An information security inspection and detection method for drone inspection
By building information security reference sets and inspection sets, and using dimensionality reduction and abnormal detection technologies, real-time information security inspection of drones is realized, solving the problem that information security inspection in drone inspection affects battery life and cannot cope with complex network attacks, and realizing the problem of information security inspection in drone inspections. Efficient information security independent detection.
Patent Information
- Application Number
- CN202411178013.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-27
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-08-27
AI Technical Summary
Information security inspection methods during drone inspection affect battery life and cannot effectively deal with complex cyber attacks.
Using information security inspection and detection methods for drone patrol, by constructing information security reference sets and information security inspection sets, using PCA and LOF algorithms for dimensionality reduction and abnormal detection, collecting drone information in real time for security detection, and based on the detection The results update the inspection set, reduce the communication frequency with the server, and realize independent information security inspection.
Real-time information security inspection of drones during flight is realized, avoiding the impact on endurance, and effectively responding to changing cyber attacks, reducing the risk of information leakage, and improving the autonomy of information security.
Smart Images

Figure CN118709115B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing technology, and in particular to an information security inspection and detection method for unmanned aerial vehicle inspections. Background Art
[0002] Drone inspections have been widely used in many fields such as electricity, petroleum, agriculture, and traffic monitoring. Since drones may be attacked by cyber attacks during the inspection process, stealing drone inspection data, or even causing drones to transmit erroneous data or lose control and deviate from the inspection trajectory, it is necessary to conduct information security inspections on drones. The usual inspection method is to obtain the drone's log information after the drone inspection is completed, and use the log information to check whether there are security vulnerabilities, such as whether the data communication traffic is abnormal, whether the file access permissions are abnormal, etc. If information security inspections are performed during the drone inspection process, the log information needs to be transmitted back in real time, which not only increases the drone's computing power and affects the drone's endurance, but also further increases the risk of drone information leakage.
[0003] One method of self-checking the information security of drones is to store risk-free log template information in the drone, and compare the log information of the drone during the inspection process with the log template information to determine whether there is a security vulnerability. However, the detection results of this information security inspection method are too dependent on the log template information and cannot adapt to the complex network attack behaviors during the inspection process. Summary of the invention
[0004] In order to solve the problem that the information security inspection method during the inspection of UAV affects the endurance of the UAV or the inspection method is unable to cope with complex network attack behaviors, the present invention provides an information security inspection and detection method for UAV inspection.
[0005] The information security inspection and detection method for drone inspection of the present invention adopts the following technical solution:
[0006] An embodiment of the present invention provides an information security inspection and detection method for drone inspection, the method comprising the following steps:
[0007] S1: The drone information at all times during the historical inspection process is recorded as the information security reference set;
[0008] S2: Generate an information security verification set based on the information security reference set, including:
[0009] Obtain the inspection trajectory of the drone during the inspection, obtain the data dimension D when the drone information at each moment in the information security reference set has the largest correlation with the inspection trajectory, and obtain the dimension reduction result when all drone information in the information security reference set is reduced to D dimensions, which is recorded as the information security test set;
[0010] S3: Use the information security verification set to perform security checks on the drone information collected in real time, and update the information security verification set, including:
[0011] The drone information collected in real time is also reduced to D dimension to obtain a new dimension reduction result, and the new dimension reduction result and the dimension reduction result in the information security test set are used together for anomaly detection to obtain an abnormal dimension reduction result, and the drone information corresponding to all abnormal dimension reduction results obtained within a preset time period is sent to the server for security judgment. When it is judged to be safe, the abnormal dimension reduction result is recorded as Class I data, otherwise it is recorded as Class II data; calculate the distribution law of Class I data relative to Class II data; and update the information security test set according to the distribution law;
[0012] The updating of the information security verification set comprises the steps of: adding the drone information corresponding to the first type of points to the information security reference set according to the distribution law, removing the drone information corresponding to the second type of points from the information security reference set, and regenerating an updated information security verification set according to the information security reference set;
[0013] S4: Repeat the steps included in S3 according to the updated information security inspection set until the drone inspection is completed.
[0014] Preferably, the drone information at each moment includes a control vector, a data access vector, and a context vector;
[0015] Among them, the control instructions received by the drone at each moment are recorded as control vectors, the behavior of reading and querying data in the drone at each moment is recorded as data access vectors, and all data collected by the drone within the preset neighborhood time period at each moment is recorded as context vectors.
[0016] Preferably, the step of obtaining the data dimension D when the correlation between the drone information and the inspection trajectory at each moment in the information security reference set is the largest includes the following specific steps:
[0017] Several data dimensions are set. For any data dimension K, the PCA algorithm is used to reduce the dimension of the drone information at all times in the information security reference set to K dimensions, and the dimension reduction result of the drone information at each time under each data dimension is obtained; the correlation coefficient between the dimension reduction result of the drone information at all times under each data dimension and the inspection trajectory is calculated, and recorded as the degree of correlation under each data dimension; the data dimension with the largest degree of correlation is recorded as data dimension D.
