CN112906746B - Multi-source track fusion evaluation method based on structural equation model - Google Patents
Multi-source track fusion evaluation method based on structural equation model Download PDFInfo
- Publication number
- CN112906746B CN112906746B CN202110092959.2A CN202110092959A CN112906746B CN 112906746 B CN112906746 B CN 112906746B CN 202110092959 A CN202110092959 A CN 202110092959A CN 112906746 B CN112906746 B CN 112906746B
- Authority
- CN
- China
- Prior art keywords
- fusion
- track
- data
- index
- sensor
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000004927 fusion Effects 0.000 title claims abstract description 147
- 238000011156 evaluation Methods 0.000 title claims abstract description 59
- 238000004364 calculation method Methods 0.000 claims abstract description 18
- 230000006872 improvement Effects 0.000 claims abstract description 3
- 238000000034 method Methods 0.000 claims description 30
- 230000015572 biosynthetic process Effects 0.000 claims description 13
- 238000012360 testing method Methods 0.000 claims description 10
- 238000010586 diagram Methods 0.000 claims description 9
- 238000012937 correction Methods 0.000 claims description 8
- 238000000926 separation method Methods 0.000 claims description 8
- 230000000694 effects Effects 0.000 claims description 7
- 230000010365 information processing Effects 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- 230000002159 abnormal effect Effects 0.000 claims description 6
- 230000000875 corresponding effect Effects 0.000 claims description 6
- 230000036314 physical performance Effects 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 238000007499 fusion processing Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 3
- 230000002596 correlated effect Effects 0.000 claims description 2
- 230000009191 jumping Effects 0.000 claims description 2
- 230000035899 viability Effects 0.000 claims description 2
- 238000013467 fragmentation Methods 0.000 claims 1
- 238000006062 fragmentation reaction Methods 0.000 claims 1
- 230000004083 survival effect Effects 0.000 claims 1
- 238000004458 analytical method Methods 0.000 abstract description 8
- 238000007689 inspection Methods 0.000 abstract description 3
- 238000013461 design Methods 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000005259 measurement Methods 0.000 description 4
- 238000013210 evaluation model Methods 0.000 description 3
- 238000012854 evaluation process Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 238000004445 quantitative analysis Methods 0.000 description 3
- 230000006978 adaptation Effects 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 238000000556 factor analysis Methods 0.000 description 2
- 239000010410 layer Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000000704 physical effect Effects 0.000 description 2
- 238000004451 qualitative analysis Methods 0.000 description 2
- 238000012827 research and development Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 238000007619 statistical method Methods 0.000 description 2
- 230000009897 systematic effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 239000002346 layers by function Substances 0.000 description 1
- 238000011158 quantitative evaluation Methods 0.000 description 1
- 238000012916 structural analysis Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/20—Instruments for performing navigational calculations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/251—Fusion techniques of input or preprocessed data
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Remote Sensing (AREA)
- Radar, Positioning & Navigation (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Automation & Control Theory (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a multisource track fusion evaluation method based on a structural equation model, which comprises the steps of providing a scientific, general and comprehensive evaluation system and a quantitative calculation model, wherein the evaluation system is divided into two layers, primary indexes are used as factors for finally describing the advantages and disadvantages of the track fusion system, the secondary indexes are quantitatively calculated by utilizing a formula, and the primary indexes are quantized by the factor load of the structural equation model. And determining the availability of an index model through fitting degree indexes and significance inspection by adopting SEM, determining the importance of each observation variable by utilizing standardized analysis to obtain the standardized factor load quantity of each observation variable, and utilizing non-standardized analysis to obtain the weight and the general assessment model of each index. The invention can improve the universality of indexes, enhance the objectivity of each observation variable to the interpretation of the latent variables, eliminate the problem of collinearity among the interpretation variables, and provide data-based improvement opinion for a multi-source information fusion system.
Description
Technical Field
The invention belongs to the technical field of multi-source information fusion, and particularly relates to a multi-source track fusion evaluation method for calculating index weights according to a structural equation model.
Background
The performance and the function of the multi-source information data fusion system can be quantitatively estimated, so that the effectiveness of the multi-source information data fusion system can be determined. However, in general, the information collected and processed by the fusion system is various, and the relationship between the information sources is complex, so that great difficulty is brought to objective and fair testing and evaluating of the fusion system. At present, the information fusion evaluation index system is unclear in hierarchy, the system is difficult to guide to improve, most of the existing information fusion evaluation indexes have the problems of contradiction, index redundancy, poor generality, difficulty in calculation and the like, a generalized and effective fusion model, algorithm and a comprehensive evaluation system are not formed at present, a qualitative analysis method is mainly adopted, subjectivity is high, and therefore a scientific quantitative evaluation method needs to be explored to accurately judge the effectiveness of the fusion system. Therefore, the effective fusion evaluation system is designed to have very important practical significance, the performance of the information fusion system is evaluated in advance through the fusion evaluation system, partial problems possibly occurring are eliminated, the research and development speed of the information fusion system is further improved, and meanwhile, the research and development cost can be reduced.
