Unmanned aerial vehicle position sensitive data identification and desensitization method based on adaptive fusion

Abstract

The invention provides an unmanned aerial vehicle position sensitive data identification and desensitization system and method based on adaptive fusion, and belongs to the technical field of data desensitization. The technical problems that in the traditional sensitive information identification and desensitization process, the identification precision is low, the processing efficiency is poor, and the safety is insufficient are solved. Firstly, the flight data of the unmanned aerial vehicle are preprocessed; secondly, performing deep learning analysis on the sensitive information by combining an adaptive convolutional neural network and a long-short-term memory network model; and finally, performing security desensitization processing on the sensitive information according to an SM4 encryption algorithm and an RSA technology. Accurate recognition of sensitive information such as flight paths, geographical location information and height data in flight data of the unmanned aerial vehicle is achieved, encryption processing is conducted on the sensitive data according to the multi-layer encryption theory in combination with the adaptive convolutional neural network and the long-short-term memory network model encryption algorithm, and therefore desensitization operation is achieved, and the sensitivity of the unmanned aerial vehicle is improved. The risk of sensitive information leakage in the flight data of the unmanned aerial vehicle is effectively reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data desensitization, and particularly relates to a UAV position-sensitive data identification and desensitization system and method based on adaptive fusion. BACKGROUND

[0002] With the rapid development of UAV technology, UAVs are increasingly widely used in various fields. UAVs not only play an important role in the military field, but also have made remarkable achievements in civil fields such as agriculture, logistics, transportation, and environmental monitoring. UAVs carry sensors and camera equipment to collect a large amount of data including flight data, environmental information, and geographic location information in real time, which plays a crucial role in the control, navigation trajectory, and task execution of UAVs. However, the flight data often contains a large amount of sensitive information, and once this information is leaked, it may cause serious safety hazards, especially in military applications, civil monitoring, and scenarios involving personal privacy. How to protect sensitive data has become a key problem that must be solved in the development process of UAV technology. Currently, the sensitive information in UAV flight data mainly includes position-sensitive information, heading data, and sensor data. In particular, in some special flight missions, the flight trajectory and sensor information of the UAV are often directly related to the success or failure of the mission, geographic location, and environmental safety. The leakage of position-sensitive information not only may expose the whereabouts and mission target of the UAV, but also may lead to tracking or interference of the UAV by the enemy, so timely identification and desensitization of sensitive information in flight data is particularly important. The existing sensitive information identification and desensitization technology still has many deficiencies. Traditional sensitive information identification methods are usually based on rules or static patterns, which mainly rely on manually preset rules or fixed matching algorithms, which are not flexible and efficient enough for handling complex and diverse flight data. Although the existing desensitization technology can encrypt data, it lacks sufficient precision and flexibility in intelligently identifying sensitive information and desensitizing it, resulting in certain risks in the data desensitization process.

[0003] In recent years, the rapid development of deep learning technology has provided new ideas for solving the above technical problems. Adaptive Convolutional Neural Networks (ACNN) is a new type of convolutional neural network that can adaptively adjust the size and shape of the convolution kernel, thus more accurately extracting important features in the data. Compared with traditional Convolutional Neural Networks (CNN), ACNN can better capture the spatial features and complex relationships of the data by adjusting the convolution kernel when faced with diverse and complex patterns in the data. This adaptability enables ACNN to be more efficient and accurate in identifying sensitive information in flight data. Long-Short Term Memory (LSTM) is a deep learning model that can handle time series data, and it performs particularly well in handling time-related information in flight data. LSTM can effectively capture the temporal relationships in flight data through its memory mechanism, providing higher accuracy in analyzing flight trajectories and dynamic information. By combining ACNN and LSTM, we can better analyze the spatial and temporal information in flight data and provide stronger support for accurate identification of sensitive information.

[0004] However, accurate identification of sensitive information is only the first step in the processing process, and how to ensure the security and privacy of data remains a core technical challenge. Traditional data encryption techniques, such as symmetric encryption and asymmetric encryption, can protect data security, but when dealing with large amounts of flight data, their computational complexity is high, which may affect the real-time performance and processing efficiency of the UAV. Currently, SM4 encryption algorithm, as a symmetric encryption algorithm independently developed by China, has been widely applied in many fields. SM4 has high encryption efficiency while ensuring data security, and is particularly suitable for resource-constrained devices such as UAVs. However, SM4 algorithm itself does not have the ability of key management and distribution, so combining RSA (Rivest-Shamir-Adleman) encryption technology to encrypt the encryption key of SM4 can further enhance the security and privacy protection of the encryption process. RSA algorithm, as a classic asymmetric encryption algorithm, has incomparable advantages in key management, which can effectively prevent the risk of key leakage.

[0005] The method of identifying location-sensitive information in UAV flight data based on adaptive deep learning and heterogeneous information fusion and the desensitization method mainly need to consider the following problems: (1) How to effectively fuse heterogeneous data to improve the accuracy and robustness of sensitive information identification.