[0018] Preferably, the calculation of the distribution law of the first type of data relative to the second type of data includes the following specific steps:
[0019] All dimensionality reduction results and first-class data and second-class data in the information security test set are regarded as data points, among which first-class data are regarded as first-class data points and second-class data are regarded as second-class data points;
[0020] In the point set composed of all the data points, the average Euclidean distance Q1 between all the first-class points and the center of the point set, the average Euclidean distance Q2 between the second-class points and the center of the point set, and the average number of second-class points Q3 in the preset neighborhood of each first-class point are obtained. As the distribution law.
[0021] Preferably, the adding of the drone information corresponding to the first type of points to the information security reference set and removing the drone information corresponding to the second type of points from the information security reference set according to the distribution law comprises the following specific steps:
[0022] Obtain the incompleteness of the information security reference set, the incompleteness is respectively related to , There is a positive correlation;
[0023] When the incompleteness is greater than the preset threshold, the drone information corresponding to the first type of points is added to the information security reference set, and the drone information corresponding to the second type of points is removed from the information security reference set.
[0024] Preferably, the method adds the drone information corresponding to the first type of points to the information security reference set according to the distribution law, removes the drone information corresponding to the second type of points from the information security reference set, and generates an updated information security verification set according to the information security reference set, and then includes the following specific steps:
[0025] The data dimension of the dimension reduction result in the updated information security inspection set is obtained, recorded as D1, and the preset threshold is reduced according to the difference between D1 and D, and the reduction amplitude is positively correlated with the difference between D1 and D.
[0026] Preferably, the specific calculation formula of the preset threshold is as follows:
[0027]
[0028] in , indicating the amplitude of the reduction, Indicates the preset threshold value, Indicates the value of th1 after being reduced.
[0029] Preferably, the logs and database of the server side contain drone information. When the drone information corresponding to the abnormal dimensionality reduction result is contained in the logs and database of the server side, it is judged to be safe; if the drone information corresponding to the abnormal dimensionality reduction result is not contained in the logs and database of the server side, it is judged to be unsafe.
[0030] Preferably, the center of the point set is the mean of all data points other than the abnormal dimensionality reduction result.
[0031] Preferably, the anomaly detection uses a LOF algorithm.
[0032] The beneficial effects of the technical solution of the present invention are:
[0033] The present invention forms an information security reference set with drone information and generates an information security verification set. The drone uses the information security verification set to perform information security self-inspection, and the server performs verification at the same time. Specifically, it includes: obtaining the data dimension D when the drone information at each moment in the information security reference set has the largest correlation with the inspection trajectory, and obtaining the dimension reduction result when all drone information in the information security reference set is reduced to D dimensions, which is recorded as the information security verification set. The drone information collected in real time is also reduced to D dimensions to obtain a new dimension reduction result, and the new dimension reduction result and the dimension reduction result in the information security verification set are used together for abnormal detection to obtain abnormal dimension reduction results, and the drone information corresponding to all abnormal dimension reduction results obtained within a preset time period is sent to the server for security judgment.
[0034] On the one hand, this process achieves the purpose of real-time information security inspection of the UAV during flight. On the other hand, the self-inspection of the UAV's information security avoids the problem of affecting the UAV's endurance when the UAV has to communicate securely with the server at all times.
[0035] Furthermore, the present invention utilizes an information security verification set to perform security detection on drone information collected in real time, and updates the information security verification set, specifically including: when judged to be safe, recording the abnormal dimensionality reduction result as Class I data, otherwise recording it as Class II data; calculating the distribution law of Class I data relative to Class II data, adding the drone information corresponding to Class I points to the information security reference set according to the distribution law, removing the drone information corresponding to Class II points from the information security reference set, and regenerating an updated information security verification set according to the information security reference set, and subsequently continuously updating the information security verification set and performing security inspections until the drone inspection is completed.
[0036] During the above process, the drone will send the drone information of abnormal dimensionality reduction results to the server for security judgment at regular intervals, so as to avoid taking up too many computing resources on the basis of knowing whether there is a network attack behavior. At the same time, the update process of the information security test set does not need to be carried out in real time. Therefore, the update of the information security test set and the data verification on the server are combined to avoid the problem of information security inspection affecting the endurance of the drone as much as possible.
[0037] More importantly, with the continuous updating of the information security test set, when the dimensionality reduction results in the information security test set can describe the most relevant relationship between the drone information and the inspection trajectory as much as possible, it is ensured that the information security test set can cope with the changing network attack behaviors, and finally the drone information corresponding to the abnormal dimensionality reduction results can be directly regarded as unsafe data, or when the drone has abnormal dimensionality reduction results, it means that the drone has a network attack behavior. At this time, it is no longer necessary to conduct communication verification with the server, and the drone can achieve completely autonomous information security inspection. While coping with the changing network attack environment, it further avoids the problem of information security inspection affecting the drone's endurance. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0039] Figure 1 A flowchart of the steps of an information security inspection and detection method for drone inspections provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0040] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the information security inspection and detection method for drone inspection proposed by the present invention, its specific implementation method, structure, characteristics and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.
[0041] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
[0042] The following is a detailed description of a specific scheme of an information security inspection and detection method for drone inspections provided by the present invention in conjunction with the accompanying drawings.
[0043] Embodiment 1:
[0044] See also Figure 1 , which shows a flowchart of an information security inspection and detection method for drone inspection provided by an embodiment of the present invention, the method comprising the following steps:
[0045] Step S001: The drone information at all times during the historical inspection process is recorded as an information security reference set.
[0046] For the same inspection task, such as inspecting a section of transmission line in a high-voltage power grid, or inspecting a section of an oil pipeline that transports oil, the drone information generated by the drone at each moment when the inspection task is executed in the historical process is obtained, and the drone information generated at all moments is recorded as an information security reference set.
[0047] When no inspection task has been executed in history, the drone is ordered to execute the current inspection task. When part of the current inspection task is completed, such as 30 minutes or half of the current inspection task, the drone information generated at all times of the executed inspection tasks is recorded as the information security reference set. When there are a large number of inspection tasks in history, a large amount of drone information may be generated. In this case, in order to reduce the amount of data, only the drone information generated by the most recent inspection task can be retained, for example, only the drone information generated by the most recent three inspection tasks can be retained.
[0048] In this embodiment, drone information is collected every 1 minute (that is, every 1 minute is a moment).
[0049] As a preferred example, the drone information at each moment includes a control vector, a data access vector, and a context vector.
[0050] Specifically, the control instructions received by the drone at each moment are obtained. The control instructions are action instructions executed by the drone when the drone is remotely controlled or automatically controlled, including movement direction, movement speed, and flight altitude. Other embodiments also include the camera gimbal angle installed on the drone, the fuselage pitch angle, etc.; these control vectors constitute a vector, which is recorded as a control vector.
[0051] The behavior of reading or querying the data in the drone at each moment is obtained, including: the address of the data being read, the database query statement used when querying the database; in other embodiments, it also includes: the permission level of the accessed data, the time when the data was last updated, the average access flow of the data at each moment, etc. These behaviors are recorded as data access vectors.
[0052] Acquire all data collected by the drone within a preset neighborhood time period at each moment, for example, within a time period consisting of two moments before and after each moment. In other embodiments, it may also be a time period consisting of two moments before each moment or two moments after each moment (including each moment). These data are arranged in the order of time generated or spliced into a vector, which is recorded as a context vector.
[0053] It should be noted that the above acquired data includes image data and text data, etc., wherein the image data is expanded into a column in the order of pixels from top to bottom and from left to right. The text data needs to be converted into word vectors (or converted into one-hot vectors) through word2vec technology. In other embodiments, the purpose of reducing the amount of data can be achieved by reducing the image data in proportion.
[0054] In some other embodiments, the drone information at each moment also includes: whether the camera is turned on, whether the temperature sensor, pressure sensor and other sensors are enabled, the data output by each sensor, etc. This embodiment will not list them one by one.
[0055] The above control vector, data access vector, and context vector are concatenated together to form the drone information at each moment.
[0056] As an optional example, the method for obtaining drone information at each moment includes:
[0057] Obtain all log information generated within the preset neighborhood time period at each moment, convert all log information into word vectors using word2vec technology, and concatenate all word vectors together to form the drone information at each moment.
[0058] Since not all information in the log information is useful for information security verification, nor does it necessarily comprehensively include all information useful for information security verification, the information security reference set obtained by this method is not optimal.
[0059] Step S002: Generate an information security verification set based on the information security reference set.
[0060] The information security reference set obtained above is used to describe the specific working conditions of the drone at each moment during the historical inspection process. In this embodiment, the drone implements self-security verification based on the information security reference set.
[0061] Considering that not all data contained in the information security reference set is related to security verification, it is necessary to generate an information security verification set based on the information security reference set.
[0062] The specific method is:
[0063] (1) Obtaining the inspection trajectory of the drone during inspection; the inspection trajectory refers to the inspection trajectory when executing the same inspection task, such as the inspection task when inspecting the same section of the oil pipeline multiple times. The inspection task of inspecting other sections of the oil pipeline is not the same inspection task.
[0064] It should be noted that before the UAV performs an inspection mission, it is necessary to set the inspection trajectory for the inspection mission. The inspection trajectory determines the approximate movement trajectory of the UAV. The control instructions executed by the UAV during the inspection process and the data access or collected data are all determined by the inspection trajectory and the scene under the inspection trajectory (such as the natural environment of the power grid transmission line or oil pipeline).
[0065] For each inspection task in the historical process, the position of the drone at each moment is obtained when the inspection task is executed, and the position sequence composed of the positions at all moments is used as the inspection sequence of each inspection task. In this embodiment, the inspection sequences of all inspection tasks in the historical process are averaged to obtain the average inspection sequence, which is recorded as the inspection trajectory of the drone during the inspection. In other embodiments, an inspection task can be manually specified from all inspection tasks in the historical process (for example, an inspection task that best meets the inspection expectations is specified), and the inspection sequence of the inspection task is recorded as the inspection trajectory of the drone during the inspection.
[0066] (2) Obtain the data dimension D when the correlation between the drone information and the inspection trajectory at each moment in the information security reference set is the largest, and obtain the dimension reduction result when all drone information in the information security reference set is reduced to D dimensions, which is recorded as the information security test set.
[0067] This process takes into account that the inspection trajectory and the natural scene under the inspection trajectory (or the inspection target) determine the drone information to a certain extent (for example, how the drone moves, what data the drone needs to read, store, access and collect), and network attacks will reduce the association or constraint relationship between drone information and the inspection trajectory.
[0068] The dimensionality reduction result in the information security test set represents the data contained in the drone information in the information security reference set that is most relevant to the inspection trajectory and can be used to test network attack behaviors. The data dimension D can ensure that: while ensuring that the information security test set contains the most relevant data to the inspection trajectory, on the one hand, it reduces the amount of data and saves computing resources, thereby ensuring the endurance of the drone, and on the other hand, more importantly, avoids introducing too much irrelevant interference information during information security inspection.
[0069] As a preferred example, obtaining the data dimension D when the correlation between the drone information and the inspection trajectory at each moment in the information security reference set is the largest includes the following steps:
[0070] This embodiment sets several data dimensions, which are 5, 8, 11, ..., 62. Other data dimensions may be set in other embodiments, and this embodiment does not specifically limit them.
[0071] For any data dimension K, the PCA algorithm is used to reduce the dimension of the drone information at all times in the information security reference set to K dimensions, and the dimension reduction results of the drone information at each time under each data dimension are obtained; the typical correlation coefficient between the dimension reduction results of the drone information at all times under each data dimension and the inspection trajectory is calculated, and recorded as the degree of correlation under each data dimension; the data dimension with the largest degree of correlation is recorded as data dimension D.
[0072] As an optional example, the steps of obtaining the data dimension D when the correlation between the drone information and the inspection trajectory at each moment in the information security reference set is the largest are as follows:
[0073] Delete one dimension of all drone information in the information security reference set (for example, randomly or from left to right), then calculate the typical correlation coefficient between the dimensionality reduction results of drone information at all times and the inspection trajectory, then delete one dimension of all drone information in the information security reference set, continue to calculate the typical correlation coefficient between the dimensionality reduction results of drone information at all times and the inspection trajectory, repeat this process until only one dimension of drone information remains, and then calculate the typical correlation coefficient between the dimensionality reduction results of drone information at all times and the inspection trajectory again. In this process, the dimension of drone information retained when the typical correlation coefficient is the largest is obtained, which is recorded as data dimension D. At this time, all drone information under data dimension D constitutes the information security inspection set.
[0074] However, since this process directly deletes some dimensions, some information that is highly correlated with the inspection trajectory will be lost, resulting in a suboptimal information security verification set. However, this process has a small amount of computation and can further ensure the endurance of the drone.
[0075] At this point, the drone information with the greatest correlation with the inspection trajectory during the historical inspection process is constructed into an information security inspection set.
[0076] Step S003: Use the information security verification set to perform security checks on the newly collected drone information, and update the information security verification set.
[0077] (1) The drone collects drone information in real time during the current inspection mission, and also uses PCA to reduce the dimension of the drone information into D dimensions to obtain a new dimension reduction result.
[0078] It should be noted that, since the process of obtaining the above-mentioned dimensionality reduction results in this embodiment uses the PCA dimensionality reduction algorithm, it can be seen from the PCA dimensionality reduction algorithm that the algorithm will obtain multiple principal component directions (each principal component direction is a unit vector), each principal component direction corresponds to a variance contribution (that is, the maximum projection variance), of which only the D principal component directions with the largest variance contribution are retained, and the above-mentioned dimensionality reduction result is the vector formed by the projection of the drone information in the D principal component directions. This embodiment uses the vector formed by the projection of the real-time collected drone information in the D principal component directions as the new dimensionality reduction result. PCA is a well-known technology, and this embodiment will not repeat its specific process.
[0079] In other embodiments, if the process of obtaining the above-mentioned dimensionality reduction result is achieved by deleting some dimensions (as described in step S002), the real-time collection of drone information can also obtain a new dimensionality reduction result by deleting the corresponding dimensions.
[0080] (2) The new dimensionality reduction results are used together with the dimensionality reduction results in the information security test set for anomaly detection to obtain the abnormal dimensionality reduction results.
[0081] The anomaly detection algorithm is the LOF algorithm. Other anomaly detection methods may also be used in other embodiments, such as performing DBSCAN clustering on the dimensionality reduction results, and treating isolated dimensionality reduction results without corresponding categories as abnormal dimensionality reduction results.
[0082] The abnormal dimensionality reduction results obtained refer to the dimensionality reduction results that are significantly different from most of the dimensionality reduction results (including new dimensionality reduction results). They describe the dimensionality reduction results that have a relatively small correlation with the inspection trajectory. This dimensionality reduction result may be due to a network attack, resulting in a weak correlation with the inspection trajectory.
[0083] For example, when a drone is inspecting, in a certain inspection track, the drone needs to query the template image stored in the database to match whether there is a specific target object (such as pedestrians, buildings, substations, power towers, etc.), but there are no specific targets in some inspection tracks, so there is no need to query the database. When the database is queried in the inspection track that does not require the database to be queried (that is, the dimensionality reduction result of the drone information in the inspection track loses its association with the inspection track), there may be a network attack (that is, the drone information corresponding to the abnormal dimensionality reduction result may be attacked by the network).
[0084] For example, some inspection trajectories do not require continuous collection of a large number of images and transmission back, but continuous data transmission occurs under this trajectory, resulting in excessive data traffic (at this time, the dimensionality reduction result of the drone information under the inspection trajectory loses its association with the inspection trajectory), and there may be a network attack (that is, the drone information corresponding to the abnormal dimensionality reduction result may be attacked by the network).
[0085] Even if the drone deviates from the inspection trajectory, the dimensionality reduction result of the drone information under the inspection trajectory loses its correlation with the inspection trajectory, and there may be a cyber attack.
[0086] In summary, the above-mentioned new dimension reduction result is used together with the information security test set for anomaly detection, and the drone information corresponding to the abnormal dimension reduction result obtained may be subject to network attacks. Therefore, in this embodiment, when the above-mentioned new dimension reduction result is an abnormal dimension reduction result, it means that the drone has self-checked its own information security loopholes.
[0087] Step S004: Send the drone information corresponding to all abnormal dimensionality reduction results obtained within a preset time period to the server for safety judgment. If it is judged to be safe, the abnormal dimensionality reduction results are recorded as Class I data, otherwise they are recorded as Class II data; calculate the distribution law of Class I data relative to Class II data; and update the information security verification set according to the distribution law;
[0088] The above process uses the drone's self-inspection to determine in real time whether the drone has any network attack behavior or security vulnerabilities (that is, the dimensionality reduction result corresponding to the drone information collected in real time is an abnormal dimensionality reduction result).
[0089] It should be noted that when the above new dimensionality reduction result is not an abnormal dimensionality reduction result, it means that the drone has not self-detected that it has a network attack behavior, but it does not necessarily mean that the drone does not really have a network attack behavior. The reason is that the dimensionality reduction results obtained in the above information security test set are not comprehensive enough, or cannot cover complex and changeable network attack behaviors. Even the drone information corresponding to the dimensionality reduction results in the information security test set is the drone information after the network attack, resulting in the dimensionality reduction results in the information security test set cannot accurately describe the most relevant relationship between the drone information and the inspection trajectory. In addition, some of the abnormal dimensionality reduction results obtained above appear in the information security test set, which is also due to the fact that the dimensionality reduction results in the information security test set cannot accurately describe the most relevant relationship between the drone information and the inspection trajectory.
[0090] Therefore, this embodiment needs to update the information security verification set to ensure that it can cope with the changing network attack behaviors and describe the most relevant association between the drone information and the inspection trajectory.
[0091] The specific method is: all the drone information corresponding to the abnormal dimensionality reduction results obtained in the last few moments (for example, within 30 moments) is sent to the server. The server refers to the terminal that sends control instructions to the drone or exchanges information with the drone, monitors the operation status of the drone, and is responsible for the safety of the entire inspection task and performs inspections according to the preset task objectives. The server records the log information and data information of any data that has interacted with the drone. After the server receives the drone information corresponding to all the abnormal dimensionality reduction results, it determines whether the drone information is stored and recorded in the server's log or database by comparing the log information or querying the database. If there is a record, it means that the drone information corresponding to the abnormal dimensionality reduction result is safe. If not, it means that the drone information corresponding to the abnormal dimensionality reduction result may have been attacked by the network.
[0092] In this embodiment, when it is judged to be safe, the abnormal dimensionality reduction result is recorded as the first category of data, otherwise it is recorded as the second category of data.
[0093] It should be noted that when communicating with the server, encryption and identity authentication technologies are used to ensure the security of communication. Although this process brings additional computing power, encryption and identity authentication technologies are not used for every drone data at every moment. It is only when there are abnormal dimensionality reduction results and communication is performed once every period of time, so it will not affect the drone's endurance. In particular, when the information security verification set is updated to be more accurate and reliable, it can be ensured that when there is no network attack, the drone's endurance will not be reduced due to communication with the server.
[0094] The first type of data obtained above is safe but judged as abnormal data, and the second type of data refers to the data that is unsafe but exists in the information security test set. That is, the dimensionality reduction results in the information security test set cannot accurately describe the most relevant relationship between drone information and inspection trajectories, and cannot cover the variable network attack behaviors, resulting in unreasonable distribution of the first and second types of data, which cannot be used to accurately self-check the information security of drones.
[0095] This embodiment updates the information security verification set according to the distribution law of the first type of data relative to the second type of data to ensure that the updated data set can cope with the changing network attack behaviors and describe the most relevant association between the drone information and the inspection trajectory as much as possible.
[0096] As an optional example, the information security verification set is updated according to the distribution law of the first type of data relative to the second type of data, and the specific process involved is:
[0097] All dimensionality reduction results and first-class data and second-class data in the information security test set are regarded as data points, among which first-class data are regarded as first-class data points and second-class data are regarded as second-class data points;
[0098] Calculate the distance between any pair of Class I data points and Class II data points. The mean of the distance between all Class I data points and Class II data points is recorded as the distribution law of Class I data relative to Class II data. The larger the distribution law, the more obvious the difference between Class I data and Class II data. This means that although the current information security test set uses both Class I data and Class II data as abnormal dimensionality reduction results, the current information security test set can still distinguish Class I data from Class II data. At this time, in order to reduce the amount of calculation, the information security test set does not need to be updated, or it is not time to update the information security test set. When the distribution law is smaller, it means that there is no obvious difference between Class I data and Class II data. The current information security test set cannot distinguish Class I data from Class II data. At this time, the information security test set can no longer be used for self-inspection of security information, and the information security test set needs to be updated.
[0099] The above process determines whether the information security verification set needs to be updated through the distribution pattern of the first type of data relative to the second type of data. On the one hand, it avoids the huge amount of calculation caused by frequent updates of the information security verification set, which affects the endurance of the drone. On the other hand, updating the information security verification set at the right time can ensure the completeness and reliability of the information security verification set.
[0100] It should be noted that when the information security verification set is not updated, the abnormal dimensionality reduction results detected by the drone's self-detection may or may not reflect the existence of a network attack on the drone. In this case, whether a network attack actually exists needs to be sent to the server for judgment.
[0101] Specifically, the method for updating the information security verification set is:
[0102] The Euclidean distance between the two data points with the longest distance in the information security verification set is obtained. When the ratio of the distribution law of the first type of data relative to the second type of data to the Euclidean distance is less than 0.4, the information security verification set is updated, otherwise the information security verification set is not updated. The specific method for updating the information security verification set is: adding the drone information corresponding to the first type of point to the information security reference set, removing the drone information corresponding to the second type of point from the information security reference set, and then regenerating the information security verification set based on the information security reference set to obtain an updated information security verification set. The specific process is the same as step S002. This process requires a large amount of calculation, so the update of the information security verification set cannot be performed frequently (that is, the distribution law of the first type of data relative to the second type of data needs to meet certain conditions, such as updating only when the ratio to the Euclidean distance is less than 0.4).
[0103] As a preferred example, the information security verification set is updated according to the distribution law of the first category data relative to the second category data, and the specific process included is: the mean of the distances between all the first category data points and the second category data points is recorded as the distribution law of the first category data relative to the second category data.
[0104] In the point set composed of all data points, the average Euclidean distance Q1 between all the first-class points and the center of the point set and the average Euclidean distance Q2 between the second-class points and the center of the point set are obtained;
[0105] Get the number of second-class points in the preset neighborhood of each first-class point, and record the mean of the number of second-class points in the preset neighborhood of all first-class points as Q3 (that is, the average number of second-class points in the preset neighborhood of each first-class point). As the distribution law.
[0106] In this embodiment, the mean of all data points outside the abnormal dimensionality reduction result is recorded as the center of the point set. In other embodiments, the mean of all data points can be recorded as the center of the point set. In this embodiment, the preset neighborhood of each type of point refers to a circular area centered on each data point, and the radius of the circular area is: one tenth of the maximum distance between all data points outside the abnormal dimensionality reduction result. In other embodiments, it can be set to other radius sizes, which are not specifically limited in this embodiment.
[0107] The distribution law obtained by the above method can more accurately describe the distribution law of the first type of data relative to the second type of data.
[0108] Further, the incompleteness of the information security reference set is obtained, and the incompleteness is respectively related to , There is a positive correlation.
[0109] The incompleteness of the information security reference set describes whether the current information security reference set can fully and completely describe the most relevant relationship between drone information and inspection trajectories. The larger the value, the more the first-class data deviates from other non-abnormal dimensionality reduction results, while the second-class data is closer to other non-abnormal dimensionality reduction results, indicating that the information security reference set cannot distinguish between the first-class data and the second-class data, resulting in the actual safe first-class data deviating more from the expected most relevant relationship (that is, deviating from the center of the point set) than the unsafe second-class data. At this time, the information security reference set cannot describe the most relevant relationship between the drone information and the inspection trajectory, and the incompleteness of the information security reference set is relatively large; when The larger it is, the more Class II points there are in the preset neighborhood for each Class I point on average, which further indicates that the information security reference set cannot distinguish between Class I and Class II data. At this time, the information security reference set cannot describe the most relevant relationship between drone information and inspection trajectories, and the information security reference set is highly incomplete.
[0110] When the incompleteness is greater than the preset threshold th1, the information security reference set is updated. The specific method is: the drone information corresponding to the first type of points is added to the information security reference set, and the drone information corresponding to the second type of points is removed from the information security reference set. Then, the information security verification set is regenerated based on the information security reference set to obtain the updated information security verification set.
[0111] This embodiment is described by taking th1 equal to 20 as an example. In other embodiments, it can be set to other values, which are not specifically limited in this embodiment. In some other embodiments, Recorded as incompleteness.
[0112] It should also be noted that when there is no second-category data, the information security verification set will not be updated.
[0113] Step S005: Continue to use the updated information security verification set to perform security checks on the newly collected drone information, and update the information security verification set again until the drone inspection is completed.
[0114] After the information security verification set is updated in the above process, the method in step S003 is executed again. The general process is:
[0115] The drone continues to collect drone information in real time, and reduces the dimension to D1 to obtain a new dimension reduction result. It should be noted that the data dimension D1 is the data dimension recalculated when the information security verification set is generated from the information security reference set in step S004, and D1 is not necessarily equal to D; in addition, the new dimension reduction result obtained here is also based on the dimension reduction method when the information security verification set is generated from the information security reference set in step S004.
[0116] Next, the new dimensionality reduction results and the dimensionality reduction results in the information security verification set are used together for anomaly detection to obtain abnormal dimensionality reduction results. The drone information corresponding to all abnormal dimensionality reduction results obtained within a preset time period is sent to the server for safety judgment. When it is judged to be safe, the abnormal dimensionality reduction results are recorded as Class I data, otherwise they are recorded as Class II data; the distribution law of Class I data relative to Class II data is calculated; and the information security verification set is updated again according to the distribution law.
[0117] The above process is then repeated until the drone inspection is completed.
[0118] To summarize, this embodiment ensures that the information security verification set is gradually complete by continuously updating the information security verification set, continuously adding safe drone information to the information security verification set and removing unsafe drone information. That is, it can accurately describe the most relevant relationship between drone information and inspection trajectories, and ensure that the information security verification set can cope with changing network attack behaviors.
[0119] At the same time, during this process, the drone will send the drone information of abnormal dimensionality reduction results to the server for security judgment at regular intervals, so as to avoid taking up too many computing resources on the basis of knowing whether there is a network attack behavior, and the update process of the information security test set does not need to be carried out in real time. Therefore, the update of the information security test set and the data verification on the server are combined to avoid the problem of information security inspection affecting the endurance of the drone as much as possible.
[0120] More importantly, as the information security verification set is continuously updated, when the dimensionality reduction results in the information security verification set can describe the most relevant relationship between the drone information and the inspection trajectory as much as possible, it is ensured that the information security verification set can cope with the changing network attack behaviors. For example, when the number of a type of data obtained each time is less than 10, or even when there is no data of a type, the drone information corresponding to the abnormal dimensionality reduction result can be directly regarded as unsafe data, or when the drone has an abnormal dimensionality reduction result, it means that the drone has a network attack behavior. At this time, it is no longer necessary to conduct communication verification with the server. The drone realizes completely autonomous information security inspection, and while coping with the changing network attack environment, it further avoids the problem of information security inspection affecting the drone's endurance.
[0121] It should be noted that when the drone detects a network attack, it can send a vulnerability patching instruction to the server, and after receiving the instruction, the server can remotely scan the drone vulnerabilities and perform vulnerability patching. In other embodiments, the ID of the network attack user can be disabled, or the information of the drone that is under network attack can be protected, such as encrypting the data or enabling identity authentication.
[0122] This embodiment does not limit the specific method for the drone to prevent network attacks.
[0123] Embodiment 2:
[0124] According to the distribution law, the drone information corresponding to the first type of points is added to the information security reference set, and the drone information corresponding to the second type of points is removed from the information security reference set. Then, an updated information security verification set is generated based on the information security reference set. The specific steps involved are as follows:
[0125] The data dimension of the dimension reduction result in the updated information security inspection set is obtained, recorded as D1, and the preset threshold th1 is reduced according to the absolute value of the difference between D1 and D (that is, the difference between D1 and D), and the reduction amplitude is positively correlated with the difference between D1 and D.
[0126] The purpose of the above process is that when the difference between D1 and D is large, it means that the most relevant relationship between the drone information and the inspection trajectory contained in the dimensionality reduction results in the information security test set before and after the adjustment has undergone a significant change, indicating that the information security test set before the update is less reliable. At this time, by reducing th1, the information security test set can be updated as soon as possible to avoid frequent communication with the server due to detecting too many abnormal dimensionality reduction results.
[0127] As an example, when the ratio of the difference between D1 and D and D1+D is denoted as a, th1 is reduced by a×100%;
[0128] That is:
[0129]
[0130] in:
[0131]
[0132] Indicates the value of th1 after being reduced. Indicates the amount of reduction.
[0133] In another example, when the difference between D1 and D is greater than 10, th1 is reduced by 10%, otherwise th1 is not adjusted.
[0134] When the information security verification set is updated next time, the reduced th1 (ie, th2) is used to determine whether the information security verification set needs to be updated.
[0135] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.
