CN110852158B - Radar human motion state classification algorithm and system based on model fusion - Google Patents
Radar human motion state classification algorithm and system based on model fusion Download PDFInfo
- Publication number
- CN110852158B CN110852158B CN201910942378.6A CN201910942378A CN110852158B CN 110852158 B CN110852158 B CN 110852158B CN 201910942378 A CN201910942378 A CN 201910942378A CN 110852158 B CN110852158 B CN 110852158B
- Authority
- CN
- China
- Prior art keywords
- support vector
- vector machine
- model
- machine model
- obtaining
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/23—Recognition of whole body movements, e.g. for sport training
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
Abstract
The invention belongs to the technical field of radars, and particularly relates to a radar human motion state classification algorithm and system based on model fusion, wherein the method comprises the following steps: obtaining a training set; constructing a support vector machine model according to the training set; obtaining a predicted value of the support vector machine model according to the support vector machine model; and constructing a limit gradient lifting tree model according to the predicted value of the support vector machine model. According to the invention, the support vector machine model and the limit gradient lifting tree model are fused by the stacking model fusion method, the support vector machine model is suitable for processing small samples with high latitude, the limit gradient lifting tree model has the advantage of strong fitting capacity, and the fused model has the advantages of the support vector machine model and the limit gradient lifting tree model, so that the model generalization capacity is stronger, the recognition precision is higher, and the model training time in deep learning is shortened.
Description
Technical Field
The invention belongs to the technical field of radars, and particularly relates to a radar human motion state classification algorithm and system based on model fusion.
Background
The radar has great advantages over other sensors in detecting the motion state of the human body, such as the optical sensor is easily affected by weather environment and light, and the radar can work all day long. In addition, the radar has certain penetrating power, can detect targets behind the obstacle, can even judge the motion state of a human body behind the obstacle, and can be used for anti-terrorism, military and post-disaster rescue and life detection by combining related technologies. In a word, the human motion state classification based on radar has a very wide application prospect.
Beginning in nineties of the twentieth century, researchers began to study body micro-motion according to the micro-doppler characteristics of radar, initially used for classification between different targets on war according to the doppler characteristic differences between the different targets, and later applied to the classification of the motion state of the body targets. Chen V.C establishes radar echo data simulated by a human body model through software, and then carries out time-frequency analysis on the simulated radar echo data, so as to compare and analyze micro Doppler characteristic differences of limbs of the model in different motion states; aiming at the defect of the traditional time-frequency analysis method on non-stationary signal processing, the Chiehping Lai et al introduces Hilbert-Huang transformation to extract micro Doppler characteristics of a human body from complex echo signals, but the operation processing time is longer; javier et al studied classification of various human activities based on micro Doppler features by using linear predictive coding, and put forward a method for extracting micro Doppler features mixed by different frequencies, wherein the classification accuracy reaches 85%.
Most of the existing human motion state classification methods based on machine learning use a single classifier, which causes insufficient model generalization capability and relatively low recognition accuracy. In deep learning, the complexity of the model is quite high, the calculated amount is too large, and the training data set is not easy to collect; training of the model often takes a lot of time and the process of automatically extracting image features by the model is not well interpretable.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a radar human motion state classification algorithm and system based on model fusion. The technical problems to be solved by the invention are realized by the following technical scheme:
a radar human motion state classification algorithm based on model fusion, comprising:
obtaining a training set;
constructing a support vector machine model according to the training set;
obtaining a predicted value of the support vector machine model according to the support vector machine model;
and constructing a limit gradient lifting tree model according to the predicted value of the support vector machine model.
In one embodiment of the present invention, constructing a support vector machine model according to the training set includes:
obtaining a direction gradient support vector machine model according to the training set;
obtaining a local binary support vector machine model according to the training set;
obtaining a haar support vector machine model according to the training set;
and carrying out combined operation on the direction gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model to obtain a support vector machine model.
