An aircraft classification method and system integrating multi-source features

Abstract

The present invention discloses an aircraft classification method and system integrating multi-source features. The aircraft classification method includes the following steps: receiving aircraft data acquired by multiple sensors; determining the type of the aircraft data; if the aircraft data is single-source data, directly preprocessing the aircraft data and performing feature extraction to obtain single-source features; if the aircraft data is multi-source data, respectively preprocessing each type of aircraft data and respectively performing feature extraction on each type of aircraft data to obtain comprehensive features; and inputting the single-source features or comprehensive features into a multi-layer perceptron network to predict the category of the aircraft. The present invention comprehensively considers various feature information, can improve the accuracy and classification effect of aircraft classification, and improves the adaptability of the aircraft classification system.

Description

Technical Field

[0001] The present invention relates to a method for classifying aircraft by fusing multi-source features, and also relates to a corresponding aircraft classification system, belonging to the technical field of aviation management. Background Art

[0002] Currently, most aircraft classification models are based on visible light images and mainly use deep learning technology. Among them, the most classic method is to use a convolutional neural network (CNN) or a bidirectional long short-term memory network (Bi-LSTM) to extract features from visible light images, and then input the high-dimensional features into a multi-layer perceptron (MLP) for aircraft classification.

[0003] In the paper "Convolutional Neural Network Method for Aircraft Target Classification in Remote Sensing Images" published by Zhou Min (Journal of Image and Graphics, Vol. 22, No. 5, 2017), the author designed a 5-layer convolutional neural network and achieved an accuracy of 97.2% on 8 types of aircraft data. However, considering factors such as the speed and altitude of the aircraft during actual flight, it is very difficult to capture visible light images, and the visible light aircraft classification model cannot be fully trained.

[0004] On the other hand, in the Chinese invention patent with the patent number ZL 201910484690.5, a method for classifying aircraft using flight tracks is disclosed, including: Step 1, processing the original data of various types of aircraft to obtain a visualized flight track image; Step 2, marking and distinguishing the aircraft types in the visualized flight track image, and using the visualized flight track image to train a convolutional neural network to generate a convolutional neural network model for prediction; Step 3, converting the flight track data to be classified into an image, and using the convolutional neural network model to determine the aircraft type. This method realizes flight track classification based on deep learning, can automatically complete feature extraction and weight assignment, and finally achieves the effect of target recognition.

[0005] However, although the current various technical solutions have achieved certain effects, the research is relatively scattered, the correlation between features and features has not been studied, information is prone to be missing, and the classification effect is not good. Summary of the Invention

[0006] The primary technical problem to be solved by the present invention is to provide a method for classifying aircraft by fusing multi-source features to improve the accuracy and classification effect of aircraft classification.

[0007] Another technical problem to be solved by the present invention is to provide an aircraft classification system that fuses multi-source features.

[0008] To achieve the above technical objectives, the present invention adopts the following technical solutions:

[0009] According to the first aspect of the embodiments of the present invention, a method for classifying aircraft by fusing multi-source features is provided, including the following steps:

[0010] Receiving aircraft data acquired by multiple sensors;

[0011] Judging the type of the aircraft data;

[0012] If the aircraft data is single-source data, directly preprocess the aircraft data and perform feature extraction to obtain single-source features; if the aircraft data is multi-source data, preprocess each type of aircraft data separately and perform feature extraction on each type of aircraft data separately to obtain comprehensive features;

[0013] Inputting the single-source features or the comprehensive features into a multi-layer perceptron network to predict the category of the aircraft.

[0014] Preferably, the aircraft data at least includes: visible light images, flight trajectories, and electromagnetic radiation.

[0015] Preferably, the preprocessing of the aircraft data includes:

[0016] Visible light image preprocessing: flipping and angle rotating the visible light image to perform data enhancement on the visible light image; selecting the maximum inscribed rectangle of the visible light image; for the enhanced visible light image data, cropping visible light images of different sizes into visible light images of a unified size;

[0017] Flight trajectory preprocessing: selecting flight data within a set time; converting the flight data into longitude coordinates, latitude coordinates, and altitude coordinates; calculating the curvature and torsion corresponding to each sampling moment using a trajectory equation; performing feature splicing on the longitude coordinates, latitude coordinates, altitude coordinates, curvature, and torsion;

[0018] Electromagnetic radiation preprocessing: selecting radar echo information within a set time; performing sparse recovery and normalization processing on the radar echo information using the wavelet transform method and the mean variance method.

[0019] Preferably, the feature extraction of the aircraft data includes:

[0020] Using a first feature extractor to perform feature extraction on visible light images, where the first feature extractor at least includes a Vision Transformer model;

[0021] Using a second feature extractor to perform feature extraction on flight trajectories and electromagnetic radiation, where the second feature extractor at least includes an IDCNN model.

[0022] Preferably, the comprehensive feature is obtained in the following manner:

[0023] After extracting features from each type of aircraft data respectively, it is determined whether feature fusion is required;

[0024] If it is required, the weight ratio of different types of features is calculated according to a cross-feature gated network, and based on the weight ratio of different types of features, the comprehensive feature is obtained by weighting different types of features; if it is not required, the different types of features are directly added to obtain the comprehensive feature.

[0025] Preferably, the calculating the weight ratio of different types of features according to a cross-feature gated network, and based on the weight ratio of different types of features, obtaining the comprehensive feature by weighting different types of features specifically includes:

[0026] Performing weighted summation on the visible light image feature and the flight trajectory feature to obtain a first weighted feature;

[0027] Performing weighted summation on the flight trajectory feature and the electromagnetic radiation feature to obtain a second weighted feature;

[0028] Performing feature splicing on the first weighted feature and the second weighted feature to obtain the comprehensive feature.

[0029] Preferably, the performing weighted summation on the visible light image feature and the flight trajectory feature to obtain a first weighted feature specifically includes:

[0030] Performing a linear transformation on the visible light image feature c i to obtain c i_linear such that the feature dimension of the c i_linear is the same as the feature dimension of the flight trajectory feature h i ;

[0031] Using the Sigmoid function, performing weighted summation through the following formula to obtain the first weighted feature C i ;

[0032] α = sigmoid(W c ·c i_linear +U c ·h i )

[0033] C i = α·c i +(I - α)·h i

[0034] where, I represents the identity matrix; · represents element-wise multiplication; α represents the feature weight after normalization by the Sigmoid function; Wc and U c represent weight matrices, which are obtained by the model during the training process.

[0035] Preferably, the weighted summation of the flight trajectory features and the electromagnetic radiation features to obtain the second weighted feature specifically includes:[[]]

[0036] Using the Sigmoid function, the second weighted feature H is obtained by weighted summation through the following formula i ;

[0037] β = sigmoid(W h ·h i +U h ·u i )

[0038] H i = α·h i +(I - α)·u i

[0039] where, I represents the identity matrix; · represents element-wise multiplication; h i represents the flight trajectory feature; u i represents the electromagnetic radiation feature; β represents the feature weight after Sigmoid normalization; W h and U h represent weight matrices, which are obtained by the model during the training process.

[0040] Preferably, the inputting the single-source feature or the comprehensive feature into a multi-layer perceptron network to predict the category of the aircraft specifically includes:[[]]

[0041] Inputting the single-source feature or the comprehensive feature into a multi-layer perceptron network for dimensionality reduction;

[0042] Performing probability value prediction through the softmax function;

[0043] Taking the category corresponding to the maximum probability as the final predicted category of the aircraft.

[0044] According to the second aspect of the embodiments of the present invention, there is provided an aircraft classification system integrating multi-source features, including a processor and a memory. The processor reads a computer program in the memory and is used to perform the following operations:

[0045] Receiving aircraft data acquired by multiple sensors;

[0046] Judging the type of the aircraft data;

[0047] If the aircraft data is single-source data, directly preprocess the aircraft data and perform feature extraction to obtain single-source features; if the aircraft data is multi-source data, preprocess each type of aircraft data separately and perform feature extraction on each type of aircraft data separately to obtain comprehensive features;

[0048] Input the single-source features or the comprehensive features into a multi-layer perceptron network to predict the category of the aircraft.

