An integrated UAV threat assessment method based on multi-sensors and LSTM

Abstract

The present invention provides a comprehensive method for evaluating the threat of drones based on multi-sensors and LSTM, which relates to the technical field of anti-drone prevention. This method utilizes the attribute information such as distance, speed, heading, size, radio frequency, etc. detected by radar, combines the LSTM algorithm to evaluate the threat level of drone targets, and sorts them from high to low threat level to obtain a drone threat level list; according to the sorting result, it guides optoelectronic and electronic reconnaissance equipment to track and detect drone targets; summarizes the target attributes of multi-sensors, comprehensively utilizes the image information detected by optoelectronic detection, the frequency, signal strength, etc. detected by electronic reconnaissance, and the attribute information detected by radar, and uses a multi-layer LSTM model for secondary threat level evaluation and rearranges the threat level list. This method combines multi-sensor and LSTM technologies, can obtain more comprehensive and accurate target attribute information, improve the accuracy of threat assessment, and meet the real-time requirements at the same time, thus providing a more effective means for drone threat prevention.

Description

Technical Field

[0001] The invention relates to the technical field of anti-drone prevention, in particular to a comprehensive method for evaluating drone threats based on multi-sensors and LSTM (Long Short-Term Memory Network). Background Art

[0002] Regarding the problem that it is difficult to evaluate the threat level of drone targets. Traditional drone threat assessment methods usually use a single sensor (such as radar or optoelectronic equipment) for data collection and processing, and then judge the threat level of the drone according to specific rules or algorithms. This method has the following disadvantages:

[0003] 1) Single data: Traditional methods usually use a single sensor to collect data, unable to obtain comprehensive data information from multiple sources and perspectives, making it difficult to conduct accurate threat assessment.

[0004] 2) Simple processing method: Traditional methods use simple rules or algorithms to process data, making it difficult to cope with the diversity and complexity of drone target features, resulting in a large deviation in the evaluation results. Summary of the Invention

[0005] In view of this, the present invention provides a comprehensive method for evaluating drone threats based on multi-sensors and LSTM. This method obtains comprehensive data information from multiple sources and perspectives through multi-sensors, and processes and analyzes the data through the LSTM deep learning model, improving accuracy and real-time performance.

[0006] The technical solution adopted by the present invention is as follows:

[0007] A comprehensive method for evaluating drone threats based on multi-sensors and LSTM, comprising the following steps:

[0008] Step 1: Construct a single-layer LSTM model and a three-layer LSTM model; each layer of the LSTM unit in the three-layer LSTM model consists of a memory unit, an input gate, a forget gate, and an output gate, and the number of hidden layers of each layer of the LSTM unit in the three-layer LSTM model is 64 layers, 32 layers, and 16 layers respectively; the input sequence length of the three-layer LSTM model is 5 time steps, the activation function uses softmax, and the multi-classification loss function is the cross-entropy loss function;

[0009] Step 2: Use radar equipment, optoelectronic equipment, and electronic reconnaissance equipment to detect UAV targets in the same space-time, and obtain information such as the distance, longitude-latitude-altitude position, heading, size, type, image, repetition frequency, carrier frequency, pulse width, and signal strength of the UAV targets; preprocess the collected data to complete data cleaning, invalid data filtering, and data format conversion. Then, divide the data into two parts. One part is the target dataset that only contains the detection data of the radar equipment, and the other part is the target dataset that contains the detection data of the radar equipment, optoelectronic equipment, and electronic reconnaissance equipment. Divide each dataset into a training set, a validation set, and a test set according to the ratio of 8:1:1. Among them, the training set is used to train the model, the validation set is used to adjust the model hyperparameters, and the test set is used to evaluate the performance of the model;

[0010] Step 3: Serialize the attribute information of each UAV target to form a sequence dataset. Set a fixed time window for the sequence dataset every 5 seconds, and each sequence in the time window contains a set of target attribute information; according to the target attribute information, use the radar dataset and the multi-sensor dataset to label each UAV target respectively, and label the threat label of the target;