[0006] (2) How to design an adaptive convolutional neural network (ACNN) and a long short-term memory network (LSTM) combined solution to handle diverse flight data patterns.

[0007] (3) How to apply encryption algorithms to the data desensitization of unmanned inspection terminals to ensure the high security and privacy protection of complex flight data.

[0008] How to solve the above technical problems is the subject faced by the present application. SUMMARY

[0009] The purpose of the present application is to provide a UAV location-sensitive data identification and desensitization system and method based on adaptive fusion, solving the technical problems of low recognition accuracy, poor processing efficiency and insufficient security in the traditional sensitive information identification and desensitization process. This mechanism is a strategic method that can efficiently fuse multi-source flight data and adaptively adjust the convolutional network structure to accurately identify and safely desensitize location-sensitive information in complex flight data, thus realizing real-time and accurate data privacy protection and flight safety assurance.

[0010] The inventive idea of the present application is: first, pre-processing the UAV flight data, including data cleaning and standardization, feature extraction, to ensure the integrity, accuracy and consistency of the data; second, combining adaptive convolutional neural network (ACNN) and long short-term memory network (LSTM) models for deep learning analysis of sensitive information, achieving efficient and accurate identification of location-sensitive information; finally, based on SM4 encryption algorithm and RSA technology, through the combination of encryption and key protection, a complete data desensitization scheme is constructed to safely desensitize sensitive information, ensuring privacy protection and security of flight data throughout its life cycle.

[0011] To achieve the above-mentioned application purpose, the technical solution adopted by the present application is as follows: a UAV location-sensitive data identification and desensitization method based on adaptive fusion, comprising the following steps: S1, data acquisition and preprocessing, This step is responsible for quality control of the original data collected by the UAV, through operations such as outlier detection, missing value filling and data standardization, to ensure the integrity and consistency of the input data and lay a foundation for subsequent feature extraction; S2, multi-dimensional feature extraction and shunting According to the heterogeneous characteristics of UAV data, the pre-processed data is divided into spatial features and time series features, and is directed to the corresponding processing path to realize targeted feature extraction strategy; S3, spatial feature analysis and detection Deep analysis of spatial dimension data using adaptive convolutional neural networks to extract spatial feature representations, and distance measurement methods to determine whether sensitive location information is included; S4, time series feature analysis and detection Using long short-term memory networks to process time series data, and using a classifier to determine whether sensitive behavior patterns exist; S5, comprehensive sensitive information determination Integrating spatial and temporal detection results, using a double verification mechanism to determine the final sensitive information determination, ensuring the accuracy of the determination result.

[0012] S6, data encryption and secure transmission Implementing double encryption protection on confirmed sensitive information, using SM4 and RSA hybrid encryption system to ensure data security during transmission and storage.

[0013] Further, in S1, 3σ criterion is used for outlier detection, and data points that meet the condition are removed, where represents the th data point, is the mean of the data set, is the standard deviation. Linear interpolation is used to fill in missing values: , where is the time point of the missing value, and are the nearest valid data time points before and after the missing value, and are the corresponding data values. Finally, the data is mapped to the interval by the min-max normalization method: , where is the original data, and are the minimum and maximum values of the data set, is the normalized data.

[0014] Further, in step S2, the system identifies spatial feature data, including latitude and longitude coordinates (where represents longitude, represents latitude), flight altitude , attitude angle (respectively representing roll angle, pitch angle and yaw angle) and other static or quasi-static geographic spatial information. These data will be directed to the spatial feature processing path.

[0015] Further, in step S2, the system identifies time series feature data, including speed , acceleration , angular velocity and other dynamic changing parameters, where denotes time variable. Such data has obvious time-dependent property and will be directed to time series feature processing path for special treatment.

[0016] Further, in step S3, in the spatial feature processing path, the system adopts adaptive convolutional neural network (ACNN) for deep feature extraction. The core advantage of ACNN lies in its adaptive convolution kernel mechanism: where is the initial convolution kernel weight matrix, is the dynamic adjustment amount based on the input data , and is the adjustment function.

[0017] First, the spatial feature map is generated by convolution operation , where is the value of the output feature map at position , is the value of the input data at position , is the weight of the adaptive convolution kernel at position .

[0018] Second, after multi-layer convolution and pooling operation: where is the feature map of the th layer, is the weight matrix of the th layer, is the bias term, denotes convolution operation, is the activation function, is the pooling operation.

[0019] Finally, sensitive information detection is performed on the extracted spatial feature map . The system calculates the Euclidean distance between it and the preset reference feature map : where is the distance measure value, denotes L2 norm (Euclidean distance), is the spatial feature map to be detected, is the pre-defined sensitive area reference feature map. When the distance measure , the system determines that the spatial data contains sensitive position information, where is the preset sensitivity determination threshold.

