Data management field checking method and system based on unmanned aerial vehicle

Abstract

The invention relates to a data management field checking method and system based on an unmanned aerial vehicle, and belongs to the technical field of power grid data checking. The method comprises the steps of obtaining a data governance site checking task; selecting an unmanned aerial vehicle for executing a field checking task requirement; planning a flight route and a task area of the unmanned aerial vehicle; collecting data of a target in the task area in real time; performing primary processing on the original data acquired by the unmanned aerial vehicle; analyzing the data after the primary processing, and extracting key information; the key information comprises coordinate positions and image features; checking the extracted key information and pre-stored target data source end information, and comparing a checking result to generate a problem data list; checking results are input into the system, problem data treatment is carried out, and target data source end information is updated. According to the invention, the unmanned aerial vehicle carries out the data treatment field checking work, reduces the labor cost of the field checking, achieves the improvement of the field checking efficiency and safety, and is easy to popularize and apply.

Description

Technical Field

[0001] The present invention belongs to the technical field of power grid data verification, and in particular relates to an on-site verification method and system for data management based on unmanned aerial vehicles. Background Art

[0002] With the rapid development of smart grids, the scale of grid data continues to expand, and the quality of data is crucial to the safe, stable and economical operation of the grid. On-site data verification, as an important part of data governance, plays an irreplaceable role in ensuring the accuracy and integrity of data.

[0003] In the current power grid data management process, the existing on-site verification method uses manual search, recording and verification one by one, which consumes a lot of labor costs. In addition, power equipment is widely distributed, the facility layout is complex, and the natural environment of the substations or overhead lines where some equipment is located is complex and harsh.

[0004] The traditional manual on-site verification method is inefficient and requires high investment. In addition, on-site verification of power equipment in complex geographical environments can threaten the safety of on-site verification personnel. Therefore, how to overcome the shortcomings of existing technologies is an urgent problem to be solved in the field of power grid data verification technology. Summary of the invention

[0005] The purpose of the present invention is to solve the deficiencies of the prior art and to provide a method and system for on-site verification of data management based on drones.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0007] A data governance on-site verification method based on drones, comprising the following steps:

[0008] S1: Obtain data governance on-site verification tasks;

[0009] S2: Select the drone that will perform the on-site verification mission requirements;

[0010] S3: Plan the flight route and mission area of ​​the UAV according to the terrain and weather conditions of the area to be verified and the distribution of the targets to be verified;

[0011] S4: The drone flies according to the flight route planned in S3 and collects data on targets in the mission area in real time;

[0012] S5: Preliminary processing of the raw data collected by the drone;

[0013] S6: parsing the data after preliminary processing to extract key information; the key information includes coordinate position and image features;

[0014] S7: Verify the extracted key information with the pre-stored information of the target data source, compare the verification results, and generate a list of problematic data;

[0015] S8: Enter the verification results into the system, conduct problem data governance, and update the target data source information.

[0016] Furthermore, preferably, in S1, the data governance on-site verification task is formulated based on the anomaly detection results or verification requirements.

[0017] Further, preferably, in S4, the target data includes images, videos and sensor data.

[0018] Further, preferably, in S5, the preliminary processing includes data cleaning, denoising and correction.

[0019] Further, preferably, in S6, the specific method for extracting the key information is:

[0020] Step 1: Construct a target detection network; the target detection network consists of three parts: a backbone network, a multiple feature fusion network and a prediction head; the backbone network is used to extract feature information of the input drone image and output feature maps of different sizes; the multiple feature fusion network is used to perform feature fusion on feature maps of different sizes; the prediction head is used to perform detection based on the fused features to obtain the final detection result; the backbone network consists of a coordinate and global information aggregation module; the coordinate and global information aggregation module includes a coordinate information extraction branch and a global information extraction branch. The coordinate information extraction branch introduces an attention mechanism to allow it to always pay attention to the coordinate information while using convolution for feature extraction. The global information extraction branch obtains the global information of the feature map through pooling operations, and finally obtains features with richer context information by fusing the features extracted by the two branches;

[0021] Step 2: Train the constructed target detection network to obtain the target detection model, and use Focal-EIOUloss as the bounding box loss of the model;

[0022] Step 3: Perform multi-scale target detection on drone aerial images based on the target detection model;

[0023] In step 1, a dual attention mechanism is introduced, where the dual attention mechanism is a bottleneck attention module and a SENet attention mechanism module, thereby obtaining an improved target detection model;

[0024] The historically collected target data is used as input and its key information is used as output for training to obtain an extraction model;

[0025] The extraction model is then used to extract key information in real time.

