Data information analysis method based on multi-source data

Abstract

The invention discloses a data information analysis method based on multi-source data, and relates to the technical field of data analysis, and the method comprises the steps: setting an unmanned aerial vehicle cluster, collecting the data information of a plurality of data sources through the unmanned aerial vehicle cluster, carrying out the information preprocessing of the data information of a plurality of data sources, and generating modal data corresponding to each data source, the method comprises the following steps: performing modal feature extraction on modal data to obtain modal information features corresponding to the modal data, setting fusion layers of different dimensions, mapping the modal information features to the different fusion layers, and combining the different fusion layers and the corresponding modal information features to generate multi-source fusion data. And performing data analysis on the multi-source fusion data to generate corresponding analysis information, judging whether dangerous information exists in the analysis information or not, and executing different data decision schemes according to a judgment result.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a data information analysis method based on multi-source data. Background Art

[0002] Data needs to be transmitted during network communication. There are two commonly used data formats: JSON and XML. The process of JOSN data parsing is: first create a JSON file, then include document.h and cocos-ext.h header files in the class, then get the JSON file path through FileUtils, and parse the JSON data through the Document object to finally get different types of data values. The process of XML data parsing is: first create an XML file, then include the header file in the class and use the named file, then get the full path of the XML file, load the XML file, and finally get the element and parse it.

[0003] With the rapid development of drone technology, drones have become an important tool in various industries and are widely used in agriculture, logistics, security and other fields. The data collected by drones comes from multiple sources, such as sensors, remote sensing instruments, and communication terminals. How to integrate data information from multiple sources and then accurately analyze it, and then locate dangerous information in a timely manner, is a problem that we urgently need to solve. Summary of the invention

[0004] In order to solve the above problems, the purpose of the present invention is to provide a data information analysis method based on multi-source data.

[0005] The object of the present invention can be achieved by the following technical solution: A data information analysis method based on multi-source data, comprising the following steps:

[0006] Step S1: setting up a drone cluster, and collecting data information from multiple data sources through the drone cluster, and performing information preprocessing on the data information from multiple data sources to generate modal data corresponding to each data source;

[0007] Step S2: extracting modal features from the modal data, and then obtaining modal information features corresponding to the modal data, setting fusion layers of different dimensions, mapping the modal information features to different fusion layers, and then merging different fusion layers and their corresponding modal information features to generate multi-source fusion data;

[0008] Step S3: Analyze the multi-source fusion data to generate corresponding analysis information, determine whether there is dangerous information in the analysis information, and execute different data decision plans based on the judgment results.

[0009] Furthermore, the process of setting up a drone cluster to collect data information from multiple data sources includes:

[0010] Obtain the communication rights of several drones, and number them as i, i=1, 2, 3, ..., n, where n is a natural number greater than 0. The communication rights of each drone are obtained through its corresponding communication code. The communication code corresponding to the drone numbered i is TX[i]. Starting from i=1, the drone with i=1 is used as the communication object, and the drones corresponding to other numbers are connected to the communication object, thereby establishing data communication between the drone corresponding to the current communication object and another n-1 drones. Continue to take the drone corresponding to i=2 as the communication object, and connect the other n-2 drones that have not been used as communication objects to the current communication object. Repeat the above operation until each drone establishes data communication with drones other than itself, thereby forming a drone cluster consisting of several drones.

[0011] Several types of sensors, remote sensing instruments and external communication terminals are deployed on the drones corresponding to the drone cluster, and data information from multiple data sources is collected through the drone cluster, wherein the data information collected by the multiple types of sensors, remote sensing instruments and external communication terminals are each used as a data source, sensor data is collected through multiple types of sensors, remote sensing image data is collected through remote sensing instruments, and external communication data is obtained through external communication terminals.

