Intelligent firepower task planning method based on convolutional neural network and storage medium

Abstract

The invention discloses an intelligent firepower task planning method based on a convolutional neural network and a storage medium, and the method comprises the following steps: constructing a data set which is marked data and historical data for a commander to plan a task scheme; dividing the data set into a training set and a test set; repeatedly inputting the training data set into an algorithm network for training to generate model parameter information; calling the model parameter file, and verifying the correctness of the model parameters by using the test data set; after verification, generating a binary model file; selecting one piece of task data to be planned, and converting the data into a matrix of [800, 800, 5]; inputting the matrix into a binary model file for prediction; and after the prediction matrix is obtained, analyzing the prediction matrix to form a task planning scheme.

Description

Technical Field

[0001] The present invention relates to the technical field of task planning, and in particular to a firepower intelligent task planning method and a storage medium based on a convolutional neural network. Background Art

[0002] The main purpose of planning artillery combat missions is to give our strike decision plan based on the current terrain, topography, weather, enemy target attributes, our combat forces, our combat capabilities, and the expected value of enemy target damage. The strike decision plan includes what kind of firepower to use and how to allocate firepower to achieve the best damage effect in theory. Therefore, the problem of artillery firepower task planning can be studied in combination with the characteristics of neural networks in deep learning. Among them, what firepower to use to strike the target can be solved by classification ideas, and the allocation of firepower to different targets can be solved by the regression characteristics of neural networks. The answer to intelligent decision-making is given through iterative training of convolutional neural networks.

[0003] Deep learning is a process in which machines learn from historical data to understand the factors that humans need to consider when making similar decisions, as well as the results of human decisions, so as to replace or assist humans in making decisions. When we consider which weapons and ammunition to use to attack relevant targets, we first focus on what type of target the target is, whether it is an armored target, a combatant, a command post, or other targets. After that, we need to determine how our superiors or we want to damage the target, whether it is to lose the ability to move or to be directly destroyed, that is, our expectations for damage. After determining the target and damage expectations, we need to determine how to reasonably plan our weapons and ammunition under the current environment (what kind of terrain and what kind of visibility conditions) to ultimately achieve our goals. And visibility conditions are often affected by terrain and weather conditions.

[0004] At present, the traditional mission planning algorithm is based on genetic algorithms. The idea of ​​genetic algorithms is to constrain the initial population data of ammunition under the constraints of the established strike targets. In order to avoid the size of the coverage space selected by the initial population directly affecting the population diversity, the system is generated by roulette according to the value of the target, thereby generating a bias in allocating ammunition to high-value targets, providing seeds for seeking optimal solutions; or using time-controlled random numbers to randomly generate and allocate targets, cyclically allocate ammunition, thereby generating a relatively average allocation plan; or using time-controlled random numbers to randomly generate and allocate ammunition numbers, cyclically allocate to each target, thereby generating a large ammunition allocation plan for individual targets when generating evenly distributed individuals, as a necessary supplement to population diversity. Traditional mission planning algorithms often fall into local optimality when facing real-time changing battlefield situations. It is difficult to cope with battlefield data with diverse data forms, high randomness, nonlinearity, and high dimensions, and cannot meet the objective needs of real-time decision-making in complex battlefield environments. Summary of the invention

[0005] The present invention proposes a firepower intelligent task planning method based on convolutional neural network, which can solve at least one of the above technical problems.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A firepower intelligent task planning method based on convolutional neural network includes the following steps: S1. First, according to the process of constructing the training data set, a data set of more than 10,000 items is constructed. The data set is the labeled data and is the historical data for the commander to plan the mission plan; S2, divide the data set into a training set and a test set, with a ratio of 9:1 between the training set and the test set; S3, repeatedly input the training set data into the algorithm network for training to generate model parameter information; S4. Call the model parameter file and use the test set data to verify the correctness of the model parameters; S5. After verification, a binary model file is generated; S6. Select one piece of task data to be planned and convert the data into a matrix of [800, 800, 5]. S7, input the matrix into the binary model file and make predictions; S8. After obtaining the prediction matrix, the prediction matrix is ​​analyzed to form a task planning plan.

[0007] Furthermore, the training data set construction steps include: First, the battlefield environment is gridded, with each square containing characteristic information and training data label information in four aspects: topography, natural environment, enemy target type, and target damage expectation; the position of the square represents the geographical space location of the current weapons, weather, and landforms. Furthermore, the training data set construction step also includes digitally encoding the terrain, landforms, weather, enemy targets, etc.; The data is encoded as follows: Weather: {Sunny: 0, Cloudy: 1, Rain: 2, Snow: 3, Fog: 4, Wind: 5, Sandstorm: 6, Hail: 7}; Terrain: {Road: 0, Lake: 1, River: 2, Hill: 3, Mountain: 4, Valley: 5, Tree: 6, Swamp: 7, Basin: 8, Grassland: 9, Bush: 10, Mudflat: 11, Ocean: 12, Coast: 13, Bridge: 14, Pier: 15}; Target type: {No target: 0, Armored target: 1, Airport: 2, Personnel: 3, Radar station: 4, Command post: 5, Bunker: 6, Oil depot: 7}; Ammunition: {GP-1 155mm laser terminal-guided artillery shell: 0, GP155B image terminal-guided artillery shell: 1, WS-35 155mm satellite-guided artillery shell: 2, GP155G 155mm terminal-sensitive projectile: 3, Hongjian-8A anti-tank missile: 4, Hongjian-9 anti-tank missile: 5, Hongjian-10 anti-tank missile: 6, Hongjian-73A: 7, AFT07C anti-tank missile: 8, hq-7B field air defense missile system: 9, Hongqi 16 air defense missile: 10, Hongqi 17 air defense missile: 11, PHL03A 300mm long-range multiple rocket launcher: 12, QW-1 individual air defense missile: 13, Hongying 6 portable air defense missile: 14, 155mm conventional ammunition 01: 15, 155mm conventional ammunition 02: 16, 155mm conventional ammunition 03: 17, 155mm conventional ammunition 04: 18, PL96 122mm howitzer: 19}; Target damage expectation: {no damage: 0, slight damage: 1, moderate damage: 2, severe damage: 3}; Similarly, the information at each level is digitally encoded, and enumeration classification is used to digitally encode all known climates, target types, target materials, and terrains.

[0008] Furthermore, the algorithm network includes: 1) A matrix of length 576, width 576, and channels 5 is input into a convolution layer with an 8*8 convolution core, a stride of 2, and channels 64 for processing. After processing, it is input into a pooling layer with a 3*3 convolution core and a stride of 2, and finally outputs a 144*144*64 matrix; 2) The matrix with a length of 144, a width of 144, and a channel of 64 is successively input into three convolutional layers of 1*1*64, 3*3*64, and 1*1*128, with a step size of 1, and is continuously input 4 times, for a total of 12 convolutional layers; then the data is input into a convolutional layer with a 3*3 convolution kernel, 128 channels, and a step size of 2, and finally input into a pooling layer with a 3*3 convolution kernel and a step size of 1, obtaining a 72*72*128 matrix; 3) The 72*72*128 matrix is ​​input into the 1*1*128 and 3*3*128 convolution layers in sequence, with a step size of 1, for 4 consecutive times, for a total of 8 layers, and then input into a convolution layer with a convolution core of 8*8, a channel of 256, and a step size of 2, and finally input into a pooling layer with a 3*3 convolution core and a step size of 2; 4) Input a 18*18*256 matrix into three convolutional layers of 3*3*256, 1*1*256, and 3*3*1024, with a step size of 1, and input it 4 times in succession, for a total of 12 layers; finally input it into a 3*3 pooling layer with a step size of 2 to obtain a 9*9*1024 matrix; 5) Input the 9*9*1024 matrix into the three convolutional layers of 3*3*1024, 1*1*1024, and 3*3*1024 in sequence, with a step size of 1, and input it 4 times in succession to obtain a 9*9*1024 matrix; 6) Straighten the 9*9*1024 matrix to form a 4096 long matrix; 7) The 4096 long vector is converted into a 9*9*30 matrix through a fully connected layer.

