Electronic Countermeasure Target Threat Assessment Method Based on Residual Convolutional Autoencoder

Abstract

The present invention discloses an electronic countermeasure target threat assessment method based on a residual convolutional autoencoder, aiming to solve the problems in the prior art that weight coefficients need to be set manually and threat assessment can be effectively carried out only with a complete data set. The implementation steps of the present invention include: generating a training set containing 9 electronic countermeasure target feature parameters; building a residual convolutional autoencoder with 17 network layers that can process incomplete data sets; inputting the training set into the residual convolutional autoencoder, and using the backpropagation gradient descent method to iteratively update the parameters of each layer of the residual convolutional autoencoder to obtain a trained residual convolutional autoencoder; (4) inputting the sample to be evaluated for threat level into the trained residual convolutional autoencoder and outputting the evaluation level of the threat level. The present invention has the advantages of not requiring manual setting of weight coefficients and being able to effectively carry out threat assessment without relying on a complete data set.

Description

Technical Field

[0001] The present invention belongs to the technical field of radar communication, and further relates to a method for evaluating the threat of electronic countermeasure targets based on a residual convolutional autoencoder in the technical field of electronic countermeasures. The present invention evaluates the threat level of electronic countermeasure targets. Background Art

[0002] In electronic countermeasures, the threat level of electronic countermeasure targets is an index that is highly concerned and urgently desired in the entire countermeasure system. With the intelligence of electronic countermeasure equipment and the complexity of the electromagnetic environment in which electronic countermeasures are carried out, it has become more difficult to evaluate the threat level of electronic countermeasure targets. Therefore, threat assessment technology has received more attention in the field of electronic countermeasures.

[0003] The Xi'an Institute of Electronic Engineering proposed a method for evaluating the threat of air defense weapon system targets in its patent document "Method for Evaluating the Threat of Air Defense Weapon System Targets" (Application No.: CN 201810411954.X; Publication No.: CN 108805406 A). The implementation scheme of this method is as follows: First step, select threat assessment indicators, including the importance of protected key areas, target types, and the urgency of target interception. Second step, calculate the evaluation indicators: according to the indicator calculation formula, calculate the quantization value of each indicator. Third step, according to the indicator weight coefficients given by experts, calculate the indicator quantization values to obtain the final threat assessment result. This method is simple to calculate and can qualitatively and quantitatively analyze the threat level to obtain the threat assessment result. However, the deficiency of this method is that the threat assessment result requires artificial determination of the indicator weight coefficients, and the influence of such human factors cannot objectively and reasonably evaluate the threat level.

[0004] Zhang Yang et al. disclosed a method for evaluating the threat level of targets using deep learning technology in their published paper "An Intelligent Threat Assessment Algorithm Based on Deep Learning" (Modern Navigation 2021, 12(04): 293 - 296 + 305). The implementation scheme of this method is as follows: First step, generate a data set: select threat assessment indicators and quantify the indicators, and use the quantified indicator values to generate a data set, which includes a training set and a test set. Second step, construct a deep learning network and input the complete training set generated in the first step into the model for training. Third step, predict the output result: input the complete test set into the trained model to predict the target threat level. This method has high prediction accuracy and strong generalization ability, and can effectively achieve target threat assessment. However, the deficiency of this method is that a complete indicator data set is required to effectively complete the training of the deep learning network. When the indicator data is missing, it is impossible to effectively extract data features, which affects the accuracy of the evaluation result. Summary of the Invention

[0005] The object of the present invention is to provide a threat assessment method for electronic countermeasure targets based on a residual convolutional autoencoder in view of the deficiencies of the above-mentioned existing technologies, so as to solve the problems that the existing technologies need to artificially determine the index weight coefficients and obtain all index information to conduct threat assessment, and to ensure the effective assessment of the threat level of targets in actual application scenarios.

[0006] The idea to achieve the object of the present invention is as follows: The present invention constructs a residual convolutional autoencoder model. Since the residual convolutional autoencoder has the ability to process classification problems, and threat assessment essentially belongs to a classification problem, threat assessment can be carried out through the characteristics of the residual convolutional autoencoder in processing classification problems. In addition, the autoencoder structure in the residual convolutional autoencoder can learn damaged data and reconstruct the damaged data to fill in the missing information indicators, ensuring the integrity of the indicator set and solving the problem that the existing technologies need to rely on a complete indicator set to conduct threat assessment. By training the residual convolutional autoencoder with the target training set, the update of the weights will adaptively change according to the characteristics of the actual training set data, making the trained residual convolutional autoencoder have good non-linear mapping ability. When conducting threat assessment on electronic countermeasure targets, passing the target test data through the trained residual convolutional autoencoder can quickly fill in the missing information and extract the target feature parameters, and the corresponding threat assessment level can be obtained to complete the threat assessment.