[0026] Further, preferably, the SENet attention mechanism module includes a Squeeze operation, an Excitation operation, and a Reweight operation;

[0027] First, the feature map is squeezed to obtain the global feature. Then the global feature is excitation operated to learn the relationship between channels and obtain the weights of different channels. Finally, the original feature map is multiplied by the reweight operation to obtain the final feature and complete the target recognition.

[0028] Further, preferably, in S6, the specific method of the Squeeze operation is as follows:

[0029] Squeeze operation, the original feature map is subjected to maximum pooling and average pooling respectively, and feature maps a(W, H, 1) and m(W, H, 1) are generated respectively, and then concatenated so that the obtained weight map has both maximum value information and average value information. Each feature in the feature map is subjected to global average pooling operation and compressed into a real number (1, 1, C), specifically:

[0030]

[0031] In the formula, z c Indicates the output of the Squeeze operation, u c represents the cth channel of the feature map, H and W represent the height and width of the feature map, and u c (i,j) represents the i-th row and j-th column element of the c-th channel.

[0032] Further, preferably, in S6, the specific method of the Excitation operation is as follows:

[0033] The global features are convolved with a kernel size of 7 and a padding size of 3, so that the size of the global features is reduced from c(W, H, 2) to v(W, H, 1). The weights for each feature channel are generated by the parameter W, specifically:

[0034] s=F ex (z,W)=σ(g(z,W))=σ(w2δ(w1z));

[0035] Where s is the weight obtained by the Excitation operation, z is the output of the Squeeze operation, W1 and W2 are the weights of the two-layer fully connected operation, and δ is the ReLU function.

[0036] Further, preferably, in S6, the specific method of the Reweight operation is as follows: the weight of the output of Excitation is weighted to the original feature map by multiplication, specifically:

[0037]

[0038] In the formula, is the final output feature map, u c is the cth channel of the feature map, s c is the cth weight in s.

[0039] The present invention also provides a data management on-site verification system based on a drone, which adopts the above-mentioned data management on-site verification method based on a drone, including:

[0040] On-site verification task management module, used for closed-loop management of on-site verification tasks, including task creation, task allocation, task status update, task progress monitoring, comparison and verification supervision and data archiving;

[0041] The UAV path planning module and the on-site verification task management module are used to plan the flight route and mission area of ​​the UAV according to the terrain and weather conditions of the verification area and the distribution of the targets to be verified;

[0042] The UAV data acquisition module is connected to the UAV path planning module and is used to collect data on targets in the mission area, including images, videos, and sensor data;

[0043] The on-site data verification module is connected to the drone data collection module and is used to perform preliminary processing on the raw data collected by the drone; analyze the data after preliminary processing to extract key information; verify the extracted key information with the pre-stored target data source information, compare the verification results, and generate a list of problematic data;

[0044] The problem data governance module is connected to the on-site data verification module and is used to give problem data governance suggestions and update the managed target data as the target data source information.

[0045] In S1 of the present invention, the data governance on-site verification task is formulated according to the anomaly detection result or verification requirement. Among them, the anomaly detection result is detected by the existing method, and the present invention does not limit this.

[0046] In the present invention S2, the drone needs to be equipped with necessary equipment such as a high-resolution camera, an infrared sensor, a laser radar and a drone cerebellum.

[0047] In the present invention, the target data source end information is the original key information of the target pre-stored in the system before executing the on-site verification.

[0048] The verification results obtained by the drone of the present invention are transmitted to the background system by using a wireless network, and the verification results are stored by using a cloud platform, or a mobile workstation is used to provide storage and computing power support.

[0049] In the present invention, the on-site verification task management module is also used for scoring and reporting.

[0050] In the present invention, the UAV path planning module can construct a low-power and long-flight UAV path that is adaptive to scene changes for static targets in on-site verification tasks and complex mixed scenes where individual target positions change due to environmental factors, further improve the flight time of the UAV path planning method during the verification process, and reduce the power wasted by multiple return charges of the UAV;

[0051] In the present invention, the UAV data acquisition module can collect data in the face of complex mountain, forest and foggy environmental backgrounds and the problem that the targets to be identified are diverse and have large size differences. Through the on-site data verification module, the detection effect of the UAV on small target nameplates of different scales during high-altitude shooting is improved, the search capability of the verification target is enhanced, and the image quality is improved after data collection.

