Cross-region unmanned aerial vehicle plant protection data privacy protection method and system based on federated learning

Abstract

The present application relates to the technical field of federal learning data protection, and proposes a cross-regional unmanned aerial vehicle plant protection data privacy protection method and system based on federal learning, which comprises the following steps: taking real-time wind speed, temperature and humidity data in the current operation voyage as a disturbance component, and uploading the disturbed and encrypted local plant protection decision parameters to a central server; performing aggregation operation on the disturbed parameters and returning them to each node; performing inverse disturbance decryption on the aggregated parameters according to the stored disturbance components to obtain individualized plant protection decision parameters of each node; and adjusting the spraying parameter instructions of the next operation voyage of the unmanned aerial vehicle according to the individualized plant protection decision parameters. The present application realizes cross-regional plant protection data privacy protection through disturbance encryption driven by environmental parameters, and each node restores individualized plant protection decision parameters that are different from each other, thereby realizing differentiated plant protection decision output.

Description

Technical Field

[0001] This invention relates to the field of federated learning data protection technology, and in particular to a method and system for protecting the privacy of cross-regional unmanned aerial vehicle (UAV) plant protection data based on federated learning. Background Technology

[0002] With the rapid development of precision agriculture, drone-based plant protection operations have evolved from independent operations in a single region to cross-regional collaborative operations. In cross-regional collaborative operation scenarios, drone edge nodes distributed in different geographical locations each collect local plant protection data, including flight altitude, flight speed, spraying flow rate, and corresponding plant protection operation effect evaluation values. This data needs to be shared and integrated between the nodes or with the central server to optimize plant protection decision parameters in each region. However, plant protection data contains sensitive information such as farmland geographical location, crop distribution, and soil conditions. Directly aggregating raw data will lead to the leakage of privacy information in each operation area. How to achieve effective integration and collaborative utilization of cross-regional plant protection data while protecting the privacy of plant protection data at each node, so as to improve the accuracy and adaptability of plant protection operations in each region, is a current technical challenge facing cross-regional drone-based plant protection collaborative operations.

[0003] Existing cross-regional plant protection data collaboration solutions typically employ a centralized data processing architecture. Each drone edge node directly uploads raw plant protection data to a central server, which then aggregates and processes the data before distributing it to the nodes. Under this architecture, the central server controls all the raw data from all nodes, posing a risk of data leakage and failing to effectively protect the privacy of plant protection data at each node. Some solutions use cryptographic techniques such as differential privacy or homomorphic encryption to encrypt and protect uploaded data; however, these cryptographic schemes involve complex key generation, distribution, and management mechanisms, incurring high computational overhead and numerous communication rounds, making them unsuitable for drone edge nodes with limited computing resources and communication bandwidth. Furthermore, in existing solutions, all nodes ultimately obtain the same global processing result, failing to generate personalized decision parameters based on the differences in plant protection operation environments across different regions, thus reducing the practical application value of cross-regional data collaboration. Therefore, how to achieve a balance between data privacy protection and personalized decision output in cross-regional drone-based plant protection collaborative operations has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a method and system for protecting the privacy of agricultural data from unmanned aerial vehicles (UAVs) across regions based on federated learning, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a cross-regional drone-based plant protection data privacy protection method based on federated learning, comprising: S1. The real-time wind speed data, real-time ambient temperature data, and real-time ambient humidity data of the edge nodes of the UAV in the current operation flight are used as disturbance components to encrypt the local plant protection decision parameters generated by the UAV plant protection data training, thereby obtaining the disturbance parameters of the UAV. The disturbance components are then stored in the local storage unit of the node according to the current operation flight number. S2. When the disturbance parameters are uploaded to the central server, the disturbance parameters are aggregated to obtain the aggregated parameters of the UAV, and then returned to each UAV edge node. S3. Based on the disturbance component corresponding to the operation flight number, perform inverse perturbation decryption on the aggregated parameters to obtain the personalized plant protection decision parameters for each node; S4. Based on the personalized plant protection decision parameters, adjust the plant protection spraying parameters for the next flight of the UAV.

[0006] Compared with the prior art, the present invention has the following beneficial effects: 1. Each drone edge node uses real-time wind speed, temperature, and humidity data collected during the current operational flight as perturbation components to apply a data perturbation uniquely bound to the current operational environment to its local plant protection decision parameters before uploading them to the central server. The central server only obtains the perturbed parameters and cannot access the original data of any node. Furthermore, each node does not need to expose its original local plant protection data to the outside world. In this way, plant protection decision parameters exist in a perturbed and obfuscated form throughout the entire transmission and aggregation chain, while the original data always resides locally at each node. This ensures the privacy and security of cross-regional plant protection data at both the data transmission and aggregation computation levels.

[0007] 2. Each node persistently stores the disturbance component in its local storage unit, indexed by the current flight number. After the central server returns the aggregation result, each node reads the disturbance component from its local storage unit based on the same flight number and completes the inverse disturbance decryption. Due to the objective differences in environmental parameters such as wind speed, temperature, and humidity in the different operational areas of each node, the personalized plant protection decision parameters recovered by each node after removing its own disturbances are different, each carrying its own unique operational environment fingerprint information. Based on this, each node updates its local plant protection decision parameter set and generates spraying parameter adjustment instructions for the next flight, ensuring that the spraying strategy for each region matches the actual environmental conditions of the current flight. This achieves differentiated plant protection decision output for different geographical areas and operational environments within the framework of cross-regional data collaborative optimization. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating a cross-regional drone-based plant protection data privacy protection method based on federated learning, provided in an embodiment of the present invention. Figure 2This is a functional block diagram of a cross-regional drone plant protection data privacy protection system based on federated learning, provided as an embodiment of the present invention.