[0049] Compared with the prior art, the aircraft classification method and system for fusing multi-source features provided by the embodiments of the present invention comprehensively consider various feature information, and considering that the importance degrees of different features are different in different scenarios, fuse multiple features through a gated network to obtain comprehensive features, and finally apply them to the aircraft classification task in the real scenario. Thereby, the accuracy and classification effect of the classification method are improved. In addition, in practical applications, even if one or several of the features are missing in this aircraft classification system, as long as one or more features of the aircraft can be input, the system can work normally, thereby enhancing the robustness of the entire aircraft classification system. Description of the Drawings

[0050] Figure 1 It is the overall flowchart of an aircraft classification method for fusing multi-source features provided by the embodiments of the present invention;

[0051] Figure 2 It is the specific flowchart of an aircraft classification method for fusing multi-source features provided by the embodiments of the present invention;

[0052] Figure 3 It is the schematic diagram of using the Vision Transformer model to extract features from image data in the embodiments of the present invention;

[0053] Figure 4 It is the schematic diagram of using the IDCNN model to extract features from sequence data in the embodiments of the present invention;

[0054] Figure 5 It is the schematic diagram of the feature fusion process in the embodiments of the present invention;

[0055] Figure 6 It is the structural schematic diagram of an aircraft classification system for fusing multi-source features provided by the embodiments of the present invention. Detailed Embodiments

[0056] The technical content of the present invention will be described in detail below with reference to the drawings and specific embodiments.

[0057] As Figure 1 and Figure 2As shown in the figure, a method for classifying aircraft by integrating multi-source features provided by an embodiment of the present invention specifically includes the following steps:

[0058] S1: Receive aircraft data obtained by multiple sensors.

[0059] In the embodiment of the present invention, the aircraft data at least includes: visible light images, flight trajectories, and electromagnetic radiation. Specifically, visible light image information of the aircraft during flight is obtained through an image sensor; the specific position of the aircraft at each set time interval during flight is obtained through a displacement sensor, so as to form flight trajectory information of the aircraft; electromagnetic signals emitted by the aircraft during flight are received through a signal sensor to obtain electromagnetic radiation information of the aircraft.

[0060] It can be understood that the aircraft data is not limited to the three types listed in the above embodiments. In other embodiments, it can be increased or decreased according to needs. At the same time, the types of each sensor can also be adaptively selected according to needs, as long as the required aircraft data can be obtained.

[0061] S2: Determine the type of the aircraft data.

[0062] Specifically, although the aircraft data in the embodiment of the present invention at least includes three types of information: visible light images, flight trajectories, and electromagnetic radiation, due to reasons such as climatic conditions, sensor failures, terrain, and communication, the information transmitted by the sensors may be missing. Therefore, after the required aircraft data is obtained, it is necessary to determine the type of the aircraft data to confirm whether the aircraft data is single-source data or multi-source data.

[0063] Among them, single-source data refers to that the aircraft data only includes one type of information. For example, the aircraft data only includes visible light images, or only includes flight trajectories, or only includes electromagnetic radiation. Multi-source data refers to that the aircraft data includes more than one type of information. For example, the aircraft data includes two types of information: visible light images and flight trajectories, or visible light images and electromagnetic radiation, or three types of information: visible light images, flight trajectories, and electromagnetic radiation.

[0064] S3: If the aircraft data is single-source data, directly preprocess the aircraft data and perform feature extraction to obtain single-source features.

[0065] After determining that the aircraft data is single-source data, directly preprocess the information included in the aircraft data, and then perform feature extraction on the preprocessed aircraft data to obtain single-source features for subsequent prediction of aircraft categories.

[0066] Among them, it can be understood that the preprocessing of information in the embodiments of the present invention includes: visible light image preprocessing, flight trajectory preprocessing, and electromagnetic radiation preprocessing. Which specific preprocessing among the three is required is determined by the information specifically included in the single-source data. For example: if the aircraft data only includes visible light images, visible light image preprocessing is required; if the aircraft data only includes flight trajectories, flight trajectory preprocessing is required.

[0067] S4: If the aircraft data is multi-source data, preprocess each type of aircraft data separately, and extract features from each type of aircraft data separately to obtain comprehensive features.

[0068] After determining that the aircraft data is multi-source data, it is necessary to preprocess each type of information separately, and then extract features from each preprocessed type of information separately, so as to obtain comprehensive features based on the features extracted from each type of information for subsequent prediction of the aircraft category.

[0069] Specifically, it includes steps S41 to S43:

[0070] S41: After extracting features from each type of aircraft data separately, determine whether feature fusion is required;

[0071] S42: If required, calculate the weight ratio of different types of features according to the gating network across features, and based on the weight ratio of different types of features, obtain comprehensive features by weighting different types of features;

[0072] S43: If not required, directly add different types of features to obtain comprehensive features.

