Unmanned aerial vehicle inspection method and system, inspection control device thereof and computer equipment

Abstract

The invention provides an unmanned aerial vehicle inspection method and system, an inspection control device thereof and computer equipment, and the method comprises the steps: carrying out the text extraction of an input unmanned aerial vehicle operation instruction, and obtaining a text vector; performing similarity matching on the text vector and an instruction vector and / or an instruction control function vector in an unmanned aerial vehicle operation instruction database to obtain an instruction vector-instruction control function vector most relevant to the text vector; querying a matched unmanned aerial vehicle operation code in an unmanned aerial vehicle operation instruction database according to the instruction vector-instruction control function vector; and controlling operation of the unmanned aerial vehicle according to the unmanned aerial vehicle operation code. According to the method, the unmanned aerial vehicle is directly dialogue through the instruction matching model, the unmanned aerial vehicle is directly controlled to inspect by using a natural language, the operation convenience is improved, a control algorithm of the unmanned aerial vehicle can be tested in combination with a simulation experiment, and the safety and reliability during flight in a real environment are ensured.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and specifically to the field of unmanned aerial vehicle inspection methods, systems, and inspection control devices and computer equipment, systems, and inspection control devices and computer equipment. Background Art

[0002] With the continuous development of intelligent manufacturing technology, drones have demonstrated efficient execution and strong survey capabilities in the field of transmission line inspection. In the inspection of high-voltage transmission lines, drones not only show great potential, but also have a broad application prospect. Drones not only play an important role in daily line tower inspections, but are also widely used in target search and precise positioning of line towers. This provides key information support for power inspection personnel, enabling them to discover and handle line faults more quickly and accurately, and improve the overall operating efficiency and safety of the power system. With the continuous advancement of big data technology, the large amount of data collected by drones during the inspection process can be stored, analyzed and utilized. Artificial intelligence-based technology can conduct in-depth analysis and intelligent prediction of these data, thereby further improving the accuracy and efficiency of inspections.

[0003] However, the efficient operation of drones depends on the operation and maintenance skills of professionals. This not only involves the training of operators, but also requires a complete equipment and technical support system. Operators need to receive systematic training, which includes not only theoretical knowledge learning, but also a lot of practical exercises, making the cost of putting drones into use high and the procedures complicated.

[0004] Based on this, there is an urgent need for an automated drone inspection method to replace manually operated drones for transmission line inspections, thereby reducing the complexity of the process and investment costs of putting drones into use. Summary of the invention

[0005] In response to the problem that the training of the above-mentioned drone operation requires a large amount of manpower and material support, the present technical solution provides a drone inspection method, system, and inspection control device and computer equipment, which can directly communicate with the drone through a command matching model and use natural language to directly control the drone for inspection; it can effectively solve the above-mentioned problem.

[0006] On the one hand, the present invention provides a drone inspection method, comprising:

[0007] Extract text from the input drone operation instructions to obtain a text vector;

[0008] After the text vector is parsed and completed by the instruction matching model, similarity matching is performed with the instruction vector and / or instruction control function vector in the drone operation instruction database of the instruction matching model to obtain the instruction vector most relevant to the text vector, namely, the instruction control function vector;

[0009] According to the instruction vector—instruction control function vector, a matching UAV operation code is searched in the UAV operation instruction database;

[0010] Control the operation of the drone according to the drone operation code.

[0011] The present invention parses the operation instructions input by the UAV operator through an instruction matching model, performs context analysis and completion on the operation instructions according to the operation instructions, so that the format is consistent with the standard format in the UAV operation instruction database of the instruction matching model, and then matches by similarity, and selects a group of operation instructions corresponding to the instruction vector-instruction control function vector ranked by similarity from the database for the UAV operator to choose, or directly selects the operation instruction corresponding to the instruction vector-instruction control function vector with the highest similarity; then the UAV is controlled to perform a flight mission according to the UAV operation code corresponding to the final instruction vector-instruction control function vector.