[0052] In the present invention, the on-site data verification module utilizes preset verification rules to automatically verify the extracted key information with the pre-stored information of the target data source, quickly identify and mark potential errors or abnormal data, and form a list of problem data; when verifying, it is preferred to verify the consistency, integrity, and compliance of the data.

[0053] In the present invention, the problem data governance module can provide problem data consistency, integrity, and compliance governance. After governance, the data is synchronously updated with the target data source information through data interface docking or electronic transfer.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] The present invention integrates different methods and makes full use of their respective advantages to obtain more accurate target detection results. This enables the application to cope with the low pixel ratio of small targets and the low precision of intelligent recognition of small target defects. The on-site verification of data management is carried out by drones, which reduces the manpower cost of on-site verification. The efficiency and safety of on-site verification are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a flow chart of the on-site verification method of data management based on drones of the present invention;

[0057] Figure 2 This is a schematic diagram of the structure of the data management on-site verification system based on drones of the present invention;

[0058] Figure 3 This is a feature map enhanced fusion map of the on-site verification method for data management based on drones of the present invention. DETAILED DESCRIPTION

[0059] The present invention is further described in detail below in conjunction with embodiments.

[0060] Those skilled in the art will appreciate that the following examples are only used to illustrate the present invention and should not be considered to limit the scope of the present invention. If no specific techniques or conditions are specified in the examples, the techniques or conditions described in the literature in the art or the product specifications are used. If the manufacturer of the materials or equipment used is not specified, they are all conventional products that can be purchased.

[0061] Example 1

[0062] A data management on-site verification method based on unmanned aerial vehicle, characterized by comprising the following steps:

[0063] S1: Obtain data governance on-site verification tasks;

[0064] S2: Select the drone that will perform the on-site verification mission requirements;

[0065] S3: Plan the flight route and mission area of ​​the UAV according to the terrain and weather conditions of the area to be verified and the distribution of the targets to be verified;

[0066] S4: The drone flies according to the flight route planned in S3 and collects data on targets in the mission area in real time;

[0067] S5: Preliminary processing of the raw data collected by the drone;

[0068] S6: parsing the data after preliminary processing to extract key information; the key information includes coordinate position and image features;

[0069] S7: Verify the extracted key information with the pre-stored information of the target data source, compare the verification results, and generate a list of problematic data;

[0070] S8: Enter the verification results into the system, conduct problem data governance, and update the target data source information.

[0071] Example 2

[0072] A data governance on-site verification method based on drones, comprising the following steps:

[0073] S1: Obtain data governance on-site verification tasks;

[0074] S2: Select the drone that will perform the on-site verification mission requirements;

[0075] S3: Plan the flight route and mission area of ​​the UAV according to the terrain and weather conditions of the area to be verified and the distribution of the targets to be verified;

[0076] S4: The drone flies according to the flight route planned in S3 and collects data on targets in the mission area in real time;

[0077] S5: Preliminary processing of the raw data collected by the drone;

[0078] S6: parsing the data after preliminary processing to extract key information; the key information includes coordinate position and image features;

[0079] S7: Verify the extracted key information with the pre-stored information of the target data source, compare the verification results, and generate a list of problematic data;

[0080] S8: Enter the verification results into the system, conduct problem data governance, and update the target data source information.

[0081] In S1, the data governance on-site verification task is formulated based on the anomaly detection results or verification requirements.

[0082] In S4, the target data includes images, videos, and sensor data.

[0083] In S5, preliminary processing includes data cleaning, denoising and correction.

[0084] In S6, the specific method for extracting key information is:

[0085] Step 1: Construct a target detection network; the target detection network consists of three parts: a backbone network, a multiple feature fusion network and a prediction head; the backbone network is used to extract feature information of an input drone image and output feature maps of different sizes; the multiple feature fusion network is used to perform feature fusion on feature maps of different sizes; the prediction head is used to perform detection based on the fused features to obtain the final detection result; the backbone network consists of a coordinate and global information aggregation module; the coordinate and global information aggregation module includes a coordinate information extraction branch and a global information extraction branch. The coordinate information extraction branch introduces an attention mechanism to allow it to always pay attention to coordinate information while using convolution for feature extraction. The global information extraction branch obtains the global information of the feature map through a pooling operation, and finally obtains features with richer context information by fusing the features extracted by the two branches; the present invention improves on the basis of the existing model, and changes the traditional attention mechanism into a dual attention mechanism, specifically a bottleneck attention module and a SENet attention mechanism module;

[0086] Step 2: Train the constructed target detection network to obtain the target detection model, and use Focal-EIOUloss as the bounding box loss of the model;

[0087] Step 3: Perform multi-scale target detection on drone aerial images based on the target detection model;

[0088] In step 1, a dual attention mechanism is introduced, where the dual attention mechanism is a bottleneck attention module and a SENet attention mechanism module, thereby obtaining an improved target detection model;

[0089] The historically collected target data is used as input and its key information is used as output for training to obtain an extraction model;

[0090] The extraction model is then used to extract key information in real time.