When the performance of the information fusion system is evaluated, a proper information fusion evaluation method is required to be selected according to a performance evaluation index system of the system, so that evaluation work is expanded, and the establishment mode of the performance evaluation index system has important significance for system performance evaluation. Meanwhile, the evaluation result of the information fusion system can be fed back, and parameters and design modes of the information fusion related algorithm can be further corrected, so that the performance of the information fusion system is optimized. How to design a reasonable and effective information fusion system evaluation method is still one of the difficult problems to be solved.
The core of the information fusion evaluation is a scene generation algorithm and a fusion evaluation index system establishment method. From the existing information fusion evaluation scheme, as the information fusion system is a multi-information processing center and multi-sensor system, and the hierarchical structure is complex, the application scene is diversified, no scientific and reasonable evaluation index system design standard is available yet, and the current evaluation method is often used for evaluating index body design by depending on a specific data set.
Disclosure of Invention
The invention provides a scientific and universal evaluation index system, which is based on a analytic hierarchy process and a fuzzy comprehensive evaluation process, adopts a SEM (Structural Equation Modeling) model to obtain the weight of each observation variable in the model and the effect value among each potential variable through factor analysis and path coefficient inspection, and obtains a final model structure through continuous fitting, thereby improving the utilization rate of data and enhancing the accuracy and the universality of an evaluation result.
A multi-source multi-data track fusion system evaluation method based on a structural equation model comprises the following steps:
s1, a scientific, general and comprehensive evaluation system is provided, the evaluation system is divided into two index levels, the first-level index is used as a factor for finally describing the advantages and disadvantages of the track fusion system, the second-level index is calculated by a systematic formula, and the first-level index is quantized by the factor load of a structural equation model.
S2, providing each index meaning and a calculation formula.
And S3, determining the latent variable, the observation variable and the interrelation among the variables according to the index level in the step S1, and constructing an initial model based on SEM.
S4, collecting data, calculating according to the formula in the step S2, importing a calculation result into amos software, performing calculate estimates operation to obtain a fitness index, and performing model fitness and significance inspection.
When the fitting degree index and the significance obtained in the step S4 meet the requirements, carrying out standardized estimation to obtain a standardized path coefficient and a factor load; and if the requirements are not met, carrying out model correction according to the meaning of each fitness index and the non-standardized coefficient value, and repeating the step S4 until a structural equation model with good performance is obtained.
And (3) substituting the obtained result of the calculation of the example data by the formula in the step (2) into the final path diagram obtained in the step (S5), so as to obtain the evaluation result of the track fusion, and providing an improved measure for the track fusion process.
The multi-source track fusion evaluation method based on the structural equation model is characterized in that index extraction is carried out according to general principles of academic evaluation on information fusion performance, including comprehensive principles, objectivity principles, independence principles, testability principles, sensitivity principles, generality principles and environment dependence principles. In addition, the following specific criteria are set for the track fusion system: firstly, in order to keep the consistency of the track and the target true value, the track correlation accuracy reaches a certain requirement; secondly, compared with a true value, the fused target state estimation shows whether the track precision of the system can meet the requirement; thirdly, whether the identification accuracy of the target attribute, the category, the model and the like meets the design requirement; fourth, the evaluation index need only include factors that can reflect the most essential, important, and representative of the system functions, and need not include all the indexes of the common sensor.
The multi-source track fusion evaluation method based on the structural equation model is characterized in that index layers are divided into two stages, wherein the first-stage index comprises physical performance, time efficiency, information processing precision, data updating capability, track quality and target recognition capability. The secondary indexes comprise storage capacity, sensor precision, sensor stability, sensor identification rate, fusion time delay, sensor track formation time, system track formation time, sensor reporting update rate, target fusion update rate, system issuing indication update rate, target fusion position precision, target fusion speed precision, target fusion heading precision, track fragility, tracking continuity, track survivability, correct fusion rate, correct separation rate, missed association rate and error association rate.
The multi-source track fusion evaluation method based on the structural equation model is characterized in that a latent variable is a first-level index in the method, namely a variable which cannot be directly measured, but can be represented by other measurable variables, and an observed variable is a second-level variable in the method, namely a variable which can be obtained through measurement. And drawing an initial model according to the index level.
The invention relates to a multisource track fusion evaluation method based on a structural equation model, wherein original data is divided into normal data and abnormal data, the abnormal data is divided into four types of outlier data, unreasonable data, data with larger jumping and missing data, the first three types of abnormal data are processed by adopting a 3 alpha data elimination method and a robust regression method, and vacant data are processed by adopting a data interpolation method. The invention adopts eight fitness indexes of CMIN/DF, CFI, RMSEA, GFI, NFI, RMR, TLI, AGFI to carry out fitness test, and adopts four indexes of Estimate, S.E., C.R., and P to carry out significance test.
A multisource track fusion evaluation method based on a structural equation model, wherein model fitting degree describes the consistency degree of a theoretical model and actual data, and the magnitude of factor load reflects whether the relation between variables is obvious or not. In the course of the saliency check, when the P value of the coefficient >0.05, it is shown that the effect between the two variables of the path is insignificant. In this case, in order to improve the fitting degree of the model, the insignificant path may be deleted.
The invention has the following beneficial effects: according to the invention, the level index is set, quantitative calculation is carried out on the index, iterative analysis is carried out on input data by utilizing the structural equation model, insignificant variables and paths are deleted, and finally, the evaluation results of the influence degree and positive and negative influence relation of six aspects of physical performance, time efficiency, information processing precision, data updating capability, trace quality and target recognition capability on the track fusion result are obtained, so that quantitative, systematic and low-cost evaluation is realized, and advice is provided for improving the track fusion quality.