[0009] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0010] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0011] This application provides a method for protecting the privacy of agricultural data from unmanned aerial vehicles (UAVs) across regions based on federated learning. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cluster of cloud servers. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0012] Reference Figure 1 The diagram shown is a flowchart illustrating a cross-regional drone-based plant protection data privacy protection method based on federated learning, according to an embodiment of the present invention. In this embodiment, the cross-regional drone-based plant protection data privacy protection method based on federated learning includes: S1. The real-time wind speed data, real-time ambient temperature data, and real-time ambient humidity data of the edge nodes of the UAV in the current operation flight are used as disturbance components to encrypt the local plant protection decision parameters generated by the UAV plant protection data training, thereby obtaining the disturbance parameters of the UAV. The disturbance components are then stored in the local storage unit of the node according to the current operation flight number. In this embodiment of the invention, the step of using real-time wind speed data, real-time ambient temperature data, and real-time ambient humidity data of the edge nodes of the UAV in the current operational flight as perturbation components to perturb and encrypt the local plant protection decision parameters generated by training the UAV plant protection data, thereby obtaining the perturbation parameters of the UAV, includes: Based on the local plant protection data of the drone's edge nodes, the locally stored set of plant protection decision parameters is trained and updated to obtain the local plant protection decision parameters of the drone. Based on the parameter dimension arrangement order of the local plant protection decision parameters, the real-time wind speed data, real-time ambient temperature data, and real-time ambient humidity data of the edge nodes of the UAV in the current operation flight are used as disturbance components and assigned to the corresponding dimension components of the local plant protection decision parameters. Based on the disturbance component, the corresponding dimension component is perturbed and encrypted, and the disturbance parameters of the UAV are obtained according to the parameter dimension arrangement order.

[0013] The step of perturbing and encrypting the corresponding dimension components based on the perturbation components, and obtaining the perturbation parameters of the UAV according to the parameter dimension arrangement order, includes: The tens digit and the units digit of the disturbance component value are combined into a two-digit additional code in the order of units digit first and tens digit last. The product of the dimensional component and the additional code is multiplied by one hundred and added to the perturbation component to obtain the encrypted component value of the corresponding dimensional component. This encrypted component value is then combined according to the parameter dimension arrangement order to form the perturbation parameter of the UAV.

[0014] The step of storing the disturbance component in the node's local storage unit according to the current mission number includes: The real-time wind speed data, the real-time ambient temperature data, and the real-time ambient humidity data are arranged in the order of the collection timestamps to form the disturbance component sequence of the UAV. The disturbance component sequence is associated with the current mission number and written into a pre-allocated storage block in the storage unit.

[0015] Based on the local plant protection data from the UAV edge node, the locally stored plant protection decision parameter set is trained and updated to obtain the UAV's local plant protection decision parameters. Specifically, the UAV edge node reads the plant protection decision parameter set from the local storage unit. This set includes spray flow rate decision components, spray height decision components, and flight speed decision components, with initial values ​​set to 2.5 liters / minute, 2.8 meters, and 5.0 meters / second, respectively. The UAV edge node reads local plant protection data, which includes flight height parameters, flight speed parameters, spray flow rate parameters, and plant protection operation effect evaluation values ​​corresponding to each parameter combination recorded in each historical operation flight. The plant protection operation effect evaluation values ​​are expressed as a percentage of coverage uniformity, ranging from 60% to 99%. The UAV edge node extracts data samples from the most recent 50 operation flights from the local plant protection data and performs linear regression operations on the flight height parameters, flight speed parameters, and spray flow rate parameters in each data sample with the corresponding plant protection operation effect evaluation values ​​to obtain the regression coefficient values. The edge node of the drone updates the values ​​of the spray flow rate decision component, spray height decision component, and flight speed decision component of the plant protection decision parameters according to the regression coefficient values, and obtains the updated local plant protection decision parameters. The updated spray flow rate decision component is 2.5 liters / minute, the spray height decision component is 2.8 meters, and the flight speed decision component is 5.0 meters / second.

[0016] In one feasible implementation, the parameter dimensions are ordered and perturbation components are assigned. The UAV edge node reads the updated local plant protection decision parameters, which contain three dimensional components, ordered as follows: spray flow rate decision component, spray altitude decision component, and flight speed decision component. The UAV edge node reads the real-time wind speed data (3.6 m / s) for the current operation flight from the flight control system, and the real-time ambient temperature data (28.5℃) and ambient humidity data (73%) from the onboard temperature and humidity sensor unit. Following the parameter dimension order, the UAV edge node assigns the real-time wind speed data (3.6 m / s) to the corresponding position of the spray flow rate decision component, the real-time ambient temperature data (28.5℃) to the corresponding position of the spray altitude decision component, and the real-time ambient humidity data (73%) to the corresponding position of the flight speed decision component.