[0012] Furthermore, before similarity matching is performed between the text vector and the instruction vector and / or instruction control function vector of a drone operation instruction database, the method further includes the steps of: supplementing the text vector in combination with the context information of the text vector, performing format standardization processing on the supplemented text vector to obtain a standardized text vector, wherein the format of the standardized text vector is consistent with a preset format.

[0013] Furthermore, the data of the drone operation instruction database is generated by the following method:

[0014] Given a standard structured input sequence and an example of a UAV operation instruction, a plurality of instruction vectors, instruction control function vectors and UAV operation instructions are generated, and all the generated data constitute a first instruction data set;

[0015] The first instruction data set is deduplicated and de-errored to obtain a UAV operation instruction database.

[0016] By generating a drone operation instruction database with the same format, different scenarios can be adapted and the matching accuracy can be improved.

[0017] Furthermore, the instruction matching model is trained in the following manner:

[0018] Extract text from preset drone operation instructions to obtain text vectors;

[0019] Input the text vector into the instruction matching model, and output the most relevant instruction vector matched by the text vector—instruction control function vector;

[0020] The instruction matching model parameters are modified according to the difference between the most relevant instruction vector-instruction control function vector and the standard instruction vector-instruction control function vector that matches the preset UAV operation instruction to complete the model training.

[0021] Furthermore, the instruction matching model is modified specifically including:

[0022] The difference calculation parameter increment between the most relevant instruction vector-instruction control function vector and the standard instruction vector-instruction control function vector matching the preset UAV operation instruction;

[0023] The weight matrix of the instruction matching model is updated according to the parameter increment;

[0024] The value of the loss function is calculated according to the updated weight matrix. When the value of the loss function is the smallest, the correction is ended and the training is completed. Furthermore, the calculation formula of the parameter increment is as follows:

[0025] ΔW=AB T

[0026] Among them, A is the first low-rank matrix; B is the second low-rank matrix.

[0027] Furthermore, the calculation formula of the loss function is as follows:

[0028] L(θ+Δθ)=L(W+AB T )

[0029] During the forward propagation of the training process, the output is calculated using the updated weight matrix W′:

[0030] y=W′x=(W+AB T )x

[0031] Among them, x is the input and y is the output.

[0032]

[0033] Among them, ΔW is the update amount of the weight matrix; η is the learning rate, which is used to control the step size of each parameter update; is the gradient of the loss function to the weight matrix; ΔA is the update amount of the first low-rank matrix; is the gradient of the loss function to the first low-rank matrix; ΔB is the update amount of the second low-rank matrix; is the gradient of the loss function with respect to the second lowest rank matrix.

[0034] Furthermore, after the first instruction data set is deduplicated and de-errored, the method further includes:

[0035] Different data in the first instruction data set are labeled and classified.

[0036] On the other hand, the present invention also provides a UAV inspection control device based on an instruction matching model, comprising:

[0037] Text vector extraction module: used to extract text from the input drone operation text instructions to obtain text vectors;

[0038] Instruction matching model: used to perform similarity matching on the text vector after parsing and completion by the instruction matching model with the instruction vector and instruction control function vector in the drone operation instruction database of the instruction matching model, so as to obtain the instruction vector most relevant to the text vector, namely, the instruction control function vector;

[0039] Operation code matching module: used to query the matching drone operation code in the drone operation instruction database according to the instruction vector-instruction control function vector;

[0040] Operation code execution module: used to control the operation of the drone according to the drone operation code.

[0041] Furthermore, the drone inspection control device also includes a drone operation instruction database generation module, which is used to generate relevant data for instruction matching, including:

[0042] Data generation submodule: given a standard structured input sequence and an example of a UAV operation instruction, it generates a variety of instruction vectors, instruction control function vectors, and UAV operation instructions, and all the generated data constitute the first instruction data set;

[0043] Data cleaning submodule: remove duplication and errors from the first instruction data set to obtain the drone operation instruction database.