Drawings
FIG. 1 is a schematic diagram of an evaluation flow of a multi-source multi-data track fusion system evaluation method based on a structural equation model.
FIG. 2 is an index level diagram of a multi-source multi-data track fusion system evaluation method based on a structural equation model.
FIG. 3 is a schematic diagram of a structural equation model of a multi-source multi-data track fusion system evaluation method based on the structural equation model.
Fig. 4 is a structural equation model standardization result diagram of a multi-source multi-data track fusion system evaluation method based on a structural equation model.
FIG. 5 is a structural equation model non-standardized result diagram of a multi-source multi-data track fusion system evaluation method based on a structural equation model.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
1. Track fusion evaluation system
As shown in fig. 1, the evaluation process is divided into two line flow lines, one is to propose an assumed evaluation model, and fit through input data, and a final evaluation model can be obtained if the fitting requirement (generally indicated by chi-square P in AMOS and generally requiring chi-square value P > 0.05) is met, and if the fitting degree is not good, the model needs to be corrected until the fitting degree meets the requirement. And the other is to perform instance analysis after obtaining a reliable evaluation model, so as to obtain an evaluation result of the instance.
2. Hypothesis assessment model
The invention uses a structural equation model (Structural Equation Model, SEM for short) to analyze the relation between variables based on covariance matrix of the variables, the structural equation model analysis is also called covariance structural analysis, the problem is treated by applying multivariate statistical analysis, two statistical methods of factor analysis and path analysis are sequentially used for carrying out integration statistics, then the relation between the apparent variable, the latent variable and the error variable in the model is checked, and finally the direct effect, the indirect effect and the total effect of the influence of the independent variable on the dependent variable are obtained.
The method utilizes a verification analysis method of a structural equation model, under the support of theory or rule of thumb, under the quantitative calculation of a summarized index formula, firstly, a hypothesized model diagram is constructed according to theory and quantitative analysis, then, the fitting degree of the model is checked, the usability of the model is observed, and meanwhile, whether each path is obvious or not is checked, so that whether the influence of independent variables on dependent variables is obvious or not is determined. According to the method, an index system and an index calculation model are obtained through a qualitative and quantitative analysis combined method, an initial structural equation model is built through AMOS software, and a final model which is well fitted with input data is obtained through continuous fitting.
3. Index selection principle
The multi-sensor information fusion system completes specific tasks through the functions of certain sensors in the system, has multiple functional layers and complex structure, and therefore, the index of the fusion system is a multi-layer index system structure. The performance evaluation is carried out on the flow and algorithm of the multi-source information data fusion, probability statistics is firstly carried out on the track related result, the target state estimation and the attribute identification result, and then a group of performance indexes with scientificity, rationality, completeness and testability are extracted.
The index extraction is carried out according to general principles of academic circles on information fusion performance evaluation, including comprehensive principles, objectivity principles, independence principles, testability principles, sensitivity principles, generality principles and environment dependence principles. On the basis, the following specific standards are set for the track fusion system: firstly, in order to keep the consistency of the track and the target true value, the track correlation accuracy reaches a certain requirement; secondly, compared with a true value, the fused target state estimation shows whether the track precision of the system can meet the requirement; thirdly, whether the identification accuracy of the target attribute, the category, the model and the like meets the design requirement; fourth, the evaluation index need only include factors that can reflect the most essential, important, and representative of the system functions, and need not include all the indexes of the common sensor.
4. Determining an index level
By learning the working process of the track fusion system and the index system summarized by the former, six primary indexes and twenty-one secondary indexes are selected in total to jointly evaluate the performance of the track fusion system, as shown in fig. 2.
In the track fusion index system, the first-level index selected by the invention comprises physical performance, time efficiency, information processing precision, data updating capability, trace point quality and target recognition capability, the six variables are called latent variables in a structural equation model, the latent variables cannot be directly measured or calculated, but can be represented by other measurable variables, and the variables represented by ellipses in an AMOS are the latent variables.
In a track fusion index system, the selected secondary index comprises storage capacity, sensor precision, sensor stability, sensor identification rate, fusion time delay, sensor track formation time, system track formation time, sensor reporting update rate, target fusion update rate, system issuing indication update rate, target fusion position precision, target fusion speed precision, target fusion heading precision, track fragility, tracking continuity, track survivability, correct fusion rate, correct separation rate, miss association rate and error association rate, wherein twenty variables are called as apparent variables in a structural equation model, also called as observation variables, can be obtained through measurement and formula calculation, and the apparent variables in Amos are represented by rectangles.
In addition, the error variable is an indispensable factor for constructing the structural equation model. The error variable is also a variable that cannot be actually measured. In the invention, the latent variables cannot be interpreted hundred percent, errors always exist, the errors are reflected in the structural equation model to be the error variables, and each of the latent variables has the error variable. In Amos the error variables are represented using circles.
5. Quantitative analysis
5.1 physical Properties of the System
Storage capacity: the amount of data stored by the fusion system.
