical coordinate refers to the vertical coordinate of the pixel matrix, the decomposition standard deviation refers to the parameter controlling the image decomposition in the initial image set, and the number of selectable values ​​of the decomposition standard deviation is the number of layers for image decomposition. For example, the value of the decomposition standard deviation is 1 or 2, and the number is 2, then the image in the initial image set is decomposed into two layers. When the decomposition standard deviation is 1, the decomposition function is used to decompose the initial image to obtain the first layer of image. When the decomposition standard deviation is 2, the decomposition function is used to decompose the initial image to obtain the second layer image. The initial image refers to the image in the initial image set. The size of the first layer image is one quarter of the size of the initial image, and the size of the second layer image is one sixteenth of the size of the initial image. The first layer image refers to the image obtained by decomposing the initial image using the decomposition function when the decomposition standard deviation is 1. The second layer image refers to the image obtained by decomposing the initial image using the decomposition function when the decomposition standard deviation is 2. The decomposition function refers to the function for decomposing the initial image. The first decomposition pixel set refers to the set formed by the first decomposition pixel and the first convolution pixel. The first decomposition pixel refers to the pixel of the first layer image corresponding to the first pixel image. The first convolution pixel refers to the pixel of the second layer image corresponding to the first pixel image. The first pixel image refers to the image corresponding to the first pixel matrix. The second decomposition pixel set refers to the set formed by the second decomposition pixel and the second convolution pixel. The second decomposition pixel refers to the pixel of the first layer image corresponding to the second pixel image. The second convolution pixel refers to the pixel of the second layer image corresponding to the second pixel image. The second pixel image refers to the image corresponding to the second pixel matrix.

[0131] It can be understood that the first pixel difference set refers to a set formed by the first pixel difference and the second pixel difference, the first pixel difference refers to the difference between the pixel point in the first expanded pixel and the corresponding pixel point in the first initial image, the first expanded pixel refers to the pixel of the image obtained by expanding the size of the image corresponding to the first decomposed pixel to the same size as the first initial image, expansion refers to repeating the pixel point in the first decomposed pixel, and repeating refers to copying the pixel point in the first decomposed pixel and repeatedly using the copied pixel point to achieve the purpose of expanding the size. For example, the first decomposed pixel is: Expand to get: Expand and copy And reuse Expand the first decomposed pixel into 255, 254, 253 and 252 are all one pixel. The second pixel difference refers to the difference between the pixel in the second expanded pixel and the corresponding pixel in the first decomposed pixel. The second expanded pixel refers to the pixel of the image obtained by expanding the size of the image corresponding to the first convolution pixel to the same size as the first decomposed pixel. The first initial image refers to the image before the image corresponding to the first decomposed pixel is decomposed. The first grayscale image refers to the image obtained by integrating the first pixel difference set. The first grayscale image is a grayscale image of the same size as the first initial image. The second grayscale image refers to a grayscale image of the same size as the second initial image. The grayscale image refers to an image containing only grayscale information. The grayscale value of each pixel in the grayscale image is an integer between 0 and 255. The grayscale information refers to the brightness of each pixel in the grayscale image. The higher the grayscale value, the stronger the brightness. The second initial image refers to the image before the image corresponding to the second decomposed pixel is decomposed. For example, the size of the first initial image is a 4×4 matrix, as shown below: Decomposing the first initial image to obtain first decomposed pixels, decomposing the first initial image means decomposing a quarter of the pixels in the upper left corner of the matrix corresponding to the first initial image, and the first decomposed pixels are: Decomposing the first decomposed pixel to obtain the first convolution pixel, decomposing the first decomposed pixel means decomposing the pixel point of one quarter of the upper left corner of the first decomposed pixel, and the second decomposed pixel is:

[254] , expanding the first decomposed pixel to the first expanded pixel of 4×4 with the same size as the first initial image: Calculate the difference between each pixel in the first initial image and each pixel in the corresponding first expanded pixel to obtain a first pixel difference: Expand the second decomposed pixel to the same size as the first decomposed pixel: The expansion method of the second decomposed pixel is the same as the expansion method of the first decomposed pixel, which will not be repeated here. The difference between each pixel point in the first decomposed pixel and each pixel point in the corresponding second expanded pixel is calculated to obtain the second pixel difference: Expand the second pixel difference to get Then and The first grayscale image is obtained by integrating the maximum function:

[0132] In detail, the decomposing the first pixel matrix and the second pixel matrix based on the decomposition function and the decomposition standard deviation to obtain a first decomposition pixel set and a second decomposition pixel set includes:

[0133] The decomposition standard deviation is set to 1 to obtain a first decomposition function, and the first pixel matrix and the second pixel matrix are convolved based on the first decomposition function to obtain first decomposition pixels and second decomposition pixels;

[0134] The decomposition standard deviation is set to 2 to obtain a second decomposition function, and the first pixel matrix and the second pixel matrix are respectively convolved based on the second decomposition function to obtain first convolution pixels and second convolution pixels;

[0135] The first decomposed pixels and the first convolution pixels are aggregated to obtain a first decomposed pixel set, and the second decomposed pixels and the second convolution pixels are aggregated to obtain a second decomposed pixel set.

[0136] It can be explained that the first decomposition function refers to the decomposition function when the value of the decomposition standard deviation is set to 1, the second decomposition function refers to the decomposition function when the value of the decomposition standard deviation is set to 2, when the decomposition standard deviation is set to 1, the size of the image after convolution is one quarter of the size of the image before convolution, when the decomposition standard deviation is set to 2, the size of the image after convolution is one sixteenth of the size of the image before convolution, convolution refers to the process of applying a convolution kernel to an input image and decomposing the input image, the convolution kernel refers to the decomposition function, and the input image refers to the first pixel matrix, the second pixel matrix, the first pixel matrix and the second pixel matrix.

[0137] In detail, the first pixel difference set is summarized to obtain a first grayscale image, including:

[0138] Extracting a first pixel difference and a second pixel difference based on the first pixel difference set, obtaining a maximum function, and calculating a grayscale pixel according to the first pixel difference, the second pixel difference, the maximum function, the pixel abscissa, and the pixel ordinate:

[0139] S(x,y)=ξ(d1(x,y),d2(x,y))

[0140] Wherein, S(x, y) refers to the grayscale pixel with the horizontal coordinate x and the vertical coordinate y, ξ refers to the maximum function, d1(x, y) refers to the first pixel difference with the horizontal coordinate x and the vertical coordinate y, and d2(x, y) refers to the second pixel difference with the horizontal coordinate x and the vertical coordinate y;

[0141] The first pixel difference set is summarized based on the grayscale pixels to obtain a first grayscale image.

[0142] It can be explained that the maximum function refers to a function that takes the maximum value, for example, ξ(2,5)=5, and the grayscale pixel refers to the grayscale value of the pixel point in the first grayscale image.

[0143] In detail, calculating the image difference value according to the first grayscale image and the second grayscale image includes:

[0144] Get the total number of pixels and calculate the image error value based on the pixel horizontal coordinate, pixel vertical coordinate, grayscale pixels and the total number of pixels:

[0145]

[0146] Wherein, F refers to the image error value, S2(x, y) refers to the grayscale pixel with the horizontal coordinate x and the vertical coordinate y in the second grayscale image, and N refers to the total number of pixels;

[0147] Calculate a first mean, a second mean, a first variance, a second variance and a comprehensive covariance based on the first grayscale image and the second grayscale image;

[0148] A first stable parameter and a second stable parameter are set, and image similarity is calculated based on the first mean, the second mean, the first variance, the second variance, the comprehensive covariance, the first stable parameter, and the second stable parameter:

[0149]

[0150] Among them, F2 refers to image similarity, δ1 refers to the first mean, δ2 refers to the second mean, σ1 refers to the first stability parameter, ε1 refers to the first variance, ε2 refers to the second variance, σ2 refers to the second stability parameter, and ε3 refers to the comprehensive covariance;

[0151] Set a difference weight parameter, and calculate the image difference value based on the image error value, image similarity and the difference weight parameter:

[0152] F3=α×F+(1-α)×F2

[0153] Among them, F3 refers to the image difference value, and α refers to the difference weight parameter.

