A flight task recognition method and device based on BDTW and sequence alignment

Abstract

The application provides a flight task recognition method and device based on a blocked dynamic time warping (BDTW) and sequence alignment, the method comprising: obtaining a hand area of interest (AOI) sequence of a pilot in a flight simulator experiment, encoding data, and pre-processing; using existing hand AOI sequence data with labels as training data, and using experimental data as test data; using a blocked dynamic time warping algorithm (BDTW) to perform iterative alignment processing on training samples under each task, obtaining a reference sequence, and forming a template library; using the BDTW to perform alignment on test data and each type of reference sequence in the template library respectively, calculating a similarity, and obtaining a recognition result of the test data according to a reference sequence selected when the maximum similarity is generated. The flight task recognition method based on the BDTW and the sequence alignment solves the problem of interference of noise on flight task recognition.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of aviation flight technology, in particular to a flight task recognition method and device based on BDW and sequence alignment. BACKGROUND

[0002] Human factors have gradually become the main cause of aviation accidents. When an abnormality occurs during flight, the pilot needs to handle a series of complex operations according to the checklist, and any operational error occurring in this process may lead to a serious aviation accident. An effective method to prevent human error is to summarize error experience according to the behavior of the pilot to avoid repeating errors. The standard operating procedure for the same flight task is determined, and human error occurs when the pilot fails to perform all the operations in the operating procedure.

[0003] The behavior data records the hand area of interest (AOI) of the pilot during flight, which is a sparse time series. The non-zero value in the sequence represents the operation performed by the pilot, and the number of repetitions of the non-zero value represents the duration of the operation behavior. The operation procedure for the same type of flight task is the same, and the AOI sequence contains an ordered and discrete distribution of subsequences. The basic problem of flight task recognition is the similarity measurement of the AOI sequence. The DTW algorithm has become a popular similarity measurement method due to its prevention of time shift effects. BDW is an improved DTW algorithm based on encoded sequences. It solves the problem of high time complexity of traditional DTW by utilizing the sparsity characteristics of behavior AOI sequence. However, due to misjudgment in the extraction process of AOI sequence and the behavior habits of the pilot, some noise unrelated to the operation procedure will be generated, which will distort the similarity measurement of the AOI sequence. SUMMARY

[0004] In view of the defects in the prior art, the purpose of the present application is to provide a flight task recognition method and device based on BDW and sequence alignment, which extracts reference data through the sequence alignment process, and recognizes the flight task according to the similarity between the reference data and the measured data, solving the problem of noise interference in flight task recognition.

[0005] To solve the above problems, the technical scheme of the present application is as follows:

[0006] A flight task recognition method based on BDW and sequence alignment, comprising the following steps:

[0007] Obtain the hand area of interest sequence of the pilot in the flight simulator experiment, encode the data, and perform preprocessing;

[0008] Use the existing labeled hand area of interest sequence data as training data, and use the experimental data for testing;

[0009] Using the block dynamic normalization algorithm BDTW to perform iterative alignment processing on the training samples under each task, to obtain reference sequences, and to form a template library;

[0010] Using BDWT to align each type of reference sequence in the template library and the test data respectively, to calculate the similarity, and to obtain the recognition result of the test data according to the reference sequence selected when the maximum similarity is generated.

[0011] Preferably, the steps of obtaining the pilot's hand region of interest sequence in the flight simulator experiment, encoding the data, and preprocessing the data specifically include:

[0012] Let the pilot's hand region of interest sequence be X raw ={x11,x12,…,x1 l1 ,x21,x22,…,x2 l2 ,…,xk1,xk2,…,xk lk}, k represents the number of value repetition subsequences, xj is a value in the sequence, represented by an integer, indicating the pilot's hand AOI, j ∈ [1, k], j represents the number of subsequences, {xj1, xj2, …, xj lj} represents a continuous value repetition subsequence in the original region of interest sequence, lj represents the total length of the value repetition subsequence, and the subscript of xj indicates the position of the element in the subsequence.

[0013] Encode the original sequence, convert all continuous and repeated subsequences {xj1, xj2, …, xj lj} in the sequence into encoded form (xj, lj), and then assemble them into a two-dimensional encoded sequence

[0014] X encoded ={(x1, l1), (x2, l2), …, (xk, lk);

[0015] Remove the noise in the original sequence by traversing the encoded data.