[0017] In one feasible implementation, digit extraction and combination operations are performed on the disturbance components to generate additional codes. The UAV edge node extracts the tens and units digits of each disturbance component value. For example, the tens digit of the real-time wind speed data (3.6 m / s) is 0 and the units digit is 3; the tens digit of the real-time ambient temperature data (28.5℃) is 2 and the units digit is 8; and the tens digit of the real-time ambient humidity data (73%) is 7 and the units digit is 3. The UAV edge node combines the tens and units digits of each disturbance component in the order of units digit first, tens digit last, to form two-digit additional codes. The additional code for the real-time wind speed data is 30, the additional code for the real-time ambient temperature data is 82, and the additional code for the real-time ambient humidity data is 37.

[0018] In one feasible implementation, encryption operations are performed on the corresponding dimensional components to obtain encrypted component values. The UAV edge node multiplies the spray flow decision component 2.5 by the corresponding additional code 30 to obtain 75, multiplies 75 by 100 to obtain 7500, and then adds 7500 to the real-time wind speed data 3.6 to obtain the encrypted component value of the spray flow decision component: 7503.6. The UAV edge node multiplies the spray altitude decision component 2.8 by the corresponding additional code 82 to obtain 229.6, multiplies 229.6 by 100 to obtain 22960, and then adds 22960 to the real-time ambient temperature data 28.5 to obtain the encrypted component value of the spray altitude decision component: 22988.5. The UAV edge node multiplies the flight speed decision component 5.0 by the corresponding additional code 37 to obtain 185, multiplies 185 by 100 to obtain 18500, and then adds 18500 to the real-time ambient humidity data 73 to obtain the encrypted component value of the flight speed decision component: 18573.

[0019] In one feasible implementation, the perturbation component sequence is associated with the current flight number and written to a storage block. The UAV edge node reads the current flight number as flight number 27, extracts the values ​​of each perturbation component, and arranges them in order of collection timestamp. Specifically, the UAV edge node arranges the real-time wind speed data (3.6 m / s), real-time ambient temperature data (28.5℃), and real-time ambient humidity data (73%) collected in the current flight into a perturbation component sequence according to the order of collection timestamps. The first element of this sequence is the real-time wind speed data (3.6 m / s), the second element is the real-time ambient temperature data (28.5℃), and the third element is the real-time ambient humidity data (73%). The UAV edge node uses the current flight number 27 as the index key value, associates the disturbance component sequence with the index key value 27, and writes it into a pre-allocated storage block in the local storage unit. The starting address of the storage block is 0x0001, and the block capacity is 256 bytes. After writing, the storage block records the flight number 27, the number of elements in the disturbance component sequence is 3, the value of each element and its collection timestamp.

[0020] The beneficial effect is that each UAV edge node uses real-time wind speed, temperature, and humidity data collected during the current operational flight as perturbation components to perform perturbation encryption on local plant protection decision parameters before uploading them to the central server. Since the environmental data of each node is different, the encrypted perturbation parameters are also different. The central server cannot separate the original plant protection decision parameters of each node from the received data. Throughout the transmission process, the original data remains locally on each node, effectively preventing privacy leaks during data transmission. Each node stores the wind speed, temperature, and humidity data of the current operational flight in its local storage unit, indexed by the flight number. This provides environmental data completely consistent with the encryption process for subsequent decryption. Encryption and decryption are completed sequentially by the same node at different times, without any key negotiation or certificate exchange between the node and the central server. The communication rounds are the same as the standard data sharing process, avoiding the computational overhead caused by complex key management mechanisms. After encryption, the disturbance parameters obtained by each node retain all the digital information of the original plant protection decision parameters. The central server can directly perform aggregation operations on the disturbance parameters. When decrypting, each node can restore the original data by reading the disturbance component based on the same flight number, without the need for additional correction or compensation operations.

[0021] S2. When the disturbance parameters are uploaded to the central server, the disturbance parameters are aggregated to obtain the aggregated parameters of the UAV, and then returned to each UAV edge node. In this embodiment of the invention, when the disturbance parameters are uploaded to the central server, the aggregation operation is performed on the disturbance parameters to obtain the aggregated parameters of the UAV, and then returned to each UAV edge node, including: The ratio of the additional code to the sum of all additional codes is used as the weighting coefficient of the edge node; Based on the weighting coefficients, the values ​​of the disturbance parameters in each dimension are weighted and summed to obtain the aggregate component value of the UAV. The aggregated component values ​​are combined into aggregated parameters according to the order of the parameter dimensions. The aggregation parameters are encapsulated into data packets and sent to each UAV edge node.

[0022] The central server receives perturbation parameters uploaded by all drone edge nodes participating in the current aggregation round. Taking three drone edge nodes as an example, node A has three perturbation parameters: 7503.6, 22988.5, and 18573, corresponding to the spray flow rate decision component, spray altitude decision component, and flight speed decision component, respectively. Node B has three perturbation parameters: 8204.2, 31567.8, and 21459. Node C has three perturbation parameters: 6912.7, 18342.1, and 15726. All three nodes' perturbation parameters contain three dimensional components in the same order. The central server performs dimensional alignment verification on each perturbation parameter. After confirming that each perturbation parameter has three dimensions and is arranged in the same order, it marks each perturbation parameter as a valid parameter for aggregation calculation.