In one embodiment of the present invention, performing a combination operation on the directional gradient support vector machine model, the local binary support vector machine model, and the haar support vector machine model to obtain a support vector machine model, including:
respectively obtaining weighting coefficients of the directional gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model;
constructing a plurality of primary support vector machine models according to a plurality of preset proportionality coefficients and weighting coefficients of the directional gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model;
obtaining a prediction result according to the plurality of primary support vector machine models;
obtaining an optimal prediction result according to the prediction results of the primary support vector machine models;
and selecting the primary support vector machine model corresponding to the optimal prediction result as a support vector machine model.
In one embodiment of the present invention, constructing a limit gradient lifting tree model according to the predicted value of the support vector machine model includes:
presetting a plurality of tree depth values;
constructing a plurality of primary limit gradient lifting tree models according to the tree depth values and the predicted values of the support vector machine model;
obtaining a plurality of model parameters according to the plurality of primary limit gradient lifting tree models;
obtaining optimal model parameters according to the model parameters;
and selecting a primary limit gradient lifting tree model corresponding to the optimal model parameters as a limit gradient lifting tree model.
The invention also provides a radar human motion state classification system based on model fusion, which comprises:
the information acquisition module is used for acquiring a training set;
the support vector machine model building module is used for building a support vector machine model according to the training set;
the predicted value acquisition module is used for acquiring a predicted value of the support vector machine model according to the support vector machine model;
and the limit gradient lifting tree model construction module is used for constructing a limit gradient lifting tree model according to the predicted value of the support vector machine model.
In one embodiment of the present invention, the support vector machine model building module includes:
the directional gradient support vector machine model building unit is used for obtaining a directional gradient support vector machine model according to the training set;
the local binary support vector machine model building unit is used for obtaining a local binary support vector machine model according to the training set;
the haar support vector machine model building unit is used for obtaining a haar support vector machine model according to the training set;
and the support vector machine model construction unit is used for carrying out combined operation on the direction gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model to obtain a support vector machine model.
In one embodiment of the present invention, the support vector machine model building unit includes:
the weighting coefficient acquisition subunit is used for respectively acquiring the weighting coefficients of the directional gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model;
the primary support vector machine model construction subunit is used for constructing a plurality of primary support vector machine models according to a plurality of preset proportionality coefficients and weighting coefficients of the directional gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model;
the prediction result obtaining subunit is used for obtaining prediction results according to the plurality of primary support vector machine models;
the optimal prediction result obtaining subunit is used for obtaining an optimal prediction result according to the prediction results of the plurality of primary support vector machine models;
and the support vector machine model construction subunit is used for selecting a primary support vector machine model corresponding to the optimal prediction result as a support vector machine model.
In one embodiment of the present invention, the limiting gradient lifting tree model building module comprises:
the primary limit gradient lifting tree model construction unit is used for constructing a plurality of primary limit gradient lifting tree models according to a plurality of preset tree depth values and predicted values of the support vector machine model;
the model parameter extraction unit is used for obtaining a plurality of model parameters according to the plurality of primary limit gradient lifting tree models;
the optimal model parameter acquisition unit is used for acquiring optimal model parameters according to the model parameters;
and the limit gradient lifting tree model construction unit is used for selecting a primary limit gradient lifting tree model corresponding to the optimal model parameters as a limit gradient lifting tree model.
The invention has the beneficial effects that:
according to the invention, the support vector machine model and the limit gradient lifting tree model are fused by a stacking method model fusion method, the support vector machine model is suitable for processing small sample high latitude, the limit gradient lifting tree model has the advantage of strong fitting capability, and the fused model has the advantages of the support vector machine model and the limit gradient lifting tree model, so that the model generalization capability is stronger, the recognition precision is higher, and the time for model training in deep learning is reduced.
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Drawings
FIG. 1 is a schematic flow chart of a radar human motion state classification algorithm based on model fusion provided by an embodiment of the invention;
FIG. 2 is a schematic diagram of specific steps of a radar human motion state classification algorithm based on model fusion according to an embodiment of the present invention;
fig. 3 is a structural block diagram of a radar human motion state classification system based on model fusion according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to specific examples, but embodiments of the present invention are not limited thereto.