[0007] The specific steps to achieve the object of the present invention are as follows:

[0008] Step 1, generate a training set:

[0009] Step 1.1, form a sample by combining 9 electronic countermeasure target feature parameters; select at least 6,000 samples to form a sample set;

[0010] Step 1.2, calculate the threat assessment level of each sample target; according to the threat assessment level of each sample target, add a threat assessment level label to each sample;

[0011] Step 1.3, perform a normalization operation on the feature parameters of each sample in the sample set;

[0012] Step 1.4, form a training set by combining all the normalized samples and the threat assessment level label corresponding to each sample;

[0013] Step 2, build a residual convolutional autoencoder:

[0014] Build a 17-layer residual convolutional autoencoder capable of processing incomplete sample sets, with its structure connected in series as follows: the first convolutional layer, the first pooling layer, the second convolutional layer, the second pooling layer, the third convolutional layer, the fourth convolutional layer, the fifth convolutional layer, the sixth convolutional layer, the seventh convolutional layer, the first deconvolutional layer, the second deconvolutional layer, the third deconvolutional layer, the first upsampling layer, the fourth deconvolutional layer, the second upsampling layer, the fifth deconvolutional layer, and the fully connected layer;

[0015] Set the number of convolutional kernels of the first to seventh convolutional layers to 128, 64, 32, 32, 16, 16, and 8 in sequence, and set the size of the convolutional kernels to 1×3; both the first and second pooling layers use the maximum pooling method, with the size of the pooling kernels set to 1×2 and the pooling stride set to 1×2; set the number of convolutional kernels of the first to fifth deconvolutional layers to 8, 8, 8, 16, and 8 in sequence, and set the size of the convolutional kernels to 1×3; the size of the pooling kernels of the first and second upsampling layers is set to 1×2, and the pooling stride is set to 1×2;

[0016] Step 3, train the residual convolutional autoencoder:

[0017] Input the training set into the residual convolutional autoencoder, and use the backpropagation gradient descent method to iteratively update the parameters of each layer of the residual convolutional autoencoder until the loss function of the residual convolutional autoencoder converges, obtaining a trained residual convolutional autoencoder;

[0018] Step 4, evaluate the threat level:

[0019] Adopt the same normalization method as in Step 1.3, normalize each sample of the threat level to be evaluated, and input the normalized sample into the trained residual convolutional autoencoder to output the evaluation grade of the threat level of the sample.

[0020] Compared with the prior art, the present invention has the following advantages:

[0021] First, for the residual convolutional autoencoder constructed by the present invention, the update of its weights is adaptively changed according to actual parameters, overcoming the deficiency of the prior art that requires artificial setting of weight coefficients for evaluation, so that the present invention has the advantage of evaluating the target threat level under objective conditions.

[0022] Second, the residual convolutional autoencoder constructed by the present invention can reconstruct missing data and can fill in the missing feature parameters of information to ensure the integrity of the sample set, overcoming the deficiency of the prior art that requires a complete data set for threat assessment, so that the present invention has the advantage of evaluating the target threat level under the condition of an incomplete sample set. Description of the Drawings

[0023] Figure 1 This is the flow chart of the present invention. Detailed implementation manners

[0024] The following will further describe in detail the specific implementation steps of the present invention in conjunction with the appended Figure 1 drawings and embodiments.

[0025] Step 1: Generate a training set.

[0026] In step 1.1, an embodiment of the present invention forms a sample by using nine characteristic parameters that can reflect the target threat level, and selects 6,000 samples to form a sample set. The nine characteristic parameters are respectively target type, target maneuverability, target flight ability, target attack ability, target stealth ability, main function of the target radar, maximum detection distance of the target radar, target radar resolution, and target radar anti-jamming ability.

