Occupant performance real-time evaluation algorithm based on armored vehicle

Abstract

The invention discloses a passenger performance real-time evaluation algorithm based on an armored vehicle, and the algorithm comprises the steps: dividing a group of complete test data into continuous time periods according to a time window, calculating a feature index value in each time period, and taking the feature index value as feature data for evaluating the passenger cognitive performance at the time; determining an optimal time window of each characteristic index; a sliding time window is used for fusing original signals, a feature sequence is generated through data standardization, then parameter adjustment is carried out on an SVM model, a real-time evaluation model of the SVM is established, and indexes of a posterior probability and a confusion matrix are output; the cognitive performance features of the armored vehicle passengers are selected and combined through feature level fusion, and comprehensive state features based on eye movement, electrocardio and behavior time are obtained; and in the decision-making level fusion stage, three evidences corresponding to eye movement, electrocardio and behavior time are combined by using a DS evidence theory, decision making is performed on the combined evidences, and an evaluation result of the passenger cognitive performance level is obtained.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, analysis and machine learning, and in particular relates to a real-time evaluation algorithm based on armored vehicle crew performance. Background Art

[0002] Guided by the major needs of national security and national defense construction, it is particularly important to cultivate talents with comprehensive qualities such as professional skills such as driving skills, shooting ability, and the ability to use equipment, and who have good judgment, adaptability, and teamwork spirit, and who can complete combat missions in complex environments. However, due to the limitations of current evaluation methods, it is impossible to conduct real-time evaluation of a large number of training personnel, provide reliable and detailed performance feedback, and achieve the goal of accurately and effectively selecting excellent crew members. To this end, the use of real-time evaluation methods to assist relevant personnel in achieving a multi-faceted and comprehensive evaluation of training personnel can achieve high quality and high efficiency in the process of talent training, so as to quickly achieve the goal of improving combat capabilities.

[0003] The "information processing device, mobile device, method and program" disclosed in the application publication number "CN112041910A" adopts the analysis results of the driver's eyeball and / or pupil movement, combined with the driver-specific awakening state evaluation dictionary, to achieve the evaluation of the driver's awakening degree. The "vehicle safety assisted driving system, including the vehicle, method and medium thereof" disclosed in the application publication number "CN112991789A" collects vehicle and driver related information, evaluates the hesitation risk value, and issues a safety reminder when the value exceeds the preset threshold. However, the above schemes still have the following problems: these processing methods are not flexible enough in actual complex situations, resulting in insufficient accuracy and comprehensiveness of performance evaluation, a large gap with actual performance, and it is difficult to meet the precise requirements for occupant cognitive performance. Secondly, these two schemes often encounter problems of incomplete information or single evaluation information when evaluating occupant cognitive performance, because the environment, tasks and states in actual driving are unpredictable, which makes it difficult for the evaluation model to cope with complex and changeable situations, affecting the reliability and practicality of the evaluation results. In addition, different data sources may have different importance and credibility, and traditional methods may not be able to weigh various data well. Summary of the invention

[0004] In view of this, the present invention provides a real-time evaluation algorithm based on armored vehicle crew performance, which can address the problem of incomplete information during crew cognitive performance evaluation and achieve high fault tolerance of the real-time evaluation model. Thus, a real-time evaluation model for crew performance is established, aiming to provide evaluation information for relevant departments to make decisions.

[0005] The technical solution for implementing the present invention is as follows:

[0006] An algorithm based on real-time evaluation of armored vehicle crew performance includes the following stages:

[0007] S1. Pre-fusion pre-processing: a set of complete test data is divided into continuous time periods according to a certain time interval (i.e., time window), and the characteristic index value in each time period is calculated respectively as the characteristic data for evaluating the occupant's cognitive performance at that moment; since time windows of different lengths may affect the prediction performance of occupant cognitive performance, the optimal time window of each characteristic index needs to be determined before the characteristic data is fused;

[0008] S2, feature-level fusion: The original signals are fused using a sliding time window, and then a feature sequence is generated through data standardization. The SVM model is then adjusted, and finally a real-time evaluation model of the SVM is established and the indicators of the posterior probability and confusion matrix required to calculate the basic probability distribution function in the DS evidence theory are output;

[0009] S3. Decision-level fusion: The cognitive performance characteristics of armored vehicle crew members are selected and combined through feature-level fusion to obtain comprehensive state characteristics based on eye movements, electrocardiogram and behavior time. In the decision-level fusion stage, the DS evidence theory is used to combine the three pieces of evidence corresponding to eye movements, electrocardiogram and behavior time, and then a decision is made on the combined evidence to obtain the evaluation results of the crew's cognitive performance level.