Referring to fig. 1, fig. 1 is a schematic flow chart of a radar human motion state classification algorithm based on model fusion according to an embodiment of the present invention, including:
obtaining a training set;
constructing a support vector machine model according to the training set;
obtaining a predicted value of the support vector machine model according to the support vector machine model;
and constructing a limit gradient lifting tree model according to the predicted value of the support vector machine model.
According to the invention, the support vector machine model and the limit gradient lifting tree model are fused by the stacking model fusion method, the support vector machine model is suitable for processing small samples with high latitude, the limit gradient lifting tree model has the advantage of strong fitting capacity, and the fused model has the advantages of the support vector machine model and the limit gradient lifting tree model, so that the model generalization capacity is stronger, the recognition precision is higher, and the model training time in deep learning is shortened.
In one embodiment of the present invention, constructing a support vector machine model according to the training set includes:
obtaining a direction gradient support vector machine model according to the training set;
obtaining a local binary support vector machine model according to the training set;
obtaining a haar support vector machine model according to the training set;
and carrying out combined operation on the direction gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model to obtain a support vector machine model.
Further, the directional gradient (HOG) support vector machine model and the HAAR (HAAR) support vector machine model have good time-frequency diagram classification effect on human body walking, the Local Binary (LBP) support vector machine model has good time-frequency diagram classification effect on human body running, the characteristics of the three support vector machine models are combined, the combined characteristics are used for training the Support Vector Machine (SVM) model, the support vector machine model can be enabled to gather the beneficial points of the three support vector machine models, and generalization capability of the support vector machine model is greatly improved.
In one embodiment of the present invention, performing a combination operation on the directional gradient support vector machine model, the local binary support vector machine model, and the haar support vector machine model to obtain a support vector machine model, including:
respectively obtaining weighting coefficients of the directional gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model;
constructing a plurality of primary support vector machine models according to a plurality of preset proportionality coefficients and weighting coefficients of the directional gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model;
obtaining a prediction result according to the plurality of primary support vector machine models;
obtaining an optimal prediction result according to the prediction results of the primary support vector machine models;
and selecting the primary support vector machine model corresponding to the optimal prediction result as a support vector machine model.
Specifically, the training set has 1586 time-frequency diagrams of human body walking and running, 80% of data in the training set is used as training data for training a model, 645 images are time-frequency diagrams of human body walking, and 624 images are time-frequency diagrams of human body running; 20% of data is used as verification data for model verification in a training stage, wherein 161 images are time-frequency images of human body running, and 156 images are time-frequency images of human body running; the iteration times are set to 1000 times, the punishment coefficient C is set to 0.3, and the kernel function selects a Gaussian kernel function RBF
To facilitate comparison of performance between different models, define: TP is a time-frequency diagram of walking model prediction; TN is a time-frequency diagram of running model prediction and running; FN is a time-frequency diagram of running predicted by the running model; FP is a time-frequency plot of the running model predicted as running.
The classification accuracy A of the model is as follows:
the precision rate P and recall rate R are:
harmonic mean F of precision P and recall R 1 The method comprises the following steps:
according to the verified classification result obtained by different feature training:
prediction results of models obtained by training SVM by different features in verification set
Prediction result | HOG | LBP | HAAR |
TP | 149 | 138 | 134 |
TN | 133 | 137 | 126 |
FP | 23 | 19 | 30 |
FN | 12 | 23 | 27 |
Model performance index obtained by training SVM with different features
Specific analysis: verifying the time-frequency diagram of the running and the starting of the verification data 317, wherein the accuracy of the overall classification of the directional gradient support vector machine model obtained by using HOG feature training is highest and can reach 88.96%, and the fact that the time-frequency diagram of the running and the starting can be well distinguished by using the HOG feature is proved to have poor classification effect by using the HAAR support vector machine model constructed by using the HAAR feature; recall of direction gradient support vector machine model and haar support vector machine modelThe rate R is larger than the accuracy rate P, so that the accuracy rate of the time-frequency graph prediction of the HOG features and the HAAR features on human body walking is better, the accuracy rate P of the LBP features is larger than the recall rate R, and the accuracy rate of the time-frequency graph prediction of the LBP features on human body running is better. F (F) 1 The overall performance of the model is reflected, the HOG characteristic effect is better, the LBP characteristic is inferior, and the HAAR characteristic effect is poorer.