[0016] Preferably, the step of removing the noise in the original sequence by traversing the encoded data specifically includes:

[0017] The non-zero same value subsequence existing in the form of cluster, set threshold T1, if a value of zero subsequence (0, l0) length l0 < T1, and the two adjacent subsequences (xa, la), (xb, lb) have the same non-zero value, that is, xa = xb, then merge the three subsequences into (xa, la+l0+lb).

[0018] Non-zero sub-sequences that do not exist in the form of clusters and are too short in length, set threshold T2, set the value of non-zero sub-sequences (xc, lc) with length lc < T2 to zero, and the processing result is (0, lc).

[0019] Preferably, in the step of using existing labeled hand ROI sequence data as training data and experimental data as test, the accuracy of the training data should be ensured, that is, it should contain complete operation procedures.

[0020] Preferably, the step of using the block dynamic warping algorithm BDWT to perform iterative alignment processing on the training samples under each task to obtain reference sequences and form a template library specifically includes:

[0021] The sequence alignment process needs to meet the monotonicity, boundary matching, continuity and optimal matching principles, and the optimal alignment path is calculated.

[0022] Iterative alignment is used to extract common operations in the data, and further construct clean sequences containing only pilot operation information as a template library of reference data.

[0023] Preferably, the step of performing sequence alignment process by BDWT method, the sequence alignment process needs to meet the monotonicity, boundary matching, continuity and optimal matching principles, and the optimal alignment path is calculated specifically includes:

[0024] Two encoding sequences X encoded ={x1, x2, …, x m} and Y encoded ={y1, y2, …, y n} are matched between each other to form a matching matrix of m x n, the element D i，j in position (i, j) in the matching matrix represents the distance between the i-th element x i =(xi, li) in the first sequence and the j-th element y j =(yj, lj) in the second sequence, and the calculation formula is: D i，j =(x i [0]-y j [0]) 2 .

[0025] Monotonicity requires that the elements in the two encoding sequences must be aligned in the order from first to last, and continuity requires that each element in the encoding sequence has a matching object. If the last matching of the path is D i，j , then the current matching can only be one of D i+1，j , D i，j+1 , D i+1，j+1 .

[0026] The boundary matching requires that the first element of the two encoded sequences must match each other, and the last element must match each other, i.e. the optimal matching path must end with matching D 0，0 Start with matching D m，n End;

[0027] The optimal matching principle requires that the last matching of the current matching D i，j can only be Di-1, j, Di, j-1, Di-1, j-1, and the minimum distance to Di, j, for which the minimum cumulative distance matrix d of m x n is calculated, and the matrix element d i，j represents the minimum cumulative distance from the (0, 0)th element to the (i, j)th element of the matching matrix, and the calculation process is as follows:

[0028] d 0，0 = 0

[0029] top = d i-1，j + x i [1] * D i，j

[0030] diagonal = d i-1，j-1 + max(x i [1], y j [1]) * D i，j

[0031] left = d i，j-1 + y i [1] * D i，j

[0032] d i，j = min(top, diagonal, left)

[0033] According to the obtained cumulative distance matrix d, starting from d m，n , backtracking until the end point d 0，0 , the optimal alignment path can be obtained, and the backtracking method is as follows:

[0034] Preferably, the step of iteratively aligning to extract common operations in the data and further constructing a clean sequence library containing only pilot operation information as a template library of reference data specifically comprises:

[0035] Randomly select two encoded sequences from a training sample of a type of flight task, and use BDWT to align and extract the optimal alignment path;

[0036] Extract matching pairs with the same AOI value in the path, and for those elements that match multiple elements at the same time, take the element with the smallest time stamp difference to form a matching pair;

[0037] The average value of the start time stamp and the average value of the duration between each other of the matching pairs selected in the previous step are calculated to form a sequence with a time length equal to the time length of the optimal alignment path;

[0038] A sequence is randomly selected from the remaining training samples and aligned with the intermediate output sequence generated in the previous step, and the process is repeated until the last training sample is used up, and the reference data of the flight task of this type is output;

[0039] Continue to extract reference data from the training samples of other flight tasks to build a template library.

[0040] Preferably, the step of using BDWT to align the test data and each type of reference sequence in the template library respectively, calculating the similarity, and obtaining the recognition result of the test data according to the reference sequence selected when the maximum similarity is generated specifically comprises:

[0041] The similarity Similarity is the basis for classifying the test data, and is calculated by the number n of non-zero elements in the reference data ref and the number n of matching pairs with a distance of zero mch , and the value range is [0, 1], which is defined as:

[0042] The similarity of the test data and all types of data in the template library is calculated respectively, and the template type selected when the maximum value is taken is the flight task type to which the test data belongs, and the recognition is completed.