Sensor accuracy: and the similarity degree between parameters such as the number of targets, the target speed, the position and the like observed by the sensor and the target true value. Let the observed number of targets, the velocity estimation vector and the position estimation vector at the target time t be respectively: n (N) t ,V t =[V x ,,V y ,V z ] T ,L t =[L x ,,L y ,L z ] T . The real number of targets, the speed estimation vector and the position estimation vector at the same moment are respectively as follows: n, v t =[v x ,,v y ,v z ] T ,l t =[l x ,,l y ,l z ] T . The estimation errors of the three are respectively as follows: n=n t -N, The estimation accuracy is respectively as follows: />
Sensor identification rate: the sensor accurately identifies the number N of the data of the target batch number i Accounting for the total number N of the observed data of the sensor e Is referred to as the sensor recognition rate.
Sensor stability: for both the data volume and the time of observation, the sensor will have a difference for each observation. The larger the difference value resulting from the first observation comparison, the poorer the stability of the multi-sensor system. Let the data quantity observed by the sensor for the first time be N 1 In different time periods t, the data quantity observed by the sensor is N a Sensor stability
System stability: the sensor observation data is uploaded to the fusion system, the capacity of the system for processing the data in each period of time is different, and the larger the difference of the processed data amount in unit time is compared with the first processed data amount, the worse the stability is. The calculation formula references sensor stability.
5.2 time Performance
Fusion time delay: the single piece of data is input into the fusion system and passes through the time spent in the whole process of the output after the association and fusion processing. The shorter the time, i.e. the smaller the fusion time delay, the higher the system efficiency and the better the system performance. Let the total time spent by n pieces of data entering the fusion system for association and fusion operation be T, then fusing the time delay T m =t/n。
Sensor track formation time: the average time required for the data observed by the multi-sensor system to form a complete trackThe shorter the time, the higher the system efficiency and the better the performance. Set at t 0 To t n The number of complete tracks observed by the sensor is N in a time period t Sensor track formation time +.>
System track formation time:the data of a plurality of track points reported by the sensor system are processed by the fusion system to form the average time required by a complete trackReferred to as the system trace formation time. Set at t 0 To t n In the time period, the number of the complete tracks formed by the data output by the fusion system is N s Then fuse the system track formation time +.>
5.3 data update Performance
The sensor reports the update rate: the single sensor reports the total amount of target information to the fusion system in a unit time. If sensor a is at t 1 To t n The total information of each target reported in the time period is m 1 The sensor reporting update rate is r 1 =m 1 /(t n -t 1 ). The number of times that each sensor reports the situation information (fused or unfused) of the same area observed by the sensor of the fusion center within unit time(s) is called the reporting update rate of the sensor. In a certain evaluation, reporting the batch number p of targets to the fusion center by the sensor a for n times, wherein the time of the first reporting is t 1 The last reporting time is t 2 The update rate of reporting the target p by the sensor a is R 1 =(n-1)/(t 2 -t 1 ) If the update rates of all m targets reported by the sensor a are r respectively 1 ,r 2 ,...,r m The average reporting update rate of the sensor a isWherein the maximum and minimum reporting update rates are r respectively max =max(R 1 ,R 2 ,...,R m ),r min =min(R 1 ,R 2 ,...,R m )。
Target fusion update rate: the number of times the same trajectory data is processed by the fusion system per unit time is referred to as the target fusion update rate. And reporting the update rate of the calculation model and the sensor.
System down-set indicates update rate: the system down-sending indication update rate refers to the update times that the fusion system sends fusion data to the next stage platform in unit time and is successfully received. And reporting the update rate of the calculation model and the sensor.
5.4 data processing precision
The data processing precision is used for evaluating the precision degree of the fusion system on each parameter of the target parameters, and comprises four aspects of target fusion position precision, speed precision, heading precision and data survivability.
Target fusion position accuracy: the sensor reports the accuracy degree of the position of the target parameter in the situation data of the fusion system, and the target position is expressed by longitude and latitude. Set in one evaluation, sensor a reports to fusion center the target with lot number p at t i The position parameters of the moment are: longitude: longitude_i, latitude: the actual target lot number corresponding to the tile_i is P, which is at t i The real position parameters of the moment are: longitude: longitude_i, latitude: latitude_i. Then the target reported by sensor a as lot p is at t i The difference between the time position parameter and the real position parameter is: longitude difference: diff_lo = longitude_i-longitude_i, difference in altitude: diff_la=latency_i-latency_i. The accuracy of the target position with the batch number p reported by the sensor a can be obtained by a standard deviation formula
Target fusion speed accuracy: and measuring an index of the fusion estimation precision of the system to the target speed. And comparing the fusion speed of the target parameter at a certain moment with the target real speed, and calculating the mean value and standard deviation of the speed deviation, thereby estimating the speed precision of the fusion system. The smaller the mean and standard deviation, the higher the accuracy. Let the speed of a point j on the track with the target lot number i output by the fusion system beTrue speed of targetIs (v) x ,v y ,v z ) The speed deviation of the point i on the track a in the directions of x, y and z is +.> Performing point estimation calculation on the speed deviation to obtain a speed deviation mean value and a variance index of the target, wherein the mean value is:/degree>Standard deviation: />
Target fusion heading accuracy: the index for measuring the fusion estimation precision of the system to the target course is called the target fusion course precision. And comparing the course index fused by the target parameters at a certain moment with the real course of the target, and calculating the mean value and standard deviation of the course deviation, thereby estimating the course precision of the fusion system. The smaller the mean and standard deviation, the higher the accuracy. Is provided withDirection angle of point j on track with target lot number p output by fusion system, +.>For the true heading value of point j, the heading deviation is +.>Performing point estimation calculation on the speed course deviation to obtain a course deviation mean value and a variance index of the target, wherein the mean value is: />Standard deviation: />
5.5 quality of stippling
The index of the track quality is used for evaluating the accuracy degree of the track generated in the identifying and fusing process, and comprises three aspects of track fragility, tracking continuity and track survivability.
Track friability: and in unit time, outputting the number of tracks of one real track after the operation of the fusion system. The same target still keeps a stable batch number after being correlated and fused by the system under an ideal state, and the higher the track fragility is, the more batch numbers are split after the observation and correlation fusion of the same target are described, and the worse the system accuracy is. Setting the real track number reported by the sensor as N in different time periods t b The number of tracks output by the fusion system is N d Track friability
Tracking continuity: when a multi-sensor system tracks a target, a gap exists between the observation ranges of two sensors, or the fusion system does not successfully identify the target, so that the track of the target is incomplete. The fewer track point defects formed by the targets observed by the sensor system and processed by the fusion system, the better tracking continuity is indicated. Let the observation radius of the sensor be r, the total area of sea area be S, and the total area of the overlapped part be S o Then trace continuity P 4 =(πr 2 -S o )/S。
Track viability: the sensor reports the target point trace data to the fusion system, and after the system association and fusion processing, the target node data quantity N of the track is formed r Accounting for the total data volume N reported by the sensor e Is a ratio of (2). P (P) 5 =N r /N e 。
5.6 fusion Property
The fusion performance is used for evaluating the actual capability of the fusion system on each operation of the target parameters, and comprises four aspects of correct fusion rate, correct separation rate, missed association rate and false association rate.
Correct and correctFusion rate: in different time periods t, the true target number N of the correct fusion c Accounting for the total number N of target lot s Is a ratio of (2).
Correct separation rate: in different time periods t, the number N of correctly separated targets in the wrong targets is related cd Accounting for the total number N of separation operation targets d Is a ratio of (2).
Leakage association rate: in different time periods t, the fusion system correlates the number of lots which are not present in the output lot after the operation with the number N of all target lots detected by the sensor s Is a ratio of (2).
Error association rate: in different time periods t, the fusion system correlates the target number N on the error batch number after the correlation operation e Quantity N with real target lot number s Is a ratio of (2).
6. Construction of preliminary model
A complete structural equation model needs to contain two aspects: a measurement model and a structural model. The measurement model refers to the frontal relationship between the latent and the apparent variables, and the structural model refers to the relationship between the latent variables.
For the number of samples, it is generally considered that it should be greater than 200, and for different models, there are different requirements. The basic requirement of the model for the input samples is N/P > 10. Where N is the number of samples and p is the number of indices. The index number of the present invention counts 21 in total, and the larger the number of samples, the better, so the number of samples is set to 300.
The raw track data was calculated as shown in fig. 5 to give a total of 300 sets of data for each observed variable, which were imported into amos software and subjected to calculate estimates operations to give unstandardized estimates and standarized estimates results.
7. Model fitting degree and significance test
The model fitting process firstly solves the model, namely estimates the model parameters, and finally reaches the standard that the difference between the hidden covariance matrix of the model and the sample covariance matrix is minimum. The fitting degree is also called fitting degree and matching degree, and is the most important index for measuring whether the model is established or not in the invention. The degree of consistency of the hypothesized theoretical model and the actual data is firstly compared by using the fitness index, and the higher the model fitness is, the higher the degree of consistency of the theoretical model and the actual data is represented. Amos software uses chi-square as the result of the fitness test, typically with chi-square value P >0.05 as the standard, however chi-square is susceptible to sample size, and therefore requires reference to other fitness indicators in addition to chi-square statistics.
Table 1 initial fitness value of model
Evaluation index | Adaptation standard | Initial fitting value |
CMIN/DF | Ideal value<3, generally 1 to 3 | 1.544 |
CFI | Ideal value>0.9, and the closer to 1, the better | 0.700 |
RMSEA | Ideal value<0.08, and the closer to 0, the better | 0.051 |
GFI | Ideal value>0.9, and the closer to 1, the better | 0.956 |
NFI | Ideal value>0.9, and the closer to 1, the better | 0.862 |
RMR | Ideal value<0.08, and the closer to 0, the better | 0.017 |
TLI | Ideal value>0.9, and the closer to 1, the better | 0.652 |
AGFI | Ideal value>0.9, and the closer to 1, the better | 0.858 |
From the table, each index of the fitting degree accords with the basic requirement of the building model.
The model fitting degree describes the consistency degree of the theoretical model and the actual data, and the magnitude of the factor load reflects whether the relation between variables is obvious or not. In the course of the saliency check, when the P value of the coefficient >0.05, it is shown that the effect between the two variables of the path is insignificant. In this case, in order to improve the fitting degree of the model, the insignificant path may be deleted.