[0154] It can be explained that the total number of pixels refers to the number of pixels in the first grayscale image and the second grayscale image, and the total number of pixels in the first grayscale image and the second grayscale image is equal. The image error value refers to the mean of the square error of the pixel values ​​between the first grayscale image and the second grayscale image, reflecting the overall difference between the first grayscale image and the second grayscale image. The larger the image error value, the greater the difference. The square error of the pixel value refers to the square of the difference between the pixel values ​​of the first grayscale image and the second grayscale image at the corresponding pixel position. The corresponding pixel position refers to the corresponding position of the first grayscale image and the second grayscale image. For example, if the position of the first grayscale image is (1,1), then the corresponding pixel position of the second grayscale image is also (1,1). The first stability parameter and the second stability parameter are both constants, which are used to avoid the denominator being zero and ensure the stability of the calculation. The first mean refers to the average value of all grayscale values ​​in the first grayscale image, and the second mean refers to the average value of all grayscale values ​​in the second grayscale image. The first variance refers to the degree of deviation of the grayscale value of the first grayscale image from the first mean, reflecting the degree of discreteness of the grayscale value. The larger the first variance, the greater the deviation of the grayscale value of the first grayscale image from the first mean, and the discreteness of the grayscale value. The higher the degree, the second variance refers to the deviation degree of the grayscale value of the second grayscale image from the second mean, reflecting the discrete degree of the grayscale value. The larger the second variance, the greater the deviation degree of the grayscale value of the second grayscale image from the second mean, and the higher the discrete degree of the grayscale value. The comprehensive covariance represents the similarity degree of grayscale value change between the first grayscale image and the second grayscale image. The grayscale value change refers to the change of grayscale value of the corresponding pixel points in the first grayscale image and the second grayscale image. The similarity degree refers to the similarity degree of grayscale value change. The greater the similarity degree, the similarity degree indicates that the corresponding pixel points in the first grayscale image and the second grayscale image are similar. The more similar the changes in grayscale values ​​are, the image similarity refers to a numerical value describing the similarity between the first grayscale image and the second grayscale image. The higher the image similarity, the more similar the first grayscale image is to the second grayscale image. The difference weight parameter refers to a parameter for adjusting the influence of the image error value on the image difference value. The larger the difference weight parameter, the greater the influence of the image error value on the image difference value. Among them, the first variance is the variance of the grayscale values ​​in the first grayscale image, the comprehensive covariance is the covariance of the grayscale values ​​in the first grayscale image and the grayscale values ​​in the second grayscale image, and the second variance is the variance of the grayscale values ​​in the second grayscale image.

[0155] S5. Calculate the evidence similarity according to the text relationship graph and the video relationship graph, and determine the evidence status of the text evidence and the video evidence based on the evidence similarity, wherein the evidence status includes: a credible status and a false status.

[0156] Explainable, evidence status refers to the status of judging whether text evidence and video evidence are credible, evidence similarity refers to the degree of similarity between text evidence and video evidence, the greater the evidence similarity, the greater the degree of similarity between text evidence and video evidence, and the more credible content in the text evidence and video evidence. Evidence status includes: credible status and false status, credible status indicates that text evidence and video evidence are credible, and false status indicates that text evidence and video evidence are uncredible.

[0157] In detail, the calculating of the evidence similarity according to the text relationship graph and the video relationship graph includes:

[0158] Calculate the similarity of time nodes, and calculate the similarity of location nodes, person nodes and event nodes;

[0159] Transform the text graph edge set and the video graph edge set respectively to obtain a text edge vector and a video edge vector;

[0160] Set the edge vector index, and calculate the graph edge similarity based on the edge vector index, text edge vector, and video edge vector:

[0161]

[0162] Among them, k refers to the graph edge similarity, i refers to the edge vector index, Refers to the text edge vector when the edge vector index is i, μ i Refers to the video edge vector when the edge vector index is i, ||*|| refers to the modulo symbol, Point to the multiplication symbol;

[0163] Set the time node weight, location node weight, person node weight, event node weight and edge similarity weight, and calculate the evidence similarity based on the time node weight, location node weight, person node weight, event node weight, edge similarity weight, time node similarity, location node similarity, person node similarity, event node similarity and graph edge similarity:

[0164] P=β1×γ1+β2×γ2+β3×γ3+β4×γ4+β5×k

[0165] Among them, P refers to the evidence similarity, β1 refers to the time node weight, γ1 refers to the time node similarity, α2 refers to the location node weight, γ2 refers to the location node similarity, β3 refers to the person node weight, γ3 refers to the person node similarity, β4 refers to the event node weight, γ4 refers to the event node similarity, and β5 refers to the edge similarity weight.