[0020] Further, in the step S4, the time series feature processing adopts a long short-term memory neural network (LSTM) for modeling, and the LSTM adopts a forgetting gate controls the retention of historical information, wherein is the output of the forgetting gate at time , is a Sigmoid activation function, is a forgetting gate weight matrix, is a hidden state at the previous time, is an input at the current time, is a bias term, and an input gate controls the update degree of new information, wherein is the output of the input gate, is an input gate weight matrix, is an input gate bias, and an output gate decides the output information, and the cell state update follows: , wherein is a current cell state, is a cell state at the previous time, is a candidate value, and a final output hidden state: is generated, and a complete sequence is generated, wherein is a sequence length.

[0021] Further, in the step S4, the hidden state sequence generated by the LSTM is input into a specially designed classifier to perform sensitivity discrimination: , wherein is a complete hidden state sequence output by the LSTM, is a trained classifier function. The classifier identifies abnormal flight behavior or sensitive trajectory features based on time series feature pattern recognition.

[0022] Further, in the step S5, the system adopts a strict double verification mechanism for final determination: , wherein is a final sensitive information determination result, is a Boolean result of spatial feature detection, indicating that spatial sensitive information is detected, is a result of time series feature detection, indicates a logical AND operation. Only when both dimensions are determined to be sensitive, the related data is confirmed as sensitive information. Further, in the step S6, the confirmed sensitive information is subjected to a first layer of encryption protection, and the system adopts an SM4 symmetric cipher algorithm: , wherein is a sensitive data plaintext to be encrypted, is a 128-bit SM4 symmetric key, SM4 is a national secret SM4 encryption function, is the encrypted ciphertext data.

[0023] Further, in the step S6, in order to protect the security of the SM4 key, the system implements a second layer of encryption protection, using the RSA asymmetric encryption algorithm: , wherein is the SM4 key to be protected, is the RSA public key parameter, is the public key exponent, is the modulus, which is generated by the product of two large prime numbers and : , RSA is the RSA encryption function, is the encrypted key ciphertext.

[0024] Further, in the step S6, the system encapsulates the encrypted sensitive data and the encrypted key into a complete encrypted information package: , wherein is the final encrypted information package, is the SM4 encrypted sensitive data ciphertext, is the RSA encrypted SM4 key ciphertext. The information package is transmitted to the cloud server through a secure communication protocol, realizing the secure storage and subsequent processing of sensitive information.

[0025] In order to achieve the above-mentioned purposes, the present application provides a kind of based on adaptive fusion's unmanned aerial vehicle position sensitive data identification and desensitization system, including data preprocessing module, sensitive data identification module and data desensitization module connected in turn; The data preprocessing module preliminarily processes the flight data collected by the unmanned aerial vehicle, carries out data acquisition, obtains original data including position, speed, heading, then carries out data cleaning, removes abnormal values and noise, then standardizes data, and extracts valuable features for sensitive information identification by feature extraction method; The sensitive data identification module uses adaptive convolutional neural network and long short-term memory network to analyze the position sensitive information in flight data by deep learning, and the adaptive convolutional neural network adaptively adjusts the size and shape of convolution kernel to extract spatial features;Long short-term memory network captures the time sequence relationship in flight data through its memory mechanism; The data desensitization module encrypts the identified sensitive information by SM4 encryption algorithm, and encrypts the encryption key of SM4 by combining RSA asymmetric encryption technology.

[0026] Compared with the prior art, the present application has the following advantages: 1、The method of the present application adopts the combination of adaptive convolutional neural network (ACNN) and long short-term memory network (LSTM), which can accurately identify the position sensitive information in flight data, and encrypt the sensitive information through SM4 encryption algorithm, and finally protect the encryption key by RSA technology to ensure the safety of flight data in the whole process. This method not only improves the accuracy and efficiency of sensitive information identification, but also ensures the high security of sensitive data, which can effectively deal with the complex and diversified sensitive information identification and protection requirements in flight data.

[0027] 2、The three functional modules of data preprocessing module, sensitive data identification module and data desensitization module constitute a complete processing pipeline through strict serial execution mechanism, the data preprocessing module provides standardized feature representation to the sensitive identification module, the judgment result of the sensitive identification module controls whether the data enters the desensitization processing flow, and the data desensitization module ensures the confidentiality of sensitive information in the transmission and storage process. The modular architecture design not only ensures the system running efficiency and avoids redundant encryption operation on non-sensitive data, but also builds an unmanned aerial vehicle data processing system with efficiency and security through multi-dimensional feature analysis and hybrid encryption mechanism. This design paradigm provides a feasible technical solution for unmanned aerial vehicle data security processing. BRIEF DESCRIPTION OF DRAWINGS

[0028] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, which together with the embodiments of the present application, serve to explain the present application, and do not constitute a limitation of the present application.

[0029] Figure 1 A system structure diagram of an unmanned aerial vehicle position sensitive data identification and desensitization system and method based on adaptive fusion provided by the present application.

[0030] Figure 2 A flowchart of an unmanned aerial vehicle position sensitive data identification and desensitization system and method based on adaptive fusion provided by the present application. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical scheme and advantages of the present application more clear and explicit, the present application will be further described in detail below in combination with the drawings and embodiments. Of course, the specific embodiments described here are only used to explain the present application, and do not limit the present application.

[0032] Embodiment 1: see Figure 1 and Figure 2 The technical scheme provided by the present embodiment is Figure 1The structure diagram of the unmanned aerial vehicle flight data location sensitive information recognition method and the desensitization system based on adaptive deep learning and heterogeneous information fusion is given, which mainly includes three parts: data preprocessing module, sensitive data recognition module and data desensitization module. Figure 1 The data preprocessing module in the application is mainly used for cleaning, standardizing and extracting features of the flight data, so as to ensure the quality and consistency of the data and provide high-quality input for subsequent analysis; the sensitive data recognition module uses an adaptive convolutional neural network (ACNN) and a long short-term memory network (LSTM) model to analyze the flight data through deep learning, thereby realizing efficient and accurate sensitive information recognition; and the data desensitization module encrypts the sensitive information according to the SM4 encryption algorithm, and protects the encryption key by combining RSA technology, realizes safe desensitization processing of the data, and ensures the privacy and security of the flight data in the storage, transmission and processing process.

[0033] The data preprocessing module: this module is mainly responsible for the preliminary processing of the flight data collected by the unmanned aerial vehicle to ensure the data quality. First, data acquisition is performed to obtain original data including position, speed, heading, etc. Then, data cleaning is performed to remove outliers and noise, ensuring the accuracy and consistency of the data. Subsequently, the data is standardized to have a unified dimension and range, and valuable features for sensitive information recognition are extracted through feature extraction methods, laying a foundation for subsequent sensitive information analysis.

[0034] The sensitive data recognition module: this module uses an adaptive convolutional neural network (ACNN) and a long short-term memory network (LSTM) model to analyze the location sensitive information in the flight data through deep learning. ACNN can adaptively adjust the size and shape of the convolution kernel to accurately extract spatial features; LSTM captures the time sequence relationship in the flight data through its memory mechanism. This module effectively identifies the sensitive information in the data by combining the advantages of the two models, ensuring the efficiency and accuracy of the recognition results.

[0035] The data desensitization module: this module encrypts the identified sensitive information through the SM4 encryption algorithm to ensure the security of the sensitive data in the storage, transmission and use process. The SM4 encryption algorithm has high efficiency and security, and is suitable for resource-constrained devices such as unmanned aerial vehicles. On this basis, the RSA asymmetric encryption technology is combined to encrypt and protect the encryption key of SM4, further enhancing the security of the key and preventing key leakage, thereby ensuring effective privacy protection and security guarantee for sensitive data throughout its life cycle.

[0036] 1、Data preprocessing module The module aims to ensure the quality and consistency of input data, providing an effective foundation for subsequent sensitive information identification and desensitization processing. First, from the raw data collected by the unmanned aerial vehicle, data cleaning and standardization processing are performed to remove invalid data and noise, and to unify the format of data from different sources. Next, through feature extraction, spatial feature data and time series feature data are processed respectively. Spatial feature data is processed through adaptive convolutional neural network (ACNN), and time series feature data is analyzed through long short-term memory network (LSTM), finally preparing for the subsequent sensitive information identification. The specific process is as follows: (1) Data collection During the execution of tasks, the unmanned aerial vehicle will collect various types of raw data, including spatial geographic data, time series data, and other information related to the task. Spatial geographic data mainly includes latitude , longitude , altitude , attitude angle , etc.; time series data includes speed , acceleration , angular velocity , etc.; task-related data may include sensor data (such as temperature, air pressure) and image data, etc. These data are the basis for subsequent analysis and processing.

[0037] (2) Data cleaning During data collection, due to sensor errors, environmental interference or transmission problems, invalid or abnormal values may be contained in the raw data. In order to improve data quality, data cleaning is needed first.

[0038] Abnormal value detection: 3σ criterion is used to detect abnormal values. Assuming that the data set , the mean and standard deviation are calculated: If the data point satisfies , it is considered as an abnormal value and is removed.

[0039] Missing value processing: For missing values, linear interpolation method is used for filling. Assuming that the missing value position is , the nearest valid values before and after it are and , the interpolation formula is: (3) Data standardization Since data from different sources can have different scales and ranges, it is necessary to standardize the data for subsequent processing. The min-max normalization method is used to map the data to the interval [0, 1]. where min and max are the minimum and maximum values of the data set, respectively.

[0040] (4) Spatial feature extraction Spatial features are crucial for identifying sensitive areas. To extract effective features from the data, an adaptive convolutional neural network (ACNN) is used.

[0041] 1) Adaptive convolution kernel Traditional convolutional neural networks use fixed convolution kernels, while ACNN can dynamically adjust the convolution kernel weights based on the input data. Suppose the convolution kernel is its adjustment formula is: where , is the adjustment function.