[0091] The SENet attention mechanism module includes Squeeze operation, Excitation operation and Reweight operation;

[0092] First, the feature map is squeezed to obtain the global feature. Then the global feature is excitation operated to learn the relationship between channels and obtain the weights of different channels. Finally, the original feature map is multiplied by the reweight operation to obtain the final feature and complete the target recognition.

[0093] In S6, the specific method of Squeeze operation is as follows:

[0094] Squeeze operation, the original feature map is subjected to maximum pooling and average pooling respectively, and feature maps a(W, H, 1) and m(W, H, 1) are generated respectively, and then concatenated so that the obtained weight map has both maximum value information and average value information. Each feature in the feature map is subjected to global average pooling operation and compressed into a real number (1, 1, C), specifically:

[0095]

[0096] In the formula, z c Indicates the output of the Squeeze operation, u c represents the cth channel of the feature map, H and W represent the height and width of the feature map, and u c (i,j) represents the i-th row and j-th column element of the c-th channel.

[0097] In S6, the specific method of Excitation operation is as follows:

[0098] The global features are convolved with a kernel size of 7 and a padding size of 3, so that the size of the global features is reduced from c(W, H, 2) to v(W, H, 1). The weights for each feature channel are generated by the parameter W, specifically:

[0099] s=F ex (z,W)=σ(g(z,W))=σ(w2δ(w1z));

[0100] Where s is the weight obtained by the Excitation operation, z is the output of the Squeeze operation, W1 and W2 are the weights of the two-layer fully connected operation, and δ is the ReLU function.

[0101] In S6, the specific method of the Reweight operation is as follows: the weight of the output of Excitation is multiplied to the original feature map, specifically:

[0102]

[0103] In the formula, is the final output feature map, u c is the cth channel of the feature map, s c is the cth weight in s.

[0104] A data management on-site verification system based on a drone, using the data management on-site verification method based on a drone of this embodiment, comprises:

[0105] On-site verification task management module, used for closed-loop management of on-site verification tasks, including task creation, task allocation, task status update, task progress monitoring, comparison and verification supervision and data archiving;

[0106] The UAV path planning module and the on-site verification task management module are used to plan the flight route and mission area of ​​the UAV according to the terrain and weather conditions of the verification area and the distribution of the targets to be verified;

[0107] The UAV data acquisition module is connected to the UAV path planning module and is used to collect data on targets in the mission area, including images, videos, and sensor data;

[0108] The on-site data verification module is connected to the drone data collection module and is used to perform preliminary processing on the raw data collected by the drone; analyze the data after preliminary processing to extract key information; verify the extracted key information with the pre-stored target data source information, compare the verification results, and generate a list of problematic data;

[0109] The problem data governance module is connected to the on-site data verification module and is used to give problem data governance suggestions and update the managed target data as the target data source information.

[0110] Example 3

[0111] Reference Figure 1-3 , a data governance on-site verification method based on drones, comprising the following steps:

[0112] S1: Obtain data governance on-site verification tasks;

[0113] S2: Select the drone that will perform the on-site verification mission requirements;

[0114] S3: Plan the flight route and mission area of ​​the UAV according to the terrain and weather conditions of the area to be verified and the distribution of the targets to be verified;

[0115] S4: The drone flies according to the flight route planned in S3 and collects data on targets in the mission area in real time;

[0116] S5: Preliminary processing of the raw data collected by the drone;

[0117] S6: parsing the processed data to extract key information, including coordinate positions and image features;

[0118] Drawing on the ideas of bottleneck attention module (BAM) and SE (sequence and excitation), a target detection model based on fusion strategy optimization selection and dual attention mechanism is designed. The dual focusing effect of feature map on target shape and position is improved through channel attention and spatial attention. The channel attention mechanism is divided into three operations. The input is feature map U(W, H, C). Through a series of convolution and pooling operations, a new feature map x(W, H, C) is obtained.