[0073] It should be noted that the division of steps S3 and S4 above is only for the convenience of description. In the specific implementation process of the present invention, there is no sequence between steps S3 and S4. Either step S3 can be executed first and then step S4, or step S4 can be executed first and then step S3.

[0074] S5: Input the single-source feature or comprehensive feature into a multi-layer perceptron network to predict the category of the aircraft.

[0075] Specifically, it includes steps S51 to S53:

[0076] S51: After obtaining the single-source feature or comprehensive feature from the aircraft data, input the single-source feature or comprehensive feature into a multi-layer perceptron network for dimensionality reduction;

[0077] S52: Perform probability value prediction through the softmax function;

[0078] S53: Take the category corresponding to the maximum probability as the final predicted category of the aircraft.

[0079] The preprocessing process, feature extraction process, and feature fusion process of the aircraft data are described in detail below:

[0080] I. Preprocessing process

[0081] The preprocessing process in the embodiments of the present invention includes: visible light image preprocessing, flight trajectory preprocessing, and electromagnetic radiation preprocessing.

[0082] (1) Visible light image preprocessing

[0083] A. Flip and rotate the visible light image to enhance the data of the visible light image.

[0084] B. Select the maximum inscribed rectangle of the visible light image.

[0085] C. For the enhanced visible light image data, crop the visible light images of different sizes into visible light images of the same size.

[0086] (2) Flight trajectory preprocessing

[0087] A. Select the flight data within a set time (for example: the most recent one hour or two hours).

[0088] B. Convert the flight data into longitude coordinates, latitude coordinates, and altitude coordinates.

[0089] C. Calculate the curvature and torsion corresponding to each sampling moment using the trajectory equation.

[0090] D. Perform feature splicing on the longitude coordinates, latitude coordinates, altitude coordinates, curvature, and torsion.

[0091] (3) Electromagnetic radiation preprocessing

[0092] A. Select the radar echo information within a set time (for example: the most recent one hour or two hours).

[0093] B. Use the wavelet transform method and the mean variance method to perform sparse recovery and normalization processing on the radar echo information.

[0094] II. Feature extraction process

[0095] In the embodiments of the present invention, two feature extractors are used to extract features of image - type data and sequence - type data respectively. Among them, the feature extraction of image - type data corresponds to visible light image information, and the feature extraction of sequence - type data corresponds to flight trajectory information and electromagnetic radiation information.

[0096] (1) Feature extraction of image - type data

[0097] In an embodiment of the present invention, a first feature extractor is used to extract features from visible light images, and the first feature extractor includes at least a Vision Transformer model.

[0098] Different from the convolutional neural network for image model extraction mostly used currently, in the embodiment of the present invention, a more advanced Vision Transformer model is used for image feature extraction. For further description of the Vision Transformer model, please refer to the following link: https: / / github.com / google-research / vision_ transformer , which will not be elaborated here.

[0099] In an embodiment of the present invention, the Vision Transformer model used is as Figure 3 shown. By splitting the image into small patches and providing a sequence of linear embeddings of these small patches as the input of the Vision Transformer model, the model is trained for image classification in a supervised manner. Due to the superiority of the Transformer architecture, the Vision Transformer model can capture the interdependencies between small patches in the image better than the convolutional neural network, helping the model learn the high-level abstract features of the image better, thereby improving the accuracy of image classification.

[0100] (2) Feature extraction of sequence data

[0101] In an embodiment of the present invention, a second feature extractor is used to extract features from flight trajectories and electromagnetic radiation, and the second feature extractor includes at least an IDCNN model.

[0102] Currently, in the prior art, Bi-LSTM is mostly used to extract features at each time step. However, the calculation of each time step of Bi-LSTM depends on the calculation result of the previous time step, resulting in a general calculation speed. Considering the requirements of real scenarios for model efficiency, in the embodiment of the present invention, an IDCNN model is used to extract features from sequence data.

[0103] The IDCNN model has the advantage of parallel computing of convolutional kernels compared to the Bi-LSTM model. As Figure 4 shown, in the original convolutional neural network, its convolutional kernel slides continuously, while the IDCNN model adds a dilation width parameter in the convolutional operation to increase the receptive field, that is, it will skip the intermediate hole area during the convolutional operation. Thus, the number of network layers can be reduced, thereby avoiding the overfitting problem of the model caused by too deep network layers.