[0043] Further, the present application also provides a flight task recognition device based on BDWT and sequence alignment, which comprises a processor and a memory for storing executable instructions of the processor, and the processor is configured to execute the executable instructions to perform the flight task recognition method based on BDWT and sequence alignment as described above.

[0044] Compared with the prior art, the present application applies the BDWT algorithm in the field of sparse time sequence recognition to the recognition of the pilot's hand region of interest sequence, eliminates the offset of the sequence on the time axis, avoids the influence of the different time spent by the pilot in performing a one-step operation program on the similarity measurement, and improves the problem of high time complexity existing in the traditional DTW. Moreover, the present application also proposes an iterative alignment method for extracting reference data to solve the problem of abnormal value interference existing in the BDWT matching of the original region of interest sequence, constructs a template library of reference data, realizes the recognition of the measured data through the sequence alignment and similarity calculation of the reference data and the measured data, and makes the recognition result more accurate. BRIEF DESCRIPTION OF DRAWINGS

[0045] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0046] Figure 1 A flowchart of a flight mission identification method based on BDTW and sequence alignment provided in an embodiment of the present invention;

[0047] Figure 2 A detailed flowchart of the flight mission identification method based on BDTW and sequence alignment provided in this embodiment of the invention;

[0048] Figure 3 This is a schematic diagram of the principle structure of the BDTW algorithm provided in an embodiment of the present invention;

[0049] Figure 4 This is a schematic diagram of the preprocessing provided in an embodiment of the present invention;

[0050] Figure 5 This is a schematic diagram of the average sequence generation process provided in an embodiment of the present invention. Detailed Implementation

[0051] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0052] Specifically, this invention provides a flight mission identification method based on BDTW and sequence alignment, such as... Figure 1 and Figure 2 As shown, the method includes the following steps:

[0053] S1: Obtain the sequence of regions of interest (ROIs) of the pilot's hands during the flight simulator experiment, encode the data, and perform preprocessing;

[0054] Specifically, the original region of interest (AOI) sequence is converted into data to be identified using the following steps:

[0055] S11: Let the pilot's hand AOI sequence be X. raw ={x11, x12, ..., x1 l1 x21, x22, ..., x2 l2 ..., xk1, xk2, ..., xk lk Let X be the sequence of regions of interest for the pilot's hand. raw ={x11, x12, ..., x1 l1 x21, x22, ..., x2 l2..., xk1, xk2, ..., xk lk}, k represents the number of repeated subsequences, xj is a value in the sequence, represented by an integer, indicating the AOI of the pilot's hand, j∈[1,k], j indicates which subsequence, {xj1, xj2, ..., xj...} lj} represents a continuous subsequence of repeated values ​​in the original AOI sequence, lj represents the total length of the subsequence of repeated values, and the subscript of xj indicates the position of the element in the subsequence. The correspondence is shown in Table 1 below.

[0056]

[0057] Table 1

[0058] In Table 1, the elements in the hand AOI sequence represent the cockpit area where the pilot's hand is located at a certain moment. The cockpit area is divided into: FCP, PF's FMS, left warning, PM's FMS, right warning, throttle lever, ECAM, and top plate, represented by sequence elements 1-8. At the same time, 0 is used to indicate that the hand is not on the device.

[0059] S12: Encode the original sequence by encoding all consecutively repeating subsequences {xj1, xj2, ..., xj...} within the sequence. lj}, convert them into the encoded form (xj, lj), and then assemble them into a two-dimensional encoded sequence according to their order.

[0060] X encoded ={(x1, l1), (x2, l2), ..., (xk, lk)}.

[0061] S13: Remove noise present in the original sequence by traversing the encoded data.

[0062] Considering the noise caused by errors in the conversion of raw video data to AOI data, the lj value of the noise subsequence is small, and its represented time range is much smaller than the time consumed by performing one operation. For example... Figure 4 The original sequence shown contains two forms of noise, which are removed by traversing the encoded data. The specific steps are as follows:

[0063] S131: Non-zero subsequences with the same value existing in clusters, that is, these subsequences are separated by short zero-value subsequences, representing a continuous operation, such as... Figure 3 As shown, a threshold T1 is set. If a subsequence (0, l0) with a value of zero has a length l0 < T1, and its two adjacent subsequences (xa, la) and (xb, lb) have the same non-zero value, i.e., xa = xb, then these three subsequences are merged into (xa, la + l0 + lb).