[0023] In one feasible implementation, the central server extracts the additional codes generated by each node during the encryption phase and calculates the weighting coefficients. According to the implementation in S1, node A's additional code is 30, node B's is 45, and node C's is 28. The central server sums all the additional codes to obtain 103. Dividing each node's additional code by 103, the weighting coefficients for node A are: 30 divided by 103 = 0.291, for node B, 45 divided by 103 = 0.437, and for node C, 28 divided by 103 = 0.272. Specifically, the weighting coefficient represents the proportion of that node in the aggregation operation, and the sum of all node weighting coefficients equals one.

[0024] In one feasible implementation, the aggregated component values ​​are calculated independently for each dimension. For the spray flow decision component, the perturbation parameter value of node A (7503.6) multiplied by the weighting coefficient 0.291 yields 2183.5, the perturbation parameter value of node B (8204.2) multiplied by the weighting coefficient 0.437 yields 3585.2, and the perturbation parameter value of node C (6912.7) multiplied by the weighting coefficient 0.272 yields 1880.3. The aggregated component value of the spray flow decision component (7649.0) is obtained by multiplying and summing these three values. For the spray height decision component, the perturbation parameter value of node A (22988.5) multiplied by 0.291 yields 6689.7, the perturbation parameter value of node B (31567.8) multiplied by 0.437 yields 13794.1, and the perturbation parameter value of node C (18342.1) multiplied by 0.272 yields 4989.1. Summing these values ​​gives the aggregated component value of the spray height decision component: 25472.9. For the flight speed decision component, the perturbation parameter value of node A (18573) multiplied by 0.291 yields 5404.7, the perturbation parameter value of node B (21459) multiplied by 0.437 yields 9377.6, and the perturbation parameter value of node C (15726) multiplied by 0.272 yields 4277.5. Summing these values ​​gives the aggregated component value of the flight speed decision component: 19059.8.

[0025] In one feasible implementation, the central server arranges the aggregated component values ​​of the spray flow decision component (7649.0), the spray altitude decision component (25472.9), and the flight speed decision component (19059.8) sequentially according to the parameter dimensions to form aggregate parameters. The central server encapsulates the aggregate parameters and the weighting coefficients of all participating nodes into a data packet. The data packet format is as follows: a 4-byte header identifier field, a 24-byte aggregate parameter field, a 12-byte weighting coefficient field, and a 4-byte checksum field. The central server reads the network address information of all participating nodes, establishes communication links based on each network address, and sends the data packet to each UAV edge node through these communication links.

[0026] The beneficial effect is that the additional codes generated by each node during the encryption phase directly participate in the calculation of the aggregation weights. Since each node's additional code is uniquely determined by its own wind speed, temperature, and humidity data, and the additional codes of different nodes are different, the weighting coefficients obtained by each node in the aggregation operation are also different, realizing a differentiated contribution aggregation mechanism and effectively avoiding the bias that may be introduced by equal-weighted averaging. The additional codes of each node have already been used as multiplication scaling factors in the original parameter values ​​during the encryption phase, and they are again used as the basis for calculating the weighting coefficients during the aggregation phase. This realizes the dual driving force of the same set of environmental data on both the encryption and aggregation stages, enhancing the ability of environmental data to control the global aggregation results. The central server only calculates the weighting coefficients based on the additional codes and performs linear weighted summation, without needing to read or parse the original information in the disturbance parameters uploaded by each node. The aggregation process is completed entirely at the numerical level, remaining isolated from the original data of each node. While ensuring that the aggregation results effectively reflect the comprehensive information of all nodes, it further isolates the central server from accessing the original data. Meanwhile, the linear weighted summation operation is fully compatible with the inverse decryption operation of each subsequent node. After receiving the aggregation parameters, each node can perform the inverse operation item by item according to the perturbation components it stores, and accurately restore the aggregation form of the original data from the aggregation component values, providing a reliable data foundation for subsequent personalized plant protection decisions.

[0027] S3. Based on the disturbance component corresponding to the operation flight number, perform inverse perturbation decryption on the aggregated parameters to obtain the personalized plant protection decision parameters for each node; In this embodiment of the invention, the step of performing inverse perturbation decryption on the aggregated parameters based on the perturbation component corresponding to the operation flight number to obtain the personalized plant protection decision parameters for each node includes: According to the parameter dimension arrangement order, the disturbance component associated with the corresponding flight number in the storage unit is sequentially allocated to the corresponding dimension component of the aggregation parameter; Subtract the value of each corresponding dimension component in the aggregation parameter from the value of the perturbation component to obtain the multiple component of the edge node; Multiply the multiplier component by one percent and remove the additional code corresponding to the edge node to obtain the personalized plant protection decision dimension component of the edge node. Based on the parameter dimension arrangement order, the personalized plant protection decision dimension components of the edge nodes are combined into personalized plant protection decision parameters.

[0028] Before sequentially allocating the disturbance components associated with the corresponding flight number in the storage unit to the corresponding dimension components of the aggregation parameter according to the parameter dimension arrangement order, the method further includes: When the time difference between the storage time of the perturbation component and the current time exceeds a preset valid time threshold, the perturbation component is rejected for inverse perturbation decryption, and a perturbation component failure report is sent to the central server.

[0029] After receiving the aggregated parameters returned by the central server, the edge node of the UAV reads the flight number of the current operation flight. Taking node A as an example, the aggregated parameters received by node A are three values: 7649.0, 25472.9, and 19059.8, which correspond to the spray flow rate decision component, spray altitude decision component, and flight speed decision component, respectively. Node A reads the current operation flight number as 27.

