CN110766140A - XGboost and LSTM-based multi-sensor real-time threat level classification method - Google Patents
XGboost and LSTM-based multi-sensor real-time threat level classification method Download PDFInfo
- Publication number
- CN110766140A CN110766140A CN201911004922.9A CN201911004922A CN110766140A CN 110766140 A CN110766140 A CN 110766140A CN 201911004922 A CN201911004922 A CN 201911004922A CN 110766140 A CN110766140 A CN 110766140A
- Authority
- CN
- China
- Prior art keywords
- threat level
- xgboost
- lstm
- network model
- real
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
- G01D21/00—Measuring or testing not otherwise provided for
- G01D21/02—Measuring two or more variables by means not covered by a single other subclass
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
Abstract
The invention discloses a XGboost and LSTM-based multi-sensor real-time threat level classification method, which relates to the technical field of radar sensor networks and comprises the following steps: 1. training an XGboost model and an LSTM network model; 2. acquiring moving target signals in real time by using multiple sensors, and transmitting the signals acquired by the sensors to an information fusion center; 3. inputting a signal acquired by an information fusion center as a feature into an XGboost model, and outputting an initial threat level category of a moving target by the XGboost model; 4. preprocessing the initial threat level type output by the XGboost model, and transmitting the preprocessing result to the LSTM network model as an initial external state; 5. transmitting a signal acquired by an information fusion center to an LSTM network model as a current input sequence; and 6, outputting the real-time threat level category of the moving target by the LSTM network model based on the initial external state and the current input sequence, and inputting the output real-time threat level category into the LSTM network model as the initial external state at the next moment.
Description
Technical Field
The invention relates to the technical field of radar sensor networks, in particular to a XGboost and LSTM-based multi-sensor real-time threat level classification method.
Background
For the areas needing to evaluate the situation of a battlefield and to use a multi-sensor for warning, the actions taken by the same target at different moments correspond to different threat levels, and how to judge the threat level of a certain target in real time is a current research direction. In complex environments, it is difficult for a single sensor to acquire complete information of the environment, and a wireless sensor network is usually used to acquire better quality and more comprehensive perception information. However, the wireless sensor network has a remarkable multisource heterogeneous characteristic, that is, information collected among different types of sensors is not uniform in time-space information among the different sensors due to reasons such as network node distribution, the characteristics of the sensors themselves, and unstable network transmission, and is difficult to be well applied.
At present, methods such as kalman filtering, Dempster-Shafer evidence reasoning, multi-bayes estimation, fuzzy logic, neural network and the like are generally adopted to perform feature extraction and target identification on information obtained by a plurality of sensors, and establish preliminary judgment and conclusion on the same target. And voting is carried out by the information fusion center to judge the final threat level of the target.
Although, the above method has excellent performance in multi-source information fusion, and can achieve more accurate single threat level judgment. However, none of the above methods introduces a relationship between time and situation, i.e., the position and state of the current target at the previous time are not considered. If the object is a variable-speed motion or the motion direction is suddenly changed at a certain moment, since the prior information used by the method is still the initial information, some threat objects can be ignored, and danger is caused. The method cannot evaluate the current state for a long time according to the state of the target at the previous moment, and has the problems of selection, duplicate removal, combination, optimization and the like.
Disclosure of Invention
The invention aims to: in order to solve the problems that the existing sensor threat level classification method uses target initial information to classify the threat level of a target for a long time, so that part of threat targets are omitted, and the classification effect is poor, a multi-sensor real-time threat level classification method based on XGboost and LSTM is provided. The method includes the steps of simply fusing information initially acquired by multiple sensors through an eXtreme Gradient Boosting (XGboost) algorithm to obtain the prior condition of the threat level of a target, serializing subsequent real-time signals, and judging the threat level of the target in real time by adopting a Long Short-Term Memory (LSTM) method.
The technical scheme adopted by the invention is as follows:
a multi-sensor real-time threat level classification method based on XGboost and LSTM comprises the following steps:
step 1: training an XGboost model and an LSTM network model;
step 2: acquiring moving target signals in real time by using multiple sensors, and transmitting the signals acquired by the sensors to an information fusion center;
and step 3: inputting a signal acquired by an information fusion center as a feature into an XGboost model, and outputting an initial threat level category of a moving target by the XGboost model;
and 4, step 4: preprocessing the initial threat level type output by the XGboost model, and transmitting the preprocessing result to the LSTM network model as an initial external state;
and 5: transmitting a signal acquired by an information fusion center to an LSTM network model as a current input sequence;
step 6: the LSTM network model outputs the real-time threat level category of the moving target based on the initial external state and the current input sequence, and simultaneously inputs the output real-time threat level category into the LSTM network model as the initial external state at the next moment;
and 7: and (5) repeating the steps 2, 5 and 6 in sequence until the classification is finished.