[0044] In another aspect, the present invention further provides a computer device, comprising:

[0045] at least one memory and at least one processor;

[0046] The memory is used to store one or more programs;

[0047] When the one or more programs are executed by the at least one processor, the at least one processor implements the steps of a method for high-speed online correction of image distortion as described in any one of the above-mentioned items.

[0048] On the other hand, the present application also provides a computer-readable storage medium, and when the computer program is executed by a processor, it implements the drone inspection method as described in any one of the above-mentioned items.

[0049] The present invention proposes a drone inspection method, system, inspection control device and computer equipment, which enable the drone operating system to directly understand the natural language of the operator, and complete the inspection without controlling the drone. Under this setting, the inspection personnel can perform various operations on the drone through simple text instructions, such as take-off, ascent, descent, forward, backward, left and right movement and automatic inspection. This method not only improves the convenience of operation, but also can be combined with simulation experiments to test and debug the control algorithm of the drone, ensure the safety and reliability when flying in a real environment, and reduce the demand for experienced drone pilots.

[0050] For better understanding and implementation, the present invention is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a structural block diagram of the unmanned aerial vehicle inspection control device based on the instruction matching model of the present invention;

[0052] Figure 2 For execution Figure 1 A flow chart of a UAV inspection method performed by the UAV inspection control device shown;

[0053] Figure 3 A flowchart for generating a UAV operation instruction database of the present invention;

[0054] Figure 4 This is a training flow chart of the instruction matching model of the present invention. DETAILED DESCRIPTION

[0055] Traditional drone inspections require professional operators and complex control systems, but the present invention attempts to build a command matching model with the help of the natural language processing capabilities of the command matching model, so that the drone can understand and execute the user's verbal commands. For example, the user can directly give commands such as "check the No. 3 transmission tower" or "take a close-up of the high-voltage line" to the drone. After receiving these commands, the drone can autonomously plan the path, perform inspection tasks, and provide real-time feedback on the inspection results. This inspection method based on the command matching model can not only improve the convenience and efficiency of operation, but also reduce the investment cost of manpower and material resources. At the same time, with the continuous optimization and learning of the model, the drone's understanding ability and task execution ability will continue to improve, thereby further improving the accuracy and reliability of the inspection.

[0056] In order to meet the above requirements, the present invention first constructs a drone inspection system, which includes a drone and a drone inspection control device for controlling the drone. Figure 1 The drone inspection control device includes a text vector extraction module 1, an instruction matching model 2, an operation code matching module 3, an operation code execution module 4 and a drone operation database; wherein the drone instruction operation database includes a variety of instruction vectors, instruction control function vectors, instruction vector-instruction control function vectors and drone operation codes, the instruction vector is used to match the input text, the instruction control function vector is used to match the input function example, and the instruction vector-instruction control function vector is used to match the input data when the input drone operation instruction includes both text and function. The drone operation code is associated with an instruction vector-instruction control function vector. After the drone operator inputs the drone operation instruction, the drone operation instruction is converted into a machine-recognizable text vector by the text vector extraction module, and the instruction matching model performs similarity matching on the text vector, finds the closest vector from the drone operation instruction database, and then outputs the corresponding drone operation code according to the matched vector, and finally controls the drone to perform the corresponding flight mission according to the drone operation code. Please refer to Figure 2 The specific execution process of each component of the UAV inspection control device is as follows:

[0057] The text vector extraction module 1 executes step S1: extracting text from the input drone operation instruction to obtain a text vector.

[0058] Drone operation instructions are text or strings entered by drone operators from the input interface. The core features of the operation instructions are extracted through the text extraction algorithm to generate a text vector with key information of the operation instructions for subsequent analysis and processing.