[0030] In one feasible implementation, the UAV edge node verifies the timeliness of the perturbation component. Node A uses flight number 27 as the index key to query and read the perturbation component sequence associated with flight number 27 from the storage block of its local storage unit. This sequence contains real-time wind speed data 3.6, real-time ambient temperature data 28.5, and real-time ambient humidity data 73. Node A reads the storage timestamp of this perturbation component sequence as 14:23 and the current time as 14:38. Subtracting the storage time from the current time yields a time difference of 15 minutes. Node A compares this 15-minute time difference with a preset valid time threshold. If the preset valid time threshold is 30 minutes, then 15 minutes is less than 30 minutes, and Node A determines the perturbation component is valid and continues with subsequent decryption steps. If the preset valid time threshold is 10 minutes, then 15 minutes is greater than 10 minutes, and Node A determines the perturbation component is invalid, refuses to use it for inverse perturbation decryption, and sends a perturbation component invalidation report to the central server.

[0031] In one feasible implementation, node A assigns the disturbance components to the corresponding dimensional components of the aggregation parameters. Node A, according to the parameter dimensional order, assigns the real-time wind speed data 3.6 in the disturbance component sequence to the corresponding position of the spray flow rate decision component of the aggregation parameters, assigns the real-time ambient temperature data 28.5 to the corresponding position of the spray height decision component of the aggregation parameters, and assigns the real-time ambient humidity data 73 to the corresponding position of the flight speed decision component of the aggregation parameters.

[0032] In one feasible implementation, node A performs inverse perturbation decryption operations on each dimension component of the aggregated parameters. For the spray flow rate decision component, node A subtracts the real-time wind speed data of 3.6 from the aggregated parameter value of 7649.0 to obtain the multiple component of the spray flow rate dimension, 7645.4. Node A divides the multiple component 7645.4 by 100 to obtain 76.454. Node A reads the additional code 30 generated by the node during the encryption phase from the locally stored additional code record, divides 76.454 by the additional code 30, and obtains the personalized plant protection decision dimension component of the spray flow rate dimension, 2.548. For the spray height decision component, node A subtracts the real-time ambient temperature data of 28.5 from the aggregated parameter value of 25472.9 to obtain the multiple component of the spray height dimension, 25444.4. Dividing this by 100 gives 254.444, which is then divided by the additional code 82 to obtain the personalized plant protection decision dimension component of the spray height dimension, 3.103. For the flight speed decision component, node A subtracts the real-time environmental humidity data of 73 from the aggregated parameter value of 19059.8 to obtain the multiple component of the flight speed dimension, 18986.8. Dividing this by 100 gives 189.868. Dividing this by the additional code 37 gives 5.132, the personalized plant protection decision dimension component of the flight speed dimension.

[0033] In one feasible implementation, node A arranges the personalized plant protection decision dimension components 2.548 (spray flow rate), 3.103 (spray height), and 5.132 (flight speed) in sequence according to the parameter dimension, and combines them into the personalized plant protection decision parameters of node A.

[0034] The beneficial effects are that each UAV edge node reads the perturbation component from its local storage unit based on the current flight number and performs inverse perturbation decryption. Since the additional code used for decryption is exactly the same as that used for encryption, the decryption operation can accurately restore the original plant protection decision parameters, achieving a symmetrical closed loop between encryption and decryption. Furthermore, the entire decryption process is completed independently on each node; the central server only returns unified aggregate parameters without participating in any decryption operation. The central server remains unaware of the perturbation components stored by each node or the personalized parameters after decryption. Before decryption, each node compares the storage time of the perturbation component with the current time. Decryption is only performed if the time difference is within a preset valid time threshold, preventing deviations caused by expired perturbation components being used for incorrect decryption. Each node has a unique additional code, resulting in different personalized plant protection decision dimension components recovered under the same aggregate parameters. This enables differentiated output based on the environmental characteristics of different operational areas, providing a personalized data foundation for each node to generate spraying parameter instructions adapted to its own region.

[0035] S4. Based on the personalized plant protection decision parameters, adjust the plant protection spraying parameters for the next flight of the UAV.

[0036] In this embodiment of the invention, the instruction to adjust the spraying parameters for the next flight of the UAV based on the personalized plant protection decision parameters includes: The personalized plant protection decision parameters are fused with the local plant protection decision parameters currently stored in the UAV edge node to obtain the updated plant protection decision parameters of the UAV edge node. Based on the spray flow rate decision component, spray height decision component, and flight speed decision component in the updated plant protection decision parameters, the UAV is regulated to generate spray flow rate control command, flight height control command, and flight speed control command for the next operation flight of the UAV. The spray flow control command, the flight altitude control command, and the flight speed control command are combined into the plant protection operation spray parameter command and written into the flight control execution queue of the UAV.

[0037] The step of fusing the personalized plant protection decision parameters with the local plant protection decision parameters currently stored in the UAV edge node to obtain the updated plant protection decision parameters of the UAV edge node includes: Based on the preset fusion weight coefficient, the component values ​​of each dimension of the personalized plant protection decision parameters and the local plant protection decision parameters are weighted and summed to obtain the fusion component values ​​of each dimension. The values ​​of the fused components are combined into the updated plant protection decision parameters according to the parameter dimension arrangement order.