[0119] The specific operations are as follows:

[0120] Squeeze operation (Fsq) performs maximum pooling and average pooling on the original feature map to generate feature maps a(W, H, 1) and m(W, H, 1) respectively, and then concatenate them so that the obtained weight map has both maximum value information and average value information. Each feature in the feature map is subjected to global average pooling and compressed into a real number (1, 1, C), specifically:

[0121]

[0122] In the formula, Z crepresents the output of the Squeeze operation, represents the cth channel of the feature map, H and W represent the height and width of the feature map, u c (i, j) represents the i-th row and j-th column element of the c-th channel;

[0123] Excitation operation (Fex) is used to convolve the concatenated weight map with a kernel size of 7 and a padding size of 3, so that the size of the weight map is reduced from c(W, H, 2) to v(W, H, 1). Similar to the mechanism of the recurrent neural gate, the parameter W is used to generate weights for each feature channel. Specifically:

[0124] s=F ex (z,W)=σ(g(z,W))=σ(w2δ(w1z));

[0125] Where s is the weight obtained by the Excitation operation, z is the output of the Squeeze operation, W1 and W2 are fully connected operations, and δ is the ReLU function;

[0126] Reweight operation (Fscale) adds the weight map after the convolution layer to the original features by multiplication to obtain the feature map x(W, H, C). By fusing and parallelizing each feature map, important detail features are strengthened in a supervised manner and useless interference features are weakened. At the same time, the spatial information in the feature map is transformed accordingly to extract important spatial position information. The output weight of Excitation is multiplied to the original features. Specifically:

[0127]

[0128] Where x is the final output feature map, uc is the cth channel in the feature, and sc is the cth weight in s;

[0129] S7: Use big data analysis and artificial intelligence algorithms to check the processed data against the target data, and compare the check results to generate a list of problematic data;

[0130] S8: Enter the on-site verification results into the system, conduct problem data governance, update the target data source information, use wireless networks for data transmission, use cloud platforms to store data, or use mobile workstations to provide storage and computing support, and integrate different two-stage target detection methods to make full use of their respective advantages to obtain more accurate target detection results. The fused method can cope with the low pixel ratio of small targets and the low precision of intelligent recognition of small target defects. The on-site verification of data governance is carried out by drones, which reduces the labor cost of on-site verification. Improve the efficiency and safety of on-site verification.

[0131] A system for a data management on-site verification method based on an unmanned aerial vehicle, the system comprising:

[0132] On-site verification task management module: used for closed-loop management of on-site verification tasks, integrating key functions such as task creation, intelligent allocation, status update, progress monitoring, reminder communication, verification supervision, data archiving, and scoring and reporting;

[0133] UAV path planning module: For the complex mixed scenarios of static targets and individual target positions changing due to environmental factors in the on-site verification mission, a low-power and long-flight UAV intelligent verification path planning method that is adaptive to scene changes is constructed to further improve the flight time of the UAV path planning method during the verification process and reduce the power wasted by the UAV's multiple return charges;

[0134] UAV data acquisition module: Facing the complex mountain, forest and foggy environmental background and the problem of diverse and large size differences in the targets to be identified, it provides small target identification and image enhancement methods to improve the detection effect of small target nameplates of different sizes when the UAV is shooting at high altitude, enhance the search ability of the verification target, and improve the image quality after collecting and verifying the data;

[0135] On-site data verification module: Use preset verification rules to automatically verify the collected data and target data, quickly identify and mark potential errors or abnormal data to form a list of problematic data;

[0136] Problem data governance module: Govern the consistency, integrity and compliance of problem data. After governance, the target data source data is synchronously updated through data interface docking or electronic transfer.

[0137] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A data management on-site verification method based on drones, characterized in that: The steps include: S1: Obtain data governance on-site verification tasks; S2: Select the drone that will perform the on-site verification mission requirements; S3: Plan the flight route and mission area of ​​the UAV according to the terrain and weather conditions of the area to be verified and the distribution of the targets to be verified; S4: The drone flies according to the flight route planned in S3 and collects data on targets in the mission area in real time; S5: Preliminary processing of the raw data collected by the drone; S6: parsing the data after preliminary processing to extract key information; the key information includes coordinate position and image features; S7: Verify the extracted key information with the pre-stored information of the target data source, compare the verification results, and generate a list of problematic data; S8: Enter the verification results into the system, conduct problem data governance, and update the target data source information.

2. The method for on-site verification of data management based on drones according to claim 1 is characterized in that: In S1, the data governance on-site verification task is formulated based on the anomaly detection results or verification requirements.

3. The method for on-site verification of data management based on drones according to claim 1 is characterized in that: In S4, the target data includes images, videos, and sensor data.