[0166] Explainable, time node similarity refers to the similarity between text time nodes and video time nodes in the text relationship graph and video relationship graph. The higher the time node similarity, the higher the similarity between the text time nodes and the video time nodes. Location node similarity refers to the similarity between text location nodes and video location nodes in the text relationship graph and video relationship graph. The higher the location node similarity, the higher the similarity between the text location nodes and the video location nodes. Character node similarity refers to the similarity between text character nodes and video character nodes in the text relationship graph and video relationship graph. The higher the character node similarity, the higher the similarity between the text character nodes and the video character nodes. Event node similarity Refers to the similarity between text event nodes and video event nodes in the text relationship graph and the video relationship graph. The higher the event node similarity, the higher the similarity between the text event node and the video event node. The text edge vector refers to the feature vector that quantifies and describes the connection type between text time nodes, text location nodes, text character nodes and text event nodes. The video edge vector refers to the feature vector that quantifies and describes the connection type between video time nodes, video location nodes, video character nodes and video event nodes. The text edge vector and the video edge vector are obtained by transforming the text graph edge set and the video graph edge set respectively. For example, the text evidence is that a meeting was held on December 12, 2024, (20 On December 12, 24, a meeting was held) is the edge connecting the text time node and the text event node. The text edge vector is: [held, 1], held refers to the connection between the text time node and the text event node, 1 refers to the number of times the connection appears in the text evidence, the edge vector index refers to the parameters for traversing the text edge vector and the video edge vector, the graph edge similarity refers to the similarity between the text edge vector and the video edge vector, the higher the graph edge similarity, the higher the similarity between the text edge vector and the video edge vector, the time node weight refers to the parameter for adjusting the influence of the time node similarity on the evidence similarity, the larger the time node weight, the greater the influence of the time node similarity on the evidence similarity, the location node weight refers to the adjustment The parameter for adjusting the degree of influence of location node similarity on evidence similarity. The larger the location node weight, the greater the influence of location node similarity on evidence similarity. The person node weight refers to the parameter for adjusting the degree of influence of person node similarity on evidence similarity. The larger the person node weight, the greater the influence of person node similarity on evidence similarity. The event node weight refers to the parameter for adjusting the degree of influence of event node similarity on evidence similarity. The larger the event node weight, the greater the influence of event node similarity on evidence similarity. The edge similarity weight refers to the parameter for adjusting the degree of influence of graph edge similarity on evidence similarity. The larger the edge similarity weight, the greater the influence of adjusting graph edge similarity on evidence similarity.

[0167] In detail, the calculation of time node similarity includes:

[0168] Summarizing the text time nodes and the video time nodes to obtain a text time node set and a video time node set;

[0169] Calculate the time node similarity based on the text time node set and the video time node set:

[0170]

[0171] Among them, H1 refers to the text time node set, H2 refers to the video time node set, ∩ refers to the intersection symbol, and ∪ refers to the union symbol.

[0172] It can be explained that the text time node set refers to the set formed by the time parameters stored in the text time node, and the video time node set refers to the set formed by the time parameters stored in the video time node. The time parameter refers to the time data, for example, January 1, 2024, January 2, 2024.

[0173] It can be understood that the method for calculating the location node similarity, the person node similarity and the event node similarity is the same as the method for calculating the time node similarity. The text location node set and the video location node set are obtained, and the location node similarity is calculated based on the text location node set and the video location node set:

[0174]

[0175] Among them, γ2 refers to the location node similarity, M1 refers to the text location node set, and M2 refers to the video location node set;

[0176] Get the text character node set and the video character node set, and calculate the character node similarity based on the text character node set and the video character node set:

[0177]

[0178] Among them, γ3 refers to the similarity of character nodes, M1′ refers to the set of text character nodes, and M2′ refers to the set of video character nodes;

[0179] Get the text event node set and the video event node set, and calculate the event node similarity based on the text event node set and the video event node set:

[0180]

[0181] Among them, γ4 refers to the event node similarity, M1” refers to the text event node set, and M2” refers to the video event node set.