[0009] Furthermore, after obtaining the prediction matrix, the prediction matrix is ​​parsed to form a task planning scheme, which specifically includes: After a piece of data is processed by the algorithm, it will form a 9*9*30 matrix, which is the prediction result matrix; Divide the original data into 9*9 grids, each grid is a prediction grid; Each grid is responsible for predicting two rectangular ammunition detection frames. Each detection frame contains the predicted geographic location (x, y) where the ammunition should be placed, the width and height of the detection frame, and a confidence level of whether the ammunition should be placed in the frame. , the confidence level is 0 or 1; For each ammunition detection box, generate a 20-category probability, representing 20 types of ammunition, each probability value is between 0 and 1, including 0.001, 0.01, 0.02, 0.03, 0.004, 0.005, 0.006, 0.9, 0.01, 0.2, 0.3,0.11, 0.13, 0.15, 0.16, 0.1, 0.12, 0.15, 0.12, 0.15; each probability value represents the probability of a specific ammunition in the detection box. Furthermore, the width and height of the detection frame, that is, the product of the width and height of the detection frame, is used as the number of ammunition. For the same type of ammunition, assuming that it is spread out near the striking platform, the size of the detection frame can be considered as the number of ammunition.

[0010] On the other hand, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.

[0011] A firepower intelligent task planning method based on convolutional neural network, comprising: It can be seen from the above technical solution that the firepower intelligent task planning method based on convolutional neural network of the present invention has the following beneficial effects: Innovation 1: The deep learning network model normalizes and dimensionlessly transforms complex battlefield situations, and can efficiently process battlefield data of various forms, strong randomness, nonlinearity, and high dimensions, and can deeply explore combat rules; Innovation 2: Through the learning of a large number of samples, the rationality analysis characteristics of combat mission planning are formed. After the sample training is completed, the deep learning network can effectively simulate the commander's thinking mode, has the ability of rationality analysis, and can ultimately realize the rationality analysis of combat mission planning; Innovation 3: The deep learning network uses the self-learning ability of the model to continuously optimize the learning effect, and the model's analysis and generalization capabilities are continuously enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 is a method flow chart of an embodiment of the present invention; Figure 2 is a schematic diagram of training data according to an embodiment of the present invention; Figure 3 is a model training input-feature data sample of an embodiment of the present invention; Figure 4 is a model training input-labeled data sample of an embodiment of the present invention; Figure 5 is a schematic diagram of the algorithm network structure of an embodiment of the present invention; Figure 6 is a schematic diagram of analyzing the prediction results of an embodiment of the present invention; Figure 7 2 is a schematic diagram for comparing the algorithm effects of the embodiments of the present invention. DETAILED DESCRIPTION

[0013] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0014] Since the generation of the task planning scheme of the artillery firepower unit is an action plan for attacking a certain target based on the current combat environment, the damage capability of our combat forces, and the vulnerability characteristics of the enemy's combat forces. The action plan will change in real time due to the geographical location, weather conditions, topography, etc. of both sides, and has a highly dynamic change characteristic. According to the thinking process of task plan planning, the present invention determines to use topography, natural environment, enemy targets, and target damage expectations as feature inputs for deep learning, and the ammunition scheme as a label for training data, so the embodiment of the invention is designed as follows Figure 1As shown, the following steps are included: 1) First, according to the process of constructing the training data set, a 10,000-item data set is constructed. The data set is labeled data and is the historical data for commanders to plan mission plans. Using 10,000 or more data will have better training effects and will not be overfitted; 2) Divide the data set into a training set and a test set. The training set contains 9,000 data items and the test set contains 1,000 data items.

[0015] 3) The 9,000 training data sets are repeatedly put into the algorithm network for training to generate model parameter information.

[0016] 4) Call the model parameter file and use 1000 test data sets to verify the correctness of the model parameters.

[0017] 5) After verification, generate a binary model file.

[0018] 6) Select one task data to be planned and convert the data into a matrix of [800,800,5].

[0019] 7) Input the matrix into the binary model file and make predictions.

[0020] 8) After obtaining the prediction matrix, the prediction matrix is ​​analyzed to form a task planning plan.

[0021] The following are detailed descriptions: Step 1: Training data construction First, the battlefield environment is gridded, and each square contains characteristic information and training data label information in four aspects: topography, natural environment, enemy target type, target damage expectation, etc. The position of the square represents the geographical space location of the current weapon, weather, and topography.