[0042] 2) Convolution operation The input data is convolved to generate a feature map. Suppose the input data is and the convolution kernel is , then the convolution operation formula is: where is the th element of the output feature map.

[0043] 3) Spatial feature map generation Through multiple layers of convolution and pooling operations, the deep features of the data are gradually extracted. After the th layer of convolution and pooling, the feature map is calculated as: where c represents the convolution operation, is the activation function, is the pooling operation.

[0044] (5) Time series feature extraction The time series data contains information such as the motion trajectory and speed change of the UAV, which is important for identifying sensitive behavior. To extract effective features from the time series data, a long short-term memory neural network (LSTM) is used. ​​

[0045] LSTM processes time series data through gating mechanisms and can capture long-term dependencies. Assuming the input time series data is , the calculation process of LSTM is as follows: Forget gate: controls the forgetting degree of the cell state at the last time . The specific formula is as follows: where is the output of the forget gate, indicating the proportion of retained information, is the Sigmoid activation function, is the weight matrix of the forget gate, is the concatenation vector of the input and the hidden state at the last time, is the bias term of the forget gate.

[0046] Input gate: controls the updating degree of the current candidate value . The specific formula is as follows: where is the output of the input gate, indicating the proportion of updated information, is the weight matrix of the input gate, used for linear transformation of input data, is the bias term of the input gate, used to adjust the baseline of the output.

[0047] Candidate value: generates the candidate information at the current time. The specific formula is as follows: where is the candidate value, indicating the candidate information at the current time, is the hyperbolic tangent activation function, is the weight matrix of the candidate value, used for linear transformation of input data, is the bias term of the candidate value, used to adjust the baseline of the output.

[0048] Update cell state: update the cell state by combining the forget gate and the input gate. The specific formula is as follows: where is the cell state at the current time, storing long-term memory, is the cell state at the last time, is the retained part of the cell state at the last time controlled by the forget gate, is the updated part of the candidate value at the current time controlled by the input gate.

[0049] Output gate: controls the output of the current hidden state The specific formula is as follows: wherein, is the output of the output gate, indicating the proportion of output information, is the weight matrix of the output gate, used for linear transformation of input data, is the bias term of the output gate, used to adjust the baseline of the output.

[0050] Output hidden state: generate the hidden state at the current time. The specific formula is as follows: Generate time series features through LSTM wherein, is the hidden state at the moment.

[0051] 2、Sensitive data identification module The function of this module is to identify the part that may contain sensitive information from the preprocessed data. For spatial feature data, adaptive convolutional neural network (ACNN) is used for spatial analysis, and spatial matching algorithm and distance threshold are used to judge whether the data belongs to sensitive information. For time series data, the system uses long short-term memory network (LSTM) to analyze the time dependence relationship, and uses detection algorithm to identify whether there is sensitive information in the data. Only when both spatial feature and time series feature data are judged as sensitive information, the data will enter the encryption processing stage. The specific process is as follows: (1) Spatial feature sensitive information detection In order to judge whether the spatial feature belongs to the sensitive area, the Euclidean distance between the feature map and the reference feature map is calculated, and the specific formula is as follows: If , it is judged as sensitive information. Wherein, represents the feature map extracted from the spatial data, represents the reference feature map, is a preset threshold value, used to judge whether the feature map belongs to the sensitive area, if the distance is less than the threshold value, it is judged as sensitive information.

[0052] (2) Time series feature sensitive information detection Use the classifier to judge whether the time series feature is sensitive information, input the hidden state sequence into the classifier, and judge whether it is sensitive information according to the output of the classifier. The specific formula is as follows: wherein, is denoted as a classifier for judging whether the time series feature contains sensitive information.

[0053] (3) Comprehensive judgment When both the spatial feature and the time series feature are identified as sensitive information, the system determines that it is sensitive information. The specific formula is as follows: wherein, is denoted as sensitive information, is the spatial feature detection result, and if it is true, it means that the spatial data contains sensitive information, is the time series feature detection result, and if it is true, it means that the time series data contains sensitive information.

[0054] 3. Data desensitization module The role is to ensure that the identified sensitive information is encrypted to prevent leakage. After identifying sensitive information, the system first generates a random key to provide a secure basis for the encryption process. Then, the SM4 encryption algorithm is used to encrypt sensitive data, and then the RSA algorithm is used to encrypt the SM4 encryption key twice, thereby enhancing the security of the data. Finally, all encrypted data is organized into an encrypted information package to ensure that the data is effectively protected during transmission, and then the encrypted data is remotely transmitted to the cloud server for storage and further processing. The specific process is as follows: (1) SM4 encryption The SM4 encryption algorithm is used to encrypt sensitive data. The specific formula is as follows: wherein, is denoted as data identified as sensitive information, is denoted as the key of the SM4 encryption algorithm, with a length of 128 bits, is denoted as the SM4 encryption function, is denoted as the encrypted data, i.e. ciphertext.