[0027] In step 1.2, calculate the threat assessment level of each sample target according to the following formula:

[0028]

[0029] where, G i represents the threat assessment level of the i-th sample in the sample set. If the threat assessment level is "poor" when 0 ≤ G i ≤ 0.3, if the threat assessment level is "general" when 0.3 < G i ≤ 0.7, and if the threat assessment level is "very good" when 0.7 < G i ≤ 1. i represents the serial number of the sample in the sample set, i = 1, 2... N, N represents the total number of samples in the training set, u i1 represents the membership degree of the target type of the i-th sample in the sample set, u i2 represents the membership degree of the target maneuverability of the i-th sample in the sample set, u i3 represents the membership degree of the target flight ability of the i-th sample in the sample set, u i4 represents the membership degree of the target attack ability of the i-th sample in the sample set, u i5 represents the membership degree of the target stealth ability of the i-th sample in the sample set, u i6 represents the membership degree of the main function of the target radar of the i-th sample in the sample set, u i7 represents the membership degree of the maximum detection distance of the target radar of the i-th sample in the sample set, u i8 represents the membership degree of the target radar resolution of the i-th sample in the sample set, u i9 represents the membership degree of the target radar anti-jamming ability of the i-th sample in the sample set.

[0030] Add threat assessment level labels to each sample. Set the label of the threat assessment level "poor" to 0, the label of the threat assessment level "average" to 1, and the label of the threat assessment level "very good" to 2.

[0031] Step 1.3: Normalize the characteristic parameters of each sample in the sample set according to the following formula.

[0032]

[0033] where x ij ' represents the attribute of the j-th target characteristic parameter of the i-th sample in the normalized sample set. j represents the serial number of the target characteristic parameter attribute. j = 1 represents the target type in the sample set, j = 2 represents the target maneuverability in the sample set, j = 3 represents the target flight ability in the sample set, j = 4 represents the target attack ability in the sample set, j = 5 represents the target stealth ability in the sample set, j = 6 represents the main function of the target radar in the sample set, j = 7 represents the maximum detection range of the target radar in the sample set, j = 8 represents the resolution of the target radar in the sample set, j = 9 represents the anti-jamming ability of the target radar in the sample set, x ij represents the attribute of the j-th target characteristic parameter of the i-th sample in the sample set, x jmax represents the maximum value of the j-th target characteristic parameter attribute in the sample set, x jmin represents the minimum value of the j-th target characteristic parameter attribute in the sample set.

[0034] Step 1.4: Combine all the normalized samples and their corresponding level labels for each sample to form a training data set.

[0035] Step 2: Build a residual convolutional autoencoder.

[0036] Build a 17-layer residual convolutional autoencoder capable of processing incomplete sample sets. Its structure is connected in series as follows: the first convolutional layer, the first pooling layer, the second convolutional layer, the second pooling layer, the third convolutional layer, the fourth convolutional layer, the fifth convolutional layer, the sixth convolutional layer, the seventh convolutional layer, the first deconvolutional layer, the second deconvolutional layer, the third deconvolutional layer, the first upsampling layer, the fourth deconvolutional layer, the second upsampling layer, the fifth deconvolutional layer, the fully connected layer;

[0037] Set the number of convolutional kernels of the first to seventh convolutional layers to 128, 64, 32, 32, 16, 16, and 8 in sequence, and set the size of the convolutional kernels to 1×3; both the first and second pooling layers adopt the maximum pooling method, the size of the pooling kernels is set to 1×2, and the pooling stride is set to 1×2; set the number of convolutional kernels of the first to fifth transposed convolutional layers to 8, 8, 8, 16, and 8 in sequence, and the size of the convolutional kernels is set to 1×3; the size of the pooling kernels of the first and second upsampling layers is set to 1×2, and the pooling stride is set to 1×2;

[0038] Step 3, train the residual convolutional autoencoder.

[0039] Input the training set into the residual convolutional autoencoder, and use the backpropagation gradient descent method to iteratively update the parameters of each layer of the residual convolutional autoencoder until the loss function of the residual convolutional autoencoder converges, obtaining the trained residual convolutional autoencoder.

[0040] The calculation formula of the loss function is as follows:

[0041]

[0042] Among them, Loss represents the loss function of the residual convolutional autoencoder, ∑· represents the summation operation, c = 0, 1..m, m represents the total number of threat assessment levels, y ic represents the relationship between the predicted value and the true value of the threat assessment level c of the i-th sample in the training dataset. When the true value is equal to the predicted value, then y ic = 1, when the true value is not equal to the predicted value, y ic = 0, p ic represents the predicted probability of the threat assessment level c of the i-th sample in the training dataset.

[0043] Step 4, evaluate the threat level.

[0044] Adopt the same normalization method as in Step 1.3. For each sample of the threat level to be evaluated after normalization processing, input it into the trained residual convolutional autoencoder, and output the evaluation level of the threat level of this sample.