[0011] Step 4: Use the attributes of the UAV targets detected by the radar to complete the training and evaluation of the single-layer LSTM model based on the mean square error index; in addition, use the attributes of the UAV targets detected by the multi-sensor to complete the training and evaluation of the three-layer LSTM model based on the accuracy, precision, and recall rate indicators;

[0012] Step 5: Collect the target data detected by the radar sensor, including information such as the distance, longitude-latitude-altitude position, heading, size, and type of the target, perform serialization processing on the data, and input the serialized target data detected by the radar into the trained single-layer LSTM model to obtain a rough ranking list of UAV threats;

[0013] Then, according to the rough ranking list of UAV threats, in the order of decreasing threat level, based on the target position detected by the radar, guide the optoelectronic equipment and electronic reconnaissance equipment to detect and confirm the target in turn; the specific method is:

[0014] Collect the target data detected by the radar equipment, optoelectronic equipment, and electronic reconnaissance equipment, perform serialization processing on the data, and input the serialized target data detected by the multi-sensor into the trained three-layer LSTM model to obtain a refined ranking list of UAV threats, and complete the comprehensive evaluation of UAV threats.

[0015] The beneficial effects of the present invention are as follows:

[0016] 1. Comprehensive utilization of multi-sensor data: This method provides more comprehensive and accurate target attribute information by fusing multi-sensor data, thereby enabling more accurate assessment of the threat level of unmanned aerial vehicles (UAVs).

[0017] 2. Real-time assessment based on the LSTM model: This method uses the LSTM model to conduct real-time assessment of UAV targets, meeting the real-time requirements of threat assessment.

[0018] 3. Secondary assessment to improve accuracy: This method guides optoelectronic and electronic reconnaissance equipment to detect targets, and uses a multi-layer LSTM model to evaluate UAV targets based on the detection information, which can further improve the accuracy of threat assessment.

[0019] 4. Scalability: Since this method utilizes multi-sensor data and an LSTM-based model, it can adapt to different types of sensors and the detection of targets with different attributes and types.

[0020] In summary, the present invention can improve the accuracy and comprehensiveness of threat assessment, providing a more effective means for UAV threat prevention. Description of the Drawings

[0021] Figure 1 is a schematic diagram of the principle of a comprehensive UAV threat assessment method based on multi-sensors and LSTM.

[0022] Figure 2 is a structure diagram of a multi-layer LSTM model.

[0023] Figure 3 is the LSTM network structure of the UAV threat comprehensive assessment method. Detailed Embodiment

[0024] The present invention will be described in detail below in conjunction with the drawings and specific embodiments.

[0025] A comprehensive UAV threat assessment method based on multi-sensors and LSTM. This method first uses the attribute information such as distance, speed, heading, size, and radio frequency detected by radar, combines the LSTM algorithm to evaluate the threat level of UAV targets, and sorts them from high to low threat level to obtain a UAV threat level list; then, according to the sorting result, guides optoelectronic and electronic reconnaissance equipment to track and detect UAV targets; finally, summarizes the multi-sensor target attributes, comprehensively utilizes the image information detected by optoelectronic detection, the frequency, signal strength, etc. detected by electronic reconnaissance detection, and the attribute information detected by radar detection, and uses a multi-layer LSTM model for secondary threat level assessment and rearranges the threat level list. This method combines multi-sensor and LSTM technologies, can obtain more comprehensive and accurate target attribute information, improve the accuracy of threat assessment, and meet the real-time requirements at the same time, thereby providing a more effective means for UAV threat prevention.

[0026] As Figure 1 shown, the method specifically includes the following steps:

[0027] Step 1: Model construction

[0028] Construct a three-layer LSTM model with an input sequence length of 5 time steps, using the softmax activation function and the cross-entropy loss function for multi-classification. The number of hidden layers in each layer is 64, 32, and 16 respectively. Each LSTM cell consists of a memory unit, an input gate, a forget gate, and an output gate.

[0029] Step 2: Model training:

[0030] Data collection: Use radar equipment, optoelectronic equipment, and electronic reconnaissance equipment to detect UAV targets in the same time and space, and obtain attribute information such as the distance, position (longitude, latitude, altitude), heading, size, type, image, pulse repetition frequency, carrier frequency, pulse width, and signal strength of the UAV targets.