[0038] The edge nodes of the drone perform fusion calculations with the currently stored local plant protection decision parameters to obtain updated plant protection decision parameters. Taking node A as an example, the personalized plant protection decision parameters obtained by node A after S3 are: spray flow rate decision component 2.548, spray height decision component 3.103, and flight speed decision component 5.132. The currently stored local plant protection decision parameters of node A are: spray flow rate decision component 2.5, spray height decision component 2.8, and flight speed decision component 5.0.

[0039] In one feasible implementation, the UAV edge node dynamically determines the fusion weight coefficients based on the stability indicators of the current operational flight. Specifically, the UAV edge node reads the flight attitude data of the current operational flight, which includes roll angle, pitch angle, and yaw angle. It calculates the flight stability score based on the ratio of the absolute values ​​of the roll angle, pitch angle, and yaw angle to their corresponding thresholds. The flight stability score ranges from 0 to 1. The UAV edge node reads the flight stability score of the previous operational flight from its local storage unit, adds the current flight stability score to the previous flight stability score, and divides the sum by 2 to obtain the average stability score. This average stability score is used as the fusion weight coefficient. Taking node A as an example, the absolute value of the roll angle of the current voyage is 2.3 degrees with a threshold of 5 degrees, the absolute value of the pitch angle is 1.8 degrees with a threshold of 5 degrees, and the absolute value of the yaw angle is 3.1 degrees with a threshold of 8 degrees. The three ratios are 0.46, 0.36, and 0.3875, respectively. The maximum value of 0.46 is taken as the stability score of the current voyage. The stability score of the previous voyage is 0.52, and the average stability score is 0.49. Node A uses 0.49 as the fusion weight coefficient.

[0040] In one feasible implementation, the UAV edge nodes perform a weighted summation of the components of each dimension according to the parameter dimension arrangement order. Specifically, for the spray flow rate decision component, node A multiplies the spray flow rate decision component 2.548 in the personalized plant protection decision parameters by the fusion weight coefficient 0.49 to obtain 1.249, multiplies the spray flow rate decision component 2.5 in the currently stored local plant protection decision parameters by 1 minus 0.49 to obtain 0.51 to obtain 1.275, and adds 1.249 and 1.275 to obtain the fused component value of the spray flow rate dimension, 2.524. For the spray height decision component, node A multiplies the spray height decision component 3.103 in the personalized plant protection decision parameters by 0.49 to obtain 1.520, multiplies the spray height decision component 2.8 in the currently stored local plant protection decision parameters by 0.51 to obtain 1.428, and adds 1.520 and 1.428 to obtain the fused component value of the spray height dimension, 2.948. For the flight speed decision component, Node A multiplies the flight speed decision component (5.132) in the personalized plant protection decision parameters by 0.49 to obtain 2.515, and multiplies the flight speed decision component (5.0) in the currently stored local plant protection decision parameters by 0.51 to obtain 2.550. Adding 2.515 and 2.550 yields the fused component value for the flight speed dimension, 5.065. Node A then arranges and combines the fused component values ​​for the spray flow rate dimension (2.524), spray height dimension (2.948), and flight speed dimension (5.065) in the order of parameter dimensions to obtain Node A's updated plant protection decision parameters.

[0041] In one feasible implementation, the UAV edge node extracts each decision component from the updated plant protection decision parameters to generate control commands. Node A extracts the spray flow rate decision component 2.524 as the target spray flow rate, the spray height decision component 2.948 as the target flight altitude, and the flight speed decision component 5.065 as the target flight speed from the updated plant protection decision parameters. Node A generates a spray flow control command based on the target spray flow rate of 2.524 liters per minute. Specifically, it divides the target spray flow rate of 2.524 by the maximum spray flow rate of 6.0 to obtain a duty cycle of 0.421, converts this duty cycle into a pulse width modulation signal duty cycle of 42.1%, and generates a spray flow control command containing the duty cycle information. Node A generates a flight altitude control command based on the target flight altitude of 2.948 meters. Specifically, it uses the difference of 0.448 meters between the target flight altitude of 2.948 meters and the current flight altitude of 2.5 meters as the vertical speed adjustment amount, limiting the vertical speed adjustment to a maximum range of 1.5 meters per second. The calculated vertical speed adjustment amount is 0.299 meters per second, and a flight altitude control command incorporating this vertical speed adjustment is generated. Node A also generates a flight speed control command based on the target flight speed of 5.065 meters per second. Specifically, it uses the difference of 0.065 meters per second between the target flight speed of 5.065 meters per second and the current flight speed of 5.0 meters per second as the horizontal speed adjustment amount, calculating the horizontal speed adjustment amount to 0.065 meters per second, and a flight speed control command incorporating this horizontal speed adjustment is generated.

[0042] In one feasible implementation, the UAV edge node combines various control commands into a plant protection spraying parameter command and writes it into the flight control execution queue. Node A combines the spray flow control command, flight altitude control command, and flight speed control command into a plant protection spraying parameter command according to the priority order of the control commands, with the priority order being spray flow control command first, then flight altitude control command first, then flight speed control command. Node A writes the combined plant protection spraying parameter command to the end of the flight control execution queue. The flight control execution queue adopts a first-in-first-out scheduling method. There are already two frames of commands to be executed in the queue, and the currently written command is third in the queue. After the first two frames of commands in the queue are executed, this command will be sent to the flight controller for execution, realizing the coordinated control of the spray flow, flight altitude, and flight speed of the UAV for the next operation flight.