[0064] S132: Non-zero subsequences that do not exist in the form of clusters and are too short are often caused by misidentification in the AOI data extraction process. Set a threshold T2 and set the value of non-zero subsequences (xc, lc) with length lc < T2 to zero. The result is (0, lc).

[0065] S2: Use existing labeled hand region of interest sequence data as training data, and experimental data as testing data;

[0066] Specifically, in step S2, the accuracy of the training data should be ensured, that is, it should include complete operating procedures.

[0067] S3: Use the Block Dynamic Tunneling (BDTW) algorithm to iteratively align the training samples for each task, obtain reference sequences, and form a template library.

[0068] Specifically, step S3 includes the following steps:

[0069] S31: The sequence alignment process performed using the BDTW method needs to satisfy the principles of monotonicity, boundary matching, continuity, and optimal matching. The optimal alignment path is then calculated, specifically including the following steps:

[0070] S311: Combine the two encoded sequences X encoded ={x1, x2, ..., x m} and Y encoded ={y1, y2, ..., y n The pairwise matching results of all elements are listed in an m×n matching matrix, and the element D at position (i, j) in the matching matrix is... i，j x represents the i-th element in the first sequence. i = (xi, li) and the j-th element y of the second sequence j The distance between (yj, lj) is calculated using the formula: D i，j =(x i [0]-y j [0]) 2 ;

[0071] S312: Monotonicity requires that elements in two encoded sequences must be aligned in chronological order, and continuity requires that each element in the encoded sequence has a matching object. If the previous match of the path is D... i，j Then the current match can only be D. i+1，j D i，j+1 D i+1，j+1 One of them;

[0072] S313: Boundary matching requires that the first elements of two coded sequences must match each other, and the last elements must match each other; that is, the optimal matching path must start with matching D.0，0 Begin by matching D m，n Finish;

[0073] S314: The optimal matching principle requires that the previous matching of the current matching Di,j can only be the case among Di-1,j, Di,j-1, Di-1,j-1 that minimizes the distance to Di,j. To achieve this, an m×n cumulative distance matrix d is calculated, where the matrix elements d... i，j Let represent the minimum cumulative distance from the (0, 0)th element to the (i, j)th element of the matching matrix. The calculation process is as follows:

[0074] d 0，0 =0

[0075] top = d i-1，j +x i [1]*D i，j

[0076] diagonal=d i-1，j-1 +max(x i [1], y j [1])*D i，j

[0077] left = d i，j-1 +y i [1]*D i，j

[0078] d i，j =min(top, diagonal, left)

[0079] The elements D located on the top left and right sides of the matching matrix 0，* and D *，0 Since the top, diagonal, and left cases cannot exist simultaneously, the calculation only needs to consider the cases that exist. In particular, the minimum cumulative distance d at (0,0) is... 0，0 The value is 0, d m，n The value represents the cumulative distance under optimal alignment of the two sequences.

[0080] S315: As Figure 3 As shown, based on the cumulative distance matrix d obtained in S314, from d... m，n Start backtracking until the endpoint d. 0，0 The optimal alignment path can be obtained by backtracking as follows:

[0081]

[0082] S32: Iterative alignment to extract common operations in the data, further constructing a template library of clean sequences containing only pilot operation information as reference data, requiring only a small number of training samples. The specific steps are as follows:

[0083] S321: Randomly select two encoded sequences X from the training samples of a certain type of flight mission. t ={x1, x2, ..., x m} and Y t ={y1, y2, ..., y n}, use the BDTW method for alignment, and extract the matching pairs {x} on the optimal alignment path. i y j}, thus obtaining the normalized sequence X t and Y t X t and Y t The length value is equal to the number of matching pairs;

[0084] S322: Extract matching pairs with the same AOI value in the path. For elements that match multiple elements at the same time, select the element with the smallest difference in timestamp to form a matching pair.

[0085] S323: For each matching pair {x} in the previous step i y j}, calculate the average of the start timestamps for each pair of elements: And the average duration: Obtain subsequence Based on the calculated average timestamp, fill the sequence with zeros (0, t), where t is determined by the starting timestamp of the subsequence, to form a sequence with a time length equal to the optimal alignment path time length ∑x. i [1](or ∑y) j [1]) average sequence;

[0086] like Figure 5 As shown, Figure 5 The process of averaging the sequences is presented in the form of a decoded one-dimensional sequence, and the output sequence contains common operational information of the two input sequences.

