A Deep Learning-Assisted Extended Kalman Array for Dual-Band Infrared Radiation Attitude Estimation

Abstract

This invention discloses a deep learning-assisted extended Kalman dual-band infrared radiation attitude estimation method. The method analyzes the impact of geometric and random errors in dual-band infrared radiation measurement signals on attitude estimation reliability. A CBGTE model is constructed to eliminate geometric and random errors in the dual-band infrared radiation signal. First, a dual-band infrared radiation signal dataset is collected and cleaned using a hardware-in-the-loop simulation platform. The dataset is then input into a BGT neural network for feature learning and optimized weights are saved. Data acquired by the dual-band infrared radiation sensor is then input into the BGT neural network with optimized weights to obtain infrared radiation data after random error compensation. This data, after random error filtering, is then input into an extended Kalman algorithm to filter out geometric errors, thereby obtaining reliable attitude angle information. This invention provides a high-precision and robust method for acquiring and processing attitude information for a spinning drone using infrared radiation as its attitude measurement method.

Description

Technical Field

[0001] This invention relates to the field of deep learning feature extraction technology, and in particular to a deep learning-assisted extended Kalman scattering method for dual-band infrared radiation attitude estimation. Background Technology

[0002] With advancements in infrared sensor technology, optimizations have been made in areas such as size, speed, and energy efficiency. Consequently, infrared detection technology is being applied in various cutting-edge fields, including infrared thermal imaging, infrared communication, and dual-band infrared radiation attitude measurement. The principle of dual-band infrared radiation attitude measurement is to extract navigation information from different radiation fields of the Earth using dual-band infrared radiation sensors. Compared to mainstream attitude measurement technologies such as inertial systems and Global Positioning Systems (GPS), dual-band infrared radiation attitude measurement technology offers significant advantages, such as resistance to electromagnetic interference, passive operation, and low cost. Therefore, dual-band infrared radiation attitude measurement systems are widely used in spin-based unmanned aerial vehicles (UAVs). However, spin-based UAVs are complex, multivariable, and strongly coupled nonlinear systems, making their attitude information easily susceptible to interference, thus affecting their reliability.

[0003] Despite the numerous advantages of dual-band infrared radiation attitude measurement technology, some drawbacks have been identified with further application. Due to the nature of dual-band infrared radiation sensors, geometric and random errors are introduced into the data. In practical applications, all sensors are affected by geometric errors, including sensor measurement noise, assembly position, and motor interference. Researchers typically employ various filtering algorithms to address geometric error issues, such as Bayesian filtering, Kalman filtering, unscented Kalman filtering, extended Kalman filtering, and particle filtering.

[0004] Since the flight background of a spinning UAV is the atmosphere, random errors are unique to dual-band infrared radiation sensors. Attitude information is inevitably affected by various atmospheric radiations, such as solar interference and cloud interference. These interferences introduce random errors into the collected data, significantly affecting attitude estimation and making the attitude information unreliable. Therefore, current research on dual-band infrared radiation attitude measurement technology focuses on further mitigating attitude estimation errors caused by random errors. Summary of the Invention

[0005] The problem to be solved by this invention is to provide a deep learning-assisted extended Kalman dual-band infrared radiation attitude estimation method to reduce attitude estimation errors caused by random errors.

[0006] This invention adopts the following technical solution: a deep learning-assisted extended Kalman spectroscopy dual-band infrared radiation attitude estimation method, comprising the following steps:

[0007] S1. Analyze the causes of geometric and random errors in the dual-band infrared radiation measurement signals of a spin-operated UAV and their impact on the reliability of attitude estimation.

[0008] S2. Collect dual-band infrared radiation data of the spin-type UAV through a hardware-in-the-loop simulation platform, and preprocess the collected dataset.

[0009] S3. Construct a CBGTE model, including a BGT neural network and an extended Kalman filter. Input the collected dataset into the CBGTE model, train the CBGTE model, and obtain the optimal weights of the CBGTE model.

[0010] S4. The optimal weights of the CBGTE model are put into the test equipment to verify the feasibility of the algorithm and to estimate the attitude of the dual-band infrared radiation of the spin-type UAV.

[0011] Preferably, in step S1, the geometric error is caused by sensor measurement noise, assembly position and motor interference; from the perspective of time series, the geometric error value is random and independent of time, and the amplitude is in the range of [-0.02, 0.02]V during the measurement process on the semi-physical platform.

[0012] Random errors are caused by solar and cloud interference. From a time series perspective, random errors are limited by time intervals, affecting a range of half a time period, but there are no limitations on their amplitude.

[0013] Preferably, in step S2, the sampling rate and sampling time of the hardware-in-the-loop simulation platform are set, dual-band infrared radiation data are collected under different atmospheric radiation backgrounds, and the collected dataset is preprocessed as follows:

[0014] S201. Classify and organize the dataset according to the type of interference, dividing the dataset into: no interference, solar interference, and cloud interference.

[0015] S202. