Helicopter flight state identification method, device, equipment and medium

Abstract

The invention relates to a helicopter flight state recognition method, device and equipment and a medium, and the method comprises the steps: obtaining a plurality of flight parameters of a helicopter at the same moment; preprocessing the plurality of flight parameters of the helicopter to obtain a flight parameter sequence; inputting the flight parameter sequence into a helicopter flight state recognition network to obtain a helicopter flight state recognition result; wherein the helicopter flight state identification network comprises an encoder module used for capturing long-distance data characteristics in the flight parameter sequence; the multi-scale feature extraction module is used for extracting deep data features of different scales from the long-distance data features; the residual module is used for finding key features of flight parameters from the deep data features of different scales; and the classification module is used for identifying the flight state of the helicopter according to the key features of the flight parameters to obtain an identification result of the flight state of the helicopter.

Description

Technical Field

[0001] The present invention relates to the technical field of aircraft status recognition, and in particular to a method, device, equipment and medium for recognizing the flight status of a helicopter. Background Art

[0002] Helicopters have strong maneuverability and can achieve a variety of different flight states. In different flight states, the damage degree of dynamic parts and life parts is different. Therefore, correctly identifying the flight state and conducting statistical analysis is of great practical significance for the life prediction of key helicopter components.

[0003] The flight state of a helicopter is affected by multiple factors, with a huge number of parameters and strong correlations between the parameters. It is difficult to process and analyze them using traditional machine learning methods, and it is also difficult to fully extract parameter features using a single network model structure, resulting in low network recognition accuracy and poor generalization. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method, device, equipment and medium for identifying the flight status of a helicopter, which can effectively improve the recognition accuracy and thus improve the safety of air transportation.

[0005] The technical solution adopted by the present invention to solve the technical problem is: to provide a method for identifying the flight status of a helicopter, comprising the following steps:

[0006] Get multiple flight parameters of the helicopter at the same time;

[0007] Preprocessing a plurality of flight parameters of the helicopter to obtain a flight parameter sequence;

[0008] The flight parameter sequence is input into the helicopter flight state recognition network to obtain the recognition result of the helicopter flight state; wherein,

[0009] The helicopter flight status recognition network includes:

[0010] An encoder module, for capturing long-range data features in the flight parameter sequence;

[0011] A multi-scale feature extraction module, used to extract deep data features of different scales from the long-distance data features;

[0012] A residual module, used for finding key features of flight parameters from the deep data features of different scales;

[0013] The classification module is used to identify the flight status of the helicopter according to the key features of the flight parameters to obtain the identification result of the flight status of the helicopter.

[0014] The preprocessing of the multiple flight parameters of the helicopter specifically includes: sequentially performing wild point removal processing, repair processing, amplitude limiting processing, smoothing processing and fitting processing on the acquired multiple flight parameters.

[0015] The helicopter flight state recognition network is trained in the following way:

[0016] Match the historical flight parameters of the helicopter with the flight status to obtain the original flight form database;

[0017] Preprocess the historical flight parameters in the flight configuration original database;

[0018] Calculate the Pearson correlation coefficient between the preprocessed historical flight parameters, and eliminate the historical flight parameters whose Pearson correlation coefficient does not meet the requirements;

[0019] The historical flight parameters in the original flight morphology database after elimination are divided into a training set, a validation set and a test set;

[0020] The helicopter flight state recognition network is trained using a training set, and the trained helicopter flight state recognition network is verified and tested using a verification set and a test set. When the test results meet the requirements, the training is completed.

[0021] The encoder module includes three consecutive encoding units, each encoding unit includes: a first normalization layer, a multi-head attention layer, a second normalization layer and a feedforward neural network layer arranged in sequence; wherein the input of the second normalization layer is a fusion feature of the output of the multi-head attention layer and the input of the first normalization layer; the fusion feature of the output of the feedforward neural network layer and the input of the second normalization layer is used as the output of the encoding unit.

[0022] The multi-scale feature extraction module comprises:

[0023] The first stage is a multi-scale feature extraction unit, which consists of four branches. Each branch performs convolution and pooling operations of different scales on the input features to extract feature information of each dimension.

[0024] A first splicing unit, used for splicing feature information of each dimension;

[0025] The second stage is a multi-scale feature extraction unit, which includes three branches. Each branch downsamples the concatenated feature information of each dimension at different scales to obtain deep features of different scales.

[0026] The second splicing unit is used to splice deep features of different scales.

[0027] The residual module includes a first channel attention layer, a first one-dimensional convolutional neural network layer, a second one-dimensional convolutional neural network layer, a third one-dimensional convolutional neural network layer, a fourth one-dimensional convolutional neural network layer and a second channel attention layer connected in sequence; wherein the input of the third one-dimensional convolutional neural network layer is a fusion feature of the input of the first channel attention layer and the output of the second one-dimensional convolutional neural network layer; the fusion feature of the output of the second channel attention layer and the input of the third one-dimensional convolutional neural network layer is used as the output of the residual module.

[0028] The first channel attention layer and the second channel attention layer both include:

[0029] An extraction unit, used to obtain the number of channels and feature length vectors of deep data features;

[0030] Two fully connected layers to obtain channel-wise weight vectors;

[0031] The output layer is used to multiply the elements of each channel of the deep data features by the corresponding weight vector.

[0032] The technical solution adopted by the present invention to solve the technical problem is: to provide a helicopter flight status identification device, comprising:

[0033] An acquisition module is used to obtain multiple flight parameters of the helicopter at the same time;

[0034] A preprocessing module, used for preprocessing a plurality of flight parameters of the helicopter to obtain a flight parameter sequence;

[0035] The recognition module is used to input the flight parameter sequence into the helicopter flight state recognition network to obtain the recognition result of the helicopter flight state; wherein,

[0036] The helicopter flight status recognition network includes:

[0037] An encoder module, for capturing long-range data features in the flight parameter sequence;

[0038] A multi-scale feature extraction module, used to extract deep data features of different scales from the long-distance data features;

[0039] A residual module, used for finding key features of flight parameters from the deep data features of different scales;

[0040] The classification module is used to identify the flight status of the helicopter according to the key features of the flight parameters to obtain the identification result of the flight status of the helicopter.

[0041] The technical solution adopted by the present invention to solve its technical problem is: to provide an electronic device, including a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the above-mentioned helicopter flight status identification method when executing the computer program.

[0042] The technical solution adopted by the present invention to solve its technical problem is: providing a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned helicopter flight status identification method are implemented.

[0043] Beneficial Effects

[0044] Due to the adoption of the above-mentioned technical scheme, the present invention has the following advantages and positive effects compared with the prior art: the improved Transformer encoder module of the present invention can accurately capture the long-distance data features in the sequence; the present invention samples the input flight parameters at different granularities through a multi-scale feature extraction module to obtain features at different scales, thereby enhancing the diversity of features and facilitating flight state recognition; the present invention uses an improved residual module to increase the network depth, and introduces a channel attention mechanism to improve the network's perception of key information, thereby enhancing the network's feature extraction capability. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flow chart of a helicopter flight status identification method according to a first embodiment of the present invention;

[0046] Figure 2 is a flow chart of flight parameter preprocessing in the first embodiment of the present invention;

[0047] Figure 3 is a schematic diagram of the structure of a helicopter flight status recognition network in the first embodiment of the present invention;

[0048] Figure 4 is a schematic structural diagram of an encoder unit in a first embodiment of the present invention;

[0049] Figure 5 is a schematic structural diagram of a multi-scale feature extraction module in a first embodiment of the present invention;

[0050] Figure 6 is a schematic structural diagram of a residual module in a first embodiment of the present invention;

[0051] Figure 7 is a schematic diagram of the structure of the channel attention layer in the first embodiment of the present invention;

[0052] Figure 8 It is a convergence curve diagram of the helicopter flight state recognition network in the first embodiment of the present invention. DETAILED DESCRIPTION

[0053] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only intended to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art may make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0054] The first embodiment of the present invention relates to a method for identifying the flight status of a helicopter, such as Figure 1 As shown, the following steps are included:

[0055] Step 1, obtain multiple flight parameters of the helicopter at the same time. There are 23 flight parameters obtained in this step, as shown in Table 1.

[0056] Table 1 Helicopter flight parameters

[0057] serial number symbol Parameter name serial number symbol Parameter name 1 Vi Indicated airspeed 13 Mfa Engine torque 2 Hp Pressure altitude 14 wxya Tail drive shaft torque 3 Nx Longitudinal overload at center of gravity 15 Hpb Radio altitude 4 Ny Lateral overload at the center of gravity 16 T20 Fuel inlet temperature 5 Nz Normal overload at center of gravity 17 P20 Fuel inlet pressure 6 Nx Tail longitudinal overload 18 Ng Engine speed 7 Nwy Tail lateral overload 19 Nr Rotor speed 8 NvZ Tail Normal Overload 20 Nf Free turbine speed 9 W Total displacement 21 Q Engine fuel consumption 10 Wy Pedal displacement 22 COCI Yaw angle 11 W Longitudinal displacement of operating lever 23 GAM Tilt Angle 12 Wx Lateral displacement of operating lever

[0058] Step 2, preprocessing the multiple flight parameters of the helicopter to obtain a flight parameter sequence. Since the environment during helicopter flight is complex and changeable, the flight parameter recorder will introduce interference such as external noise and cross-linking coupling effect during its collection and recording, which is prone to abnormal phenomena such as data loss and data distortion. These abnormal data will directly affect the recognition rate of the flight status, so before flight status recognition, it is necessary to preprocess the flight parameter data to repair distorted data, fill in lost data, and reduce unnecessary noise and unpredictable data during the status recognition process. Figure 2 As shown, in this step, when pre-processing the flight parameters, wild point removal, repair, limiting, smoothing and fitting are performed in sequence.

[0059] Step 3: Input the flight parameter sequence into the helicopter flight state recognition network to obtain the recognition result of the helicopter flight state. The recognition result includes 35 flight states, as shown in Table 2.

[0060] Table 2 Flight status table to be identified

[0061] serial number Flight status serial number Flight status 1 Hover in ground effect 19 Long-distance flight speed, level flight, left glide angle 10° 2 Hovering without ground effect 20 Long-distance flight speed, level flight, right glide angle 10° 3 Backfly 21 Maximum cruising speed level flight, sideslip angle 0° 4 Fly right 22 Maximum cruising speed, level flight, left slip angle 5° 5 Left Fly 23 Maximum cruising speed, level flight, right slip angle 5° 6 Vertical Ascent 24 Horizontal deceleration 7 Inclined climb 25 Horizontal acceleration 8 Maximum continuous power ramp climb 26 Dive 9 Ascending turn or spiral rise 27 Dive and pull up 10 Takeoff speed 28 Long flight speed horizontal turn or hover 11 Transition speed level flight 29 180km / h speed level flight turn 12 Long flight speed level flight, sideslip angle 0° 30 Maximum cruising speed turn 13 Long flight speed, level flight, left glide angle 10° 31 rotation 14 Long flight speed level flight, right glide angle 10° 32 Approach, decline, deceleration 15 180km / h level flight, sideslip angle 0° 33 Flare landing 16 180km / h level flight, left slip angle 10° 34 Vertical Descent 17 180km / h level flight, right angle 10° 35 Ground slow train 18 Long-distance flight speed, side slip angle 0°

[0062] like Figure 3As shown, the helicopter flight state recognition network (CTFM-Net) in this embodiment includes: an encoder module, used to capture the long-distance data features in the flight parameter sequence; a multi-scale feature extraction module, used to extract deep data features of different scales from the long-distance data features; a residual module, used to find the key features of the flight parameters from the deep data features of different scales; a classification module, used to identify the helicopter flight state according to the key features of the flight parameters, and obtain the recognition result of the helicopter flight state.

[0063] Among them, the encoder module includes three consecutive encoding units, such as Figure 4 As shown, each encoding unit includes: a first normalization layer, a multi-head attention layer, a second normalization layer and a feedforward neural network layer arranged in sequence; wherein the input of the second normalization layer is a fusion feature of the output of the multi-head attention layer and the input of the first normalization layer; the fusion feature of the output of the feedforward neural network layer and the input of the second normalization layer is used as the output of the encoding unit.

[0064] It can be seen that the encoding unit is mainly composed of a multi-head self-attention mechanism and a feedforward neural network. In addition to the above two network layers, the structure of each encoding unit also contains residual connections and layer normalization to accelerate and stabilize the training of the model. The residual connection adds the input and output of the encoder unit, and the layer normalization standardizes the output of each sub-layer, making the model easier to converge.

[0065] The multi-head self-attention layer calculates self-attention for all positions in the input sequence to obtain a weighted representation of each position, where the weight is calculated using three vectors: query, key, and value. The query vector represents the features of the current time step, and the key vector and value vector represent the features of other time steps in the sequence. The multi-head attention mechanism maps the input query vector and key vector to different vector spaces, and calculates the similarity between them to obtain the attention distribution. The attention distribution is then multiplied by the value vector to obtain a weighted value vector, and finally the weighted vectors of multiple heads are concatenated as the output. The calculation formula for each self-attention is as follows:

[0066]

[0067] Where, d k represents the channel dimension, Q, K, and V represent the query vector, key vector, and value vector, respectively, and Softmax is the activation function. Multiple self-attention weighted vectors mapped to different vector spaces are concatenated to obtain the final multi-head attention vector. The final multi-head attention output is obtained by mapping back to the original vector dimension through matrix transformation.

[0068] In the encoding unit, the multi-head self-attention mechanism layer performs a weighted calculation on each input vector to capture the relationship between all vectors in the sequence.

[0069] The feedforward neural network layer performs a separate fully connected feedforward neural network transformation on each vector representation after the multi-head self-attention mechanism layer. It consists of two consecutive linear layers with a nonlinear activation function ReLU in the middle, which enables the network to learn more complex nonlinear relationships and enhance the expressiveness of the model.

[0070] Convolutions of different scales extract features with different degrees of precision. The smaller the convolution scale, the easier it is to extract local sensitive features. If a single layer only uses convolutions of the same scale, it is easy to ignore other precision features, resulting in failure to fully extract the feature information of each scale in the input data. In order to improve the feature extraction capability of the network and improve the accuracy of flight status recognition, the multi-scale feature extraction module in this embodiment is as follows: Figure 5 As shown, it includes: a first-stage multi-scale feature extraction unit, including four branches, each branch performs convolution and pooling operations of different scales on the input features to extract feature information of each dimension; a first splicing unit, used to splice the feature information of each dimension; a second-stage multi-scale feature extraction unit, including three branches, each branch performs downsampling of different scales on the spliced ​​feature information of each dimension to obtain deep features of different scales; a second splicing unit, used to splice deep features of different scales.

[0071] The multi-scale feature extraction module superimposes two feature splicing units to extract features of different degrees of precision from the input flight parameters, improve the richness and diversity of features, and enhance the network's perception of targets of different scales in the flight parameters. When processing, first, the input features pass through a four-branch multi-scale convolution layer. On each branch, a series of convolution and pooling operations of different scales are performed on the input features to extract feature information of each dimension; then, these feature information of different dimensions are spliced ​​to complete the first stage of multi-scale feature extraction; then, the merged data features are passed through a three-branch multi-scale downsampling layer to further extract deeper data features while reducing the data dimension; finally, the deep features of three different scales are spliced.

[0072] In order to enable CTFM-Net to obtain deeper data features while avoiding the problems of gradient vanishing and gradient exploding, this embodiment introduces a residual module into CTFM-Net. Since the residual module introduces a skip connection, that is, the input is directly added to the output, the neural network can propagate the gradient from the back to the deeper network without losing information, effectively alleviating the degradation problem of the neural network. At the same time, at the input and output of the residual module, the channel attention mechanism is used to enable the network to better utilize the features of different channels, thereby improving the network's expressiveness and generalization. The residual module structure in this embodiment is as follows: Figure 6 As shown, it includes a first channel attention layer, a first one-dimensional convolutional neural network layer, a second one-dimensional convolutional neural network layer, a third one-dimensional convolutional neural network layer, a fourth one-dimensional convolutional neural network layer and a second channel attention layer connected in sequence; wherein the input of the third one-dimensional convolutional neural network layer is a fusion feature of the input of the first channel attention layer and the output of the second one-dimensional convolutional neural network layer; the fusion feature of the output of the second channel attention layer and the input of the third one-dimensional convolutional neural network layer is used as the output of the residual module.

[0073] Among them, the channel attention layer can make the network focus on the key features of the flight parameters through the channel attention mechanism, making the model more suitable for the flight status classification task. Figure 7 As shown in the figure, the channel attention layer includes: an extraction unit for obtaining the number of channels and feature length vectors of deep data features; two fully connected layers for obtaining the weight vectors in the channel direction; and an output layer for multiplying the elements of each channel of the deep data features with the corresponding weight vectors. When processing the input features, first, the number of channels T and the vector of feature length L of the deep data features are obtained; then, the weight vectors in the channel direction are obtained through two fully connected layers, where the number of nodes in the first fully connected layer is T, and the Swish activation function is used, and the number of nodes in the second fully connected layer is also T, and the Sigmoid activation function is used; finally, the elements of each channel of the input feature are multiplied with the corresponding weight vectors, and each channel has a different weight, so it can have different attention to the features of each channel.

[0074] The output layer includes two fully connected hidden layers and the final Softmax layer. The fully connected hidden layer is used to connect the input data with the weights and apply the activation function for calculation. The Softmax layer converts the output into a probability distribution, thereby realizing the recognition of the helicopter's flight status.

[0075] The training method of the helicopter flight state recognition network in this embodiment is as follows:

[0076] First, the historical flight parameters of the helicopter are matched with the flight state to obtain the original flight state database, where each flight state corresponds to a period of continuously collected flight parameter signals.

[0077] Next, the historical flight parameters in the original flight configuration database are preprocessed in the same manner as in step 2 above.

[0078] Then, the Pearson correlation coefficient is used to select sensitive flight parameters, that is, the Pearson correlation coefficient between the preprocessed historical flight parameters is calculated, and the historical flight parameters whose Pearson correlation coefficient does not meet the requirements are eliminated.

[0079] In order to determine the correlation between data items, this embodiment uses the Pearson correlation coefficient and its linear correlation coefficient. Suppose the two groups of sample corresponding data {(x i ,y i ), i = 1, 2, ... n} (n is the number of samples) is a bivariate normal distribution The correlation coefficient is:

[0080]

[0081] in, and are the sample means of X and Y respectively. Correlation coefficient The closer the absolute value of is to 1, the stronger the correlation between the two is; the closer it is to 0, the weaker the correlation between the two is.

[0082] Afterwards, the historical flight parameters in the processed flight state original database are divided into a training set, a validation set, and a test set. The total sample size of the flight parameter data set used in this embodiment is 782,400, and data are randomly selected from each flight state sample at a ratio of 7:2:1 for training, validation, and testing.

[0083] In this implementation, the structural parameters of the helicopter flight state recognition network are shown in Table 3. In order to increase the nonlinearity of the model, the output of each layer uses a ReLu nonlinear activation function to enhance the feature expression ability of the network.

[0084] Table 3 CTFM-Net network model parameters

[0085]

[0086] The Adam optimizer is used for network training. The initial learning rate is set to 0.01, the batch sample data size is 1000, the number of training iterations is 100, and the cross entropy loss function is selected. The input of the network is a 21×1 column vector. The network is trained using training samples. After one iteration, the network is tested using test samples to obtain the test accuracy. After the neural network model iteration is completed, the network is tested using the test set. The experimental results are shown in Table 4 and Figure 8 shown.

[0087] Table 4 CTFM-Net recognition accuracy table

[0088]

[0089] Combining Table 4 and Figure 8 It can be seen that as the network iteration process continues, the training accuracy and test accuracy continue to increase. After the number of iterations reaches 15 times, the three curves of training accuracy, verification accuracy and training loss all tend to be stable, and the training accuracy and verification accuracy are stable at 99.99% and 99.51% respectively. The network has completed convergence. By testing the network's test set accuracy, the test set accuracy reaches 99.50%, proving that there is no overfitting during network training, and the network has good stability and generalization; the average test recognition accuracy of 35 flight states is 99.33%, proving that after data enhancement, the network has overcome the problem of low recognition rate caused by data imbalance to a certain extent. The recognition results of each flight state in the test set are shown in Table 5.

[0090] Table 5 Flight status recognition results

[0091] Flight status number Total number of data Identify the correct number Number of recognition errors Recognition accuracy (%) 1 1760 1760 0 100.0 2 1880 1880 0 100.0 3 2190 2160 22 98.9 4 5260 5228 32 99.3 5 1930 1910 20 98.9 6 1750 1750 0 100.0 7 1400 1400 0 100.0 8 960 930 30 96.8 9 920 915 5 99.4 10 1220 1197 23 98.1 11 1520 1520 0 100.0 12 680 680 0 100.0 13 230 230 0 100.0 14 3440 3440 0 100.0 15 1810 1747 63 96.5 16 360 360 0 100.0 17 1320 1320 0 100.0 18 4070 4070 0 100.0 19 4270 4260 10 99.7 20 4280 4280 0 100.0 21 2820 2820 0 100.0 22 2780 2780 0 100.0 23 2720 2700 20 99.2 24 1360 1360 0 100.0 25 1330 1330 0 100.0 26 1330 1330 0 100.0 27 4000 3990 10 99.7 28 3950 3942 8 99.7 29 3990 3982 8 99.7 30 2190 2070 120 94.5 31 2540 2533 7 99.7 32 5820 5782 38 99.3 33 950 922 28 97.0 34 670 670 0 100.0 35 5680 5641 39 99.3

[0092] It can be seen from Table 5 that among the recognition results of 35 flight states on the test set, the recognition accuracy of 18 flight states reaches 100%, and only the recognition rate of the flight state with state number 30 is 94.5%, which is relatively low. The recognition rates of most states are above 99%, which proves the superiority of the network model CTFM-Net proposed in this embodiment in the flight state recognition task.

Claims

1. A method for identifying a helicopter flight state, characterized in that: The following steps are involved: Get multiple flight parameters of the helicopter at the same time; Preprocessing a plurality of flight parameters of the helicopter to obtain a flight parameter sequence; The flight parameter sequence is input into the helicopter flight state recognition network to obtain the recognition result of the helicopter flight state; wherein, The helicopter flight status recognition network includes: An encoder module, for capturing long-distance data features in the flight parameter sequence; A multi-scale feature extraction module, used to extract deep data features of different scales from the long-distance data features; A residual module, used for finding key features of flight parameters from the deep data features of different scales; The classification module is used to identify the flight status of the helicopter according to the key features of the flight parameters to obtain the identification result of the flight status of the helicopter.

2. The helicopter flight status identification method according to claim 1, characterized in that: The preprocessing of the multiple flight parameters of the helicopter specifically includes: sequentially performing wild point removal processing, repair processing, amplitude limiting processing, smoothing processing and fitting processing on the acquired multiple flight parameters.

3. The helicopter flight status identification method according to claim 1, characterized in that: The helicopter flight state recognition network is trained in the following way: Match the historical flight parameters of the helicopter with the flight status to obtain the original flight form database; Preprocess the historical flight parameters in the original flight configuration database; Calculate the Pearson correlation coefficient between the preprocessed historical flight parameters, and eliminate the historical flight parameters whose Pearson correlation coefficient does not meet the requirements; The historical flight parameters in the original flight form database after elimination are divided into a training set, a verification set and a test set; the training set is used to train the helicopter flight state recognition network, and the verification set and the test set are used to verify and test the trained helicopter flight state recognition network. When the test results meet the requirements, the training is completed.

4. The helicopter flight status recognition method according to claim 1, characterized in that: The encoder module includes three consecutive encoding units, each encoding unit includes: a first normalization layer, a multi-head attention layer, a second normalization layer and a feedforward neural network layer arranged in sequence; wherein the input of the second normalization layer is a fusion feature of the output of the multi-head attention layer and the input of the first normalization layer; the fusion feature of the output of the feedforward neural network layer and the input of the second normalization layer is used as the output of the encoding unit.

5. The helicopter flight status identification method according to claim 1, characterized in that: The multi-scale feature extraction module comprises: The first stage is a multi-scale feature extraction unit, which consists of four branches. Each branch performs convolution and pooling operations of different scales on the input features to extract feature information of each dimension. A first splicing unit, used for splicing feature information of each dimension; The second stage is a multi-scale feature extraction unit, which includes three branches. Each branch downsamples the concatenated feature information of each dimension at different scales to obtain deep features of different scales. The second splicing unit is used to splice deep features of different scales.

6. The helicopter flight status identification method according to claim 1, characterized in that: The residual module includes a first channel attention layer, a first one-dimensional convolutional neural network layer, a second one-dimensional convolutional neural network layer, a third one-dimensional convolutional neural network layer, a fourth one-dimensional convolutional neural network layer and a second channel attention layer connected in sequence; wherein the input of the third one-dimensional convolutional neural network layer is a fusion feature of the input of the first channel attention layer and the output of the second one-dimensional convolutional neural network layer; the fusion feature of the output of the second channel attention layer and the input of the third one-dimensional convolutional neural network layer is used as the output of the residual module.

7. The helicopter flight status identification method according to claim 6, characterized in that: The first channel attention layer and the second channel attention layer both include: An extraction unit, used to obtain the number of channels and feature length vectors of deep data features; Two fully connected layers to obtain channel-wise weight vectors; The output layer is used to multiply the elements of each channel of the deep data features by the corresponding weight vector.

8. A helicopter flight status recognition device, characterized in that: include: An acquisition module is used to obtain multiple flight parameters of the helicopter at the same time; A preprocessing module, used for preprocessing a plurality of flight parameters of the helicopter to obtain a flight parameter sequence; The recognition module is used to input the flight parameter sequence into the helicopter flight state recognition network to obtain the recognition result of the helicopter flight state; wherein, The helicopter flight status recognition network includes: An encoder module, for capturing long-distance data features in the flight parameter sequence; A multi-scale feature extraction module, used to extract deep data features of different scales from the long-distance data features; A residual module, used for finding key features of flight parameters from the deep data features of different scales; The classification module is used to identify the flight status of the helicopter according to the key features of the flight parameters to obtain the identification result of the flight status of the helicopter.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the helicopter flight status identification method as described in any one of claims 1-8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the helicopter flight status identification method as described in any one of claims 1-8 are implemented.

Images