Further, the sensor for acquiring the signal of the moving target includes a radar sensor, an infrared sensor, and an ultrasonic sensor.
Further, the signal of the moving object includes a spatial position, a moving speed, and a moving direction of the moving object.
Further, in step 4, the specific method for preprocessing the initial threat level category is as follows: and vectorizing the initial threat level category by utilizing one-hot coding.
Further, the XGboost model is trained by using the existing big data information, and the LSTM network model is trained by using the existing big data information. The big data information refers to a large number of samples containing features and labels, the feature part contains signals or other information (such as speed and pitch angle) of the moving target acquired by multiple sensors, and the label part is the decision level of the target under the features.
In summary, due to the adoption of the technical scheme, the invention has the beneficial effects that:
1. in the invention, a more accurate prior probability is provided by using a machine learning model XGboost and combining a large amount of historical data, and real-time judgment is carried out by using an LSTM network model in deep learning, so that the problem that the calculation quantity increases exponentially along with the increase of time is avoided. The method combines machine learning, deep learning and big data, and better obtains the classification effect by learning the conditions under other similar environments.
2. In the method, the influence of the current previous moment state on the current moment is considered, and the prior information and the real-time information obtained by the multiple sensors are used for judging the threat of the target in real time, so that better monitoring effect and classification accuracy are obtained, and the aim of classifying the threat level of the target in real time by using the threat level of the target at the previous moment and the information obtained by the multiple sensors at the current moment for a long time is fulfilled.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
FIG. 1 is an overall flow chart of the method of the present invention;
FIG. 2 is a detailed flow chart of the method of the present invention;
fig. 3 is a diagram of the LSTM principle of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the detailed description and specific examples, while indicating the preferred embodiment of the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.
It is noted that relational terms such as "first" and "second," and the like, may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The features and properties of the present invention are described in further detail below with reference to examples.
Example one
The preferred embodiment of the present invention provides a XGBoost and LSTM-based multi-sensor real-time threat level classification method, as shown in fig. 1 and 2, comprising the following steps:
step 1: and training the XGboost model and the LSTM network model. In this embodiment, the XGBoost model is trained using the existing big data information, and the LSTM network model is trained using the existing big data information.
The XGBoost is an ensemble learning algorithm, generally using a decision tree as a base learner, and summing the results of a plurality of base learners to obtain a final result. Compared with the traditional ensemble learning algorithm, such as GBDT, XGboost adds a regular term in the objective function to prevent overfitting. And a first derivative and a second derivative are used as pseudo-residual errors to train each tree. When the tree is trained, a new mode of finding the best segmentation point is also adopted.
The formula is as follows:
wherein i represents the sequence number of the decision tree, M represents the total number of the decision tree, x represents the signal acquired by the sensor, fiDecision function, f, representing the ith decision treei(x) Represents the decision result of the i-th decision tree, and f (x) represents the initial threat level category.
During the training process, the loss function is as follows:
wherein, L is a loss function, theta is a parameter of the XGboost model, L represents the loss of each sample and can be a mean square error, yiA label representing the ith training sample,
representing the predicted result of the ith sample after the model. Omega (f)k) A term of the regularization is represented,
gamma and lambda are self-defined constants with values less than 1, | T | represents the number of leaves,
the leaf score is expressed as a two-norm sum.
For the t-th tree, the penalty function is:
wherein f istRepresenting the complexity of the t-th tree.
And performing second-order Taylor expansion on the above formula to obtain:
wherein the content of the first and second substances,denotes derivation, giIs a first derivative, hiIs a second derivative, xiThe ith sample is represented.
Unlike conventional GBDTs, the best partitioning point for each tree in XGBoost is not measured by the "minimum mean square error", but rather is determined by the following function:
wherein, ILRepresenting the leaf node of the left subtree, IRRepresenting the leaf nodes of the right subtree and I representing all the leaf nodes.
The LSTM is a deep learning structure, and is an algorithm invented for solving the problem that the traditional RNN cannot have long-time memory capacity due to gradient disappearance. The principle of LSTM is shown in FIG. 3, which is composed of an input gate itForgetting to leave the door ftAnd an output gate otThree doors. In addition, it introduces a new internal state ctThe method is specially used for linear circulation information transmission, and nonlinear output information is simultaneously transmitted to an external state h of a hidden layertThe formula is as follows:
it＝σ(Wi·[xt,ht-1]+bi)
ft＝σ(Wf·[xt,ht-1]+bf)
ot＝σ(Wo·[xt,ht-1]+bo)
where σ represents a sigmoid function, and the output interval is (0, 1). x is the number of0Is a threat level vector after one-hot coding, h0And c0Random initialization; x is the number of0、h0、c0Respectively represent xt、ht、ctWhen the time t is 0, the initial value is set. x is the number oftRepresenting the input at time t, i.e. the target signal captured by the sensor at time t, Wc、bc、bi、Wf、bf、Wo、boAre parameters of the neural network. f. oftControlling the internal state c of the previous momentt-1How much information needs to be forgotten; i.e. itCandidate state c E C E CtHow much information needs to be saved; otControlling the internal state c at the present momenttHow much information needs to be output to the external state ht。
Through training of a large amount of data, the LSTM model will be fixed. Specifically, W and b in the model are both fixed to a certain value, thereby controlling the threshold values of the forgetting gate, the input gate and the output gate. The forgetting door can forget some useless information and information with overlong time at the previous moment, and the problem that the time exponentially rises in the operation process is solved.