[0010] Furthermore, the implementation steps include:

[0011] The following steps are involved:

[0012] Step 1: With the goal of maximizing the classification effect of the armored vehicle crew's cognitive performance, the optimal time window is selected for the extracted eye movement indicators, electrocardiogram indicators and time performance indicators as the input indicators of the real-time evaluation algorithm; a set of complete test data is used to calculate the characteristic indicator values ​​in each time period according to each input indicator, which are used as the characteristic data for evaluating the crew's cognitive performance at that moment;

[0013] Step 2: For the optimal time windows of different features, a sliding time window method is used to ensure multi-source fusion of different feature indicators at the same time;

[0014] Step 3: Scale the occupant cognitive performance characteristic indicators of different physical meanings and dimensional units to a small specific interval, remove the dimensions, and convert them into unitless values;

[0015] Step 4: Use the Gaussian kernel function to optimize the parameters with the accuracy of the SVM model as the optimization target, classify the characteristic indicators such as eye movement and behavior time according to the divided cognitive performance state, and divide them into training set and prediction set in a ratio of 3:1 for model training and prediction;

[0016] Step 5: According to the output results, obtain the recognition rate, rejection rate and accuracy index predicted by the model;

[0017] Step 6: By transforming the output posterior probability, the basic probability distribution of each piece of evidence required in DS evidence theory can be obtained;

[0018] Step 7: DS evidence theory integration.

[0019] Furthermore, the formula used in step 1 is specifically:

[0020]

[0021] Among them, the maximum optimization target is based on the F value in variance analysis, and the classification ability of the occupant's cognitive performance status increases with the increase of F value. A is the mean square error of indicators at different cognitive performance levels; MS E is the mean square error of the indicator at different cognitive performance levels.

[0022] Furthermore, the specific contents of step 4 are:

[0023] Step 4-1: Set the boundary of C to [0.01, 100] and the boundary of γ to [0.1, 10], take the model accuracy as the optimization target, use the exhaustive search algorithm for optimization, and finally determine the optimal parameters C = 0.1, γ = 1;

[0024] Step 4-2: Use the training set data to train an effective SVM model. The established SVM model includes an eye movement SVM evaluation model, an electrocardiogram SVM evaluation model, and a behavior time SVM evaluation model of the armored vehicle occupant's cognitive performance;

[0025] Step 4-3: Use the trained SVM model to predict the data in the prediction set and output the prediction results of the confusion matrix and posterior probability.

[0026] Furthermore, through the eye movement SVM model, electrocardiogram SVM model and behavioral time SVM model of the cognitive performance of armored vehicle occupants established in step 4, the posterior probability vector and confusion matrix of the prediction results can be output. By converting the output posterior probability, the basic probability distribution of each evidence required in the DS evidence theory can be obtained.

[0027] Beneficial effects:

[0028] 1) The present invention introduces the concept of time window and uses sliding time window to process the occupant cognitive performance data, so that the evaluation system has real-time and high fault tolerance. The value in variance analysis is innovatively used as the optimization target to determine the optimal time window of the feature, so as to improve the responsiveness of the evaluation model to different time periods.

[0029] (2) The present invention fuses the original signals through a sliding time window and performs data normalization. The innovation lies in improving the fusion effect of the evaluation model on multi-source data by reasonably selecting and processing the optimal time window with different characteristics.

[0030] (3) The present invention innovatively uses the DS evidence theory to perform decision-level fusion of multiple aspects such as eye movements, electrocardiograms, and behavioral time. In this process, the system comprehensively considers the weights of different evidences, improves the accuracy and credibility of the evaluation, and enables the system to better guide decision-making and actions. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is based on the real-time evaluation algorithm of armored vehicle crew performance;

[0032] Figure 2 is the F value of the lower eyelid opening in different time windows;

[0033] Figure 3 It is a schematic diagram of a sliding time window;

[0034] Figure 4 It is the sliding time window processing of different indicators;

[0035] Figure 5 It is SVM model training;

[0036] Figure 6 is the SVM model prediction. DETAILED DESCRIPTION

[0037] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0038] The present invention provides an armored vehicle crew performance real-time evaluation algorithm, comprising:

[0039] S1. Pre-processing before fusion. In order to make the characteristic indicators of real-time evaluation of the cognitive performance of armored vehicle crew members more accurately predict the level of their cognitive performance, the concept of time window is introduced, that is, a set of complete test data is divided into continuous time periods according to a certain "time interval", and the characteristic indicator values ​​in each time period are calculated respectively as characteristic data for evaluating the cognitive performance of the crew at that moment. Since time windows of different lengths may affect the predictive performance of the crew's cognitive performance, it is necessary to determine the "optimal" time window of each characteristic indicator before fusion of the characteristic data, such as Figure 1 As shown, the specific steps are as follows:

[0040] Step 1: Determine the "optimal" time window and characteristic indicators. With the goal of maximizing the classification effect of the armored vehicle crew's cognitive performance, the optimal time window is selected for the extracted eye movement indicators, electrocardiogram indicators and time performance indicators as the input indicators of the real-time evaluation algorithm. A set of complete test data is used to calculate the characteristic indicator values ​​in each time period according to each input indicator, which are used as the characteristic data for evaluating the crew's cognitive performance at that moment;

[0041] The formula used in step 1 is:

[0042]

[0043] Among them, the maximum optimization target is based on the F value in variance analysis, and the classification ability of the occupant's cognitive performance status increases with the increase of F value. A is the mean square error of indicators at different cognitive performance levels; MS E is the mean square error of the indicator at different cognitive performance levels.

[0044] Take the eyelid opening index as an example. Based on the actual situation, the time window size is 10s, the step size is 2s, and the F value of the difference analysis in each time window is calculated as follows: Figure 2 As shown in Figure 2, when the time window T = 10s, the F value is the largest, indicating that the classification ability of cognitive performance level is the strongest. Therefore, the optimal time window for determining the standard deviation of eyelid opening is T = 10s.

[0045] Based on the above method, the final determined real-time evaluation algorithm characteristic indicators and optimal time window are shown in Table 1.

[0046] Table 1 Optimal time window for characteristic indicators

[0047]

[0048] Step 2: Sliding time window processing. For the optimal time window of different features, the sliding time window method is used to ensure multi-source fusion of different feature indicators at the same time, such as Figure 3 As shown in the figure, take eyelid opening as an example. The optimal time window for eyelid opening is 10s, while the optimal time window for pupil diameter standard deviation is 12s. Since the time window lengths of the two indicators are different, it is necessary to perform a "sliding" process on the window, and take the sliding distance Δt of the sliding time window as 2s each time, as shown in the figure. Figure 4 As shown in the figure, it can be seen that the sliding time window can realize the fusion processing between indicators of different time windows at any moment.

[0049] Step 3: Data standardization. Scale the occupant cognitive performance characteristic indicators with different physical meanings and dimensional units to a small specific range, remove the dimensions, and convert them into unitless values;

[0050] The specific formula of the Z-score standardization method used is:

[0051]

[0052] Among them, x is some original data to be standardized; is the mean of the entire original data set; σ is the standard deviation of the entire original data set. The standardized data will be input into the classifier.

[0053] S2, feature-level fusion. The original signal is fused using a sliding time window, and then a feature sequence is generated through data standardization. The SVM model is then adjusted, and finally a real-time evaluation model of the SVM is established and outputs indicators such as the posterior probability and confusion matrix required to calculate the basic probability distribution function in the DS evidence theory. The specific steps are as follows:

[0054] Step 4: SVM model training and prediction. Using the Gaussian kernel function, the parameters are optimized with the SVM model accuracy as the optimization target. The characteristic indicators such as eye movement and behavior time are classified according to the cognitive performance state, and the training set and prediction set are divided into a ratio of 3:1 for model training and prediction.

[0055] Step 4-1: SVM parameter optimization. For the SVM model using the Gaussian kernel function, the penalty coefficient C and the Gaussian kernel parameter γ are crucial to the recognition accuracy and generalization ability of the SVM algorithm. The larger the C, the lower the error tolerance, and the more prone to overfitting; the larger the γ value, the better the linear separability, and the lower the generalization ability of the model. Set the boundaries of C to [0.01, 100] and γ to [0.1, 10]. Take the model accuracy as the optimization goal, use the exhaustive search algorithm for optimization, and finally determine the optimal parameters C = 0.1, γ = 1.

[0056] Step 4-2: SVM model training. The training of the SVM model is to use the training set data to train an effective SVM model. The implementation process is as follows: Figure 5 The established SVM models include the eye movement SVM evaluation model, the electrocardiogram SVM evaluation model and the behavior time SVM evaluation model of the armored vehicle occupant's cognitive performance;

[0057] Step 4-3: SVM model prediction. SVM model prediction is to use the trained SVM model to predict the data of the prediction set and output the prediction results such as confusion matrix and posterior probability. The implementation process is as follows: Figure 6 shown.