According to the analysis result, different features are different in time-frequency diagram classification of human walking and running, HOG features and HAAR features are better in time-frequency diagram classification of human walking, LBP features are better in time-frequency diagram classification of human running, therefore the three features are combined for training of a support vector machine model, and the combined features are w: w=xhog+ybbp+zhaar, x, y, z are different characteristic coefficients, x+y+z=1. And adjusting the values of x, y and z to obtain different combined characteristics and a support vector machine model, respectively verifying verification data by using the obtained support vector machine model to obtain prediction results under different proportion coefficients, wherein the prediction results are shown in the following table:
prediction result table of model obtained by training support vector machine model by different combination features in verification set
Prediction result | (0.6,0.3,0.1) | (0.5,0.4,0.1) | (0.5,0.5,0) | (0.5,0.3,0.2) | (0.4,0.3,0.3) |
TP | 143 | 150 | 149 | 146 | 145 |
TN | 142 | 141 | 138 | 140 | 138 |
FP | 14 | 15 | 18 | 16 | 18 |
FN | 18 | 11 | 12 | 15 | 16 |
Model performance index table obtained by training support vector machine models by different combination features
Performance index | Accuracy A | Accuracy P | Recall rate R | F 1 Value of |
(0.6,0.3,0.1) | 89.91% | 91.08% | 88.82% | 89.92% |
(0.5,0.4,0.1) | 91.80% | 90.91% | 93.17% | 92.16% |
(0.5,0.5,0) | 90.54% | 94.90% | 92.55% | 93.76% |
(0.5,0.3,0.2) | 90.22% | 92.99% | 90.68% | 91.83% |
(0.4,0.3,0.3) | 89.27% | 88.96% | 90.06% | 89.53% |
Specific analysis: as can be seen from a model performance index table obtained by training the support vector machine model from different combined features, combining the HOG features, the LBP features and the HAAR features and constructing the support vector machine model can improve the classification accuracy of the model, when x=0.5, y=0.4 and z=0, 1, the classification performance of the model is optimal, after the features are combined, the difference between the accuracy and the recall rate is also reduced, and the model is more stable for predicting different types of time-frequency diagrams. The different features extract information of different classes and layers of the time-frequency diagram, and the combination of the features can enable the model to learn the features of the time-frequency diagram more comprehensively, so that the combined support vector machine model has stronger generalization capability and higher robustness due to the combination of the different features.
In one embodiment of the invention, constructing a limit gradient lifting tree (XGBOOST) model from the predicted values of the support vector machine model comprises:
presetting a plurality of tree depth values;
constructing a plurality of primary limit gradient lifting tree models according to the tree depth values and the predicted values of the support vector machine model;
obtaining a plurality of model parameters according to the plurality of primary limit gradient lifting tree models;
obtaining optimal model parameters according to the model parameters;
and selecting a primary limit gradient lifting tree model corresponding to the optimal model parameters as a limit gradient lifting tree model.
Specifically, the parameters of the XGBOOST model are shown in the following table:
training limit gradient lifting tree parameter table
Parameter class | booster | eta | lambda | silent | max_depth |
Parameter value | gbtree | 0.1 | 1 | 0 | 4-9 |
As can be seen from the table above, the base classifier selected when training the limit gradient lifting tree model is a tree model, namely a classification and regression tree (CART), the tree model has stronger fitting capability than a linear model, in this embodiment, the learning rate eta is set to 0.1, and the leaf nodes of the XGBOOST model are multiplied by the learning rate, so that the following tree has a larger learning space; lambda=1, i.e. L2 regularization was added to prevent model overfitting; the silent=0 shows the information such as iteration times, model loss and the like on a control console, so that parameters can be conveniently adjusted in the training process; the tree depth, max_depth is set to 4-9, and XGBOOST models with different depths can be obtained by adjusting the parameter; the base classifier type boost is a tree model gbtree.