[0022] Since computers can only calculate numbers but cannot recognize text, it is necessary to digitally encode terrain, landforms, weather, enemy targets, etc. The data encoding is as follows: Weather: {Sunny: 0, Cloudy: 1, Rain: 2, Snow: 3, Fog: 4, Wind: 5, Sandstorm: 6, Hail: 7} Terrain: {Road: 0, Lake: 1, River: 2, Hill: 3, Mountain: 4, Valley: 5, Tree: 6, Swamp: 7, Basin: 8, Grassland: 9, Bush: 10, Mudflat: 11, Ocean: 12, Coast: 13, Bridge: 14, Pier: 15} Target type: {No target: 0, Armored target: 1, Airport: 2, Personnel: 3, Radar station: 4, Command post: 5, Bunker: 6, Oil depot: 7} Ammunition: {GP-1 155mm laser terminal-guided artillery shell: 0, GP155B image terminal-guided artillery shell: 1, WS-35 155mm satellite-guided artillery shell: 2, GP155G 155mm terminal-sensitive projectile: 3, Hongjian-8A anti-tank missile: 4, Hongjian-9 anti-tank missile: 5, Hongjian-10 anti-tank missile: 6, Hongjian-73A: 7, AFT07C anti-tank missile: 8, hq-7B field air defense missile system: 9 , Hongqi 16 air defense missile: 10, Hongqi 17 air defense missile: 11, PHL03A 300mm long-range multiple rocket launcher: 12, QW-1 individual air defense missile: 13, Hongying 6 portable air defense missile: 14, 155mm conventional ammunition 01: 15, 155mm conventional ammunition 02: 16, 155mm conventional ammunition 03: 17, 155mm conventional ammunition 04: 18, PL96 122mm howitzer: 19} Target damage expectation: {No damage: 0, slight damage: 1, moderate damage: 2, severe damage: 3} Similarly, the information at each level is digitally encoded. Currently, enumeration classification is used to digitally encode all known climates, target types, target materials, and terrains. Figure 2 The training dataset shown.

[0023] Label data, use XML to record annotation information, and input all data into the algorithm network. One piece of data contains 5 modules, and each module describes the matrix information of each attribute, such as Figure 3 As shown; Figure 4 It is the model training input - labeled data samples.

[0024] Step 2: Algorithm network structure design Figure 5 As shown; 1) A matrix of length 576, width 576, and channels 5 is input into a convolutional layer with an 8*8 convolutional core, a stride of 2, and channels 64 for processing. After processing, it is input into a pooling layer with a 3*3 convolutional core and a stride of 2, and finally outputs a 144*144*64 matrix.

[0025] 2) A matrix with a length of 144, a width of 144, and a channel of 64 is input into three convolutional layers of 1*1*64, 3*3*64, and 1*1*128, with a step size of 1, and input it 4 times in succession, for a total of 12 convolutional layers. Then the data is input into a convolutional layer with a 3*3 convolution kernel, 128 channels, and a step size of 2, and finally input into a pooling layer with a 3*3 convolution kernel and a step size of 1, resulting in a 72*72*128 matrix.

[0026] 3) A 72*72*128 matrix is ​​input into the 1*1*128 and 3*3*128 convolution layers with a step size of 1 for 4 consecutive times, for a total of 8 layers, and then input into a convolution layer with a convolution kernel of 8*8, 256 channels, and a step size of 2. Finally, it is input into a pooling layer with a 3*3 convolution kernel and a step size of 2 to generate a 18*18*256 matrix.

[0027] 4) Input a 18*18*256 matrix into three convolutional layers of 3*3*256, 1*1*256, and 3*3*1024, with a step size of 1, and input it 4 times in succession, for a total of 12 layers. Finally, input it into a 3*3 pooling layer with a step size of 2 to obtain a 9*9*1024 matrix.

[0028] 5) Input a 9*9*1024 matrix into three convolutional layers: 3*3*1024, 1*1*1024, and 3*3*1024, with a step size of 1, and input it 4 times in succession to obtain a 9*9*1024 matrix 6) Straighten a 9*9*1024 matrix to form a 4096 long matrix.

[0029] 7) Convert the 4096-long matrix into a 9*9*30 matrix through a fully connected layer.

[0030] Step 3: Prediction matrix analysis The prediction results are analyzed as follows: Figure 6 As shown; 1) After a piece of data is processed by the above network algorithm, it will form a 9*9*30 matrix, which is the prediction result matrix. In order to facilitate the analysis of the results, the original data is divided into 9*9 grids, each of which is a prediction grid.

