Satellite mission planning control and image intelligent processing embedded implementation method

Abstract

The application discloses a kind of satellite mission planning control and image intelligent processing embedded implementation method, comprising the following steps: S1, constructs remote sensing satellite mission planning mathematical model and target identification algorithm model based on YOLO v3 network;S2, the task planning mathematical model and target identification algorithm model are transplanted into the embedded platform of preestablished;S3, determine the attitude of remote sensing satellite and obtain the attitude control algorithm of remote sensing satellite, based on remote sensing satellite attitude information and task planning mathematical model, obtain the remote sensing image of remote sensing satellite;S4, input remote sensing image in target identification algorithm model and carry out target identification, and then output identification result.The application creatively transplants the task planning mathematical model and target identification algorithm model constructed to the embedded platform of preestablished, based on the embedded platform of preestablished, realizes the whole process operation of remote sensing satellite's mission planning and control, remote sensing image acquisition and remote sensing image processing, to meet the demand of processing efficiency.

Description

Technical Field

[0001] This invention relates to the field of remote sensing satellite technology, and in particular to an embedded implementation method for satellite mission planning and control and intelligent image processing. Background Technology

[0002] With the development of satellite technology, modern remote sensing technology has greatly enhanced Earth observation capabilities. The remote sensing data acquired can play an important role in both military and civilian applications. In civilian applications, spaceborne remote sensing can be used for disaster monitoring, engineering construction progress tracking, agricultural yield estimation surveys, air crash search and rescue, and land use planning, making significant contributions to national economic construction and development.

[0003] Currently, optical remote sensing satellites are characterized by fast acquisition speed, high image quality, high resolution, and short revisit cycles. These characteristics result in the acquisition of massive amounts of information in a short time. How to quickly and accurately extract the effective information from this vast amount of remote sensing data is one of the major challenges in current remote sensing image processing applications. At present, existing remote sensing satellites primarily focus on imaging, with relatively little post-processing, and even then, processing is mainly performed on the ground. Furthermore, the complexity and inefficiency of existing algorithms limit their ability to apply target detection and recognition in engineering applications, making it difficult to meet the needs of rapid intelligence acquisition. In addition, mission planning and control of remote sensing satellites are crucial prerequisites for obtaining high-quality remote sensing imagery and are currently an important research direction. However, current methods all involve mission planning on the ground followed by uploading control commands to the satellite for control, resulting in limited satellite autonomy. Summary of the Invention

[0004] In view of this, the present invention proposes an embedded implementation method for satellite mission planning and control and intelligent image processing to solve the above-mentioned problems.

[0005] The embedded implementation method for satellite mission planning and control and intelligent image processing provided by this invention includes the following steps:

[0006] S1. Construct a mathematical model for remote sensing satellite mission planning and a target recognition algorithm model based on the YOLO v3 network;

[0007] S2. Port the task planning mathematical model and target recognition algorithm model to a preset embedded platform;

[0008] S3. Determine the attitude of the remote sensing satellite and obtain the attitude control algorithm of the remote sensing satellite. Obtain the remote sensing image of the remote sensing satellite based on the attitude information of the remote sensing satellite and the mission planning mathematical model.

[0009] S4. Input the acquired remote sensing image into the target recognition algorithm model for target recognition, and then output the recognition result.

[0010] Preferably, the specific implementation method of constructing the mathematical model for remote sensing satellite mission planning in step S1 includes:

[0011] S11. Construct a mathematical model for mission planning of remote sensing satellites based on constraints, wherein the constraints include the set of observation targets, satellite orbit, platform characteristics, payload type, energy, storage and illumination, and the objectives include maximizing the priority of observation targets, maximizing the number of observation targets and maximizing resource load balancing.

[0012] S12. The mathematical model for task planning is solved and optimized based on genetic algorithm and tabu search algorithm.

[0013] Preferably, the mathematical model for task planning in step S11 is expressed by the following formula:

[0014]

[0015] In the formula, This indicates maximizing the priority of the observed target. This indicates maximizing the number of observed targets. This indicates maximizing resource load balancing. , and They represent , and The weight value, ,in:

[0016] ;

[0017] ;

[0018] ;

[0019] In the formula, Indicates the target observed by the satellite. Represents the target priority set. Indicates the task number. Indicates the satellite number, This indicates the total number of tasks. Indicates the total number of satellites. This indicates the energy consumed by a satellite when performing a mission. This represents the standard deviation function.