The depth of the tree may cause excessive learning training data, lead to overfitting, obtain different limit gradient lifting tree models by adjusting the depth of the XGBOOST tree, and perform walking and running time-frequency diagram verification on images in verification data, and the verification results are shown in the following table:
prediction result table of limit gradient lifting tree model of different tree depths s in verification set
Prediction result | 4 | 5 | 6 | 7 | 8 | 9 |
TP | 145 | 147 | 152 | 154 | 141 | 139 |
TN | 141 | 142 | 143 | 138 | 143 | 140 |
FP | 15 | 14 | 13 | 18 | 13 | 16 |
FN | 16 | 14 | 9 | 7 | 20 | 22 |
Meanwhile, the performance index table of the limit gradient lifting tree model with different tree depths is shown as follows:
performance index table of limit gradient lifting tree model of different tree depths
Performance index | Accuracy A | Accuracy P | Recall rate R | F 1 Value of |
4 | 90.22% | 90.63% | 90.06% | 90.33% |
5 | 91.17% | 91.30% | 91.30% | 91.30% |
6 | 93.06% | 92.12% | 94.41% | 93.25% |
7 | 92.11% | 89.53% | 95.65% | 92.49% |
8 | 89.59% | 91.56% | 87.58% | 89.53% |
9 | 88.01% | 89.67% | 86.34% | 87.97% |
From the two tables, the verification data are used for verifying the limit gradient lifting tree models with different tree depths, the classification accuracy of the models is steadily improved along with the increase of the tree depths, the classification accuracy of the models is highest when the tree depths are 6, and can reach 93.06%, and when the tree depths reach 7-9, the accuracy is gradually decreased, and the fitting phenomenon occurs, so that the model has the highest fitting capability and the best performance when the tree depths are 6; when the tree depth is 4-6 layers, the deviation between the accuracy rate and recall rate of the limit gradient lifting tree model is small, so the limit gradient lifting tree model at the moment is relatively robust. In summary, when the tree depth is 6, the performance of the model is optimal.
Further, referring to fig. 2, fig. 2 is a schematic diagram of specific steps of a radar human motion state classification algorithm based on model fusion according to an embodiment of the present invention, coefficients of an optimal feature combination are found by comparing three different features, a support vector machine model is obtained by combining the three features using the set of coefficients, and a limit gradient lifting tree is trained by using a prediction result of the support vector machine model to finally obtain the limit gradient lifting tree. Model fusion based on stacking equally divides training data into k parts, k models with different parameters are obtained through k-fold cross validation training, the predicted values of the k models are used as new characteristic values to train another model, and output values of the different models are combined to be used as output values after the models are fused. In this embodiment, features of a time-frequency diagram of image walking and running of a training data set are divided into 5 parts, a 5-fold cross validation method is used for training a support vector machine model to obtain 5 support vector machine models, predicted values of the 5 support vector machine models are used as training data of a limit gradient lifting tree model to obtain a limit gradient lifting tree, a predicted value E of the limit gradient lifting tree is obtained, the support vector machine model is validated through validation data to obtain 5 predicted values, an average is taken to obtain a predicted value D of the support vector machine model, and the predicted value E of the limit gradient lifting tree and the predicted value D of the support vector machine model are combined to obtain a predicted value F of a final model:
F=x 1 D+y 1 E,x 1 +y 1 =1,
wherein x is 1 And y 1 Is the proportionality coefficient of the model, different final models are obtained by adjusting the proportionality coefficient, and the final model after the support vector machine model and the limit gradient lifting tree model are fused by a stacking model fusion methodAnd verifying the different final models through the verification data set, wherein the verification results are shown in the following table:
prediction result table of different final models in verification set
Further, a performance index table of different final models is obtained, as follows:
performance index of different fusion models
Performance index | Accuracy A | Accuracy P | Recall rate R | F 1 Value of |
(0.3,0.7) | 92.11% | 90.00% | 95.03% | 92.45% |
(0.35,0.65) | 94.32% | 92.81% | 96.27% | 94.51% |
(0.4,0.6) | 93.06% | 93.17% | 93.17% | 93.17% |
(0.5,0.5) | 92.43% | 93.08% | 91.93% | 92.50% |
(0.55,0.45) | 92.74% | 95.39% | 90.06% | 92.65% |
Specific analysis:
from the two tables, the fused model has better performance than a single support vector machine model, the classification performance of the fused model is improved to a certain extent, the difference between the accuracy and the recall is smaller, the model performance is stable, and the proportionality coefficient x of the support vector machine model is stable 1 0.35, limit gradient lifting tree model y 1 The final model classification performance was optimal at 0.65. However, as the scaling factor of the support vector machine model increases, the performance of the final model is gradually reduced, so that an optimal model corresponding to the optimal scaling factor is obtained.