[0059] See also Figure 2 , the instruction matching model 2 includes a text vector normalization module 21 and a vector similarity matching module 22;

[0060] The text vector standardization module 21 executes step S21: supplementing the text vector in combination with the context information of the text vector, performing format standardization processing on the supplemented text vector to obtain a standardized text vector, wherein the format of the standardized text vector is consistent with a preset format.

[0061] In the present invention, a command matching model based on LLM (large language model) is used to receive the input of drone operation instructions given by the drone operator, which are specific task instructions. The command matching model parses the instructions input by the drone pilot through natural language processing technology, extracts core keywords and key information, and finally matches the instructions based on the extracted information. For some complex instructions, the command matching model needs to supplement the drone operation instructions in combination with context information, so that the generated instructions are consistent with the standard structured input format to improve the efficiency and accuracy of drone execution. At the same time, the drone operation instructions are semantically analyzed and keyword expanded to complete the supplementation and standardization of the text vector of the input drone operation instructions, making the query more targeted and extensive.

[0062] In order to meet the standardization requirements, the data stored in the drone operation instruction database of the present invention also has a fixed format and standard, but manual setting or input is inefficient and prone to errors. Therefore, the present invention automatically generates a large amount of standardized data based on the Ai model. Figure 3 ,The UAV operation instruction database is generated by the following method: given a standard structured input sequence and a standard example, a variety of instruction vectors, instruction control function vectors, instruction vector-instruction control function vectors, and UAV operation codes are generated, and all the generated data constitute a first instruction data set;

[0063] The present invention automatically generates the first instruction data set with ChatGPT3.5prompt, specifically providing ChatGPT3.5prompt with standard structured input sequence examples and drone operation instruction examples, so that ChatGPT3.5 generates its own data set and determines the content and type of the data set. The generated instructions include drone operations, such as takeoff, landing, flight path, obstacle avoidance, and surrounding line towers. For example: aw.takeoff(): takeoff instruction, let the drone take off; aw.land(): landing instruction, let the drone land; aw.fly_to([x,y,z]): fly to the specified position, the drone will fly to the [X,Y,Z] coordinates specified by the list; aw.get_position(object_name): get the position of the specified object, etc. ChatGPT3.5 will generate a large amount of data in a short time. The instruction-instruction control function combination is the basis of the database. Generate instructions and instruction control function vectors: convert the instructions and answers in the data source into instruction vectors and instruction control function vectors respectively, and obtain instruction vectors, instruction control function vectors, and instruction vector-instruction control function vector combinations. For example:

[0064] Move 5 units to the right: aw.fly_to([aw.get_drone_position()[0]+5,aw.get_drone_position()[1],aw.get_drone_position()[2]]), where the vector converted by moving 5 units to the right is the command vector, and the vector converted by aw.fly_to([aw.get_drone_position()[0]+5,aw.get_drone_position()[1],aw.get_drone_position()[2]] is the command control function vector. The two vectors are concatenated into a command vector-command control function vector combination.

[0065] The first instruction data set is deduplicated and de-errored to obtain a UAV operation instruction database.

[0066] ChatGPT3.5 will generate a large amount of data in a short period of time, and there are repeated and erroneous instructions in these data. Therefore, it is necessary to check the generated data set to remove repeated and erroneous instructions to ensure data quality.

[0067] In order to associate the instruction vector, the instruction control function vector and the drone operation code and improve the efficiency of subsequent queries, the present invention also includes labeling and classifying different data in the first instruction data set after deduplication and error removal. Different types of instructions are marked for subsequent use. For example, take-off instructions are stored in different libraries, and take-off to the right, take-off upwards and take-off forward in the take-off instructions are stored in different tables. Improve execution efficiency.

[0068] The vector similarity matching module 22 executes step S22: performing similarity matching between the standardized text vector and the instruction vector and / or instruction control function vector of the drone operation instruction database to obtain the most relevant instruction vector matched by the text vector—the instruction control function vector.