[0031] 2) Each grid is responsible for predicting two rectangular ammunition detection frames. Each detection frame contains the predicted geographic location (x, y) where the ammunition should be placed, the width and height of the detection frame (the product of the width and height of the detection frame is used as the number of ammunition. For the same type of ammunition, assuming that it is laid out near the strike platform, the size of the detection frame can be considered as the number of ammunition) and a confidence level of whether ammunition should be placed in the frame. , with a confidence level of 0 or 1.

[0032] 3) For each ammunition detection frame, generate a probability of 20 categories (representing 20 types of ammunition), each probability value is between 0 and 1, such as [0.001, 0.01, 0.02, 0.03, 0.004, 0.005, 0.006, 0.9, 0.01, 0.2, 0.3,0.11, 0.13, 0.15, 0.16, 0.1, 0.12, 0.15, 0.12, 0.15]. Each probability value represents the probability of a specific ammunition in the detection frame. For example, 0.9 represents the classification of 155 howitzer, and the confidence of the detection frame having ammunition in the second step is 1, then it is considered that the probability of the detection frame detecting that the target is hit by a 155 howitzer is 0.9. If the confidence of the detection frame having ammunition is 0, it is also considered that there is no ammunition at this position, that is, this type of ammunition should not be fired here.

[0033] Step 4: Algorithm effect comparison For these 10,000 data, we used both the traditional genetic algorithm and the invented method for training and prediction, and tested their convergence time and effect. We found that the task planning method based on deep learning has better convergence effect and better prediction accuracy than the genetic algorithm. (e.g. Figure 7 As shown in the figure, for the 1000 test data sets, the accuracy of the genetic algorithm is 56.8%, and the prediction accuracy of the algorithm in this project is 79.5%).

[0034] In yet another aspect, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of any of the above methods.

[0035] On the other hand, the present invention further discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of any of the above methods.

[0036] In another embodiment provided by the present application, a computer program product including instructions is also provided, which, when executed on a computer, enables the computer to execute the steps of any method in the above embodiments.

[0037] It is understandable that the system provided by the embodiment of the present invention corresponds to the method provided by the embodiment of the present invention, and the explanation, examples and beneficial effects of the relevant contents can refer to the corresponding parts in the above method.

[0038] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0039] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A firepower intelligent task planning method based on convolutional neural network, characterized in that: The following steps are included: S1. First, according to the process of constructing the training data set, a data set of more than 10,000 items is constructed. The data set is the labeled data and is the historical data for the commander to plan the mission plan; S2, divide the data set into a training set and a test set, with a ratio of 9:1 between the training set and the test set; S3, repeatedly input the training set data into the algorithm network for training to generate model parameter information; S4. Call the model parameter file and use the test set data to verify the correctness of the model parameters; S5. After verification, a binary model file is generated; S6. Select one piece of task data to be planned and convert the data into a matrix of [800, 800, 5]. S7, input the matrix into the binary model file and make predictions; S8. After obtaining the prediction matrix, the prediction matrix is ​​analyzed to form a task planning plan.

2. The firepower intelligent task planning method based on convolutional neural network according to claim 1 is characterized in that: The steps to construct the training dataset include: First, the battlefield environment is gridded, with each square containing characteristic information and training data label information in four aspects: topography, natural environment, enemy target type, and target damage expectation; the position of the square represents the geographical space location of the current weapons, weather, and landforms.