[0043] The beneficial effect is that each UAV edge node performs a weighted fusion of personalized plant protection decision parameters with the currently stored local plant protection decision parameters. The fusion weight coefficient is dynamically determined based on the flight attitude stability of the current operation flight, ensuring that personalized decision parameters with high flight stability have a higher proportion in the fusion, while personalized decision parameters with low flight stability are appropriately suppressed, thereby generating more reliable updated plant protection decision parameters. The UAV edge node extracts three decision components—spray flow rate, spray altitude, and flight speed—from the updated plant protection decision parameters, and generates corresponding spray flow rate control commands, flight altitude control commands, and flight speed control commands, respectively. These three commands are combined into a complete plant protection operation spray parameter command and written into the flight control execution queue, realizing a complete closed-loop link from cross-regional data collaboration to local operation execution. Because the individual plant protection decision parameters of each node are different, the updated plant protection decision parameters obtained after fusion are also different. Finally, the spraying parameter instructions written into the flight control queue of each node are also different. This makes the plant protection operation strategy of each node match the actual environment such as crop distribution, terrain undulation and climate conditions of its geographical area. Under the framework of cross-regional data collaborative optimization, global information sharing and local decision personalization are realized at the same time.

[0044] like Figure 2 The diagram shown is a functional block diagram of a cross-regional UAV plant protection data privacy protection system based on federated learning, provided by an embodiment of the present invention.

[0045] The federated learning-based cross-regional UAV plant protection data privacy protection system 100 described in this invention can be installed in an electronic device. Depending on the functions implemented, the federated learning-based cross-regional UAV plant protection data privacy protection system 100 may include an environmental disturbance encryption and storage module 101, a disturbance parameter aggregation module 102, a personalized inverse disturbance decryption module 103, and a spraying parameter decision adjustment module 104. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0046] In this embodiment, the functions of each module / unit are as follows: The environmental disturbance encryption and storage module 101 is used to use the real-time wind speed data, real-time ambient temperature data and real-time ambient humidity data of the edge node of the UAV in the current operation flight as disturbance components to encrypt the local plant protection decision parameters generated by the UAV plant protection data training, so as to obtain the disturbance parameters of the UAV, and store the disturbance components in the local storage unit of the node according to the current operation flight number. The disturbance parameter aggregation module 102 is used to perform aggregation calculations on the disturbance parameters when they are uploaded to the central server, to obtain the aggregated parameters of the UAV, and return them to each UAV edge node. The personalized inverse perturbation decryption module 103 is used to perform inverse perturbation decryption on the aggregated parameters according to the perturbation component corresponding to the operation flight number, so as to obtain the personalized plant protection decision parameters of each node. The spraying parameter decision adjustment module 104 is used to adjust the spraying parameter instructions for the next operation flight of the UAV according to the personalized plant protection decision parameters.

[0047] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0048] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0049] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0050] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0051] The embodiments of this application can acquire and process relevant data based on an artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0052] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for protecting the privacy of cross-regional unmanned aerial vehicle (UAV) plant protection data based on federated learning, characterized in that: Includes the following steps: S1. The real-time wind speed data, real-time ambient temperature data, and real-time ambient humidity data of the edge nodes of the UAV in the current operation flight are used as disturbance components to encrypt the local plant protection decision parameters generated by the UAV plant protection data training, thereby obtaining the disturbance parameters of the UAV. The disturbance components are then stored in the local storage unit of the node according to the current operation flight number. S2. When the disturbance parameters are uploaded to the central server, the disturbance parameters are aggregated to obtain the aggregated parameters of the UAV, and then returned to each UAV edge node. S3. Based on the disturbance component corresponding to the operation flight number, perform inverse perturbation decryption on the aggregated parameters to obtain the personalized plant protection decision parameters for each node; S4. Based on the personalized plant protection decision parameters, adjust the plant protection spraying parameters for the next flight of the UAV.

2. The method for protecting the privacy of cross-regional unmanned aerial vehicle (UAV) plant protection data based on federated learning as described in claim 1, characterized in that, The method involves using real-time wind speed data, real-time ambient temperature data, and real-time ambient humidity data from the edge nodes of the UAV in the current operational flight as perturbation components to perturb and encrypt the local plant protection decision parameters generated by training the UAV plant protection data, thereby obtaining the perturbation parameters of the UAV, including: Based on the local plant protection data of the drone's edge nodes, the locally stored set of plant protection decision parameters is trained and updated to obtain the local plant protection decision parameters of the drone. Based on the parameter dimension arrangement order of the local plant protection decision parameters, the real-time wind speed data, real-time ambient temperature data, and real-time ambient humidity data of the edge nodes of the UAV in the current operation flight are used as disturbance components and assigned to the corresponding dimension components of the local plant protection decision parameters. Based on the disturbance component, the corresponding dimension component is perturbed and encrypted, and the disturbance parameters of the UAV are obtained according to the parameter dimension arrangement order.

3. The method for protecting the privacy of cross-regional unmanned aerial vehicle (UAV) plant protection data based on federated learning as described in claim 2, characterized in that, The step of perturbing and encrypting the corresponding dimension components based on the perturbation components, and obtaining the perturbation parameters of the UAV according to the parameter dimension arrangement order, includes: The tens digit and the units digit of the disturbance component value are combined into a two-digit additional code in the order of units digit first and tens digit last. The product of the dimensional component and the additional code is multiplied by one hundred and added to the perturbation component to obtain the encrypted component value of the corresponding dimensional component. This encrypted component value is then combined according to the parameter dimension arrangement order to form the perturbation parameter of the UAV.