[0069] By calculating the Euclidean distance between the input command and the command vector and / or command control function vector in the UAV operation command database, examples similar to the text vector of the UAV operation command are found, where the Euclidean distance formula is:

[0070]

[0071] Where: a is the text vector of the UAV operation instruction, and b is the vector of the UAV operation instruction database. After completing the instruction matching, the text content corresponding to the matched instruction vector and / or instruction control function vector is sorted according to the similarity and displayed on the input interface for the UAV operator to select, or the text content corresponding to the most similar instruction vector and / or instruction control function vector is directly selected as the input instruction.

[0072] The code matching module 3 executes step S3: according to the instruction vector—instruction control function vector, a matching drone operation code is searched in the drone operation instruction database.

[0073] The instruction vector processed by the instruction matching model - the instruction control function vector is retrieved in the drone operation instruction database, and the corresponding drone operation code is matched and replied.

[0074] The operation code execution module 4 executes step S4: controlling the operation of the drone according to the matched drone operation code.

[0075] Finally, the matched drone operation code is transmitted to the drone to control the drone and complete the given drone operation.

[0076] In another embodiment, after having a drone operation instruction database, the instruction matching model is modified based on the database so that the instruction matching model is adapted to the actual application, thereby improving the accuracy of drone operation. Figure 4 , the training steps include:

[0077] Perform text extraction on the preset drone operation instructions to obtain text vectors.

[0078] The text vector is supplemented in combination with the context information of the text vector, and the supplemented text vector is subjected to format standardization processing to obtain a standardized text vector, wherein the format of the standardized text vector is consistent with a preset format.

[0079] The standardized text vector is matched with the instruction vector and / or instruction control function vector of the drone operation instruction database for similarity to obtain the most relevant instruction vector matched by the text vector, namely the instruction control function vector.

[0080] Because the training process is to make the final model parameters more accurate in predicting the results of the input drone operation instructions, the model needs to parse and complete the text vectors closer to the real standard data. Therefore, it is necessary to perform similarity matching on each instruction vector and instruction control function vector.

[0081] The instruction matching model parameters are modified according to the difference between the most relevant instruction vector-instruction control function vector and the standard instruction vector-instruction control function vector that matches the preset UAV operation instruction, thereby completing the instruction matching model training.

[0082] Calculate the loss between the generated vector and the standard vector, and correct the model parameters according to the loss to achieve correction. Specifically, the embodiment of the present invention uses ChatGLM3-6b as the text vector standardization module 21. ChatGLM3-6b is an artificial intelligence assistant based on a large language model. It is a variant of the language model GLM-130B jointly trained by the KEG Laboratory of Tsinghua University and Zhipu AI Company in 2023. Generated quality. In the present invention, the generated data set is used to perform LoRA (Low-Rank Adaptation of Large Language Models) correction on ChatGLM3-6b. Through LoRA correction, the model parameter matrix W′ is adjusted to complete the model training. In order to correct the model without updating all parameters, the first low-rank matrix A and the second low-rank matrix B are introduced to express the parameter increment as:

[0083] ΔW=AB T

[0084] Where: A is the first low-rank matrix; B is the second low-rank matrix.

[0085] The updated weight matrix is:

[0086] W′=W+ΔW=W+AB T

[0087] Loss function L:

[0088] L(θ+Δθ)=L(W+AB T )

[0089] During the forward propagation process during training, the output is calculated using the updated weight matrix W′:

[0090] y=W′x=(W+AB T )x

[0091] Among them, x is the input and y is the output.

[0092]

[0093] Among them, ΔW is the update amount of the weight matrix; η is the learning rate, which is used to control the step size of each parameter update; is the gradient of the loss function to the weight matrix; ΔA is the update amount of the first low-rank matrix; is the gradient of the loss function to the first low-rank matrix; ΔB is the update amount of the second low-rank matrix; is the gradient of the loss function with respect to the second lowest rank matrix.