[0055] (2) RSA encryption The RSA algorithm is used to encrypt the SM4 encryption key twice, thereby enhancing the security of the data. The specific formula is as follows: wherein, is denoted as the key of the SM4 encryption algorithm, which needs to be further encrypted, is denoted as the RSA encryption function, is denoted as the encrypted SM4 key, i.e. ciphertext, The public key of the RSA encryption algorithm is represented as follows: where, is the public key exponent, is the modulus, generated by the product of two large prime numbers and

[0056] (3) Encryption information package construction All encrypted data is organized into an encrypted information package to ensure effective protection during transmission. The specific formula is as follows: where, represents the SM4-encrypted sensitive information, represents the RSA-encrypted SM4 key, represents the final generated encrypted information package, containing encrypted sensitive information and encryption keys, for secure transmission or storage.

[0057] Data preprocessing module: First, this module cleans and standardizes the original data collected by the unmanned aerial vehicle, providing high-quality data input for subsequent feature extraction and sensitive information identification. Second, the module extracts features according to categories for processed data. For spatial geographic data, an adaptive convolutional neural network (ACNN) is used to extract spatial features. ACNN can dynamically adjust the convolution kernel weight according to the input data to extract deep spatial features of the data. Through multiple convolution and pooling operations, a spatial feature map is generated. For time series data, a long short-term memory neural network (LSTM) is used to extract time series features. LSTM captures the temporal dependence of time series data through a gating mechanism to generate a hidden state sequence. Through feature extraction, the system can extract key information from the original data, laying a solid foundation for subsequent sensitive information identification.

[0058] Sensitive data identification module: This module is the core module of the system, mainly responsible for identifying sensitive information from preprocessed data. Based on the comprehensive judgment of spatial features and time series features, the module realizes accurate identification of sensitive information. In spatial feature recognition, the spatial matching algorithm is used to calculate the Euclidean distance between the feature map and the reference feature map. If the distance is less than the preset threshold , it is determined to be sensitive information. In time series feature recognition, the hidden state sequence is input into the classifier, and the output is the determination result. Finally, the system integrates the recognition results of spatial features and time series features. When both are determined to be sensitive information, the system marks it as sensitive data and enters the data desensitization module.

[0059] ​Data desensitization module: The data desensitization module is a security module of the system, mainly responsible for encrypting the identified sensitive information to ensure data security and privacy. This module uses a multi-layer encryption mechanism combining SM4 and RSA encryption algorithms to desensitize sensitive information. First, the SM4 encryption algorithm is used to encrypt sensitive information to generate ciphertext . SM4 is a block encryption algorithm with a key length of 128 bits, providing high encryption strength. Next, the RSA encryption algorithm is used to encrypt the SM4 key to generate ciphertext . RSA is an asymmetric encryption algorithm that ensures the security of the key. Finally, the encrypted data and key are packaged to generate a complete encrypted information package . Through the data desensitization module, the system can effectively prevent sensitive data from being stolen or misused during transmission or storage, greatly improving data security and privacy protection levels. Embodiment 2: This embodiment uses a data preprocessing module, a sensitive data identification module, and a data desensitization module connected in sequence to build a complete unmanned aerial vehicle data security processing framework. The system follows the execution process of data preprocessing, sensitive data identification, and data desensitization.

[0060] The data preprocessing module implements standardized processing of multi-source heterogeneous raw data collected by unmanned aerial vehicles, including spatial geographic data, time series data, and task-related data. In the data quality control section, the module uses the 3σ criterion for outlier detection, and when the data point meets the condition , it is removed, and the linear interpolation method is used to complete the missing values. After standardization, the system implements a dual-channel parallel feature extraction strategy: first, spatial features are extracted through an adaptive convolutional neural network (ACNN), which generates spatial feature maps through convolution operations ; second, time series features are modeled using the gating mechanism of a long short-term memory network (LSTM) to implement long-term dependency modeling, outputting a hidden state sequence . The above dual-channel feature representation provides standardized input for subsequent sensitive information identification.

[0061] The sensitive data identification module performs key discrimination functions. This module receives spatial feature maps and time series features from the preprocessing module and implements dual-dimensional sensitivity evaluation. In spatial dimension analysis, the system determines the Euclidean distance between the feature map to be tested and the reference feature map, and when the distance measure is less than the preset threshold , it is marked as spatial sensitive information. In the time dimension analysis, the system inputs the hidden state sequence generated by LSTM into the classifier to implement sensitivity discrimination . This module uses logical AND operation to achieve comprehensive judgment: That is, only when both the spatial dimension and the time dimension are determined to be sensitive, the corresponding data is finally determined to be sensitive information. This double verification mechanism effectively improves the system discrimination accuracy and reduces the false positive rate.

[0062] The data desensitization module realizes the cryptographic protection of sensitive information. This module specially processes the sensitive data confirmed by the identification module, and uses a hybrid encryption system to ensure information security. The system first uses the SM4 symmetric cipher algorithm to implement encryption transformation on sensitive data: , wherein is a 128-bit symmetric key. To protect the security of the key itself, the system further uses the RSA cipher algorithm to encrypt the SM4 key. Finally, the system encapsulates the encrypted data and the encryption key into a secure transmission package: , realizing the double protection of data confidentiality and key security.