[0087] S324: Randomly select a sequence from the remaining training samples and align it with the intermediate output sequence generated in the previous step. Repeat this process until the last training sample is used up, and output the reference data for this type of flight mission.

[0088] S325: Continue to extract reference data from training samples of other flight missions to build a template library.

[0089] S4: Use BDTW to align the test data and reference sequences of each type in the template library, calculate the similarity, and obtain the recognition result of the test data based on the reference sequence selected when generating the maximum similarity.

[0090] Specifically, step S4 includes the following steps:

[0091] S41: Similarity is the classification criterion for test data, determined by the number of non-zero elements n in the reference data. ref The number of matching pairs n with a distance of zero mch The range is [0,1], and it is defined as:

[0092] S42. Calculate the similarity between the test data and all types of data in the template library respectively. The template type selected when the maximum value is taken is the flight mission type to which the test data belongs. Identification is complete.

[0093] Compared with existing technologies, this invention applies the BDTW algorithm from the field of sparse time series recognition to the recognition of pilot hand AOI sequences. This eliminates sequence offsets on the time axis, avoids the impact of varying time spent by the pilot performing each step on similarity measurement, and improves upon the high time complexity of traditional DTW. Furthermore, this invention addresses the outlier interference problem in BDTW matching of the original AOI sequence by proposing an iterative alignment method for extracting reference data. A template library of reference data is constructed, and the latter is identified through sequence alignment and similarity calculation between the reference data and the test data, resulting in more accurate recognition results.

[0094] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A flight mission identification method based on BDTW and sequence alignment, characterized in that, The method includes the following steps: Obtain the sequence of the hand region of interest of the pilot in the flight simulator experiment, encode the data, and perform preprocessing; Use the existing labeled sequence data of the hand region of interest as training data, and use the experimental data as the test; Use the block dynamic time warping algorithm BDTW to perform iterative alignment processing on the training samples under each task, obtain the reference sequences, and form a template library; Use BDTW to align the test data and each type of reference sequence in the template library respectively, calculate the similarity, and obtain the recognition result of the test data according to the reference sequence selected when generating the maximum similarity.

2. The flight mission identification method based on BDTW and sequence alignment according to claim 1, characterized in that, The steps of obtaining the sequence of the hand region of interest of the pilot in the flight simulator experiment, encoding the data, and performing preprocessing specifically include: Let X be the sequence of regions of interest for the pilot's hand. raw ={x 11 ,x 12 ,…,x1 l1 ,x 21 ,x 22 ,…,x2 l2 ,…,xk1,xk2,…,xk lk }, k represents the number of repeated subsequences, xj is a value in the sequence, represented by an integer, indicating the AOI of the pilot's hand, j∈[1,k], j indicates which subsequence, {xj1,xj2,…,xj lj } represents a continuous subsequence of repeated values ​​in the original region of interest sequence, lj represents the total length of the subsequence of repeated values, and the subscript of xj indicates the position of the element in the subsequence within the region of interest sequence; Encode the original sequence by encoding all consecutive repeating subsequences {xj1, xj2, ..., xj...}. lj }, convert them into the encoded form (xj, lj), and then assemble them into a two-dimensional encoded sequence X in the order they appear. encoded ={(x1,l1),(x2,l2),…,(xk,lk)}; Remove the noise existing in the original sequence by traversing the encoded data.

3. The flight mission identification method based on BDTW and sequence alignment according to claim 2, characterized in that, The steps of removing the noise existing in the original sequence by traversing the encoded data specifically include: For the subsequence of non-zero same values existing in the form of a cluster, set a threshold T1. If the length l0 of a subsequence (0, l0) with a value of zero is less than T1, and its two adjacent subsequences (xa, la), (xb, lb) have the same non-zero value, i.e., xa = xb, then merge these three subsequences into (xa, la + l0 + lb); For the non-zero subsequences that do not exist in the form of a cluster and have too small lengths, set a threshold T2, and set the value of the non-zero subsequence (xc, lc) with a length lc < T2 to zero, and the processing result is (0, lc).

4. The flight mission identification method based on BDTW and sequence alignment according to claim 1, characterized in that, In the step of using the existing labeled sequence data of the hand region of interest as training data and using the experimental data as the test, it is necessary to ensure the accuracy of the training data, that is, it should include a complete operation procedure.

5. The flight mission identification method based on BDTW and sequence alignment according to claim 1, characterized in that, The steps of using the block dynamic time warping algorithm BDTW to perform iterative alignment processing on the training samples under each task, obtain the reference sequences, and form a template library specifically include: Execute the sequence alignment process through the BDTW method. The sequence alignment process needs to meet the principles of monotonicity, boundary matching, continuity, and optimal matching, and calculate the optimal alignment path; Perform iterative alignment to extract the common operations in the data, and further construct a clean sequence containing only the operation information of the pilot as the template library of the reference data.

6. The flight mission identification method based on BDTW and sequence alignment according to claim 5, characterized in that, The steps of executing the sequence alignment process through the BDTW method, where the sequence alignment process needs to meet the principles of monotonicity, boundary matching, continuity, and optimal matching, and calculate the optimal alignment path specifically include: Two encoded sequences X encoded ={x1, x2, ..., x m } and Y encoded ={y1, y2, ..., y n The pairwise matching results of all elements are listed in an m×n matching matrix. The element D at position (i, j) in the matching matrix is... i，j x represents the i-th element in the first sequence. i = (xi, li) and the j-th element y of the second sequence j The distance between (yj, lj) is calculated using the formula: D i，j =(x i [0]-y j [0]) 2 ; Monotonicity requires that elements in two encoded sequences must be aligned in chronological order, while continuity requires that each element in the encoded sequence has a matching object if the previous match of the path is D. i，j Then the current match can only be D. i,j+1 D i,j+1 D i+1,j+1 One of them; Boundary matching requires that the first elements of two encoded sequences must match each other, and the last elements must match each other; that is, the optimal matching path must start with matching D. 0,0 Begin by matching D m,n Finish; The optimal matching principle requires that the current match D i,j The previous match can only be D. i-1,j D i,j-1 D i-1,j-1 In the middle, arrive at D i,j To minimize the time distance, an m×n cumulative distance matrix d is calculated, with matrix elements d... i，j Let represent the minimum cumulative distance from the (0, 0)th element to the (i, j)th element of the matching matrix. The calculation process is as follows: d 0,0 ＝0 top=d i-1，j +x i [1]*D i,j diagonal=d i-1，j-1 +max(x i [1],y j [1])*D i，j left=d i，j-1 +y i [1]*D i，j d i，j =min(top,diagonal,left) Based on the obtained cumulative distance matrix d, from d m,n Start backtracking until the endpoint d. 0，0 The optimal alignment path can be obtained by backtracking as follows:

7. The flight mission identification method based on BDTW and sequence alignment according to claim 5, characterized in that, The steps of performing iterative alignment to extract the common operations in the data, and further construct a clean sequence containing only the operation information of the pilot as the template library of the reference data specifically include: Randomly select two encoded sequences from the training samples of a type of flight task, use BDTW for alignment, and extract the optimal alignment path; Extract the matching pairs with the same AOI value in the path. For those elements that match multiple elements at the same time, take the element with the smallest time stamp difference from it to form a matching pair; For the matching pairs selected in the previous step, calculate the average value of the start time stamps and the average value of the durations between each pair, and form a sequence with a time length equal to the time length of the optimal alignment path; Randomly select a sequence from the remaining training samples and align it with the intermediate output sequence generated in the previous step. Repeat this process until the last training sample is used up, and output the reference data for this type of flight mission. Continue to extract reference data from training samples of other flight missions to build a template library.

8. The flight mission identification method based on BDTW and sequence alignment according to claim 1, characterized in that, The steps of using BDTW to align the test data and each type of reference sequence in the template library, calculating the similarity, and obtaining the recognition result of the test data based on the reference sequence selected when generating the maximum similarity specifically include: Similarity is the classification criterion for test data, determined by the number of non-zero elements n in the reference data. ref The number of matching pairs n with a distance of zero mch The range is [0,1], and it is defined as: Calculate the similarity between the tested data and all types of data in the template library. The template type selected when the maximum similarity is taken is the flight mission type to which the test data belongs. Identification is complete.

9. A flight mission identification device based on BDTW and sequence alignment, characterized in that, The apparatus includes a processor and a memory for storing executable instructions of the processor, the processor being configured to perform the flight mission identification method based on BDTW and sequence alignment as described in any one of claims 1 to 8 by executing the executable instructions.

Images