Claims
1. An information security inspection and detection method for drone inspection, characterized in that: The method comprises the following steps: S1: The drone information at all times during the historical inspection process is recorded as the information security reference set; S2: Generate an information security verification set based on the information security reference set, including: Obtain the inspection trajectory of the drone during the inspection, obtain the data dimension D when the drone information at each moment in the information security reference set has the largest correlation with the inspection trajectory, and obtain the dimension reduction result when all drone information in the information security reference set is reduced to D dimensions, which is recorded as the information security test set; S3: Use the information security verification set to perform security checks on the drone information collected in real time, and update the information security verification set, including: The drone information collected in real time is also reduced to D dimension to obtain a new dimension reduction result, and the new dimension reduction result and the dimension reduction result in the information security test set are used together for anomaly detection to obtain an abnormal dimension reduction result, and the drone information corresponding to all abnormal dimension reduction results obtained within a preset time period is sent to the server for security judgment. When it is judged to be safe, the abnormal dimension reduction result is recorded as Class I data, otherwise it is recorded as Class II data; calculate the distribution law of Class I data relative to Class II data; and update the information security test set according to the distribution law; The updating of the information security verification set comprises the steps of: adding the drone information corresponding to the first type of points to the information security reference set according to the distribution law, removing the drone information corresponding to the second type of points from the information security reference set, and regenerating an updated information security verification set according to the information security reference set; S4: Repeat the steps included in S3 according to the updated information security inspection set until the drone inspection is completed; The specific steps of calculating the distribution law of the first type of data relative to the second type of data are as follows: All dimensionality reduction results and first-class data and second-class data in the information security test set are regarded as data points, among which first-class data are regarded as first-class data points and second-class data are regarded as second-class data points; In the point set composed of all the data points, the average Euclidean distance Q1 between all the first-class points and the center of the point set, the average Euclidean distance Q2 between the second-class points and the center of the point set, and the average number of second-class points Q3 in the preset neighborhood of each first-class point are obtained. As the distribution law; The specific steps of adding the drone information corresponding to the first type of points into the information security reference set and removing the drone information corresponding to the second type of points from the information security reference set according to the distribution law are as follows: Obtain the incompleteness of the information security reference set, the incompleteness is respectively related to , There is a positive correlation; When the incompleteness is greater than the preset threshold, the drone information corresponding to the first type of points is added to the information security reference set, and the drone information corresponding to the second type of points is removed from the information security reference set; The method adds the drone information corresponding to the first type of points to the information security reference set according to the distribution law, removes the drone information corresponding to the second type of points from the information security reference set, and generates an updated information security verification set according to the information security reference set. The specific steps included thereafter are as follows: Obtain the data dimension of the dimension reduction result in the updated information security inspection set, recorded as D1, and reduce the preset threshold according to the difference between D1 and D, and the reduction amplitude is positively correlated with the difference between D1 and D; The specific calculation formula of the preset threshold is as follows: in , indicating the amplitude of the reduction, Indicates the preset threshold value, Indicates the value of th1 after being reduced.
2. According to claim 1, the information security inspection and detection method for drone inspection is characterized in that: The drone information at each moment includes a control vector, a data access vector, and a context vector; Among them, the control instructions received by the drone at each moment are recorded as control vectors, the behavior of reading and querying data in the drone at each moment is recorded as data access vectors, and all data collected by the drone within the preset neighborhood time period at each moment is recorded as context vectors.
3. According to claim 1, the information security inspection and detection method for drone inspection is characterized in that: The specific steps of obtaining the data dimension D when the correlation between the drone information and the inspection trajectory at each moment in the information security reference set is the largest include: Set several data dimensions. For any data dimension K, use the PCA algorithm to reduce the dimension of the drone information at all times in the information security reference set to K dimensions, and obtain the dimension reduction result of the drone information at each time in each data dimension; calculate the correlation coefficient between the dimension reduction result of the drone information at all times in each data dimension and the inspection trajectory, and record it as the correlation degree in each data dimension; The data dimension with the greatest correlation is recorded as data dimension D.
4. According to claim 1, the information security inspection and detection method for drone inspection is characterized in that: The log and database of the server side contain the drone information. When the drone information corresponding to the abnormal dimensionality reduction result is contained in the log and database of the server side, it is judged to be safe; If the drone information corresponding to the abnormal dimensionality reduction result is not included in the server-side log and database, it is judged to be unsafe.
5. According to claim 1, the information security inspection and detection method for drone inspection is characterized in that: The point set center is the mean of all data points outside the abnormal dimensionality reduction results.
6. According to claim 1, the information security inspection and detection method for drone inspection is characterized in that: The anomaly detection uses the LOF algorithm.
Citation Information
Patent Citations
Power grid full-specialty differentiated operation and maintenance method based on unmanned aerial vehicle patrol technology
CN115864223A
System for providing internet of behavior based intelligent data security platform service for zero trust security
KR102542720B1
Urban elevator intelligent dispatching inspection system based on big data drive
WO2020093969A1
Unmanned aerial vehicle inspection method and apparatus, and unmanned aerial vehicle
WO2021189456A1