[0063] In summary, the data preprocessing module, the sensitive data identification module and the data desensitization module constitute a complete processing pipeline through a strict serial execution mechanism. The data preprocessing module provides standardized feature representation to the sensitive identification module, the judgment result of the sensitive identification module controls whether the data enters the desensitization processing flow, and the data desensitization module ensures the confidentiality of sensitive information in the transmission and storage process. This modular architecture design not only guarantees the system running efficiency and avoids redundant encryption operation on non-sensitive data, but also builds an unmanned aerial vehicle data processing system with efficiency and security through multi-dimensional feature analysis and hybrid encryption mechanism. This design paradigm provides a feasible technical solution for unmanned aerial vehicle data security processing.

[0064] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for identifying and desensitizing UAV location-sensitive data based on adaptive fusion, characterized in that, Includes the following steps: S1. Data Acquisition and Preprocessing Quality control is performed on the raw data collected by drones, including outlier detection, missing value imputation, and data standardization, to ensure the integrity and consistency of the input data. S2, Multidimensional Feature Extraction and Streaming Based on the heterogeneous nature of UAV data, the preprocessed data is divided into two categories: spatial features and time series features, and each is guided to a corresponding processing path to achieve a targeted feature extraction strategy. S3. Spatial Feature Analysis and Detection An adaptive convolutional neural network is used to perform in-depth analysis of spatial dimension data, extract spatial feature representations, and determine whether it contains sensitive location information by using a distance metric method. S4. Time Series Feature Analysis and Detection Long Short-Term Memory (LSTM) networks are used to process time-series data, and a classifier is used to determine whether there are sensitive behavioral patterns. S5. Comprehensive Judgment of Sensitive Information By combining the detection results from both spatial and temporal dimensions, a dual verification mechanism is employed to determine the final sensitive information, ensuring the accuracy of the determination results. S6, Data Encryption and Secure Transmission Confirmed sensitive information is protected by double-layer encryption, using a hybrid encryption system of SM4 and RSA to ensure data security during transmission and storage.

2. The method for identifying and desensitizing UAV location-sensitive data based on adaptive fusion according to claim 1, characterized in that, In step S1, the 3σ criterion is used for outlier detection. For values ​​that satisfy the condition |x i Data points with -μ|>3σ were discarded, where x i This represents the i-th data point. The mean of the dataset. To find the standard deviation, missing values ​​were imputed using linear interpolation. Where t is the time point where the missing value is located, t1 and t2 are the nearest valid data time points before and after the missing value, and x(t1) and x(t2) are the corresponding data values. The data is mapped to the interval [0,1] using the min-max normalization method. Where X represents the original data, X min and X max Let X be the minimum and maximum values ​​of the dataset, respectively. norm This is the standardized data.

3. The method for identifying and desensitizing UAV location-sensitive data based on adaptive fusion according to claim 1, characterized in that, In step S2, the system identifies spatial feature data, including latitude and longitude coordinates (x, y), flight altitude z, and attitude angles (φ, θ, ψ) static or quasi-static geospatial information. This data is guided to the spatial feature processing path. Here, x represents longitude, y represents latitude, and φ, θ, and ψ represent roll angle, pitch angle, and yaw angle, respectively. In step S2, the system identifies time-series feature data, including dynamic parameters of velocity v(t), acceleration a(t), and angular velocity w(t), where t represents the time variable.

4. The method for identifying and desensitizing UAV location-sensitive data based on adaptive fusion according to claim 1, characterized in that, In step S3, within the spatial feature processing path, the system employs an adaptive convolutional neural network for deep feature extraction. The adaptive convolutional kernel mechanism of the adaptive convolutional neural network is: W adaptive =W + ΔW, where W is the initial convolution kernel weight matrix, ΔW = f(x) is the dynamic adjustment based on the input data x, and f is the adjustment function; Includes the following steps: S31, through convolution operation Generate a spatial feature map, where F ij To output the value of the feature map at position (i,j), X i+m,j+n For the value of the input data at position (i+m,j+n), W adaptive (m,n) represents the weights of the adaptive convolutional kernel at position (m,n); S32, After multiple convolution and pooling operations: F (l) =Pooling(Relu(W (l) F (l-1) +b (l) )), where F (l) For the feature map of layer l, W (l) Let b be the weight matrix of the l-th layer. (l) For bias terms, * indicates convolution operation, ReLU is the activation function, and Pooling is the pooling operation; S33, Extracted spatial feature map F spatial To perform sensitive information detection, the system calculates its value against a preset reference feature map F. reference Euclidean distance between them: d(F) spatial ,F reference )=||F spatial -F reference ||2, where d is the distance metric, ||·||2 represents the L2 norm, i.e., the Euclidean distance, and F spatial F is the spatial feature map to be detected. reference For a predefined sensitive region reference feature map, when the distance metric d < τ spatial At that time, the system determines that the spatial data contains sensitive location information, where τ spatial This is the preset sensitivity threshold.