[0104] III. Feature fusion process

[0105] Referring to the above step S4, only when the aircraft data is multi-source data, is it necessary to judge feature fusion, and this feature fusion judgment is carried out according to the user instruction. When the user instruction is that feature fusion is not required, the features extracted from multiple pieces of information are directly added; when the user instruction is that feature fusion is required, the feature fusion process (corresponding to the above step S42) is carried out.

[0106] As Figure 5 shown, this feature fusion process specifically includes steps S421 to S423:

[0107] S421: Perform weighted summation on the visible light image feature and the flight trajectory feature to obtain a first weighted feature.

[0108] In the embodiment of the present invention, since the visible light image feature and the flight trajectory feature are respectively extracted by two different feature extractors, therefore, the feature dimension of the visible light image feature c i is different from the feature dimension of the flight trajectory feature h i . Before performing weighted summation on the two, it is necessary to unify their feature dimensions.

[0109] Specifically, first perform a linear transformation on the visible light image feature c i to obtain c i_linear , so that the feature dimension of c i_linear is the same as the feature dimension of the flight trajectory feature h i . Then, adopt the Sigmoid function to perform weighted summation through the following formula to obtain the first weighted feature C i .

[0110] α = sigmoid(W c ·c i_linear +U c ·h i )

[0111] C i = α·c i +(I - α)·h i

[0112] Among them, I represents the identity matrix; · represents element multiplication; α represents the feature weight after normalization by the Sigmoid function; W c and U c represent weight matrices, which are obtained during the training process of the model.

[0113] S422: Perform weighted summation on the flight trajectory feature and the electromagnetic radiation feature to obtain a second weighted feature.

[0114] In an embodiment of the present invention, since both the flight trajectory feature and the electromagnetic radiation feature are extracted by the second feature extractor, their feature dimensions are the same, and no linear transformation is required. The Sigmoid function is directly used to perform weighted summation through the following formula to obtain the second weighted feature H i .

[0115] β = sigmoid(W h ·h i +U h ·u i )

[0116] H i = β·h i +(I - β)·u i

[0117] wherein, I represents the identity matrix; · represents element-wise multiplication; h i represents the flight trajectory feature; u i represents the electromagnetic radiation feature; β represents the feature weight after Sigmoid normalization; W h and U h represent weight matrices, which are obtained by the model during the training process.

[0118] S423: Concatenate the first weighted feature and the second weighted feature to obtain a comprehensive feature.

[0119] Based on the above aircraft classification method for fusing multi-source features, the present invention further provides an aircraft classification system for fusing multi-source features. As Figure 6 shown, the aircraft classification system includes one or more processors 21 and a memory 22. Among them, the memory 22 is coupled to the processor 21 and is used to store one or more programs. When the one or more programs are executed by the one or more processors 21, the one or more processors 21 implement the aircraft classification method for fusing multi-source features as in the above embodiment.

[0120] Among them, the processor 21 is used to control the overall operation of the aircraft classification system to complete all or part of the steps of the above-mentioned aircraft classification method that fuses multi-source features. The processor 21 can be a central processing unit (CPU), a graphics processing unit (GPU), a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a digital signal processing (DSP) chip, etc. The memory 22 is used to store various types of data to support the operation of the aircraft classification system. These data can include, for example, instructions for any application program or method operating on the aircraft classification system, as well as application program-related data. The memory 22 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, etc.

[0121] In an exemplary embodiment, the aircraft classification system can be specifically implemented by a computer chip or an entity, or by a product with certain functions, and is used to execute the above-mentioned aircraft classification method that fuses multi-source features and achieve the same technical effects as the above method. A typical embodiment is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, an in-vehicle human-machine interaction device, a cellular phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0122] In another exemplary embodiment, the present invention also provides a computer-readable storage medium including program instructions. When the program instructions are executed by a processor, the steps of the aircraft classification method that fuses multi-source features in any one of the above embodiments are implemented. For example, the computer-readable storage medium can be the above-mentioned memory including program instructions. The above program instructions can be executed by the processor of the aircraft classification system to complete the above-mentioned aircraft classification method that fuses multi-source features and achieve the same technical effects as the above method.