Table 2 Regression Weights
Table 2 reflects the factor load and significance in the non-normalized state, where estinate represents the non-normalized coefficient, i.e., the non-normalized factor load, whose value is referenced to 1, and whose absolute value is typically between 0 and 1. S.e. represents the standard error of the estimated parameters, the smaller the standard error, the closer the sample statistic is to the value of the overall parameter. C.r. is a test statistic, also known as a critical ratio, the absolute value of which, if greater than 1.96, indicates that a level of significance of 0.05 is achieved. P values represent significance levels, and if P <0.01, P >0.01 is indicated by "×", the magnitude of the value is directly presented.
The regression coefficient of this model is between-1.793 and 1.676, 76% of the data is present between-1 and 1 ideal; the critical ratio is all greater than 1.96, reaching a 95% confidence level; except for the corresponding relation of physical performance- > system stability, the rest P values are all smaller than 0.05, which indicates that the significance meets the requirement. Therefore, the significance of the corresponding relation among the indexes of the model is good, and only a small amount of corresponding relation is needed to be modified.
8. Model correction
The invention has two methods for model correction, one is model expansion, and uses the correction index to release part to limit the path or add new path, so that the model structure is more reasonable, and the model is usually used for improving the model fitting degree; and secondly, model limitation, the critical ratio is used for deleting or limiting part of paths, so that the model structure is simpler, and the model identification is generally improved. The invention finally adopts a model limiting method to correct the model by considering the complexity and the data quantity of the model, and deletes the path with the P value of 0.156 in the table 2 to recalculate the fitting degree and the significance of the model.
Table 3 model corrected fitness values
Evaluation index | Adaptation standard | Correction of fitting values |
CMIN/DF | Ideal value<3, greater than 5 is poor | 2.634 |
CFI | Ideal value>0.9, and the closer to 1, the better | 0.712 |
RMSEA | Ideal value<0.08, and the closer to 0, the better | 0.049 |
GFI | Ideal value>0.9, and the closer to 1, the better | 0.965 |
NFI | Ideal value>0.9, and the closer to 1, the better | 0.873 |
RMR | Ideal value<0.08, and the closer to 0, the better | 0.021 |
TLI | Ideal value>0.9, and the closer to 1, the better | 0.652 |
AGFI | Ideal value>0.9, and the closer to 1, the better | 0.843 |
The fitting index after the correction has slight difference from the first calculation, and is reflected in the second three positions after the decimal point, but the total still meets the requirements of each fitting index, so that the model can still well explain the relation among variables. P values were all less than 0.05, indicating that the model was good in significance.
The residuals of the corrected standardized path diagrams are all positive, range from 0.35 to 0.88, meet the significance requirement, and have no collinearity. The absolute value of the standardized coefficient is larger than 0.6 except the corresponding relation of 'information processing precision- > target fusion heading precision'. On the premise that the model fitting degree meets the significance requirement, the path with the standardized factor load larger than or close to 0.6 is reserved as much as possible, so that the model can be evaluated more accurately and comprehensively.
9. Analysis of results
Table 4 normalized and non-normalized coefficients and significance after model modification
As can be seen from table 4, the significance after path correction all met the 95% confidence level and the model performed well.
The importance degree of the observation variable to the latent variable can be known from the standardized coefficient, and an improvement opinion can be provided for track fusion. Among four observation variables affecting physical properties, the standardized coefficient of sensor accuracy is the largest, the standardized coefficient of storage capacity is the smallest, and the most important index when a sensor is selected for observation is the accuracy index; among the three factors affecting the time efficiency, the maximum load is the sensor track formation time, which indicates that the time used for the sensor observation takes the dominant role in the whole observation and evaluation process; the load capacity of the target fusion update rate is the largest among three factors affecting the data update capacity, and the influence of the speed of target fusion on the data update in unit time is the largest; for the information processing precision, the influence degree of three factors on the information processing precision is not great; for track quality, the influence of track fragility index on the track fragility index is inversely related, and the influence is the largest, so that the situation that the batch number of one track is combined into a plurality of tracks is avoided as much as possible, and the point can be preferentially considered in the measure of improving the performance of a fusion system; among the four indexes affecting the target recognition performance, the influence of the missed association rate and the false association rate on the target recognition performance is negative, and the influence factor of the correct fusion rate is as high as 94%, which indicates that the track fusion system should take the correct rate for improving the fusion as the main direction.
The non-standardized coefficient can calculate the variation of the latent variable when the observed variable changes by one unit, so that the size of each level index can be calculated quantitatively, and the evaluation result of the track fusion system is obtained.
The invention is not limited to the specific embodiments described above. The invention extends to any novel one, or any novel combination, of the features disclosed in this specification, as well as to any novel one, or any novel combination, of the steps of the method or process disclosed. It is intended that insubstantial changes or modifications from the invention as described herein be covered by the claims below, as viewed by a person skilled in the art, without departing from the true spirit of the invention.