3. The firepower intelligent task planning method based on convolutional neural network according to claim 2 is characterized in that: The training data set construction steps also include digital coding of terrain, landforms, weather, enemy targets, etc.; The data is encoded as follows: Weather: {Sunny: 0, Cloudy: 1, Rain: 2, Snow: 3, Fog: 4, Wind: 5, Sandstorm: 6, Hail: 7}; Terrain: {Road: 0, Lake: 1, River: 2, Hill: 3, Mountain: 4, Valley: 5, Tree: 6, Swamp: 7, Basin: 8, Grassland: 9, Bush: 10, Mudflat: 11, Ocean: 12, Coast: 13, Bridge: 14, Pier: 15}; Target type: {No target: 0, Armored target: 1, Airport: 2, Personnel: 3, Radar station: 4, Command post: 5, Bunker: 6, Oil depot: 7}; Ammunition: {GP-1 155mm laser terminal-guided artillery shell: 0, GP155B image terminal-guided artillery shell: 1, WS-35 155mm satellite-guided artillery shell: 2, GP155G 155mm terminal-sensitive projectile: 3, Hongjian-8A anti-tank missile: 4, Hongjian-9 anti-tank missile: 5, Hongjian-10 anti-tank missile: 6, Hongjian-73A: 7, AFT07C anti-tank missile: 8, hq-7B field air defense missile system: 9, Hongqi 16 air defense missile: 10, Hongqi 17 air defense missile: 11, PHL03A 300mm long-range multiple rocket launcher: 12, QW-1 individual air defense missile: 13, Hongying 6 portable air defense missile: 14, 155mm conventional ammunition 01: 15, 155mm conventional ammunition 02: 16, 155mm conventional ammunition 03: 17, 155mm conventional ammunition 04: 18, PL96 122mm howitzer: 19}; Target damage expectation: {no damage: 0, slight damage: 1, moderate damage: 2, severe damage: 3}; Similarly, the information at each level is digitally encoded, and enumeration classification is used to digitally encode all known climates, target types, target materials, and terrains.

4. The firepower intelligent task planning method based on convolutional neural network according to claim 1 is characterized in that: The algorithm network includes: 1) A matrix of length 576, width 576, and channels 5 is input into a convolution layer with an 8*8 convolution core, a stride of 2, and channels 64 for processing. After processing, it is input into a pooling layer with a 3*3 convolution core and a stride of 2, and finally outputs a 144*144*64 matrix; 2) The matrix with a length of 144, a width of 144, and a channel of 64 is successively input into three convolutional layers of 1*1*64, 3*3*64, and 1*1*128, with a step size of 1, and is continuously input 4 times, for a total of 12 convolutional layers; then the data is input into a convolutional layer with a 3*3 convolution kernel, 128 channels, and a step size of 2, and finally input into a pooling layer with a 3*3 convolution kernel and a step size of 1, obtaining a 72*72*128 matrix; 3) The 72*72*128 matrix is ​​input into the 1*1*128 and 3*3*128 convolution layers in sequence, with a step size of 1, for 4 consecutive times, for a total of 8 layers, and then input into a convolution layer with a convolution core of 8*8, a channel of 256, and a step size of 2, and finally input into a pooling layer with a 3*3 convolution core and a step size of 2; 4) Input a 18*18*256 matrix into three convolutional layers of 3*3*256, 1*1*256, and 3*3*1024, with a step size of 1, and input it 4 times in succession, for a total of 12 layers; finally input it into a 3*3 pooling layer with a step size of 2 to obtain a 9*9*1024 matrix; 5) Input the 9*9*1024 matrix into the three convolutional layers of 3*3*1024, 1*1*1024, and 3*3*1024 in sequence, with a step size of 1, and input it 4 times in succession to obtain a 9*9*1024 matrix; 6) Straighten the 9*9*1024 matrix to form a 4096 long matrix; 7) The 4096 long vector is converted into a 9*9*30 matrix through a fully connected layer.

5. The firepower intelligent task planning method based on convolutional neural network according to claim 1 is characterized in that: After obtaining the prediction matrix, the prediction matrix is ​​analyzed to form a task planning scheme, which specifically includes: After a piece of data is processed by the algorithm, it will form a 9*9*30 matrix, which is the prediction result matrix; Divide the original data into 9*9 grids, each grid is a prediction grid; Each grid is responsible for predicting two rectangular ammunition detection frames. Each detection frame contains the predicted geographic location (x, y) where the ammunition should be placed, the width and height of the detection frame, and a confidence level of whether the ammunition should be placed in the frame. , the confidence level is 0 or 1; For each ammunition detection box, generate a 20-category probability, representing 20 types of ammunition, each probability value is between 0 and 1, including 0.001, 0.01, 0.02, 0.03, 0.004, 0.005, 0.006, 0.9, 0.01, 0.2, 0.3,0.11, 0.13, 0.15, 0.16, 0.1, 0.12, 0.15, 0.12, 0.15; each probability value represents the probability of a specific ammunition in the detection box.

6. The firepower intelligent task planning method based on convolutional neural network according to claim 5 is characterized in that: The width and height of the detection box are the product of the width and height of the detection box as the number of ammunition. For the same type of ammunition, assuming that it is spread out near the striking platform, the size of the detection box can be considered as the number of ammunition.

7. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 6.

Images