Referring to fig. 3, fig. 3 is a block diagram of a radar human motion state classification system based on model fusion according to an embodiment of the present invention, including:
the information acquisition module is used for acquiring a training set;
the support vector machine model building module is used for building a support vector machine model according to the training set;
the predicted value acquisition module is used for acquiring a predicted value of the support vector machine model according to the support vector machine model;
and the limit gradient lifting tree model construction module is used for constructing a limit gradient lifting tree model according to the predicted value of the support vector machine model.
In one embodiment of the present invention, the support vector machine model building module includes:
the directional gradient support vector machine model building unit is used for obtaining a directional gradient support vector machine model according to the training set;
the local binary support vector machine model building unit is used for obtaining a local binary support vector machine model according to the training set;
the haar support vector machine model building unit is used for obtaining a haar support vector machine model according to the training set;
and the support vector machine model construction unit is used for carrying out combined operation on the direction gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model to obtain a support vector machine model.
In one embodiment of the present invention, the support vector machine model building unit includes:
the weighting coefficient acquisition subunit is used for respectively acquiring the weighting coefficients of the directional gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model;
the primary support vector machine model construction subunit is used for constructing a plurality of primary support vector machine models according to a plurality of preset proportionality coefficients and weighting coefficients of the directional gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model;
the prediction result obtaining subunit is used for obtaining prediction results according to the plurality of primary support vector machine models;
the optimal prediction result obtaining subunit is used for obtaining an optimal prediction result according to the prediction results of the plurality of primary support vector machine models;
and the support vector machine model construction subunit is used for selecting a primary support vector machine model corresponding to the optimal prediction result as a support vector machine model.
In one embodiment of the present invention, the limiting gradient lifting tree model building module comprises:
the primary limit gradient lifting tree model construction unit is used for constructing a plurality of primary limit gradient lifting tree models according to a plurality of preset tree depth values and predicted values of the support vector machine model;
the model parameter extraction unit is used for obtaining a plurality of model parameters according to the plurality of primary limit gradient lifting tree models;
the optimal model parameter acquisition unit is used for acquiring optimal model parameters according to the model parameters;
and the limit gradient lifting tree model construction unit is used for selecting a primary limit gradient lifting tree model corresponding to the optimal model parameters as a limit gradient lifting tree model.
The foregoing is a further detailed description of the invention in connection with the preferred embodiments, and it is not intended that the invention be limited to the specific embodiments described. It will be apparent to those skilled in the art that several simple deductions or substitutions may be made without departing from the spirit of the invention, and these should be considered to be within the scope of the invention.
Claims (4)
1. The radar human motion state classification algorithm based on model fusion is characterized by comprising the following steps of:
obtaining a training set; the training set has 1586 time-frequency diagrams of human body walking and running, 80% of data in the training set is used as training data for training a model, 645 images are time-frequency diagrams of human body walking, and 624 images are time-frequency diagrams of human body running; 20% of data is used as verification data for model verification in a training stage, wherein 161 images are time-frequency images of human body running, and 156 images are time-frequency images of human body running;
constructing a support vector machine model according to the training set;
obtaining a predicted value of the support vector machine model according to the support vector machine model;
constructing a limit gradient lifting tree model according to the predicted value of the support vector machine model;
constructing a support vector machine model according to the training set, including:
obtaining a direction gradient support vector machine model according to the training set;
obtaining a local binary support vector machine model according to the training set;
obtaining a haar support vector machine model according to the training set;
combining the direction gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model to obtain a support vector machine model;
constructing a limit gradient lifting tree model according to the predicted value of the support vector machine model, wherein the method comprises the following steps:
presetting a plurality of tree depth values;
constructing a plurality of primary limit gradient lifting tree models according to the tree depth values and the predicted values of the support vector machine model;
obtaining a plurality of model parameters according to the plurality of primary limit gradient lifting tree models;
obtaining optimal model parameters according to the model parameters;
and selecting a primary limit gradient lifting tree model corresponding to the optimal model parameters as a limit gradient lifting tree model.