[0058] Step 5: SVM model output results. Based on the output results, the recognition rate, rejection rate, accuracy rate and other indicators predicted by the model can be calculated;

[0059] Table 2 Confusion matrix

[0060]

[0061] Table 2 shows the confusion matrix, which is a four-dimensional matrix that evaluates the classification performance of the model. The prediction results are divided into two categories: one is the occupant cognitive performance state predicted by the model; the other is the state evaluation confusion matrix and posterior probability used to calculate the probability distribution function in the DS evidence theory.

[0062] Taking the high cognitive performance state as an example, the above table is converted into a two-dimensional matrix as shown in Table 3. In general, the confusion matrix shows that there are two situations in which the prediction is correct: one is that the high cognitive performance state is predicted and the actual state is also high cognitive performance state, which we call true positive (TP); the other is that the non-high cognitive performance state is predicted and the actual state is also non-high cognitive performance state, which is called true negative (TN). There are also two situations in which the prediction error is: one is false positive (FP), which means that the prediction is high cognitive performance state, but the actual state is not high cognitive performance state; the other is false negative (FN), which means that the prediction is non-high cognitive performance state but the actual state is high cognitive performance state.

[0063] Table 3 Conversion confusion matrix

[0064]

[0065] Where TP = a 1 , FN=b 1 +c 1 +d 1 , FP = a 2 +a 3 +a 4 ,

[0066] From a vertical perspective, the confusion matrix reflects the number of high cognitive performance states and non-high cognitive performance states in the actual situation. In Table 3, the actual number of high cognitive performance states is TP+FN, and the number of non-high cognitive performance states is FP+TN; similarly, from a horizontal perspective, the confusion matrix reflects the number of high cognitive performance states predicted by the model as TP+FP, and the number of non-high cognitive performance states predicted as FN+TN.

[0067] (1) Accuracy

[0068] The model prediction accuracy is defined as the proportion of the model predictions that are correct to the total:

[0069]

[0070] (2) Sensitivity

[0071] The model recognition rate is defined as the proportion of states predicted to be high cognitive performance among states marked as high cognitive performance:

[0072]

[0073] (3) Specificity

[0074] The model rejection rate is defined as the proportion of states that are predicted to be non-high cognitive performance among states that are labeled as non-high cognitive performance:

[0075]

[0076] S3, decision-level fusion. Through feature-level fusion, the cognitive performance characteristics of armored vehicle crews are selected and combined to obtain comprehensive state characteristics based on eye movement, electrocardiogram and behavior time. In the decision-level fusion stage, the DS evidence theory is used to merge the three pieces of evidence corresponding to eye movement, electrocardiogram and behavior time, and then a decision is made on the merged evidence to obtain the evaluation results of the crew's cognitive performance level. The specific steps are as follows:

[0077] Step 6: Calculation of basic probability distribution. By converting the output posterior probability, the basic probability distribution of each piece of evidence required in DS evidence theory can be obtained;

[0078] Through the eye movement SVM model, electrocardiogram SVM model and behavior-time SVM model of the armored vehicle occupant's cognitive performance established in step 5, the posterior probability vector and confusion matrix of the prediction results can be output. By converting the output posterior probability, the basic probability distribution of each piece of evidence required in the DS evidence theory can be obtained.

[0079] Taking the high cognitive performance state as an example, suppose the posterior probability vector predicted by the established i-th SVM model for a certain test data is V i =(a i1 ,a i2 ), where i=1, 2, 3, 4 corresponds to the i-th SVM model, a i1 represents the probability value of predicting high cognitive performance, a i2 represents the probability value of predicting non-high cognitive performance, and the sum of the two is 1. Let the confusion matrix vector of the i-th SVM model be Q i =(q i1 ,q i2 ), q i1 Represents the recognition rate of the SVM model, namely Sensitivity, q i2 It represents the rejection rate of the SVM model, namely Specificity. The formula for converting the posterior probability and confusion matrix of the above SVM model into the BPA function in DS evidence theory is as follows:

[0080]

[0081] In the formula, m i (A) is the BPA function of each piece of evidence; m i (A 1 ) is the BPA function value of high cognitive performance; m i (A 2 ) is the BPA function value of non-high cognitive performance; m i (Θ) is the value of the BPA function with uncertainty.