Step 2: and acquiring moving target signals in real time by using multiple sensors, and transmitting the signals acquired by the sensors to an information fusion center. In this embodiment, the sensor for acquiring the signal of the moving target includes a radar sensor, an infrared sensor, and an ultrasonic sensor, and the signal of the moving target includes a spatial position, a moving speed, and a moving direction of the moving target.
And step 3: and inputting the signals acquired by the information fusion center into the XGboost model as features, and outputting the initial threat level category of the moving target by the XGboost model.
And 4, step 4: and vectorizing the initial threat level type output by the XGboost model by utilizing one-hot coding, and transmitting a processing result as an initial external state to the LSTM network model.
And 5: and transmitting the signals acquired by the information fusion center to the LSTM network model as the current input sequence.
Step 6: the LSTM network model outputs the real-time threat level category of the moving target based on the initial external state and the current input sequence, and simultaneously inputs the output real-time threat level category into the LSTM network model as the initial external state at the next moment, thereby realizing the classification real-time property.
And 7: and (5) repeating the steps 2, 5 and 6 in sequence until the classification is finished.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (5)
1. A multi-sensor real-time threat level classification method based on XGboost and LSTM is characterized by comprising the following steps:
step 1: training an XGboost model and an LSTM network model;
step 2: acquiring moving target signals in real time by using multiple sensors, and transmitting the signals acquired by the sensors to an information fusion center;
and step 3: inputting a signal acquired by an information fusion center as a feature into an XGboost model, and outputting an initial threat level category of a moving target by the XGboost model;
and 4, step 4: preprocessing the initial threat level type output by the XGboost model, and transmitting the preprocessing result to the LSTM network model as an initial external state;
and 5: transmitting a signal acquired by an information fusion center to an LSTM network model as a current input sequence;
step 6: the LSTM network model outputs the real-time threat level category of the moving target based on the initial external state and the current input sequence, and simultaneously inputs the output real-time threat level category into the LSTM network model as the initial external state at the next moment;
and 7: and (5) repeating the steps 2, 5 and 6 in sequence until the classification is finished.
2. The XGboost and LSTM-based multi-sensor real-time threat level classification method according to claim 1, wherein the sensors for acquiring signals of moving objects comprise radar sensors, infrared sensors and ultrasonic sensors.
3. The XGboost and LSTM-based multi-sensor real-time threat level classification method according to claim 1, wherein the signal of the moving object comprises a spatial position, a moving speed and a moving direction of the moving object.
4. The XGboost and LSTM-based multi-sensor real-time threat level classification method according to claim 1, wherein in step 4, the specific method for preprocessing the initial threat level category is as follows: and vectorizing the initial threat level category by utilizing one-hot coding.
5. The XGboost and LSTM-based multi-sensor real-time threat level classification method as claimed in claim 1, wherein the XGboost model is trained by using existing big data information, and the LSTM network model is trained by using existing big data information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911004922.9A CN110766140A (en) | 2019-10-22 | 2019-10-22 | XGboost and LSTM-based multi-sensor real-time threat level classification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911004922.9A CN110766140A (en) | 2019-10-22 | 2019-10-22 | XGboost and LSTM-based multi-sensor real-time threat level classification method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110766140A true CN110766140A (en) | 2020-02-07 |
Family
ID=69332531
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911004922.9A Pending CN110766140A (en) | 2019-10-22 | 2019-10-22 | XGboost and LSTM-based multi-sensor real-time threat level classification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110766140A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112036556B (en) * | 2020-11-06 | 2021-01-26 | 西南交通大学 | Target intention inversion method based on LSTM neural network |
| WO2022041225A1 (en) * | 2020-08-31 | 2022-03-03 | 中山大学 | Server for use in assessing cardiovascular state, wearable device, and method for cardiovascular state assessment |
| CN116050515A (en) * | 2022-12-30 | 2023-05-02 | 中国电子科技集团公司信息科学研究院 | XGBoost-based parallel deduction multi-branch situation prediction method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170206464A1 (en) * | 2016-01-14 | 2017-07-20 | Preferred Networks, Inc. | Time series data adaptation and sensor fusion systems, methods, and apparatus |