4. The method for protecting the privacy of cross-regional UAV agricultural data based on federated learning as described in claim 2, characterized in that, The step of storing the disturbance component in the node's local storage unit according to the current mission number includes: The real-time wind speed data, the real-time ambient temperature data, and the real-time ambient humidity data are arranged in the order of the collection timestamps to form the disturbance component sequence of the UAV. The disturbance component sequence is associated with the current mission number and written into a pre-allocated storage block in the storage unit.

5. The method for protecting the privacy of cross-regional unmanned aerial vehicle (UAV) plant protection data based on federated learning as described in claim 3, characterized in that, When the disturbance parameters are uploaded to the central server, the disturbance parameters are aggregated to obtain the aggregated parameters of the UAV, and then returned to each UAV edge node, including: The ratio of the additional code to the sum of all additional codes is used as the weighting coefficient of the edge node; Based on the weighting coefficients, the values ​​of the disturbance parameters in each dimension are weighted and summed to obtain the aggregate component value of the UAV. The aggregated component values ​​are combined into aggregated parameters according to the order of the parameter dimensions. The aggregation parameters are encapsulated into data packets and sent to each UAV edge node.

6. The method for protecting the privacy of cross-regional unmanned aerial vehicle (UAV) plant protection data based on federated learning as described in claim 5, characterized in that, The step of performing inverse perturbation decryption on the aggregated parameters based on the perturbation component corresponding to the operation flight number to obtain the personalized plant protection decision parameters for each node includes: According to the parameter dimension arrangement order, the disturbance component associated with the corresponding flight number in the storage unit is sequentially allocated to the corresponding dimension component of the aggregation parameter; Subtract the value of each corresponding dimension component in the aggregation parameter from the value of the perturbation component to obtain the multiple component of the edge node; Multiply the multiplier component by one percent and remove the additional code corresponding to the edge node to obtain the personalized plant protection decision dimension component of the edge node. Based on the parameter dimension arrangement order, the personalized plant protection decision dimension components of the edge nodes are combined into personalized plant protection decision parameters.

7. The method for protecting the privacy of cross-regional unmanned aerial vehicle (UAV) plant protection data based on federated learning as described in claim 6, characterized in that, Before sequentially allocating the disturbance components associated with the corresponding flight number in the storage unit to the corresponding dimension components of the aggregation parameter according to the parameter dimension arrangement order, the method further includes: When the time difference between the storage time of the perturbation component and the current time exceeds a preset valid time threshold, the perturbation component is rejected for inverse perturbation decryption, and a perturbation component failure report is sent to the central server.

8. The method for protecting the privacy of cross-regional unmanned aerial vehicle (UAV) plant protection data based on federated learning as described in claim 6, characterized in that, The instruction to adjust the spraying parameters for the next drone flight based on the personalized plant protection decision parameters includes: The personalized plant protection decision parameters are fused with the local plant protection decision parameters currently stored in the UAV edge node to obtain the updated plant protection decision parameters of the UAV edge node. Based on the spray flow rate decision component, spray height decision component, and flight speed decision component in the updated plant protection decision parameters, the UAV is regulated to generate spray flow rate control command, flight height control command, and flight speed control command for the next operation flight of the UAV. The spray flow control command, the flight altitude control command, and the flight speed control command are combined into the plant protection operation spray parameter command and written into the flight control execution queue of the UAV.

9. The method for protecting the privacy of cross-regional unmanned aerial vehicle (UAV) plant protection data based on federated learning as described in claim 8, characterized in that, The step of fusing the personalized plant protection decision parameters with the local plant protection decision parameters currently stored in the UAV edge node to obtain the updated plant protection decision parameters of the UAV edge node includes: Based on the preset fusion weight coefficient, the component values ​​of each dimension of the personalized plant protection decision parameters and the local plant protection decision parameters are weighted and summed to obtain the fusion component values ​​of each dimension. The values ​​of the fused components are combined into the updated plant protection decision parameters according to the parameter dimension arrangement order.

10. A cross-regional drone-based plant protection data privacy protection system based on federated learning, characterized in that: The system for implementing the cross-regional drone-based plant protection data privacy protection method based on federated learning as described in claim 1 includes: The environmental disturbance encryption and storage module is used to use the real-time wind speed data, real-time ambient temperature data and real-time ambient humidity data of the edge node of the UAV in the current operation flight as disturbance components to encrypt the local plant protection decision parameters generated by the UAV plant protection data training, obtain the disturbance parameters of the UAV, and store the disturbance components in the local storage unit of the node according to the current operation flight number. The disturbance parameter aggregation module is used to perform aggregation calculations on the disturbance parameters when they are uploaded to the central server, obtain the aggregated parameters of the UAV, and return them to each UAV edge node. The personalized inverse perturbation decryption module is used to perform inverse perturbation decryption on the aggregated parameters according to the perturbation component corresponding to the operation flight number, so as to obtain the personalized plant protection decision parameters of each node. The spraying parameter decision adjustment module is used to adjust the spraying parameter instructions for the next operation flight of the UAV based on the personalized plant protection decision parameters.

Images