[0082] When the basic probability M-degrees of the three evidences of eye movement, electrocardiogram and performance are determined, the evidences can be merged according to the DS evidence merging formula 13 to obtain the probability M-degrees of different cognitive performance levels after fusion.

[0083] Step 7: DS evidence theory integration;

[0084] Step 7-1: Identification framework. First, you need to set the corresponding identification framework according to the actual application scenario. The current application scenario is the level of cognitive performance of armored vehicle occupants. The corresponding identification framework can be expressed in the following form:

[0085] Θ={H(high cognitive performance), N(medium cognitive performance), L(low cognitive performance), F(cognitive failure)} (8)

[0086] Step 7-2: Basic probability allocation. Assume 2 Θ The set of all subsets of Θ is also called the power set of Θ:

[0087] 2 Θ ={Φ,{H},{N},{L},{F},{H,N},{H,L},{H,F},{N,L},{N,F},{L,F},{H,N,L},{H,N,F},{H,L,F},{N,L,F},Θ} (9)

[0088] 2 Θ Each subset in corresponds to a basic probability M, which is defined as:

[0089]

[0090] According to the basic probability M, for any Calculate the trust function Bel(X) and the likelihood function Pl(X) respectively to form the trust interval [Bel(X), Pl(X)], where:

[0091]

[0092]

[0093] [Bel(X), Pl(X)] represents the probability range of an event X, and Pl(X)-Bel(X) is the uncertainty of event X.

[0094] Step 7-3: Evidence merging. DS evidence theory merges two or more pieces of evidence into one piece through the Dempster rule and uses this as the main basis for decision making. Assume that m 1 ,m 2 ,m 3 ,m 4 is the basic probability M degree (BPA) of different evidences in the recognition framework Θ. In this study, it corresponds to the BPA values ​​of eye movement, electrocardiogram and time performance respectively. The specific merging rules between the evidences are shown in the following formula:

[0095]

[0096]

[0097]

[0098] After the evidence is merged, a decision can be made on the occupant's cognitive performance level based on the BPA function of the merged evidence and certain decision rules.

[0099] Step 7-4: Fusion decision. When the decision layer is integrated, the commonly used decision rules are the maximum support (likelihood) rule, the absolute support rule, the minimum risk rule, etc. The maximum support rule is based on the likelihood function as the decision basis, and the hypothesis corresponding to the maximum likelihood function is used as the decision result; the absolute support rule means that when the confidence function corresponding to a hypothesis is the largest and its confidence interval is less than a certain threshold, the hypothesis is selected as the decision result; the minimum risk rule considers the risks that different decisions may bring, and selects the decision with the lowest corresponding risk as the final decision result.

[0100] Table 4 Classification of real-time evaluation situations

[0101]

[0102] Among them, during the actual operation of the real-time evaluation algorithm for the cognitive performance of armored vehicle crews, various situations may occur as shown in Table 4. When the judgment result of the algorithm is consistent with the actual situation of the crew, it is a correct judgment of the algorithm; when the crew is actually in a certain cognitive performance state, and the cognitive performance judged by the algorithm is low, it is regarded as a false alarm of the algorithm; when the cognitive performance of the crew is actually at a certain level, and the level judged by the algorithm is high, it is regarded as a false alarm of the algorithm.

[0103] Obviously, when the occupants are at a certain cognitive performance level, and the cognitive performance judged by the algorithm is low, which triggers early warning intervention, it will cause greater interference to the occupants and increase their tension; on the other hand, when the occupants are at a certain cognitive performance level, and the cognitive performance judged by the algorithm is high and no early warning intervention is performed, it may cause a major safety accident. It can be seen that for different situations, the evaluation results of the algorithm correspond to different decision risks. In order to fully consider the risks brought by false detection in the algorithm decision-making stage and ensure the safety of the occupants as much as possible, the minimum risk rule is finally adopted as the final decision rule.

[0104] According to the application scenario requirements, define the concept of risk function that characterizes the corresponding risks of various decisions in the evaluation algorithm, and assign certain risk factors to different decision situations. Finally, choose the decision with the smallest corresponding risk function as the decision result:

[0105]

[0106] Where R(k) is the risk function corresponding to decision k; ik is the risk factor; m(i) is the probability distribution of different cognitive performance levels.