[0031] Data processing: Preprocess the collected radar data, optoelectronic data, and electronic reconnaissance data, including operations such as data cleaning, invalid data filtering, and data format conversion.

[0032] Dataset production: The dataset is produced in two parts. One part is the target dataset detected by radar, and the other part is the target dataset containing radar, optoelectronic, and electronic reconnaissance detections. The dataset is divided into a 0.8 training set, a 0.1 validation set, and a 0.1 test set. The training set is used to train the model, the validation set is used to adjust the model hyperparameters, and the test set is used to evaluate the performance of the model.

[0033] Serialize data: Serialize the attribute information of each UAV target to form a sequence dataset. A fixed time window is set for the sequence data every 5 seconds, and each sequence contains a set of target attribute information.

[0034] Data annotation: According to the attribute information of the UAV targets, label each UAV target in the radar dataset and the multi-sensor dataset respectively, and label the threat label of the target.

[0035] Use a deep learning model based on LSTM to complete the training and evaluation of the model for the attributes of UAV targets detected by radar based on metrics such as mean square error.

[0036] Use a multi-layer LSTM deep learning model to complete the training and evaluation of the model for the attributes of UAV targets detected by multi-sensors based on metrics such as accuracy, precision, and recall.

[0037] Step 3: Model application:

[0038] Collect the target data detected by the radar sensor, including information such as the distance, position (longitude, latitude, altitude), heading, size, and type of the target. Serialize the data, and use the serialized target data detected by the radar to input into the LSTM model to obtain a rough ranking list of UAV threats.

[0039] According to the rough ranking list of UAV threats, the model user, in the order of decreasing threat level, based on the target position detected by the radar, sequentially guides the optoelectronic and electronic reconnaissance equipment to detect and confirm the target.

[0040] Collect the target data detected by the radar, electronic reconnaissance, and optoelectronic equipment, including information such as target distance, position (longitude, latitude, altitude), heading, size, type, image, pulse repetition frequency, carrier frequency, pulse width, signal strength, etc. Serialize the data, and use the serialized target data detected by multiple sensors to input into the trained three-layer LSTM model to obtain a refined ranking list of UAV threats.

[0041] The three-layer LSTM model in Step 1 is shown in Figure 2 As shown, the number of hidden layers in each layer is 64 layers, 32 layers, and 16 layers respectively, which can reduce the number of model parameters and computational complexity, and can help the model better utilize the features of different layers and improve the generalization ability of the model. The LSTM network structure described in Step 1 is shown in Figure 3 , at time t, a single LSTM contains 3 inputs and 2 outputs, where x t represents the input at time t, h t-1 is the output at time t - 1, c t-1 represents the state of the cell at time t - 1, h t represents the output at time t, and c t represents the state of the cell at time t. The input gate controls the update method of memory, the forget gate controls whether information is retained or forgotten, and the output gate generates an output based on the current input and the state of previous memory. The gating mechanism is implemented through activation functions and element-wise multiplication operations:

[0042] f t = δ(w fh h t-1 + w fx x t + b f ) (1)

[0043] i t = δ(w ih h t-1 + w ix x t + b i ) (2)

[0044]

[0045]

[0046] o t = δ(w oh h t-1 + w ox x t + b o )(5)

[0047]

[0048] In the formula: f t represents the output of the forget gate; i t represents the output of the input gate; is the current input cell state; x t is the input data at time t in the time series; o t represents the output of the output gate; h t-1 is the output result at time t - 1; w fh is the output weight from the forget gate to the cell; w fx is the input weight from the forget gate to the cell; w oh is the weight from the cell to the input gate; w ox represents the weight from the current output gate to the cell output; b f represents the bias of the forget gate; b i represents the bias of the input gate; b o represents the bias of the output gate; δ() refers to the sigmod function; c t represents the cell memory of the cell state.

[0049] In step 2, data collection is carried out at a specific test site. When the radar, electronic reconnaissance, and optoelectronic equipment are powered on, professional pilots fly multiple drones of different models from different directions and distances, and record attribute data such as the track of the target drone detected by multiple sensors.