| CN108510113A (en) * | 2018-03-21 | 2018-09-07 | 中南大学 | A kind of application of XGBoost in short-term load forecasting |
| CN109890002A (en) * | 2019-01-04 | 2019-06-14 | 中国电子科技集团公司电子科学研究院 | A kind of method and device of the prediction Risk Degree of Maneuvering Environment of knowledge-based inference |
| CN109919685A (en) * | 2019-03-18 | 2019-06-21 | 苏州大学 | Customer churn prediction method, apparatus, equipment and computer readable storage medium |
| CN110197282A (en) * | 2019-06-10 | 2019-09-03 | 电子科技大学 | A kind of threat estimating and method for situation assessment based on Genetic-fuzzy logic tree |
-
2019
- 2019-10-22 CN CN201911004922.9A patent/CN110766140A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170206464A1 (en) * | 2016-01-14 | 2017-07-20 | Preferred Networks, Inc. | Time series data adaptation and sensor fusion systems, methods, and apparatus |
| CN108510113A (en) * | 2018-03-21 | 2018-09-07 | 中南大学 | A kind of application of XGBoost in short-term load forecasting |
| CN109890002A (en) * | 2019-01-04 | 2019-06-14 | 中国电子科技集团公司电子科学研究院 | A kind of method and device of the prediction Risk Degree of Maneuvering Environment of knowledge-based inference |
| CN109919685A (en) * | 2019-03-18 | 2019-06-21 | 苏州大学 | Customer churn prediction method, apparatus, equipment and computer readable storage medium |
| CN110197282A (en) * | 2019-06-10 | 2019-09-03 | 电子科技大学 | A kind of threat estimating and method for situation assessment based on Genetic-fuzzy logic tree |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022041225A1 (en) * | 2020-08-31 | 2022-03-03 | 中山大学 | Server for use in assessing cardiovascular state, wearable device, and method for cardiovascular state assessment |
| CN112036556B (en) * | 2020-11-06 | 2021-01-26 | 西南交通大学 | Target intention inversion method based on LSTM neural network |
| CN116050515A (en) * | 2022-12-30 | 2023-05-02 | 中国电子科技集团公司信息科学研究院 | XGBoost-based parallel deduction multi-branch situation prediction method |
| CN116050515B (en) * | 2022-12-30 | 2023-11-07 | 中国电子科技集团公司信息科学研究院 | XGBoost-based parallel deduction multi-branch situation prediction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110826602B (en) | Image classification method and system based on membrane potential regulation and control pulse neural network | |
| CN108416394A (en) | Multi-target detection model building method based on convolutional neural networks | |
| CN110766140A (en) | XGboost and LSTM-based multi-sensor real-time threat level classification method | |
| CN112434643A (en) | Classification and identification method for low-slow small targets | |
| CN113408392B (en) | Flight path completion method based on Kalman filtering and neural network | |
| Venskus et al. | Unsupervised marine vessel trajectory prediction using LSTM network and wild bootstrapping techniques | |
| US20200125930A1 (en) | Artificial neural network and method of training an artificial neural network with epigenetic neurogenesis | |
| CN111382686A (en) | Lane line detection method based on semi-supervised generation confrontation network | |
| CN114155270A (en) | Pedestrian trajectory prediction method, device, equipment and storage medium | |
| CN115731411A (en) | Small sample image classification method based on prototype generation | |
| Bækkegaard et al. | Target classification using kinematic data and a recurrent neural network | |
| Balasubramanian et al. | Open-set recognition based on the combination of deep learning and ensemble method for detecting unknown traffic scenarios | |
| CN115130357A (en) | GRU-based air target combat intention prediction system and method | |
| Murino | Structured neural networks for pattern recognition | |
| CN109887004A (en) | A kind of unmanned boat sea area method for tracking target based on TLD algorithm | |
| Vuković et al. | Thermal image degradation influence on R-CNN face detection performance | |
| CN113406623A (en) | Target identification method, device and medium based on radar high-resolution range profile | |
| CN113129336A (en) | End-to-end multi-vehicle tracking method, system and computer readable medium | |
| CN110111358B (en) | Target tracking method based on multilayer time sequence filtering | |
| CN115761240B (en) | Image semantic segmentation method and device for chaotic back propagation graph neural network | |
| CN117195031A (en) | Electromagnetic radiation source individual identification method based on neural network and knowledge-graph dual-channel system | |
| CN115310559A (en) | Target identification method based on Bayes-Transformer neural network model | |
| CN114943016A (en) | Cross-granularity joint training-based graph comparison representation learning method and system | |
| CN113989560A (en) | Online semi-supervised learning classifier for radar gesture recognition and classification method thereof | |
| Wei et al. | Accretionary learning with deep neural networks |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200207 |
|
| RJ01 | Rejection of invention patent application after publication |