[0107] The setting of risk factors needs to comprehensively consider the impact of the algorithm's missed alarms on the safety of the crew's mission execution process, as well as the interference caused by the algorithm's false alarms on the crew's normal operations. When setting, first set the risk factors of false alarms and missed alarms to 1, and then adjust them according to actual needs. If you start from the perspective of reducing the impact of the algorithm's missed alarms on the safety of the crew, you can do so by increasing the corresponding missed alarm risk factor; if you start from the perspective of reducing the interference of the algorithm's false alarms on the crew, you can do so by increasing the corresponding false alarm risk factor.

[0108] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A real-time evaluation algorithm based on armored vehicle crew performance, characterized in that: The following phases are included: S1. Pre-fusion pre-processing: a set of complete test data is divided into continuous time periods according to the set time intervals, and the characteristic index values ​​in each time period are calculated respectively as the characteristic data for evaluating the occupant's cognitive performance at that moment; since time windows of different lengths may affect the prediction performance of occupant cognitive performance, the optimal time window of each characteristic index needs to be determined before the characteristic data is fused; S2, feature-level fusion: The original signals are fused using a sliding time window, and then a feature sequence is generated through data standardization. The SVM model is then adjusted, and finally a real-time evaluation model of the SVM is established and the indicators of the posterior probability and confusion matrix required to calculate the basic probability distribution function in the DS evidence theory are output; S3. Decision-level fusion: The cognitive performance characteristics of armored vehicle crew members are selected and combined through feature-level fusion to obtain comprehensive state characteristics based on eye movements, electrocardiogram and behavior time. In the decision-level fusion stage, the DS evidence theory is used to combine the three pieces of evidence corresponding to eye movements, electrocardiogram and behavior time, and then a decision is made on the combined evidence to obtain the evaluation results of the crew's cognitive performance level.

2. The real-time evaluation algorithm based on armored vehicle crew performance according to claim 1 is characterized in that: The specific implementation steps include the following: Step 1: With the goal of maximizing the classification effect of the armored vehicle crew's cognitive performance, the optimal time window is selected for the extracted eye movement indicators, electrocardiogram indicators and time performance indicators as the input indicators of the real-time evaluation algorithm; a set of complete test data is used to calculate the characteristic indicator values ​​in each time period according to each input indicator, which are used as the characteristic data for evaluating the crew's cognitive performance at that moment; Step 2: For the optimal time windows of different features, a sliding time window method is used to ensure multi-source fusion of different feature indicators at the same time; Step 3: Scale the occupant cognitive performance characteristic indicators of different physical meanings and dimensional units to a small specific interval, remove the dimensions, and convert them into unitless values; Step 4: Use the Gaussian kernel function to optimize the parameters with the accuracy of the SVM model as the optimization target, classify the characteristic indicators such as eye movement and behavior time according to the divided cognitive performance state, and divide them into training set and prediction set in a ratio of 3:1 for model training and prediction; Step 5: According to the output results, obtain the recognition rate, rejection rate and accuracy index predicted by the model; Step 6: By transforming the output posterior probability, the basic probability distribution of each piece of evidence required in DS evidence theory can be obtained; Step 7: DS evidence theory integration.

3. The real-time evaluation algorithm based on armored vehicle crew performance as claimed in claim 2 is characterized in that: The formula used in step 1 is: Among them, the maximum optimization target is based on the F value in variance analysis, and the classification ability of the occupant's cognitive performance status increases with the increase of the F value; MS A is the mean square error of indicators at different cognitive performance levels; MS E is the mean square error of the indicator at different cognitive performance levels.

4. The real-time evaluation algorithm based on armored vehicle crew performance as claimed in claim 2 is characterized in that: The specific contents of step 4 are: Step 4-1: Set the boundary of C to [0.01, 100] and the boundary of γ to [0.1, 10], take the model accuracy as the optimization target, use the exhaustive search algorithm for optimization, and finally determine the optimal parameters C = 0.1, γ = 1; Step 4-2: Use the training set data to train an effective SVM model. The established SVM model includes an eye movement SVM evaluation model, an electrocardiogram SVM evaluation model, and a behavior time SVM evaluation model of the armored vehicle occupant's cognitive performance; Step 4-3: Use the trained SVM model to predict the data in the prediction set and output the prediction results of the confusion matrix and posterior probability.

5. The real-time evaluation algorithm based on armored vehicle crew performance according to claim 2 or 4, characterized in that: Through the eye movement SVM model, electrocardiogram SVM model and behavior time SVM model of the cognitive performance of armored vehicle occupants established in step 4, the posterior probability vector and confusion matrix of the prediction results can be output. By converting the output posterior probability, the basic probability distribution of each evidence required in the DS evidence theory can be obtained.

Images