[0012] Furthermore, the process of preprocessing the data information of multiple data sources to generate modal data corresponding to each data source includes:

[0013] The information preprocessing includes data pre-screening, data cleaning and modal file encapsulation, data pre-screening is performed on data information corresponding to different data sources, and then data information of multiple data sources that do not meet the preset qualified value range is screened out, and a data cleaning program is set to clean sensor data, remote sensing image data and external communication data, and cleaning fields of sensor data, remote sensing image data and external communication data are defined, and the data content part corresponding to the cleaning field is the data information that needs to be eliminated through data cleaning; and then the data cleaning program cleans and eliminates part of the data information that does not meet the requirements of the sensor data, remote sensing image data and external communication data according to the defined cleaning fields, and then generates corresponding standard sensor data, standard image data and standard communication data;

[0014] The modal text is set to perform modal file encapsulation of standard sensor data, standard image data and standard communication data, and then generate modal data corresponding to each data source. The modal data includes sensor modal data, image modal data and communication modal data. The file data volume of standard sensor data, standard image data and standard communication data before modal file encapsulation is obtained, and recorded as D1, D2 and D3 respectively. The corresponding file data volume after each modal file encapsulation is obtained, and recorded as D1`, D2` and D3`. Then, the respective encapsulation loss data volume is obtained, and recorded as Sh1, Sh2 and Sh3 respectively. Sh1=|D1-D1`|, Sh2=|D2-D2`|, Sh3=|D3-D3`|, and a re-encapsulation threshold is set, recorded as C. When Sh1≥C, Sh2≥C, and Sh3≥C exist, the data information of the corresponding data source is re-encapsulated in modal file. Otherwise, no operation is performed.

[0015] Furthermore, the process of extracting modal features from the modal data and obtaining modal information features corresponding to the modal data includes:

[0016] A feature extraction library is set up for modal feature extraction. The feature extraction library is pre-entered with feature extraction fields corresponding to the sensor modal data, imaging modal data, and communication modal data in the modal data. The feature extraction field corresponding to each modal data is associated with a corresponding extraction similarity. The modal data is input into the feature extraction library, and then the feature extraction library divides the modal data of each data source into several modal fields. The feature extraction field matched to the modal data in the feature extraction library is used as the benchmark field, and the field similarity between each modal field and the benchmark field is obtained. If the field similarity is greater than or equal to the extraction similarity of the benchmark field, the corresponding modal field is used as a feature extraction element, and all feature extraction elements are summarized as the modal information features corresponding to the modal data. The above operation is repeated to obtain the modal feature information of the sensor modal data, imaging modal data, and communication modal data.

[0017] Furthermore, the process of setting fusion layers of different dimensions and mapping modal information features to different fusion layers includes:

[0018] Set fusion layers of different dimensions. The dimensions of the fusion layers include high-dimensional fusion layers, general-dimensional fusion layers, and low-dimensional fusion layers. Perform convolution processing on the sensor modality data, image modality data, and communication modality data through the convolutional neural network algorithm to obtain the corresponding convolution mapping coefficients, which are recorded as J1, J2, and J3 respectively. Set the mapping qualified values ​​of the high-dimensional fusion layer, general-dimensional fusion layer, and low-dimensional fusion layer, which are recorded as R1, R2, and R3 respectively.

[0019] When J1≥R1, the modal information features corresponding to the sensor modal data are mapped to the high-latitude fusion layer, and the current modal information features are used as the first fusion features. When J2≥R2, the modal information features corresponding to the image modal data are mapped to the general latitude fusion layer, and the current modal information features are used as the second fusion features. When J3≥R3, the modal information features corresponding to the image modal data are mapped to the low-latitude fusion layer, and the current modal information features are used as the third fusion features.

[0020] Furthermore, the process of generating multi-source fusion data includes:

[0021] Obtain the number of layer channels corresponding to the high-latitude fusion layer, the general-latitude fusion layer, and the low-latitude fusion layer, and record them as TD1, TD2, and TD3 respectively. Cut each fusion layer into three equal sub-layer areas, and mark them as the left half layer, the middle layer, and the right half layer in the order of the fusion layers from left to right. Obtain the data information volume of the modal information features corresponding to the fusion layer of each dimension in each sub-layer area, and use the same sub-layer area of ​​fusion layers of different dimensions as the object area to be fused;

[0022] That is, the left half layer of the high-dimensional fusion layer, the general latitude fusion layer and the low-latitude fusion layer is used as an object area to be fused, the middle layer of each is used as an object area to be fused, and the right half layer of each is used as an object area to be fused. The fusion layer of the dimension with the highest data information in each object area to be fused is used as the reference layer, and the number of channels of the fusion layers of the other two dimensions is converted to the number of channels corresponding to the reference layer. The number of channels after conversion satisfies TD1=TD2=TD3. After the number of channels is unified, the first fusion features, the second fusion features and the third fusion features corresponding to different fusion layers are merged and fused into multi-source fusion data.

[0023] Furthermore, the process of parsing the multi-source fusion data and generating parsing information includes:

[0024] A data analysis program is set up, in which a number of analysis segments are pre-entered, and the types of analysis segments include feature keyword analysis segments, target recognition analysis segments, and time series trend analysis segments. Multi-source fusion data is input into the data analysis program, and then data analysis of the multi-source fusion data is performed. The data analysis program sets the segment analysis length, and divides the multi-source fusion data into a number of data segments to be analyzed according to the segment analysis length;

[0025] All data segments to be parsed that meet the feature keyword parsing segment are summarized into text parsing information, all data segments to be parsed that meet the target recognition parsing segment are summarized into image recognition parsing information, all data segments to be parsed that meet the time series trend parsing segment are summarized into communication evolution trend parsing information, and the text parsing information, image recognition parsing information and communication evolution trend parsing information are integrated to generate the final parsing information.

[0026] Furthermore, the process of determining whether there is dangerous information in the parsed information and executing different data decision-making solutions according to the determination result includes:

[0027] Set the corresponding danger determination conditions of the text analysis information, image recognition analysis information and communication evolution trend analysis information, set the danger determination model, input the analysis information and all the danger determination conditions into the danger determination model, and then the danger determination model outputs the corresponding determination code after determination according to each danger determination condition. The determination code includes "0" and "1". When the text analysis information, image recognition analysis information and communication evolution trend analysis information do not meet their own danger determination conditions, their respective determination codes are set to "0". When any one of them meets its corresponding danger determination condition, its determination code value is set to "1";

[0028] The judgment code is spliced ​​to generate a judgment result string, and the judgment result string takes the values ​​of "000", "001", "011" and "111". When the judgment result string takes the value of "001" or "011" or "111", it is judged that there is dangerous information in the parsed information, and the executed data decision is: locate the specific position of the drone in the drone cluster where the dangerous information exists, and inspect the drone. When the judgment result string takes the value of "000", it is judged that there is no dangerous information in the current parsed information, and the executed data decision is: continuously monitor the operation of the drone cluster and obtain multi-source fusion data.

[0029] Compared with the prior art, the beneficial effects of the present invention are as follows: by performing data communication between several drones, several drones are set as a drone cluster, and by deploying sensors, remote sensing instruments and external communication terminals in the drone cluster to collect sensor data, remote sensing image data and external communication data corresponding to different data sources, and using them as modal data of their respective data sources, modal feature extraction is performed on the modal data to generate modal information features, fusion layers of different dimensions are set to merge different modal information features to generate final multi-source fusion data, and the multi-source fusion data that comprehensively considers multiple data sources has a wider data dimension, which provides a good prerequisite for subsequent data analysis, and by performing data analysis on the multi-source fusion data, it is determined whether there is dangerous information in the analysis information generated after the data analysis, and a data decision plan is executed according to the existence of dangerous information, so as to locate the dangerous information and eliminate it in time. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0031] like Figure 1 As shown, a data information analysis method based on multi-source data includes the following steps:

[0032] Step S1: setting up a drone cluster, and collecting data information from multiple data sources through the drone cluster, and performing information preprocessing on the data information from multiple data sources to generate modal data corresponding to each data source;

[0033] Step S2: extracting modal features from the modal data, and then obtaining modal information features corresponding to the modal data, setting fusion layers of different dimensions, mapping the modal information features to different fusion layers, and then merging different fusion layers and their corresponding modal information features to generate multi-source fusion data;

[0034] Step S3: Analyze the multi-source fusion data to generate corresponding analysis information, determine whether there is dangerous information in the analysis information, and execute different data decision plans based on the judgment results.

[0035] It should be further explained that, in the specific implementation process, the process of setting up a drone cluster and collecting data information from multiple data sources through the drone cluster includes:

[0036] Obtain the communication rights of several drones, and number them, denoted by i, then i = 1, 2, 3, ..., n, where n is a natural number greater than 0. The communication rights of each drone are obtained through their corresponding communication codes. The communication code corresponding to the drone numbered i is TX[i];

[0037] Starting from i=1, take the drone with i=1 as the communication object, connect the drones with other numbers to the communication object, and then establish data communication between the drone corresponding to the current communication object and the other n-1 drones. Continue to take the drone corresponding to i=2 as the communication object, and connect the other n-2 drones that have not been used as communication objects to the current communication object. Repeat the above operation until each drone establishes a data communication with drones other than itself, thereby forming a drone cluster consisting of several drones.

[0038] Several types of sensors, remote sensing instruments and external communication terminals are deployed on the drones corresponding to the drone cluster, and data information from multiple data sources are collected through the drone cluster, wherein the data information collected by the multiple types of sensors, remote sensing instruments and external communication terminals are each used as a data source;

[0039] Sensor data is collected through various types of sensors, remote sensing image data is collected through remote sensing instruments, and external communication data is obtained through external communication terminals.

[0040] It should be further explained that, in a specific implementation process, the process of preprocessing the data information of multiple data sources to generate modal data corresponding to each data source includes:

[0041] The information preprocessing performed on the data information of the multiple data sources includes data pre-screening, data cleaning and modal file encapsulation, and the execution order of data pre-screening, data cleaning and modal file encapsulation is to execute them in sequence;

[0042] Pre-screen the data information corresponding to different data sources, and then filter out the data information of various data sources that do not meet the preset qualified value range, namely sensor data, remote sensing image data and external communication data;

[0043] A data cleaning program is set, wherein the data cleaning program is used to clean the sensor data, remote sensing image data and external communication data, and cleaning fields of the sensor data, remote sensing image data and external communication data are defined, wherein the data content portion corresponding to the cleaning field is data information that needs to be removed through data cleaning; and then the data cleaning program cleans and removes the sensor data, remote sensing image data and external communication data according to the defined cleaning fields, and then generates corresponding standard sensor data, standard image data and standard communication data;

[0044] Setting a modal text, the initial text content of which is empty, and the modal text is used to encapsulate the modal files of the standard sensing data, the standard imaging data, and the standard communication data, and then generate modal data corresponding to each data source, wherein the modal data includes sensing modal data, imaging modal data, and communication modal data;

[0045] Among them, the sensing modality data is generated by encapsulating the standard sensing data into a modality file, while the image modality data and the communication modality data are generated by encapsulating the standard image data and the standard communication data into a modality file respectively;

[0046] Obtain the file data volumes of the standard sensing data, the standard image data, and the standard communication data before modal file encapsulation, and record them as D1, D2, and D3, respectively; obtain the corresponding file data volumes after the modal file encapsulation, and record them as D1`, D2`, and D3`;

[0047] Then, the respective packaging loss data volumes are obtained and recorded as Sh1, Sh2 and Sh3 respectively. Sh1=|D1-D1`|, Sh2=|D2-D2`|, Sh3=|D3-D3`|, and a re-packaging threshold is set, recorded as C. When Sh1≥C, Sh2≥C, Sh3≥C exists, the data information of the corresponding data source is re-encapsulated in the modal file. Otherwise, no operation is performed.

[0048] It should be further explained that, in a specific implementation process, the process of extracting modal features from modal data and then obtaining modal information features corresponding to the modal data includes:

[0049] A feature extraction library is set up for modal feature extraction, wherein the feature extraction library is pre-entered with feature extraction fields corresponding to the sensor modal data, the image modal data, and the communication modal data in the modal data, and the feature extraction field corresponding to each modal data is associated with a corresponding extraction similarity;

[0050] The modal data is input into the feature extraction library, which then divides the modal data of each data source into several modal fields. The feature extraction field matched by the modal data in the feature extraction library is used as the benchmark field, and the field similarity between each modal field and the benchmark field is obtained. If the field similarity is greater than or equal to the extraction similarity of the benchmark field, the corresponding modal field is used as the feature extraction element, and all the feature extraction elements are summarized as the modal information features corresponding to the modal data. The above operation is repeated to obtain the modal feature information of the sensor modal data, the imaging modal data, and the communication modal data.

[0051] It should be further explained that, in the specific implementation process, the process of setting fusion layers of different dimensions and mapping modal information features to different fusion layers includes:

[0052] Setting fusion layers of different dimensions, the fusion layers of different dimensions serve as data carriers of modal feature information corresponding to different modal data, and the dimensions of the fusion layers include high-dimensional fusion layers, general-dimensional fusion layers, and low-dimensional fusion layers;

[0053] The mapping object of the high-dimensional fusion layer is sensor modality data, the mapping object of the general-dimensional fusion layer is image modality data, and the mapping object of the low-dimensional fusion atlas is communication modality data;

[0054] The sensor modality data, the imaging modality data, and the communication modality data are convolved by a convolutional neural network algorithm to obtain the corresponding convolution mapping coefficients, which are respectively recorded as J1, J2, and J3. The high-dimensional fusion layer, the general-dimensional fusion layer, and the low-dimensional fusion layer mapping qualified values ​​are set and recorded as R1, R2, and R3, respectively.

[0055] When J1≥R1, the modal information features corresponding to the sensor modal data are mapped to the high-latitude fusion layer, and the current modal information features are used as the first fusion features. When J2≥R2, the modal information features corresponding to the image modal data are mapped to the general latitude fusion layer, and the current modal information features are used as the second fusion features. When J3≥R3, the modal information features corresponding to the image modal data are mapped to the low-latitude fusion layer, and the current modal information features are used as the third fusion features.

[0056] It should be further explained that, in the specific implementation process, the process of merging different fusion layers and their corresponding modal information features to generate multi-source fusion data includes:

[0057] Get the layer channel numbers corresponding to the high-latitude fusion layer, the general-latitude fusion layer, and the low-latitude fusion layer, and record them as TD1, TD2, and TD3 respectively. Cut each fusion layer into three equal sub-layer areas, and mark them as the left half layer, the middle layer, and the right half layer in the order of the fusion layers from left to right.

[0058] Obtain the data information amount of the modal information features corresponding to the fusion layer of each dimension in each sub-layer area, and use the same sub-layer area of ​​the fusion layers of different dimensions as the object area to be fused, that is, the left half layer of the high-dimensional fusion layer, the general latitude fusion layer and the low-latitude fusion layer as an object area to be fused, the middle area layer as an object area to be fused, and the right half layer as an object area to be fused;

[0059] The fusion layer of the dimension with the highest data information content in each object area to be fused is used as the reference layer, and the number of channels of the fusion layers of the other two dimensions is converted into the number of channels corresponding to the reference layer. The number of channels after conversion satisfies TD1=TD2=TD3. After the number of channels is unified, the first fusion features, the second fusion features and the third fusion features corresponding to different fusion layers are merged and fused into multi-source fusion data.

[0060] It should be further explained that, in the specific implementation process, the process of parsing the multi-source fusion data and generating corresponding parsing information includes:

[0061] Setting a data analysis program, wherein a plurality of analysis segments are pre-entered in the data analysis program, and the types of the plurality of analysis segments include feature keyword analysis segments, target recognition analysis segments, and time series trend analysis segments;

[0062] Input the multi-source fusion data into a data analysis program, and then perform data analysis on the multi-source fusion data. The data analysis program sets a fragment analysis length, and divides the multi-source fusion data into a number of data fragments to be analyzed according to the fragment analysis length;

[0063] Aggregate all the data segments to be parsed that meet the feature keyword parsing segment into text parsing information, aggregate all the data segments to be parsed that meet the target recognition parsing segment into image recognition parsing information, and aggregate all the data segments to be parsed that meet the time series trend parsing segment into communication evolution trend parsing information;

[0064] The final analysis information is generated by integrating text analysis information, image recognition analysis information and communication evolution trend analysis information.

[0065] It should be further explained that, in the specific implementation process, the process of judging whether there is dangerous information in the parsed information and executing different data decision-making schemes according to the judgment results includes:

[0066] Set the corresponding danger determination conditions of the text analysis information, the image recognition analysis information and the communication evolution trend analysis information in the analysis information, set the danger determination model, input the analysis information and all the danger determination conditions into the danger determination model, and then the danger determination model outputs the corresponding determination code after making a determination according to each danger determination condition;

[0067] The determination code includes "0" and "1". When the text analysis information, image recognition analysis information and communication evolution trend analysis information do not meet their own risk determination conditions, their respective determination codes are set to "0". When any one of them meets its corresponding risk determination condition, its determination code value is set to "1".

[0068] splicing the judgment codes output by the danger judgment model according to the text analysis information, the image recognition analysis information and the communication evolution trend analysis information, thereby generating a judgment result character string, wherein the judgment result character string has values ​​of "000", "001", "011" and "111";

[0069] It should be noted that "010", "100" and "001" are equivalent, and the order of arrangement is not distinguished here. It only means that there is only one text analysis information, image recognition analysis information and communication evolution trend analysis information in the analysis information that meets the current danger judgment condition. Similarly, "011" is equivalent to "110" and "101";

[0070] When the judgment result string value is "001" or "011" or "111", it is judged that there is dangerous information in the parsed information, and the data decision to be executed is: locate the specific position of the drone in the drone cluster with dangerous information, and inspect the drone;

[0071] When the judgment result string value is "000", it is judged that there is no dangerous information in the current parsed information, and the executed data decision is: continuously monitor the operation of the drone cluster and obtain multi-source fusion data.

[0072] The above embodiments are only used to illustrate the technical method 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 method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A data information analysis method based on multi-source data, characterized in that: The following steps are involved: Step S1: setting up a drone cluster, and collecting data information from multiple data sources through the drone cluster, and performing information preprocessing on the data information from multiple data sources to generate modal data corresponding to each data source; Step S2: extracting modal features from the modal data, and then obtaining modal information features corresponding to the modal data, setting fusion layers of different dimensions, mapping the modal information features to different fusion layers, and then merging different fusion layers and their corresponding modal information features to generate multi-source fusion data; Step S3: Analyze the multi-source fusion data to generate corresponding analysis information, determine whether there is dangerous information in the analysis information, and execute different data decision plans based on the judgment results.

2. A data information analysis method based on multi-source data according to claim 1, characterized in that: The process of setting up a drone swarm to collect data from multiple data sources includes: Obtain the communication rights of several drones, and number them as i, i=1, 2, 3, ..., n, where n is a natural number greater than 0. The communication rights of each drone are obtained through its corresponding communication code. The communication code corresponding to the drone numbered i is TX[i]. Starting from i=1, the drone with i=1 is used as the communication object, and the drones corresponding to other numbers are connected to the communication object, thereby establishing data communication between the drone corresponding to the current communication object and another n-1 drones. Continue to take the drone corresponding to i=2 as the communication object, and connect the other n-2 drones that have not been used as communication objects to the current communication object. Repeat the above operation until each drone establishes data communication with drones other than itself, thereby forming a drone cluster consisting of several drones. Several types of sensors, remote sensing instruments and external communication terminals are deployed on the drones corresponding to the drone cluster, and data information from multiple data sources is collected through the drone cluster, wherein the data information collected by the multiple types of sensors, remote sensing instruments and external communication terminals are each used as a data source, sensor data is collected through multiple types of sensors, remote sensing image data is collected through remote sensing instruments, and external communication data is obtained through external communication terminals.

3. The data information analysis method based on multi-source data according to claim 2 is characterized in that: The process of preprocessing the data information of multiple data sources to generate modal data corresponding to each data source includes: The information preprocessing includes data pre-screening, data cleaning and modal file encapsulation, data pre-screening is performed on data information corresponding to different data sources, and then data information of multiple data sources that do not meet the preset qualified value range is screened out, and a data cleaning program is set to clean sensor data, remote sensing image data and external communication data, and cleaning fields of sensor data, remote sensing image data and external communication data are defined, and the data content part corresponding to the cleaning field is the data information that needs to be eliminated through data cleaning; and then the data cleaning program cleans and eliminates part of the data information that does not meet the requirements of the sensor data, remote sensing image data and external communication data according to the defined cleaning fields, and then generates corresponding standard sensor data, standard image data and standard communication data; The modal text is set to perform modal file encapsulation of standard sensor data, standard image data and standard communication data, and then generate modal data corresponding to each data source. The modal data includes sensor modal data, image modal data and communication modal data. The file data volume of standard sensor data, standard image data and standard communication data before modal file encapsulation is obtained, and recorded as D1, D2 and D3 respectively. The corresponding file data volume after each modal file encapsulation is obtained, and recorded as D1`, D2` and D3`. Then, the respective encapsulation loss data volume is obtained, and recorded as Sh1, Sh2 and Sh3 respectively. Sh1=|D1-D1`|, Sh2=|D2-D2`|, Sh3=|D3-D3`|, and a re-encapsulation threshold is set, recorded as C. When Sh1≥C, Sh2≥C, and Sh3≥C exist, the data information of the corresponding data source is re-encapsulated in modal file. Otherwise, no operation is performed.

4. The data information analysis method based on multi-source data according to claim 3 is characterized in that: The process of extracting modal features from modal data and obtaining modal information features corresponding to the modal data includes: A feature extraction library is set up for modal feature extraction. The feature extraction library is pre-entered with feature extraction fields corresponding to the sensor modal data, imaging modal data, and communication modal data in the modal data. The feature extraction field corresponding to each modal data is associated with a corresponding extraction similarity. The modal data is input into the feature extraction library, and then the feature extraction library divides the modal data of each data source into several modal fields. The feature extraction field matched to the modal data in the feature extraction library is used as the benchmark field, and the field similarity between each modal field and the benchmark field is obtained. If the field similarity is greater than or equal to the extraction similarity of the benchmark field, the corresponding modal field is used as a feature extraction element, and all feature extraction elements are summarized as the modal information features corresponding to the modal data. The above operation is repeated to obtain the modal feature information of the sensor modal data, imaging modal data, and communication modal data.

5. The data information analysis method based on multi-source data according to claim 4 is characterized in that: The process of setting fusion layers of different dimensions and mapping modal information features to different fusion layers includes: Set fusion layers of different dimensions. The dimensions of the fusion layers include high-dimensional fusion layers, general-dimensional fusion layers, and low-dimensional fusion layers. Perform convolution processing on the sensor modality data, image modality data, and communication modality data through the convolutional neural network algorithm to obtain the corresponding convolution mapping coefficients, which are recorded as J1, J2, and J3 respectively. Set the mapping qualified values ​​of the high-dimensional fusion layer, general-dimensional fusion layer, and low-dimensional fusion layer, which are recorded as R1, R2, and R3 respectively. When J1≥R1, the modal information features corresponding to the sensor modal data are mapped to the high-latitude fusion layer, and the current modal information features are used as the first fusion features. When J2≥R2, the modal information features corresponding to the image modal data are mapped to the general latitude fusion layer, and the current modal information features are used as the second fusion features. When J3≥R3, the modal information features corresponding to the image modal data are mapped to the low-latitude fusion layer, and the current modal information features are used as the third fusion features.