[0182] It can be explained that the text location node set refers to the set formed by the location parameters stored in the text location node, and the video location node set refers to the set formed by the location parameters stored in the video location node. The location parameter refers to the data of the location, for example, London. The text character node set refers to the set formed by the character parameters stored in the text character node, and the video character node set refers to the set formed by the character parameters stored in the video character node. The character parameters refer to the data of the character, and the data of the character includes but is not limited to: name, age. The text event node set refers to the set formed by the event parameters stored in the text event node, and the video event node set refers to the set formed by the event parameters stored in the video event node. The event parameters refer to the data of the event, for example, planning meeting.

[0183] S6. Determine whether the evidence similarity is less than a preset similarity threshold.

[0184] Explainable, similarity threshold refers to a threshold for judging the similarity of evidence. When the similarity of evidence is not less than the similarity threshold, it means that the evidence is credible. When the similarity of evidence is less than the similarity threshold, it means that the evidence is unreliable.

[0185] If the evidence similarity is less than the similarity threshold, S7 is executed to confirm that the evidence status is a false status.

[0186] If the evidence similarity is not less than the similarity threshold, execute S8 to confirm that the evidence status is a credible status.

[0187] Therefore, this embodiment firstly extracts data by structural analysis of the processed text to effectively remove irrelevant information and interference, such as removing redundant spaces, punctuation marks, and stop words, and performs accurate word segmentation. Removing redundant spaces, punctuation marks, and stop words can effectively reduce the noise of the processed text and reduce the computational complexity of the entity recognition model.

[0188] Furthermore, through accurate word segmentation, text evidence is converted into independent words or phrases, which is conducive to entity recognition model recognition and can focus more on specific content. Especially in multilingual or professional texts, accurate word segmentation helps to identify important information without being disturbed by irrelevant symbols or words. Structural analysis can better understand the hierarchical structure and grammatical relationship of text evidence, provide clear data for entity recognition model, and improve the recognition accuracy and efficiency of entity recognition model.

[0189] And further, a text relationship map is constructed according to the text node set. The construction of the text relationship map can clearly represent all the data in the text evidence and the relationship between the data, can effectively organize and store information, and provide a clear node structure for the calculation of evidence similarity; then, based on the dynamic acquisition frequency and the convolutional neural network model, image extraction is performed on the video evidence to adjust according to the content of the video evidence, improve the acquisition efficiency, avoid the extraction of redundant images, save computing resources, and provide accurate relationship data for the subsequent calculation of evidence similarity through the video relationship map, so that the calculation of evidence similarity is more efficient and accurate; finally, by automatically calculating the evidence similarity and judging the similarity threshold, the errors of human operation are reduced, and the accuracy and efficiency of evidence evaluation are improved.

[0190] Embodiment 2:

[0191] like Figure 2 As shown, this embodiment provides a functional module diagram of an evidence credibility intelligent assessment system based on a neural network, and the evidence credibility intelligent assessment system 100 based on a neural network can be installed in an electronic device. According to the functions implemented, the evidence credibility intelligent assessment system 100 based on a neural network can include a text relationship graph construction module 101, a video relationship graph construction module 102, an evidence similarity calculation module 103 and an evidence status judgment module 104. The module described in the present invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0192] The text relationship map construction module 101 is used to obtain text evidence and video evidence, perform text preprocessing based on the text evidence to obtain processed text, perform structural analysis on the processed text and extract data from the processed text after structural analysis based on a pre-built entity recognition model to obtain time data, location data, character data and event data; construct text time nodes, text location nodes, text character nodes and text event nodes based on the time data, location data, character data and event data; construct a text node set based on the text time nodes, text location nodes, text character nodes and text event nodes, and construct a text relationship map based on the text node set;

[0193] The video relationship graph construction module 102 is used to set a dynamic acquisition frequency, extract images from the video evidence based on the dynamic acquisition frequency and a pre-constructed convolutional neural network model to obtain an extracted image set, construct video time nodes, video location nodes, video character nodes, video event nodes and a video graph edge set based on the extracted image set, and construct a video relationship graph based on the video time nodes, video location nodes, video character nodes, video event nodes and the video graph edge set;

[0194] The evidence similarity calculation module 103 is used to calculate the evidence similarity according to the text relationship graph and the video relationship graph, and judge the evidence status of the text evidence and the video evidence based on the evidence similarity, wherein the evidence status includes: a credible state and a false state;

[0195] The evidence status judgment module 104 is used to determine whether the evidence similarity is less than a preset similarity threshold; if the evidence similarity is less than the similarity threshold, the evidence status is confirmed to be a false status; if the evidence similarity is not less than the similarity threshold, the evidence status is confirmed to be a credible status.