5. The method for identifying and desensitizing UAV location-sensitive data based on adaptive fusion according to claim 1, characterized in that, In step S4, the time series feature processing uses a long short-term memory neural network for modeling. The long short-term memory neural network uses a forgetting gate f t =σ(W f ·[h t-1 ,x t ]+b f ) Control the retention of historical information, where f t Let W be the output of the forget gate at time t, σ be the Sigmoid activation function, and W be the output of the forget gate at time t. f h is the forget gate weight matrix. t-1 The hidden state of x in the previous moment t Input for the current time, b f For the bias term, input gate i t =σ(W i ·[h t-1 ,x t ]+b i ) controls the degree of updating of new information, where i t For input gate output, W i Let b be the input gate weight matrix. i For input gate bias, output gate o t =σ(W o ·[h t-1 ,x t ]+b o The output information is determined, and cell state updates follow these rules: Where C t For the current cell state, C t-1 This represents the cell state at the previous moment. As candidate values, the final output is the hidden state: h t =o t ·tanh(C t Generate a complete sequence H = {h1, h2, ..., h}. T }, where T is the sequence length.

6. The method for identifying and desensitizing UAV location-sensitive data based on adaptive fusion according to claim 1, characterized in that, In step S4, the hidden state sequence H = {h1, h2, ..., h...} generated by the Long Short-Term Memory Neural Network is... T Input a specially designed classifier and perform sensitivity discrimination: y = Classifier(H), where H is the complete hidden state sequence output by the long short-term memory neural network, and Classifier is the trained classifier function. The classifier identifies abnormal flight behavior or sensitive trajectory features based on temporal feature patterns.

7. The method for identifying and desensitizing UAV location-sensitive data based on adaptive fusion according to claim 1, characterized in that, In step S5, the system uses a dual verification mechanism for the final determination: Sensitive = Sensitive spatial ∩Sensitive temporal Sensitive represents the final result of the sensitive information determination. spatial This is a Boolean result for spatial feature detection, indicating that spatially sensitive information has been detected. temporal The result of time series feature detection is shown. ∩ represents a logical AND operation. When both dimensions are judged to be sensitive, the relevant data is confirmed as sensitive information.

8. The method for identifying and desensitizing UAV location-sensitive data based on adaptive fusion according to claim 1, characterized in that, In step S6, the confirmed sensitive information is subjected to the first layer of encryption protection. The system uses the SM4 symmetric cryptographic algorithm: C SM4 =SM4(X Sensitive ,K SM4 ), where X Sensitive K represents the plaintext of the sensitive data to be encrypted. SM4 It uses a 128-bit SM4 symmetric key, where SM4 is the Chinese national standard SM4 encryption function. (C) SM4 This is the encrypted ciphertext data.

9. The method for identifying and desensitizing UAV location-sensitive data based on adaptive fusion according to claim 1, characterized in that, In step S6, to ensure the security of the SM4 key, the system implements a second layer of encryption protection, using the RSA asymmetric encryption algorithm: C RSA =RSA(K SM4 ,K RSA ), where K SM4 For the SM4 key that needs protection, K RSA = (e,n) represents the RSA public key parameters, where e is the public key exponent, and n is the modulus, generated by the product of two large prime numbers p and q: n = p × q. RSA is the RSA encryption function. C RSA The encrypted key ciphertext; In step S6, the system encapsulates the encrypted sensitive data and the encrypted key into a complete encrypted information packet: P encrypted ={C SM4 C RSA }, where Pencrypted is the final encrypted packet, and C SM4 C is the ciphertext of sensitive data encrypted with SM4. RSA The SM4 key ciphertext is encrypted with RSA. This information packet is transmitted to the cloud server through a secure communication protocol to achieve secure storage and subsequent processing of sensitive information.

10. The UAV location-sensitive data identification and desensitization system based on adaptive fusion according to claim 1, characterized in that, It includes a data preprocessing module, a sensitive data identification module, and a data desensitization module connected in series; The data preprocessing module performs preliminary processing on the flight data collected by the UAV. It collects raw data including position, speed, and heading, then cleans the data to remove outliers and noise, standardizes the data, and extracts features that are valuable for identifying sensitive information through feature extraction methods. The sensitive data identification module uses an adaptive convolutional neural network and a long short-term memory network to perform deep learning analysis on location-sensitive information in flight data. The adaptive convolutional neural network adaptively adjusts the size and shape of the convolutional kernel to extract spatial features. Long Short-Term Memory Networks (LSTM) capture temporal relationships in flight data through their memory mechanisms. The data desensitization module uses the SM4 encryption algorithm to encrypt the identified sensitive information, and combines RSA asymmetric encryption technology to encrypt and protect the SM4 encryption key.

Images