2. The radar human motion state classification algorithm based on model fusion according to claim 1, wherein performing a combination operation on the directional gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model to obtain a support vector machine model comprises:
respectively obtaining weighting coefficients of the directional gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model;
constructing a plurality of primary support vector machine models according to a plurality of preset proportionality coefficients and weighting coefficients of the directional gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model;
obtaining a prediction result according to the plurality of primary support vector machine models;
obtaining an optimal prediction result according to the prediction results of the primary support vector machine models;
and selecting the primary support vector machine model corresponding to the optimal prediction result as a support vector machine model.
3. A radar human motion state classification system based on model fusion, comprising:
the information acquisition module is used for acquiring a training set; the training set has 1586 time-frequency diagrams of human body walking and running, 80% of data in the training set is used as training data for training a model, 645 images are time-frequency diagrams of human body walking, and 624 images are time-frequency diagrams of human body running; 20% of data is used as verification data for model verification in a training stage, wherein 161 images are time-frequency images of human body running, and 156 images are time-frequency images of human body running;
the support vector machine model building module is used for building a support vector machine model according to the training set;
the predicted value acquisition module is used for acquiring a predicted value of the support vector machine model according to the support vector machine model;
the limit gradient lifting tree model construction module is used for constructing a limit gradient lifting tree model according to the predicted value of the support vector machine model;
the support vector machine model building module comprises:
the directional gradient support vector machine model building unit is used for obtaining a directional gradient support vector machine model according to the training set;
the local binary support vector machine model building unit is used for obtaining a local binary support vector machine model according to the training set;
the haar support vector machine model building unit is used for obtaining a haar support vector machine model according to the training set;
the support vector machine model construction unit is used for carrying out combined operation on the direction gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model to obtain a support vector machine model;
the limit gradient lifting tree model building module comprises:
the primary limit gradient lifting tree model construction unit is used for constructing a plurality of primary limit gradient lifting tree models according to a plurality of preset tree depth values and predicted values of the support vector machine model;
the model parameter extraction unit is used for obtaining a plurality of model parameters according to the plurality of primary limit gradient lifting tree models;
the optimal model parameter acquisition unit is used for acquiring optimal model parameters according to the model parameters;
and the limit gradient lifting tree model construction unit is used for selecting a primary limit gradient lifting tree model corresponding to the optimal model parameters as a limit gradient lifting tree model.