[0020] Preferably, the specific implementation of step S12 includes:

[0021] S121. Initialize the parameters of the task planning mathematical model, use a genetic algorithm to perform a global search on the initial population until the initial population converges to a preset value, and then output the optimal solution of the population.

[0022] S122. The optimal solution of the population output in step S121 is used as the initial solution of the tabu search algorithm, and a heuristic factor is set to perform local iteration on key areas until the preset number of iterations is reached, and then the planning result is output to complete the solution optimization of the mathematical model of the task planning.

[0023] Preferably, the specific implementation of constructing the target recognition algorithm model based on the YOLO v3 network in step S1 includes:

[0024] S31. Add an spp layer to the existing YOLO v3 network topology to optimize multi-scale target detection;

[0025] S32. Perform sparse training on the YOLO v3 network from step S31, and apply L1 regularization to the BN layer, as expressed by the formula:

[0026]

[0027] In the formula, , Indicates the parameters of the BN layer. Indicates the scaling factor. This represents the truth value of the minimum batch. This represents the mean of the smallest batch. Indicates the penalty factor. Indicates the offset. This represents the variance of the minimum batch size. Represents a constant. Represents the training objective function. Represents the training loss function;

[0028] S33. Based on the BN layer parameter values ​​in step S32, channel pruning is performed to achieve model compression, thereby obtaining the target recognition algorithm model based on the YOLOv3 network.

[0029] Preferably, the specific implementation method of porting the task planning mathematical model to a preset embedded platform in step S2 includes:

[0030] S13. Convert the task planning mathematical model into the corresponding Matlab program using Matlab tools;

[0031] S14. Use the Matlab-Code Generator tool to convert the Matlab program into a C++ project;

[0032] S15. Compile a top-level class to implement the model interface logic, and compile a plugin class as an auxiliary tool.

[0033] S16. Standardize the interface of the program to obtain a standardized platform model.

[0034] Preferably, the specific implementation method of porting the target recognition algorithm model to a preset embedded platform in step S2 includes:

[0035] S34. The programming framework is used to load the pre-trained target recognition algorithm model file and weight parameter file;

[0036] S35. Use the CNML library to parse it and generate model data containing instruction sets and model parameters;

[0037] S36. Use the CNRT library to transfer the generated model data to the hardware DRAM;

[0038] S37. Input the application data to be processed and start hardware computing;

[0039] S38. Read the calculation result and return it layer by layer upwards to the programming framework.

[0040] Preferably, the preset embedded platform is the Cambricon MLU220 embedded computing platform, comprising:

[0041] The x86 main control board is used for overall process control of the embedded platform and data processing of various external interfaces. It is equipped with the first M.2 interface.

[0042] The MLU220 intelligent computing card is used to accelerate the operation of algorithms and has a second M.2 interface that matches and connects to the first M.2 interface.

[0043] Preferably, in step S3, the attitude of the remote sensing satellite is determined by a combination of an infrared horizon sensor, a sun sensor, and a gyroscope, and then the attitude control algorithm based on the PD controller can be obtained using the determined attitude of the remote sensing satellite.

[0044] This invention provides an embedded implementation method for satellite mission planning, control, and intelligent image processing. The method embeds a constructed mission planning mathematical model and a target recognition algorithm model into a pre-set embedded platform. It then uses the satellite attitude control algorithm and the mission planning mathematical model to acquire remote sensing images from a remote sensing satellite. The acquired remote sensing images are then input into the target recognition algorithm model within the embedded platform for image processing, ultimately outputting the target recognition result. This invention creatively portes the constructed mission planning mathematical model and target recognition algorithm model to a pre-set embedded platform, enabling the entire process of mission planning and control, remote sensing image acquisition, and remote sensing image processing for remote sensing satellites, thereby meeting the requirements for processing efficiency. Attached Figure Description

[0045] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0046] Figure 1 This is a flowchart of an embedded implementation method for satellite mission planning and control and intelligent image processing in this invention;

[0047] Figure 2 This is a flowchart illustrating the process of porting the task planning mathematical model to a pre-defined embedded platform in this invention.

[0048] Figure 3 This is a flowchart illustrating the porting of the target recognition algorithm model in this invention to a preset embedded platform;

[0049] Figure 4 These are the aircraft and ship target identification results in the embodiments of the present invention;

[0050] Figure 5 To utilize the target recognition algorithm model in this invention for... Figure 4 The accuracy and detection rate of the calculations. Detailed Implementation

[0051] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0052] like Figure 1 As shown, an embedded implementation method for satellite mission planning and control and intelligent image processing includes the following steps:

[0053] S1. Construct a mathematical model for remote sensing satellite mission planning and a target recognition algorithm model based on the YOLO v3 network (YOLO v3 is the third version of the YOLO algorithm in target detection algorithms, where YOLO is short for You Look once);

[0054] In this step, the specific implementation methods for constructing the mathematical model for remote sensing satellite mission planning include:

[0055] S11. Construct a multi-target mission planning mathematical model for remote sensing satellites based on constraints, wherein the constraints include the observation target set, satellite orbit, platform characteristics, payload type, energy, storage, and illumination, and the objectives include maximizing the priority of observation targets, maximizing the number of observation targets, and maximizing resource load balancing; wherein the mission planning mathematical model is expressed by the formula:

[0056]

[0057] In the formula, This indicates maximizing the priority of the observed target. This indicates maximizing the number of observed targets. This indicates maximizing resource load balancing. , and They represent , and The weight value, ,in:

[0058] ;

[0059] ;

[0060] ;

[0061] In the formula, Indicates the target observed by the satellite. Represents the target priority set. Indicates the task number. Indicates the satellite number, This indicates the total number of tasks. Indicates the total number of satellites. This indicates the energy consumed by a satellite when performing a mission. This represents the standard deviation function; in this embodiment, prioritizing the protection of important objectives is the primary factor, i.e., the weight values ​​are: , , ;

[0062] Among these, energy constraint means that the cumulative energy consumed by a single satellite in performing a mission cannot exceed the satellite's energy storage limit; storage constraint means that the cumulative storage consumed by a single satellite in performing a mission cannot exceed the satellite's storage limit; illumination constraint means that visible light cameras can only observe within a certain solar altitude angle; payload type constraint means that the target must be observed by a payload of the required type; and it also includes the following constraints:

[0063] 1. Visibility window constraint: The satellite payload and the target must be visible to each other, and the duration of the window must not be less than the mission observation time;

[0064] 2. Side swing angle constraint: When the satellite switches tasks, its side swing cannot exceed the maximum side swing capability;

[0065] 3. Task preparation time constraint: The interval between two tasks (the time interval between the end of the previous task and the start of the next task) must be greater than or equal to the load preparation time.

[0066] S12. The task planning mathematical model is solved and optimized based on genetic algorithm and tabu search algorithm, specifically:

[0067] S121. Initialize the parameters of the task planning mathematical model, use a genetic algorithm to perform a global search on the initial population until the initial population converges to a preset value, and then output the optimal solution for the population; the specific process is as follows:

[0068] Step 1: Initialize the population. The initial population size is NP, which is randomly generated by the computer. The length of the chromosome is related to the target number. The initial number of generations is 1, and the upper limit of evolution is set to G. The exchange probability and mutation probability of the population are represented by pc and pm, respectively.

[0069] Step 2: Determine the termination condition. If the current generation is less than the maximum generation, increment the current generation by 1; otherwise, output the current optimal solution.

[0070] Step 3: Fitness function calculation. The optimization objective set in this embodiment will be used as the evaluation index of the fitness function to compare all individuals in the current population and save each individual in the population and its corresponding fitness value.

[0071] Step 4: Population selection operation. In this embodiment, the selection operation is carried out according to the "roulette wheel" strategy. Individuals are selected according to the proportion of fitness values ​​calculated in the previous step. The larger the fitness value, the greater the chance of the individual being retained. If the fitness value of an individual is fi, then the probability of it being selected is pi=fi / ∑fi.

[0072] Step 5: Population crossover operation. Randomly combine the selected individuals in the current population. Each pair of individuals performs a crossover operation on one or more genes in the chromosome according to the exchange probability pc.