4. The method for on-site verification of data management based on drones according to claim 1 is characterized in that: In S5, preliminary processing includes data cleaning, denoising and correction.

5. The method for on-site verification of data management based on drones according to claim 1 is characterized in that: In S6, the specific method for extracting key information is: Step 1: Construct a target detection network; the target detection network consists of three parts: a backbone network, a multiple feature fusion network and a prediction head; the backbone network is used to extract feature information of the input drone image and output feature maps of different sizes; the multiple feature fusion network is used to perform feature fusion on feature maps of different sizes; the prediction head is used to perform detection based on the fused features to obtain the final detection result; the backbone network consists of a coordinate and global information aggregation module; the coordinate and global information aggregation module includes a coordinate information extraction branch and a global information extraction branch. The coordinate information extraction branch introduces an attention mechanism to allow it to always pay attention to the coordinate information while using convolution for feature extraction. The global information extraction branch obtains the global information of the feature map through pooling operations, and finally obtains features with richer context information by fusing the features extracted by the two branches; Step 2: Train the constructed target detection network to obtain the target detection model, and use Focus-EIOUloss as the bounding box loss of the model; Step 3: Perform multi-scale target detection on drone aerial images based on the target detection model; In step 1, a dual attention mechanism is introduced, where the dual attention mechanism is a bottleneck attention module and a SE Net attention mechanism module, thereby obtaining an improved target detection model; The target data collected historically is used as input, and its key information is used as output for training to obtain an extraction model; The extraction model is then used to extract key information in real time.

6. The method for on-site verification of data management based on drones according to claim 1 is characterized in that: The SENet attention mechanism module includes Squeeze operation, Excitation operation and Reweight operation; First, the feature map is squeezed to obtain the global feature. Then the global feature is excitation operated to learn the relationship between channels and obtain the weights of different channels. Finally, the original feature map is multiplied by the re-weight operation to obtain the final feature and complete the target recognition.

7. The method for on-site verification of data management based on drones according to claim 6 is characterized in that: In S6, the specific method of Squeeze operation is as follows: Squeeze operation, the original feature map is subjected to maximum pooling and average pooling respectively, and feature maps a(W, H, 1) and m(W, H, 1) are generated respectively, and then concatenated so that the obtained weight map has both maximum value information and average value information. Each feature in the feature map is subjected to global average pooling operation and compressed into a real number (1, 1, C), specifically: In the formula, z c Indicates the output of the Squeeze operation, u c represents the cth channel of the feature map, H and W represent the height and width of the feature map, and u c (i,j) represents the i-th row and j-th column element of the c-th channel.

8. The method for on-site verification of data management based on drones according to claim 7 is characterized in that: In S6, the specific method of Excitation operation is as follows: The global features are convolved with a kernel size of 7 and a padding size of 3, so that the size of the global features is reduced from c(W, H, 2) to v(W, H, 1). The weights for each feature channel are generated by the parameter W, specifically: s=F ex (z,W)=σ(g(z,W))=σ(w2δ(w1z)); Where s is the weight obtained by the Excitation operation, z is the output of the Squeeze operation, W1 and W2 are the weights of the two-layer fully connected operation, and δ is the ReLU function.

9. The method for on-site verification of data management based on drones according to claim 8 is characterized in that: In S6, the specific method of the Reweight operation is as follows: the weight of the output of Excitation is multiplied to the original feature map, specifically: In the formula, is the final output feature map, u c is the cth channel of the feature map, s c is the cth weight in s.

10. A data management on-site verification system based on drones, characterized in that: The method for on-site verification of data management based on a drone according to any one of claims 1 to 9 comprises: On-site verification task management module, used for closed-loop management of on-site verification tasks, including task creation, task allocation, task status update, task progress monitoring, comparison and verification supervision and data archiving; The UAV path planning module and the on-site verification task management module are used to plan the flight route and mission area of ​​the UAV according to the terrain and weather conditions of the verification area and the distribution of the targets to be verified; The UAV data acquisition module is connected to the UAV path planning module and is used to collect data on targets in the mission area, including images, videos, and sensor data; The on-site data verification module is connected to the drone data collection module and is used to perform preliminary processing on the raw data collected by the drone; analyze the data after preliminary processing to extract key information; verify the extracted key information with the pre-stored target data source information, compare the verification results, and generate a list of problematic data; The problem data governance module is connected to the on-site data verification module and is used to give problem data governance suggestions and update the managed target data as the target data source information.

Images