Claims (7)
1. The multi-source multi-data track fusion system evaluation method based on the structural equation model is characterized by comprising the following steps of:
step S1, the multi-source multi-data track fusion system evaluation system is divided into two levels of indexes, wherein the first level of indexes serve as factors for finally describing the advantages and disadvantages of the track fusion system, and the first level of indexes are quantized through the factor load of the structural equation model after the second level of indexes are calculated;
step S2, providing index meanings and calculation formulas of all levels; the core calculation formula is summarized as follows:
in the physical performance of the system, the estimation precision of the sensor is as follows Wherein n, e l 、e v The method comprises the steps of obtaining the difference value between the true number of track targets, the position estimation vector, the speed estimation vector and the observed data; the sensor recognition rate is->Sensor stability is +.>Wherein N is i Number of data pieces of target lot number, N e Is the total number of the observed data of the sensor, N 1 Is the first observed data quantity of the sensor, N a Is the amount of data observed by the sensor during different time periods t;
in terms of time efficiency, the total time consumed by the n pieces of data entering the fusion system to be correlated and fused is T, and the fusion time delay is T m =t/n, at t 0 To t n The number of complete tracks observed by the sensor in the time period is N t The sensor track formation time is
On data update performanceThe sensor a reports n times of targets to the fusion center, and the time of the first report is t 1 The last reporting time is t 2 The sensor reporting update rate is R 1 =(n-1)/(t 2 -t 1 ) The method comprises the steps of carrying out a first treatment on the surface of the In the data processing precision index, the precision of the target fusion position is as follows Wherein diff_lo and diff_la are differences between longitude and latitude parameters and real positions of a target with a lot number p1 reported by a sensor a at the moment k;
in the track quality, the track fragility isTracking continuity P 4 =(πr 2 -S o ) Track viability of P 5 =N r /N e Wherein N is b True track number reported for sensor, N d For the number of tracks output by the fusion system, r is the observation radius of the sensor, S is the total area of the observed sea area, S o To observe the total area of the overlapping portion, N r The data quantity which can form a track in the target node reported by the sensor;
in the fusibility index, the true target number N which is correctly fused in the time period t c Accounting for the total number N of target lot s The ratio of (2), i.e. the correct fusion rate, isNumber of correctly separated objects N in associated erroneous objects cd Accounting for the total number N of separation operation targets d The ratio of (2), i.e. the correct separation rate, is given by +.>Batch that the fusion system does not appear in the output batch number after the association operationNumber and number N of all target lot numbers detected by the sensor s The ratio of (1), namely the leakage association rate, isThe fusion system associates the target number N to the wrong lot number after the association operation e Quantity N with real target lot number s The ratio of (2), i.e. the false correlation rate is +.>
Step S3, determining path relations among the latent variables, the observed variables and the latent variables and the observed variables according to the index level in the step S1, and constructing an initial model based on SEM;
step S4, processing the original data, calculating according to the formula in the step S2, importing a calculation result into amos software, performing calculate estimates operation to obtain a fitness index, and performing model fitness and significance test;
step S5, when the fitting degree index and the significance test obtained in the step S4 meet the requirements, standardized estimation is carried out to obtain a standardized path coefficient and factor load; if the requirements are not met, carrying out model correction according to the meaning of each fitness index and the non-standardized coefficient value, and repeating the step S4 until the fitness index CMIN/DF is smaller than 3, the RMSEA and RMR residual error item index is smaller than 0.08, and the CFI, NFI and TLI fitness index is larger than 0.9; a significance level P of less than 0.01 is ideal, less than 0.05 meets the requirements;
and S6, writing out one-to-one correspondence coefficients between the latent variables and the observed variables through the path diagram, calculating the example data through the formula in the step S2, multiplying the obtained result by the coefficient corresponding to the index, finally obtaining the evaluation result of track fusion, and providing improvement measures for the track fusion process.
2. The method for estimating multi-source track fusion based on structural equation model according to claim 1, wherein in step S1, the specific criteria are set according to the track fusion system by performing index extraction: first, to maintain the consistency of the track and the target true value, the track-related accuracy index is the primary consideration; secondly, taking the difference between the fused target state estimation and the true value into consideration, namely a track precision related index; thirdly, the identification accuracy index of the target attribute, the position information and the navigational speed information is considered.
3. The method for estimating multi-source track fusion based on structural equation model according to claim 1, wherein in step S2, the index level is divided into two levels, and the first level index is latent variable in the structural equation model, including physical performance, time efficiency, information processing precision, data updating capability, track quality and target recognition capability.
4. The method for estimating multi-source track fusion based on the structural equation model according to claim 1, wherein in the step S2, the index hierarchy is divided into two levels, and the two levels are the observed variables in the structural equation model, including storage capacity, sensor precision, sensor stability, sensor recognition rate, fusion delay, sensor track formation time, system track formation time, sensor reporting update rate, target fusion update rate, system issue indication update rate, target fusion position precision, target fusion speed precision, target fusion heading precision, track fragmentation degree, tracking continuity, track survival ability, correct fusion rate, correct separation rate, miss association rate, and error association rate.
5. The method for estimating multi-source track fusion based on the structural equation model according to claim 1, wherein in the step S3, the latent variable is the first-level index in S1; the observed variable is the second-level index in S1, and the observed variable is calculated by the formula in the step S2.
6. The multi-source track fusion evaluation method based on the structural equation model according to claim 1, wherein in the step S4, the original data is divided into normal data and abnormal data, the abnormal data is divided into four types of outlier data, unreasonable data, data with larger jumping and missing data, the first three types of abnormal data are processed by adopting a 3 alpha data removal method and a robust regression method, and the blank data are processed by adopting a data interpolation method; and selecting eight key indexes of CMIN/DF, CFI, RMSEA, GFI, NFI, RMR, TLI, AGFI from the multiple fitness indexes for testing, and adopting four indexes of Estimate, S.E., C.R., and P for testing the significance.