[0123] In summary, the aircraft classification method and system for fusing multi-source features provided by the embodiments of the present invention comprehensively consider various feature information, and considering that the importance degrees of different features are different in different scenarios, the visible light image features, flight trajectory features, and electromagnetic radiation features are fused through a gated network to obtain comprehensive features, and finally applied to the aircraft classification task in the real scenario. Thereby, the accuracy and classification effect of the classification method are improved. In addition, in practical applications, even if one or several of the features are missing in the aircraft classification system provided by the present invention, as long as one or more features of the aircraft can be input, the system can work normally, thereby enhancing the robustness of the entire aircraft classification system.

[0124] The above has described in detail the aircraft classification method and system for fusing multi-source features provided by the present invention. For those of ordinary skill in the art, any obvious changes made to it without departing from the essence of the present invention will constitute an infringement of the patent right of the present invention and will bear corresponding legal responsibilities.

Claims

1. A method for classifying aircraft by fusing multi-source features, characterized in that Including the following steps: Receiving aircraft data acquired by multiple sensors, where the aircraft data at least includes: visible light images, flight trajectories, and electromagnetic radiation; Judging the type of the aircraft data; If the aircraft data is single-source data, directly preprocess the aircraft data and perform feature extraction to obtain single-source features; if the aircraft data is multi-source data, respectively preprocess and perform feature extraction on each type of aircraft data to obtain comprehensive features; where the comprehensive features are obtained through the following method: after respectively performing feature extraction on each type of aircraft data, judge whether feature fusion is required; if so, perform weighted summation on the visible light image features and the flight trajectory features to obtain a first weighted feature; perform weighted summation on the flight trajectory features and the electromagnetic radiation features to obtain a second weighted feature; perform feature splicing on the first weighted feature and the second weighted feature to obtain the comprehensive features; if not, directly add different types of features to obtain the comprehensive features; Inputting the single-source features or the comprehensive features into a multi-layer perceptron network to predict the category of the aircraft; Wherein, the first weighted feature is obtained through the following steps: For the visible light image features perform a linear transformation to obtain so that the feature dimension of the is the same as the feature dimension of the flight trajectory feature ; Using the Sigmoid function, the first weighted feature is obtained by weighted summation through the following formula ; Among them, represents the identity matrix; represents element-wise multiplication; represents the feature weights after Sigmoid function normalization; and represents the weight matrix, which is obtained by the model during the training process; The second weighted feature is obtained through the following steps: Using the Sigmoid function, the second weighted feature is obtained by weighted summation through the following formula ; Among them, represents the flight trajectory feature; represents the electromagnetic radiation feature; represents the feature weight after Sigmoid normalization; and represents the weight matrix, which is obtained by the model during the training process.

2. The aircraft classification method according to claim 1, wherein Preprocessing the aircraft data, specifically including: Visible light image preprocessing: flipping and angle rotating the visible light image to perform data enhancement on the visible light image; selecting the maximum inscribed rectangle of the visible light image; for the enhanced visible light image data, cropping visible light images of different sizes into visible light images of a unified size; Flight trajectory preprocessing: selecting flight data within a set time; converting the flight data into longitude coordinates, latitude coordinates, and altitude coordinates; calculating the curvature and torsion corresponding to each sampling moment using a trajectory equation; performing feature splicing on the longitude coordinates, latitude coordinates, altitude coordinates, curvature, and torsion; Electromagnetic radiation preprocessing: selecting radar echo information within a set time; performing sparse recovery and normalization processing on the radar echo information using the wavelet transform method and the mean variance method.

3. The aircraft classification method according to claim 1, characterized in that Performing feature extraction on the aircraft data, specifically including: Using a first feature extractor to perform feature extraction on the visible light image, where the first feature extractor at least includes a Vision Transformer model; Using a second feature extractor to perform feature extraction on the flight trajectory and electromagnetic radiation, where the second feature extractor at least includes an IDCNN model.

4. The aircraft classification method according to claim 1, wherein Inputting the single-source features or the comprehensive features into a multi-layer perceptron network to predict the category of the aircraft, specifically including: Inputting the single-source features or the comprehensive features into a multi-layer perceptron network for dimensionality reduction; Performing probability value prediction through a softmax function; Taking the category corresponding to the maximum probability value as the final predicted category of the aircraft.

5. An aircraft classification system integrating multi-source features, characterized in that Including a processor and a memory, the processor reads a computer program in the memory and is used to execute the aircraft classification method according to any one of claims 1 to 4.

Images