[0073] Step 6: Population mutation operation, which involves inverting the binary representation of one or more genes on each individual in the population according to the mutation probability pm to achieve gene mutation;

[0074] Step 7: Determine the termination condition. If the current generation is less than the maximum generation, increment the current generation by 1; otherwise, output the current optimal solution.

[0075] S122. The optimal solution for the population output in step S121 is used as the initial solution for the tabu search algorithm. A heuristic factor is set to perform local iterations on key areas until a preset number of iterations is reached, and then the planning result is output to complete the solution optimization of the mathematical model for the task planning. The specific process is as follows:

[0076] Step 1': Input the initial solution. Input the optimal solution after optimization by the genetic algorithm as the initial solution;

[0077] Step 2': Iterate through the algebras. If the maximum number of algebras has not been reached, continue optimization; otherwise, proceed to Step 8'.

[0078] Step 3': Completion rate assessment. If the target task completion rate is not reached, continue optimization; otherwise, proceed to Step 8'.

[0079] Step 4': Algorithm parameter initialization. Analyze the current solution to determine the current neighborhood search direction, and initialize the tabu list and other parameters;

[0080] Step 5': Retain candidate solutions. Generate neighborhood solutions according to the search direction of Step 1', calculate fitness values, and retain candidate solutions;

[0081] Step 6': Determine the candidate solution's contempt criterion. If satisfied, replace the current solution and the current optimal solution with the candidate solution that satisfies the contempt criterion, and update the taboo list with its corresponding taboo object, then proceed to Step 2'; if not satisfied, proceed to Step 7'.

[0082] Step 7': Tabu attribute judgment of candidate solutions, replace the current solution with the non-tabu table of candidate solutions, update the tabu table with the corresponding tabu objects, and then go to Step 2';

[0083] Step 8': Output the planning results.

[0084] The specific implementation methods for constructing a target recognition algorithm model based on the YOLO v3 network include:

[0085] S31. Add an SPP (spatial pyramid pooling) layer to the existing YOLO v3 network topology to optimize multi-scale target detection;

[0086] S32. Perform sparse training on the YOLO v3 network from step S31, and apply L1 regularization to the BN (Batch Normalization) layer, expressed by the formula:

[0087]

[0088] In the formula, , Indicates the parameters of the BN layer. Indicates the scaling factor. This represents the truth value of the minimum batch. This represents the mean of the smallest batch. Indicates the penalty factor. Indicates the offset. This represents the variance of the minimum batch size. Represents a constant. Represents the training objective function. Represents the training loss function;

[0089] S33. Based on the BN layer parameter values ​​in step S32, channel pruning is performed to compress the model, thereby obtaining the target recognition algorithm model based on the YOLOv3 network.

[0090] S2. Port the task planning mathematical model and target recognition algorithm model to a preset embedded platform;

[0091] In this step, the specific implementation method of porting the task planning mathematical model to the preset embedded platform includes:

[0092] S13. Based on Matlab (matrix laboratory, a commercial software developed by Mathworks, is a high-level programming language and programming environment for numerical computation, algorithm development, data analysis and visualization, which can be mixed with other high-level languages ​​(such as C, C++ and FORTRAN)) tools, convert the task planning mathematical model into the corresponding Matlab program;

[0093] S14. Use the Matlab-Code Generator tool to convert the Matlab program into a C++ project;

[0094] S15. Compile a top-level class to implement the model interface logic, and compile a plugin class as an auxiliary tool.

[0095] S16. Standardize the interface encapsulation of the program to obtain a standardized platform model;

[0096] The specific implementation methods for porting the target recognition algorithm model to a preset embedded platform include:

[0097] S34. The programming framework is used to load the pre-trained target recognition algorithm model file and weight parameter file;

[0098] S35. Use the CNML library (machine learning programming library) to parse it and generate model data containing instruction sets and model parameters;

[0099] S36. Use the CNRT library (Cambricon High Performance Runtime Library) to transfer the generated model data to hardware DRAM (Dynamic Memory);

[0100] S37. Input the application data to be processed and start hardware computing;

[0101] S38. Read the calculation result and return it layer by layer upwards to the programming framework;

[0102] In this embodiment, to ensure the numerical solution accuracy and computational reliability of the task planning mathematical model, the top-level model software program is written in standard C++, while Matlab is used to develop some complex algorithm code.

[0103] The model conversion process semi-automatically converts Matlab-Simulink (a block diagram design environment based on Matlab that can be used to model, analyze, and simulate various dynamic systems) simulation files (mdl format) or Matlab files (m format) into COM components (executable code distributed in the form of WIN32 dynamic link libraries (DLLs) or executable files (EXEs) that can be used in C++ programs. For Matlab simulation files in mdl format, C / C++ code can be generated after passing through Matlab's built-in real-time workshop (RTW, called Matlab coder in Matlab versions after 2011). Finally, a COM component that can be used for C++ programming on the Visuo Studio platform is formed. Then, the algorithm model is ported to the MLU220 (a model of an embedded computing platform) embedded computing platform using cross-compilation.

[0104] S3. Determine the attitude of the remote sensing satellite and obtain the attitude control algorithm of the remote sensing satellite. Obtain the remote sensing image of the remote sensing satellite based on the attitude information of the remote sensing satellite and the mission planning mathematical model.

[0105] In this step, the attitude of the remote sensing satellite is determined by a combination of infrared horizon sensor, sun sensor and gyroscope. Then, the attitude control algorithm based on PD controller can be obtained using the determined attitude of the remote sensing satellite. Determining the attitude of the remote sensing satellite based on the combination of infrared horizon sensor, sun sensor and gyroscope, as well as obtaining the attitude control algorithm based on PD controller using the known attitude of the remote sensing satellite, are both existing technologies and will not be described in detail here.

[0106] S4. Input the acquired remote sensing image into the target recognition algorithm model for target recognition, and then output the recognition result.

[0107] The preset embedded platform is the Cambricon MLU220 embedded computing platform, which includes:

[0108] The x86 main control board is used for overall process control of the embedded platform and data processing of various external interfaces. It is equipped with the first M.2 interface (a new host interface solution that is compatible with multiple communication protocols, such as SATA, PCIe, USB, HSIC, UART, SMBus, etc.).

[0109] The MLU220 intelligent computing card is used to accelerate the operation of algorithms and has a second M.2 interface that matches and connects to the first M.2 interface.

[0110] In this embodiment, the constructed mission planning mathematical model and target recognition algorithm model are embedded into a preset embedded platform. The satellite attitude control algorithm and mission planning mathematical model are used to acquire remote sensing images from the remote sensing satellite. The acquired remote sensing images are then input into the target recognition algorithm model in the embedded platform for image processing, and the target recognition result is output. This invention creatively transplants the constructed mission planning mathematical model and target recognition algorithm model into a preset embedded platform, and realizes the entire process of mission planning and control, remote sensing image acquisition and remote sensing image processing of the remote sensing satellite based on the preset embedded platform, thereby meeting the requirements of processing efficiency.

[0111] To better illustrate the working principle and beneficial effects of the present invention, the following description uses the present invention to identify aircraft and ship targets.

[0112] like Figure 4 and Figure 5 As shown, Figure 4 The accuracy and detection rate of the target recognition algorithm model are shown. Figure 5 The present invention provides the results of aircraft and ship target identification, wherein 5a shows the results of aircraft target identification and 5b shows the results of ship target identification.

[0113] Depend on Figure 4 and Figure 5 As can be seen, the embedded implementation method for satellite mission planning and control and intelligent image processing provided by the present invention can identify key targets such as aircraft and ships with an accuracy of 72% and a detection rate of 56%, which can meet mission requirements.

[0114] The above description is merely a preferred embodiment of the embedded implementation method for satellite mission planning and control and intelligent image processing provided by the present invention, and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An embedded implementation method for satellite mission planning and control and intelligent image processing, characterized in that, Includes the following steps: S1. Construct a mathematical model for remote sensing satellite mission planning and a target recognition algorithm model based on the YOLO v3 network; S2. Port the task planning mathematical model and target recognition algorithm model to a preset embedded platform; S3. Determine the attitude of the remote sensing satellite and obtain the attitude control algorithm of the remote sensing satellite. Obtain the remote sensing image of the remote sensing satellite based on the attitude information of the remote sensing satellite and the mission planning mathematical model. S4. Input the acquired remote sensing image into the target recognition algorithm model for target recognition, and then output the recognition result; The specific implementation methods for constructing the mathematical model for remote sensing satellite mission planning in step S1 include: S11. Construct a mathematical model for mission planning of remote sensing satellites based on constraints, wherein the constraints include the set of observation targets, satellite orbit, platform characteristics, payload type, energy, storage and illumination, and the objectives include maximizing the priority of observation targets, maximizing the number of observation targets and maximizing resource load balancing. S12. Solve and optimize the mathematical model of the task planning based on genetic algorithm and tabu search algorithm; The mathematical model for task planning in step S11 is expressed by the following formula: In the formula, This indicates maximizing the priority of the observed target. This indicates maximizing the number of observed targets. This indicates maximizing resource load balancing. , and They represent , and The weight value, ,in: ； ； ； In the formula, Indicates the target observed by the satellite. Represents the target priority set. Indicates the task number. Indicates the satellite number, This indicates the total number of tasks. Indicates the total number of satellites. This indicates the energy consumed by a satellite when performing a mission. This represents the standard deviation function.

2. The embedded implementation method for satellite mission planning and control and intelligent image processing according to claim 1, characterized in that, The specific implementation of step S12 includes: S121. Initialize the parameters of the task planning mathematical model, use a genetic algorithm to perform a global search on the initial population until the initial population converges to a preset value, and then output the optimal solution of the population. S122. The optimal solution of the population output in step S121 is used as the initial solution of the tabu search algorithm, and a heuristic factor is set to perform local iteration on key areas until the preset number of iterations is reached, and then the planning result is output to complete the solution optimization of the mathematical model of the task planning.

3. The embedded implementation method for satellite mission planning and control and intelligent image processing according to claim 2, characterized in that, The specific implementation of constructing the target recognition algorithm model based on the YOLO v3 network in step S1 includes: S31. Add an spp layer to the existing YOLO v3 network topology to optimize multi-scale target detection; S32. Perform sparse training on the YOLO v3 network from step S31, and apply L1 regularization to the BN layer, as expressed by the formula: In the formula, , Indicates the parameters of the BN layer. Indicates the scaling factor. This represents the truth value of the minimum batch. This represents the mean of the smallest batch. Indicates the penalty factor. Indicates the offset. This represents the variance of the minimum batch size. Represents a constant. Represents the training objective function. Represents the training loss function; S33. Based on the BN layer parameter values ​​in step S32, channel pruning is performed to achieve model compression, thereby obtaining the target recognition algorithm model based on the YOLO v3 network.

4. The embedded implementation method for satellite mission planning and control and intelligent image processing according to claim 3, characterized in that, The specific implementation method of porting the task planning mathematical model to the preset embedded platform in step S2 includes: S13. Convert the task planning mathematical model into the corresponding Matlab program using Matlab tools; S14. Use the Matlab-Code Generator tool to convert the Matlab program into a C++ project; S15. Compile a top-level class to implement the model interface logic, and compile a plugin class as an auxiliary tool. S16. Standardize the interface of the program to obtain a standardized platform model.

5. The embedded implementation method for satellite mission planning and control and intelligent image processing according to claim 4, characterized in that, The specific implementation method of porting the target recognition algorithm model to the preset embedded platform in step S2 includes: S34. The programming framework is used to load the pre-trained target recognition algorithm model file and weight parameter file; S35. Use the CNML library to parse it and generate model data containing instruction sets and model parameters; S36. Use the CNRT library to transfer the generated model data to the hardware DRAM; S37. Input the application data to be processed and start hardware computing; S38. Read the calculation result and return it layer by layer upwards to the programming framework.

6. The embedded implementation method for satellite mission planning and control and intelligent image processing according to claim 5, characterized in that, The preset embedded platform is the Cambricon MLU220 embedded computing platform, including: The x86 main control board is used for overall process control of the embedded platform and data processing of various external interfaces. It is equipped with the first M.2 interface. The MLU220 intelligent computing card is used to accelerate the operation of algorithms and has a second M.2 interface that matches and connects to the first M.2 interface.

7. The embedded implementation method for satellite mission planning and control and intelligent image processing according to claim 6, characterized in that, In step S3, the attitude of the remote sensing satellite is determined by a combination of infrared horizon sensor, sun sensor and gyroscope. Then, the attitude control algorithm based on PD controller can be obtained using the determined attitude of the remote sensing satellite.

Images