4. The model fusion-based radar human motion state classification system of claim 3, wherein said support vector machine model construction unit comprises:
the weighting coefficient acquisition subunit is used for respectively acquiring the weighting coefficients of the directional gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model;
the primary support vector machine model construction subunit is used for constructing a plurality of primary support vector machine models according to preset proportionality coefficients and weighting coefficients of the directional gradient support vector machine model, the local binary support vector machine model and the haar support vector machine model;
the prediction result obtaining subunit is used for obtaining prediction results according to the plurality of primary support vector machine models;
the optimal prediction result obtaining subunit is used for obtaining an optimal prediction result according to the prediction results of the plurality of primary support vector machine models;
and the support vector machine model construction subunit is used for selecting a primary support vector machine model corresponding to the optimal prediction result as a support vector machine model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910942378.6A CN110852158B (en) | 2019-09-30 | 2019-09-30 | Radar human motion state classification algorithm and system based on model fusion |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910942378.6A CN110852158B (en) | 2019-09-30 | 2019-09-30 | Radar human motion state classification algorithm and system based on model fusion |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110852158A CN110852158A (en) | 2020-02-28 |
CN110852158B true CN110852158B (en) | 2023-09-22 |
Family
ID=69596209
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910942378.6A Active CN110852158B (en) | 2019-09-30 | 2019-09-30 | Radar human motion state classification algorithm and system based on model fusion |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110852158B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111812643B (en) * | 2020-06-03 | 2023-03-24 | 中国科学院空天信息创新研究院 | Radar imaging method, device, equipment and storage medium |
CN112926390A (en) * | 2021-01-26 | 2021-06-08 | 国家康复辅具研究中心 | Gait motion mode recognition method and model establishment method |
CN113159447B (en) * | 2021-05-12 | 2022-03-29 | 中国人民解放军陆军工程大学 | Laser radar electromagnetic environment effect prediction method and system |
CN113611404A (en) * | 2021-07-09 | 2021-11-05 | 哈尔滨智吾康软件开发有限公司 | Plasma sample cancer early screening method based on ensemble learning |
CN114463014B (en) * | 2022-02-23 | 2023-07-07 | 河南科技大学 | Mobile payment risk early warning method based on SVM-Xgboost |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109409426A (en) * | 2018-10-23 | 2019-03-01 | 冶金自动化研究设计院 | A kind of extreme value gradient promotion logistic regression classification prediction technique |
-
2019
- 2019-09-30 CN CN201910942378.6A patent/CN110852158B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109409426A (en) * | 2018-10-23 | 2019-03-01 | 冶金自动化研究设计院 | A kind of extreme value gradient promotion logistic regression classification prediction technique |
Also Published As
Publication number | Publication date |
---|---|
CN110852158A (en) | 2020-02-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110852158B (en) | Radar human motion state classification algorithm and system based on model fusion | |
CN110245608B (en) | Underwater target identification method based on half tensor product neural network | |
CN110807365B (en) | Underwater target identification method based on fusion of GRU and one-dimensional CNN neural network | |
CN107290741B (en) | Indoor human body posture identification method based on weighted joint distance time-frequency transformation | |
CN111627419B (en) | Sound generation method based on underwater target and environmental information characteristics | |
CN110007366B (en) | Life searching method and system based on multi-sensor fusion | |
EP3690741A2 (en) | Method for automatically evaluating labeling reliability of training images for use in deep learning network to analyze images, and reliability-evaluating device using the same | |
CN111368671A (en) | SAR image ship target detection and identification integrated method based on deep learning | |
CN109977968B (en) | SAR change detection method based on deep learning classification comparison | |
Wei et al. | A method of underwater acoustic signal classification based on deep neural network | |
CN116230019A (en) | Deep emotion clustering method based on semi-supervised speech emotion recognition framework | |
CN113780521B (en) | Radiation source individual identification method based on deep learning | |
CN109272036B (en) | Random fern target tracking method based on depth residual error network | |
CN115438708A (en) | Classification and identification method based on convolutional neural network and multi-mode fusion | |
CN110361709B (en) | Vehicle-mounted millimeter wave radar target identification method based on dynamic false alarm probability | |
Ju et al. | A new low SNR underwater acoustic signal classification method based on intrinsic modal features maintaining dimensionality reduction | |
Lim et al. | Temporal early exiting with confidence calibration for driver identification based on driving sensing data | |
CN115656958A (en) | Detection method and detection device for real-time track initiation and track classification | |
CN115547347A (en) | Whale acoustic signal identification method and system based on multi-scale time-frequency feature extraction | |
CN112668443A (en) | Human body posture identification method based on two-channel convolutional neural network | |
Li et al. | A fast detection method for polynomial fitting lane with self-attention module added | |
Gong et al. | Lightcnn: A compact cnn for moving maritime targets detection | |
US11830239B1 (en) | Systems and methods for automatic extraction and alignment of labels derived from camera feed for moving sound sources recorded with a microphone array | |
DE102019009130A1 (en) | Approximating compression method for ultrasonic sensor data | |
CN116908806B (en) | Multi-dimensional feature target detection method based on self-supervised learning |
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 |