[0196] In detail, the modules in the neural network-based evidence credibility intelligent evaluation system 100 in the embodiment of the present invention are used in the same manner as described above. Figure 1 The technical means are the same as the neural network-based evidence credibility intelligent evaluation method described in, and can produce the same technical effects, so they will not be repeated here.

[0197] like Figure 3 , is a schematic diagram of the structure of an electronic device for implementing a method for intelligently evaluating the credibility of evidence based on a neural network according to an embodiment of the present invention.

[0198] The electronic device 1 may include a processor 10, a memory 11 and a bus 12, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as the neural network-based evidence credibility intelligent assessment method program described in Example 1.

[0199] The memory 11 includes at least one type of readable storage medium, including flash memory, mobile hard disk, multimedia card, card-type memory (such as SD or DX memory, etc.), magnetic memory, disk, optical disk, etc.

[0200] In some embodiments, the processor 10 may be composed of an integrated circuit, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, and executes various functions of the electronic device 1 and processes data by running or executing programs or modules stored in the memory 11 (such as the neural network-based evidence credibility intelligent assessment method program described in Example 1, etc.), and calling data stored in the memory 11.

[0201] The bus 12 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus.

[0202] Figure 3 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 3 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0203] The neural network-based evidence credibility intelligent assessment method program stored in the memory 11 in the electronic device 1 is a combination of multiple instructions. When running in the processor 10, the neural network-based evidence credibility intelligent assessment method described in Example 1 can be implemented.

[0204] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to Figures 1 to 3 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0205] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor of an electronic device, it can implement the neural network-based evidence credibility intelligent assessment method described in Example 1.

[0206] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0207] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for intelligently evaluating the credibility of evidence based on a neural network, characterized in that: The method comprises: Acquire text evidence and video evidence, perform text preprocessing based on the text evidence to obtain processed text, perform structural analysis on the processed text, and perform data extraction on the processed text after structural analysis based on a pre-built entity recognition model to obtain time data, location data, person data, and event data; Constructing text time nodes, text location nodes, text character nodes and text event nodes according to the time data, location data, character data and event data; Constructing a text node set based on the text time node, text location node, text character node and text event node, and constructing a text relationship graph according to the text node set; A dynamic acquisition frequency is set, and images are extracted from the video evidence based on the dynamic acquisition frequency and a pre-built convolutional neural network model to obtain an extracted image set, and a video time node, a video location node, a video character node, a video event node, and a video graph edge set are constructed based on the extracted image set, and a video relationship graph is constructed based on the video time node, the video location node, the video character node, the video event node, and the video graph edge set; Calculating evidence similarity according to the text relationship graph and the video relationship graph, and judging the evidence status of the text evidence and the video evidence based on the evidence similarity, wherein the evidence status includes: a credible status and a false status; Determine whether the evidence similarity is less than a preset similarity threshold; If the evidence similarity is less than the similarity threshold, the evidence status is confirmed to be a false status; If the evidence similarity is not less than the similarity threshold, the evidence status is confirmed to be credible.

2. The method for intelligently evaluating the credibility of evidence as claimed in claim 1, characterized in that: The step of constructing a text relationship graph according to a text node set includes: Defining a connection type based on the text evidence, wherein the connection types include: time connection, location connection, character connection and event connection, the time connection is used to connect a text time node with a text event node, the location connection is used to connect a text location node with a text event node, the character connection is used to connect a text character node with a text event node, and the event connection is used to connect different text event nodes; Constructing a text graph edge set according to the connection type, time connection, location connection, person connection and event connection; Construct a text relationship graph based on the text node set and the text graph edge set: Q = (W, E) Among them, Q refers to the text relationship graph, W refers to the text node set, and E refers to the text graph edge set.

3. The method for intelligently evaluating the credibility of evidence as claimed in claim 2, characterized in that: The step of setting the dynamic acquisition frequency includes: Set the collection time interval and calculate the initial collection frequency based on the collection time interval: Among them, R1 refers to the initial acquisition frequency, and t refers to the acquisition time interval; Extracting images from the video evidence based on the initial acquisition frequency to obtain an initial image set, and calculating an image difference value based on the initial image set; Set the highest difference value and the lowest difference value, and set the dynamic acquisition frequency based on the highest difference value, image difference value, lowest difference value and initial acquisition frequency: Among them, R2 refers to the dynamic acquisition frequency, T refers to the image difference value, T1 refers to the highest difference value, and T2 refers to the lowest difference value.

4. The method for intelligently evaluating the credibility of evidence as claimed in claim 3, characterized in that: The calculating of the image difference value based on the initial image set comprises: An initial image sequence is acquired based on the initial image set, wherein the initial image sequence includes a plurality of initial images, a first initial image is sequentially extracted from the initial image sequence, and the following operations are performed on the extracted first initial image: Based on the extracted first initial image, identifying a second initial image in the initial image sequence, wherein the second initial image is an initial image adjacent to the first initial image and lagging behind the first initial image; Performing pixel detection on the first initial image and the second initial image respectively to obtain a first pixel matrix and a second pixel matrix; Set the pixel horizontal coordinate, pixel vertical coordinate and decomposition standard deviation, where the value of the decomposition standard deviation is 1 or 2; Construct a decomposition function based on the pixel horizontal coordinate, pixel vertical coordinate and decomposition standard deviation: Wherein, A(x, y, θ) refers to the decomposition function when the pixel abscissa is x, the pixel ordinate is y, and the decomposition standard deviation is θ, x refers to the pixel abscissa, y refers to the pixel ordinate, θ refers to the decomposition standard deviation, exp refers to the exponential function, and π refers to pi; Decomposing the first pixel matrix and the second pixel matrix based on the decomposition function and the decomposition standard deviation respectively to obtain a first decomposition pixel set and a second decomposition pixel set; Calculating a first pixel difference set based on the first decomposed pixel set, and calculating a second pixel difference set based on the second decomposed pixel set, wherein the first pixel difference set consists of a first pixel difference and a second pixel difference; Summarizing the first pixel difference set to obtain a first grayscale image, and summarizing the second pixel difference set to obtain a second grayscale image; An image difference value is calculated according to the first grayscale image and the second grayscale image.

5. The method for intelligently evaluating the credibility of evidence as claimed in claim 4, characterized in that: The first pixel matrix and the second pixel matrix are decomposed based on the decomposition function and the decomposition standard deviation to obtain a first decomposed pixel set and a second decomposed pixel set, including: The decomposition standard deviation is set to 1 to obtain a first decomposition function, and the first pixel matrix and the second pixel matrix are convolved based on the first decomposition function to obtain first decomposition pixels and second decomposition pixels; The decomposition standard deviation is set to 2 to obtain a second decomposition function, and the first pixel matrix and the second pixel matrix are respectively convolved based on the second decomposition function to obtain first convolution pixels and second convolution pixels; The first decomposed pixels and the first convolution pixels are aggregated to obtain a first decomposed pixel set, and the second decomposed pixels and the second convolution pixels are aggregated to obtain a second decomposed pixel set.

6. The method for intelligently evaluating the credibility of evidence as claimed in claim 5, characterized in that: The step of summarizing the first pixel difference set to obtain a first grayscale image includes: Extracting a first pixel difference and a second pixel difference based on the first pixel difference set, obtaining a maximum function, and calculating a grayscale pixel according to the first pixel difference, the second pixel difference, the maximum function, the pixel abscissa, and the pixel ordinate: S(x,y)=ξ(d1(x,y),d2(x,y)) Wherein, S(x, y) refers to the grayscale pixel with the horizontal coordinate x and the vertical coordinate y, ξ refers to the maximum function, d1(x, y) refers to the first pixel difference with the horizontal coordinate x and the vertical coordinate y, and d2(x, y) refers to the second pixel difference with the horizontal coordinate x and the vertical coordinate y; The first pixel difference set is summarized based on the grayscale pixels to obtain a first grayscale image.