6. A data information analysis method based on multi-source data according to claim 5, characterized in that: The process of generating multi-source fusion data includes: Obtain the number of layer channels corresponding to the high-latitude fusion layer, the general-latitude fusion layer, and the low-latitude fusion layer, and record them as TD1, TD2, and TD3 respectively. Cut each fusion layer into three equal sub-layer areas, and mark them as the left half layer, the middle layer, and the right half layer in the order of the fusion layers from left to right. Obtain the data information volume of the modal information features corresponding to the fusion layer of each dimension in each sub-layer area, and use the same sub-layer area of ​​fusion layers of different dimensions as the object area to be fused; That is, the left half layer of the high-dimensional fusion layer, the general latitude fusion layer and the low-latitude fusion layer is used as an object area to be fused, the middle layer of each is used as an object area to be fused, and the right half layer of each is used as an object area to be fused. The fusion layer of the dimension with the highest data information in each object area to be fused is used as the reference layer, and the number of channels of the fusion layers of the other two dimensions is converted to the number of channels corresponding to the reference layer. The number of channels after conversion satisfies TD1=TD2=TD3. After the number of channels is unified, the first fusion features, the second fusion features and the third fusion features corresponding to different fusion layers are merged and fused into multi-source fusion data.

7. The data information analysis method based on multi-source data according to claim 6, characterized in that: The process of parsing multi-source fusion data and generating parsing information includes: A data analysis program is set up, in which a number of analysis segments are pre-entered, and the types of analysis segments include feature keyword analysis segments, target recognition analysis segments, and time series trend analysis segments. Multi-source fusion data is input into the data analysis program, and then data analysis of the multi-source fusion data is performed. The data analysis program sets the segment analysis length, and divides the multi-source fusion data into a number of data segments to be analyzed according to the segment analysis length; All data segments to be parsed that meet the feature keyword parsing segment are summarized into text parsing information, all data segments to be parsed that meet the target recognition parsing segment are summarized into image recognition parsing information, all data segments to be parsed that meet the time series trend parsing segment are summarized into communication evolution trend parsing information, and the text parsing information, image recognition parsing information and communication evolution trend parsing information are integrated to generate the final parsing information.

8. The data information analysis method based on multi-source data according to claim 7, characterized in that: The process of judging whether there is dangerous information in the parsed information and executing different data decision-making plans according to the judgment results includes: Set the corresponding danger determination conditions for the text analysis information, image recognition analysis information and communication evolution trend analysis information, set the danger determination model, input the analysis information and all the danger determination conditions into the danger determination model, and then the danger determination model outputs the corresponding determination code after determination according to each danger determination condition. The determination code includes "0" and "1". When the text analysis information, image recognition analysis information and communication evolution trend analysis information do not meet their own danger determination conditions, their respective determination codes are set to "0". When any one of them meets its corresponding danger determination condition, its determination code value is set to "1"; The judgment code is spliced ​​to generate a judgment result string. The judgment result string takes values ​​of "000", "001", "011" and "111". When the judgment result string takes the value of "001" or "011" or "111", it is judged that there is dangerous information in the parsed information, and the data decision to be executed is: locate the specific position of the drone in the drone cluster where the dangerous information exists, and inspect the drone. When the judgment result string takes the value of "000", it is judged that there is no dangerous information in the current parsed information, and the data decision to be executed is: continuously monitor the operation of the drone cluster and obtain multi-source fusion data.

Images