[0045] The following further illustrates the effect of the present invention in combination with simulation experiments:

[0046] 1. Simulation experiment conditions:

[0047] The hardware platform for the simulation experiment of the present invention is: the processor is an 11th Gen Intel(R) Core(TM) i7-11700F@2.50GHz CPU, and the memory is 32GB.

[0048] The software platform for the simulation experiment of the present invention is: Windows 10 operating system and python 3.7.

[0049] The sizes of the training set and the test set used in the simulation experiment of the present invention are set as follows: The training set contains 4,000 samples and the corresponding threat assessment level values, and the test set contains 2,000 samples and the corresponding threat assessment level values. The threat assessment results of the present invention are evaluated using the test set.

[0050] The configuration parameters of the residual convolutional autoencoder for the simulation experiment of the present invention are: The training batch size is set to 100, the learning rate is 0.0005, the dropout rate is 0.5, and the total number of training epochs is 1,000. The network parameters of the residual convolutional autoencoder trained for 200 epochs, 400 epochs, 600 epochs, 800 epochs, and 1,000 epochs using the training set are recorded respectively.

[0051] 2. Simulation content and its result analysis:

[0052] In the simulation experiment of the present invention, after normalizing the 2,000 samples in the test set using the method of the present invention, random data loss is performed, and the test set after random data loss is input into the trained residual convolutional autoencoder to obtain the threat assessment level values of the present invention.

[0053] In the simulation experiment, the existing technology residual convolutional autoencoder classification method adopted refers to the image classification method combining a convolutional autoencoder and a residual network proposed by Cai Yuhan in his published paper "Image Classification Algorithm Based on Convolutional Autoencoding and Residual Network" (Information Technology & Informatization 2022, No. 268(07): 73 - 76), which is abbreviated as the residual convolutional autoencoder classification method.

[0054] The threat assessment level of each sample in the test set is obtained using the method of the present invention. The threat assessment levels are divided into three categories: "poor", "average", and "very good". The test set is tested using the residual convolutional autoencoder trained for 200 epochs, 400 epochs, 600 epochs, 800 epochs, and 100 epochs. The accuracy rate of the threat assessment level "poor" is the number of correctly predicted samples in the test set divided by the total number of samples with the threat assessment level "poor"; the accuracy rate of the threat assessment level "average" is the number of correctly predicted samples in the test set divided by the total number of samples with the threat assessment level "average"; the accuracy rate of the threat assessment level "very good" is the number of correctly predicted samples in the test set divided by the total number of samples with the threat assessment level "very good". Finally, the threat assessment level prediction results are shown in Table 1.

[0055] Table 1 Threat Assessment Level Prediction Table (unit: percentage)

[0056]

[0057]

[0058] As can be seen from Table 1, when the residual convolutional autoencoder is trained for 1000 rounds, the accuracy rate of the present invention in predicting the threat assessment level of "poor" is 96.60%, the accuracy rate of predicting the threat assessment level of "general" is 97.26%, and the accuracy rate of predicting the threat assessment level of "very good" is 96.53%, which proves that the present invention is relatively accurate in the threat assessment results.

Claims

1. An electronic countermeasure target threat assessment method based on a residual convolutional autoencoder, characterized in that Generate a training set containing the characteristic parameters of electronic countermeasure targets, and use the training set to train the constructed residual convolutional autoencoder with classification function; the specific steps of this evaluation method are as follows: Step 1, generate the training set: Step 1.1, form a sample by combining 9 characteristic parameters of electronic countermeasure targets, and select at least 6000 samples to form a sample set; Step 1.2, calculate the threat assessment level of each sample target; according to the threat assessment level of each sample target, add a threat assessment level label to each sample; Step 1.3, perform a normalization operation on the characteristic parameters of each sample in the sample set; Step 1.4, form a training set by combining all the samples after the normalization operation and the corresponding level label of each sample; Step 2, build a residual convolutional autoencoder: Build a 17-layer residual convolutional autoencoder capable of processing incomplete sample sets, and its structure is connected in series in turn: the first convolutional layer, the first pooling layer, the second convolutional layer, the second pooling layer, the third convolutional layer, the fourth convolutional layer, the fifth convolutional layer, the sixth convolutional layer, the seventh convolutional layer, the first deconvolutional layer, the second deconvolutional layer, the third deconvolutional layer, the first upsampling layer, the fourth deconvolutional layer, the second upsampling layer, the fifth deconvolutional layer, the fully connected layer; Set the number of convolutional kernels of the first to seventh convolutional layers to 128, 64, 32, 32, 16, 16, and 8 in turn, and set the size of the convolutional kernels to 1×3; both the first and second pooling layers use the maximum pooling method, and the size of the pooling kernels is set to 1×2, and the pooling stride is set to 1×2; set the number of convolutional kernels of the first to fifth deconvolutional layers to 8, 8, 8, 16, and 8 in turn, and the size of the convolutional kernels is set to 1×3; the size of the pooling kernels of the first and second upsampling layers is set to 1×2, and the pooling stride is set to 1×2; Step 3, train the residual convolutional autoencoder: Input the training set into the residual convolutional autoencoder, and use the backpropagation gradient descent method to iteratively update the parameters of each layer of the residual convolutional autoencoder until the loss function of the residual convolutional autoencoder converges, and obtain the trained residual convolutional autoencoder; Step 4, evaluate the threat level: Adopt the same normalization method as in Step 1.3, input the samples after normalizing each sample whose threat level is to be evaluated into the trained residual convolutional autoencoder, and output the evaluation level of the threat level of this sample.