[0094] Through LoRA fine-tuning, when the loss function L is minimized, the model parameter matrix W′ is finally obtained as the final parameter of the model. The corrected model can better adapt to the corresponding actual application scenarios of drones, and the text vector matching accuracy is higher, thereby improving the accuracy of drone operations.

[0095] In a specific embodiment of the present invention, the use of the present invention is as follows: the unmanned aerial vehicle inspection method combining simulation software and the command matching model of the present invention is used to perform unmanned aerial vehicle inspection: prepare the unmanned aerial vehicle operation instruction database, the trained command matching model model, and establish the connection between the command matching model model and the Microsoft Airsim simulation system, establish several line poles and unmanned aerial vehicles in the Microsoft Airsim simulation system, and then the unmanned aerial vehicle driver inputs the instruction, the instruction matching model receives the input and parses the instruction input by the unmanned aerial vehicle driver, extracts the core keywords and key information, supplements it with context information, and generates an instruction vector. The instruction vector is used to search in the selected database, and the system adopts an advanced search algorithm to quickly locate and query the most relevant unmanned aerial vehicle operation instructions. Finally, the instruction matching model sorts the relevance scores of all the filtered instructions, and selects a group of unmanned aerial vehicle operation instructions sorted by relevance to be displayed on the interface for the unmanned aerial vehicle driver to select. After selecting the operation instruction, the unmanned aerial vehicle operation code associated with the operation instruction in the unmanned aerial vehicle operation instruction database is extracted, and the unmanned aerial vehicle operation code is transmitted to the Airsim simulation system, the corresponding function in Airsim is called to pass in the corresponding parameters, and the unmanned aerial vehicle is simulated and controlled by using the Airsim environment.

[0096] In summary, the drone inspection method, system, and inspection control device proposed in the present invention enable the drone inspection system to directly understand the natural language of the operator and complete the inspection without controlling the drone. Under this setting, the inspection personnel can perform various operations on the drone through simple text instructions, such as take-off, ascent, descent, forward, backward, left and right movement, and automatic inspection. This method not only improves the convenience of operation, but also can be combined with simulation experiments to test and debug the control algorithm of the drone, ensure safety and reliability when flying in a real environment, and reduce the need for experienced drone pilots.

[0097] An embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, an unmanned aerial vehicle inspection method described in any one of the above embodiments is implemented.

[0098] The present invention may take the form of a computer program product implemented on one or more storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing program code. Computer-readable storage media include permanent and non-permanent, removable and non-removable media, and information storage can be achieved by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include but are not limited to: phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

[0099] The above-mentioned embodiments only express several implementation methods of the present invention, and the description is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, and the present invention is also intended to include these modifications and modifications.

Claims

1. A drone inspection method, system, inspection control device and computer equipment, characterized in that: include: S1: Extract text from the input drone operation instructions to obtain a text vector; S2: performing similarity matching between the text vector and a command vector and / or a command control function vector in a drone operation command database to obtain a command vector matched by the text vector, that is, a command control function vector; S3: According to the instruction vector—instruction control function vector, searching for a matching drone operation code in the drone operation instruction database; S4: Control the operation of the drone according to the drone operation code.

2. The unmanned aerial vehicle inspection method based on the command matching model according to claim 1 is characterized in that: Before the step S2 performs similarity matching between the text vector and the instruction vector and / or instruction control function vector of a drone operation instruction database, the method further includes the steps of supplementing the text vector in combination with the context information of the text vector, performing format standardization processing on the supplemented text vector, and obtaining a standardized text vector, wherein the format of the standardized text vector is consistent with a preset format.

3. The unmanned aerial vehicle inspection method based on the command matching model according to claim 2 is characterized in that: The data of the drone operation instruction database is generated by the following method: Given a standard structured input sequence and an example of a UAV operation instruction, a plurality of instruction vectors, instruction control function vectors and UAV operation instructions are generated, and all the generated data constitute a first instruction data set; The first instruction data set is deduplicated and de-errored to obtain a UAV operation instruction database.

4. The unmanned aerial vehicle inspection method, system, inspection control device and computer equipment according to claim 3, characterized in that: The step S2 is performed by an instruction matching model, and the instruction matching model is trained in the following manner: Extract text from preset drone operation instructions to obtain text vectors; Input the text vector into the instruction matching model, and output the most relevant instruction vector matched by the text vector—instruction control function vector; The instruction matching model parameters are modified according to the difference between the most relevant instruction vector-instruction control function vector and the standard instruction vector-instruction control function vector that matches the preset UAV operation instruction to complete the model training.

5. The unmanned aerial vehicle inspection method, system, inspection control device and computer equipment according to claim 4, characterized in that: The modification of the instruction matching model specifically includes: Calculate the parameter increment according to the difference between the most relevant instruction vector-instruction control function vector and the standard instruction vector-instruction control function vector matching the preset drone operation instruction; The weight matrix of the instruction matching model is updated according to the parameter increment; The value of the loss function is calculated based on the updated weight matrix. When the value of the loss function is minimized, the training is completed.

6. The unmanned aerial vehicle inspection method, system, inspection control device and computer equipment according to claim 5, characterized in that: The calculation formula of the parameter increment is as follows: ΔW=AB T Among them: A is the first low rank matrix; B is the second low rank matrix.

7. The unmanned aerial vehicle inspection method based on the command matching model according to claim 6 is characterized in that: The calculation formula of the loss function is as follows: L(θ+Δθ)=L(W+AB T ) During the forward propagation of the training process, the output is calculated using the updated weight matrix W′: y=W′x=(W+AB T )x Among them, x is the input and y is the output. Among them, ΔW is the update amount of the weight matrix; η is the learning rate, which is used to control the step size of each parameter update; is the gradient of the loss function to the weight matrix; ΔA is the update amount of the first low-rank matrix; is the gradient of the loss function to the first low-rank matrix; ΔB is the update amount of the second low-rank matrix; is the gradient of the loss function with respect to the second lowest rank matrix.

8. The unmanned aerial vehicle inspection method based on the command matching model according to claim 7 is characterized in that: After the first instruction data set is deduplicated and error-free, the following steps are also included: Different data in the first instruction data set are labeled and classified.

9. A drone inspection control device based on an instruction matching model, characterized in that: include: Text vector extraction module: used to extract text from the input drone operation text instructions to obtain text vectors; Instruction matching model: used to perform similarity matching on the text vector after parsing and completion by the instruction matching model with an instruction vector and / or instruction control function vector in a drone operation instruction database, to obtain an instruction vector matched by the text vector, namely, an instruction control function vector; Operation code matching module: used to query the matching drone operation code in the drone operation instruction database according to the instruction vector-instruction control function vector; Operation code execution module: used to control the operation of the drone according to the drone operation code.

10. The unmanned aerial vehicle inspection control device based on the instruction matching model according to claim 9 is characterized in that: It also includes a UAV operation instruction database generation module, which is used to generate relevant data for instruction matching, including: Data generation submodule: given a standard structured input sequence and an example of a UAV operation instruction, it generates a variety of instruction vectors, instruction control function vectors, and UAV operation instructions, and all the generated data constitute the first instruction data set; Data cleaning submodule: remove duplication and errors from the first instruction data set to obtain the drone operation instruction database.

11. A drone inspection system based on a command matching model, characterized in that: include: A drone and a drone inspection control device based on a command matching model as described in any one of claims 9 to 10.

12. A computer device, characterized in that: include: at least one memory and at least one processor; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, the at least one processor implements the steps of the drone inspection method of the instruction matching model as described in any one of claims 1 to 8.

Images