[0050] In step 3, the rough ranking list of drone threats refers to the result obtained by importing the target data detected by the radar into the LSTM model. Generally, the radar is more sensitive to the position information of the target, but not sensitive to the size, type, model, etc. of the target. Therefore, generally, only the target attributes detected by the radar and the area of the protection zone can be used to roughly rank the threats of drone targets. The guidance of the optoelectronic equipment and electronic reconnaissance equipment according to the target detected by the radar means that the radar transmits the detected target position information to the optoelectronic equipment and electronic reconnaissance equipment in real time; the optoelectronic equipment and electronic reconnaissance equipment aim at the target position by rotating the camera platform. The target is found and tracked through a certain search strategy, and the attribute information of the complete target is determined through image recognition and library matching of the target.

[0051] This method adopts the combined technology of multi-sensors and LSTM deep learning model. The LSTM model can make full use of past information and is outstanding in solving sequence modeling. At the same time, this method deeply studies the diversity and complexity of the characteristics of UAV targets, and provides a ranking algorithm to rank the threat levels of targets, which can more accurately evaluate UAV threats and provide a more effective means for UAV threat prevention. The present invention can solve the problem of real-time and accurate evaluation of UAV threat levels by anti-UAV command and control software, and realize intelligent, real-time and accurate perception of UAV threats.

Claims

1. An integrated UAV threat assessment method based on multi-sensors and LSTM, characterized in that, The steps are as follows: Step 1: Construct a single-layer LSTM model and a three-layer LSTM model; each layer of LSTM units in the three-layer LSTM model consists of a memory unit, an input gate, a forget gate, and an output gate, and the number of hidden layers of each layer of LSTM units is 64 layers, 32 layers, and 16 layers respectively; the input sequence length of the three-layer LSTM model is 5 time steps, the activation function is softmax, and the multi-classification loss function is the cross-entropy loss function; Step 2: Use radar equipment, optoelectronic equipment, and electronic reconnaissance equipment to detect UAV targets in the same space-time, and obtain information on the distance, longitude-latitude-altitude position, heading, size, type, image, pulse repetition frequency, carrier frequency, pulse width, and signal strength of the UAV targets; preprocess the collected data to complete data cleaning, filtering of invalid data, and data format conversion. Then, divide the data into two parts, one is the target data set containing only the detection data of the radar equipment, and the other is the target data set containing the detection data of the radar equipment, optoelectronic equipment, and electronic reconnaissance equipment. Each data set is divided into a training set, a validation set, and a test set according to the ratio of 8:1:

1. Among them, the training set is used to train the model, the validation set is used to adjust the model hyperparameters, and the test set is used to evaluate the performance of the model; Step 3: Serialize the attribute information of each UAV target to form a sequence data set. Set a fixed time window for the sequence data set every 5 seconds, and each sequence in the time window contains a set of target attribute information; according to the target attribute information, use the radar data set and the multi-sensor data set to label each UAV target respectively, and label the threat label of the target; Step 4: Use the UAV target attributes detected by radar to complete the training and evaluation of the single-layer LSTM model based on the mean square error index; in addition, use the UAV target attributes detected by multi-sensors to complete the training and evaluation of the three-layer LSTM model based on the accuracy, precision, and recall rate indexes; Step 5: Collect the target data detected by the radar sensor, including information on the distance, longitude-latitude-altitude position, heading, size, and type of the target, perform serialization processing on the data, and input the serialized target data detected by the radar into the trained single-layer LSTM model to obtain a rough ranking list of UAV threats; Then, according to the rough ranking list of UAV threats, in the order of decreasing threat level, based on the target position detected by the radar, guide the optoelectronic equipment and electronic reconnaissance equipment to detect and confirm the target in turn; the specific method is as follows: Collect the target data detected by the radar equipment, optoelectronic equipment, and electronic reconnaissance equipment, perform serialization processing on the data, and input the serialized target data detected by the multi-sensors into the trained three-layer LSTM model to obtain a refined ranking list of UAV threats, and complete the comprehensive evaluation of UAV threats.

Images