2. The electronic countermeasure target threat assessment method based on a residual convolutional autoencoder according to claim 1, characterized in that The 9 target characteristic parameters mentioned in Step 1.1 refer to target type, target maneuverability, target flight ability, target attack ability, target stealth ability, main function of target radar, maximum detection range of target radar, target radar resolution, and target radar anti-jamming ability.

3. The method for evaluating the threat of electronic countermeasure targets based on the residual convolutional autoencoder according to claim 1, characterized in that The threat assessment level of each sample target mentioned in Step 1.2 is calculated by the following formula: Among them, G i represents the threat assessment level of the i-th sample in the sample set. If 0 ≤ G i ≤ 0.3, the threat assessment level is "poor". If 0.3 < G i ≤ 0.7, the threat assessment level is "average". If 0.7 < G i ≤ 1, the threat assessment level is "very good". i represents the serial number of the sample in the sample set, i = 1, 2... N, and N represents the total number of samples in the sample set. u i1 represents the target type of the i-th sample in the sample set, u i2 represents the target maneuverability of the i-th sample in the sample set, u i3 represents the target flight ability of the i-th sample in the sample set, u i4 represents the target attack ability of the i-th sample in the sample set, u i5 represents the target stealth ability of the i-th sample in the sample set, u i6 represents the main function of the radar of the i-th sample in the sample set, u i7 represents the maximum detection range of the radar of the i-th sample in the sample set, u i8 represents the radar resolution of the i-th sample in the sample set, u i9 represents the anti-jamming ability of the radar of the i-th sample in the sample set.

4. The method for evaluating the threat of an electronic countermeasure target based on a residual convolutional autoencoder according to claim 1, wherein The threat assessment level label mentioned in Step 1.2 refers to: the label for the threat assessment level of "poor" is set to 0, the label for the threat assessment level of "general" is set to 1, and the label for the threat assessment level of "very good" is set to 2.

5. The method for evaluating the threat of an electronic countermeasure target based on a residual convolutional autoencoder according to claim 1, characterized in that The normalization operation mentioned in Step 1.3 is realized by the following formula: Among them, x ij ' represents the j-th target feature parameter of the i-th sample in the normalized sample set. j represents the serial number of the target feature parameter. j = 1 represents the target type in the sample set, j = 2 represents the target maneuverability in the sample set, j = 3 represents the target flight ability in the sample set, j = 4 represents the target attack ability in the sample set, j = 5 represents the target stealth ability in the sample set, j = 6 represents the main function of the target radar in the sample set, j = 7 represents the maximum detection range of the target radar in the sample set, j = 8 represents the resolution of the target radar in the sample set, j = 9 represents the anti-jamming ability of the target radar in the sample set, x ij represents the j-th target feature parameter in the i-th sample in the sample set, x jmax represents the maximum value of the j-th target feature parameter in the sample set, x jmin represents the minimum value of the j-th target feature parameter in the sample set.

6. The electronic countermeasure target threat assessment method based on a residual convolutional autoencoder according to claim 1, characterized in that The loss function of the residual convolutional autoencoder described in step 3 is as follows: Among them, Loss represents the loss function of the residual convolutional autoencoder, ∑· represents the summation operation, c = 0, 1..m, where m represents the total number of threat assessment levels, and y ic represents the relationship between the predicted value and the true value of the threat assessment level c of the i-th sample in the training dataset. When the true value is equal to the predicted value, then y ic = 1, and when the true value is not equal to the predicted value, y ic = 0, and p ic represents the predicted probability of the threat assessment level c of the i-th sample in the training dataset.

Images