7. The method for intelligently evaluating the credibility of evidence as claimed in claim 6, characterized in that: The calculating the image difference value according to the first grayscale image and the second grayscale image includes: Get the total number of pixels and calculate the image error value based on the pixel horizontal coordinate, pixel vertical coordinate, grayscale pixels and the total number of pixels: Wherein, F refers to the image error value, S2(x, y) refers to the grayscale pixel with the horizontal coordinate x and the vertical coordinate y in the second grayscale image, and N refers to the total number of pixels; Calculate a first mean, a second mean, a first variance, a second variance and a comprehensive covariance based on the first grayscale image and the second grayscale image; A first stable parameter and a second stable parameter are set, and image similarity is calculated based on the first mean, the second mean, the first variance, the second variance, the comprehensive covariance, the first stable parameter, and the second stable parameter: Among them, F2 refers to image similarity, δ1 refers to the first mean, δ2 refers to the second mean, σ1 refers to the first stability parameter, ε1 refers to the first variance, ε2 refers to the second variance, σ2 refers to the second stability parameter, and ε3 refers to the comprehensive covariance; Set a difference weight parameter, and calculate the image difference value based on the image error value, image similarity and the difference weight parameter: F3=α×F+(1-α)×F2 Among them, F3 refers to the image difference value, and α refers to the difference weight parameter.

8. The method for intelligently evaluating the credibility of evidence as claimed in claim 7, characterized in that: The calculating of the evidence similarity according to the text relationship graph and the video relationship graph includes: Calculate the similarity of time nodes, and calculate the similarity of location nodes, person nodes and event nodes; Transform the text graph edge set and the video graph edge set respectively to obtain a text edge vector and a video edge vector; Set the edge vector index, and calculate the graph edge similarity based on the edge vector index, text edge vector, and video edge vector: Among them, k refers to the graph edge similarity, i refers to the edge vector index, Refers to the text edge vector when the edge vector index is i, μ i Refers to the video edge vector when the edge vector index is i, ||*|| refers to the modulo symbol, Point to the multiplication symbol; Set the time node weight, location node weight, person node weight, event node weight and edge similarity weight, and calculate the evidence similarity based on the time node weight, location node weight, person node weight, event node weight, edge similarity weight, time node similarity, location node similarity, person node similarity, event node similarity and graph edge similarity: P=β1×γ1+β2×γ2+β3×γ3+β4×γ4+β5×k Among them, P refers to the evidence similarity, β1 refers to the time node weight, γ1 refers to the time node similarity, β2 refers to the location node weight, γ2 refers to the location node similarity, β3 refers to the person node weight, γ3 refers to the person node similarity, β4 refers to the event node weight, γ4 refers to the event node similarity, and β5 refers to the edge similarity weight.

9. The method for intelligently evaluating the credibility of evidence as claimed in claim 8, characterized in that: The calculating time node similarity includes: Summarizing the text time nodes and the video time nodes to obtain a text time node set and a video time node set; Calculate the time node similarity based on the text time node set and the video time node set: Among them, H1 refers to the text time node set, H2 refers to the video time node set, ∩ refers to the intersection symbol, and ∪ refers to the union symbol.

10. An intelligent evaluation system for evidence credibility based on neural network, characterized in that: The system comprises: A text relationship map construction module is used to obtain text evidence and video evidence, perform text preprocessing based on the text evidence to obtain processed text, perform structural analysis on the processed text and perform data extraction on the processed text after structural analysis based on a pre-built entity recognition model to obtain time data, location data, character data and event data; construct text time nodes, text location nodes, text character nodes and text event nodes based on the time data, location data, character data and event data; construct a text node set based on the text time nodes, text location nodes, text character nodes and text event nodes, and construct a text relationship map based on the text node set; A video relationship graph construction module is used to set a dynamic acquisition frequency, perform image extraction on the video evidence based on the dynamic acquisition frequency and a pre-built convolutional neural network model to obtain an extracted image set, construct video time nodes, video location nodes, video character nodes, video event nodes and a video graph edge set based on the extracted image set, and construct a video relationship graph based on the video time nodes, video location nodes, video character nodes, video event nodes and the video graph edge set; The evidence similarity calculation module is used to calculate the evidence similarity according to the text relationship map and the video relationship map, and judge the evidence status of the text evidence and the video evidence based on the evidence similarity, wherein: Evidence status includes: credible status and false status; An evidence status judgment module is used to determine whether the evidence similarity is less than a preset similarity threshold; If the evidence similarity is less than the similarity threshold, the evidence status is confirmed to be a false status; if the evidence similarity is not less than the similarity threshold, the evidence status is confirmed to be a credible status.

Images