7. The multi-source track fusion evaluation method based on the structural equation model according to claim 1, wherein in the step S5, the model fitting degree describes the consistency degree of the theoretical model and the actual data, and the magnitude of the factor load reflects whether the relationship between the variables is significant; in the saliency check process, when the P value of the coefficient >0.05, it is shown that the effect between the two variables of the path is insignificant.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110092959.2A CN112906746B (en) | 2021-01-25 | 2021-01-25 | Multi-source track fusion evaluation method based on structural equation model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110092959.2A CN112906746B (en) | 2021-01-25 | 2021-01-25 | Multi-source track fusion evaluation method based on structural equation model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112906746A CN112906746A (en) | 2021-06-04 |
CN112906746B true CN112906746B (en) | 2024-04-02 |
Family
ID=76118594
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110092959.2A Active CN112906746B (en) | 2021-01-25 | 2021-01-25 | Multi-source track fusion evaluation method based on structural equation model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112906746B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115859695B (en) * | 2023-02-27 | 2023-05-26 | 北京千种幻影科技有限公司 | Method, system and equipment for analyzing simulated driving test data |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7719461B1 (en) * | 2008-08-05 | 2010-05-18 | Lockheed Martin Corporation | Track fusion by optimal reduced state estimation in multi-sensor environment with limited-bandwidth communication path |
CN110849372A (en) * | 2019-11-28 | 2020-02-28 | 哈尔滨工程大学 | Underwater multi-target track association method based on EM clustering |
CN110907903A (en) * | 2019-10-30 | 2020-03-24 | 中电科技(合肥)博微信息发展有限责任公司 | Self-adaptive tracking processing method based on multiple sensors |
CN111539625A (en) * | 2020-04-23 | 2020-08-14 | 中国北方车辆研究所 | Ground target-oriented multi-source information fusion evaluation method |
-
2021
- 2021-01-25 CN CN202110092959.2A patent/CN112906746B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7719461B1 (en) * | 2008-08-05 | 2010-05-18 | Lockheed Martin Corporation | Track fusion by optimal reduced state estimation in multi-sensor environment with limited-bandwidth communication path |
CN110907903A (en) * | 2019-10-30 | 2020-03-24 | 中电科技(合肥)博微信息发展有限责任公司 | Self-adaptive tracking processing method based on multiple sensors |
CN110849372A (en) * | 2019-11-28 | 2020-02-28 | 哈尔滨工程大学 | Underwater multi-target track association method based on EM clustering |
CN111539625A (en) * | 2020-04-23 | 2020-08-14 | 中国北方车辆研究所 | Ground target-oriented multi-source information fusion evaluation method |
Also Published As
Publication number | Publication date |
---|---|
CN112906746A (en) | 2021-06-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107992968B (en) | Electric energy meter metering error prediction method based on integrated time series analysis technology | |
CN102393881B (en) | A kind of high-precision detecting method of real-time many sensing temperatures data fusion | |
CN104931960A (en) | Trend message and radar target state information whole-track data correlation method | |
CN111950627B (en) | Multi-source information fusion method and application thereof | |
CN106354995A (en) | Predicting method based on Lagrange interpolation and time sequence | |
CN107679734A (en) | It is a kind of to be used for the method and system without label data classification prediction | |
CN112906746B (en) | Multi-source track fusion evaluation method based on structural equation model | |
WO2022213600A1 (en) | Method for recognizing type of vortex signal of evaporator of nuclear power plant on basis of lstm-cnn | |
CN114971345B (en) | Quality measuring method, equipment and storage medium for built environment | |
CN111144462B (en) | Unknown individual identification method and device for radar signals | |
CN117670256B (en) | BIM technology-based engineering cost accurate control system and method | |
CN113295421B (en) | Engine fault diagnosis method based on improved conflict coefficient and reliability entropy | |
CN113095579B (en) | Daily-scale rainfall forecast correction method coupled with Bernoulli-gamma-Gaussian distribution | |
CN113553319A (en) | LOF outlier detection cleaning method, device and equipment based on information entropy weighting and storage medium | |
WO2019196437A1 (en) | Index decision method | |
CN110766248B (en) | Workshop artificial factor reliability assessment method based on SHEL and interval intuitionistic fuzzy assessment | |
CN113673609B (en) | Questionnaire data analysis method based on linear hidden variables | |
CN113946953B (en) | Calculation method for solving global sensitivity under probability box framework | |
CN117933832B (en) | Index weight evaluation method for spacecraft ground equivalence test | |
CN117436770B (en) | BIM-based building engineering construction management method and system | |
CN116957361B (en) | Ship task system health state detection method based on virtual-real combination | |
CN118070231B (en) | Multi-source multi-target measurement data real-time fusion method based on cluster association | |
CN118033691B (en) | Surveying and mapping reference supervision and management method and system based on satellite navigation positioning reference | |
WO2022217568A1 (en) | Daily precipitation forecast correction method coupled with bernoulli-gamma-gaussian distributions | |
CN117933832A (en) | Index weight evaluation method for spacecraft ground equivalence test |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |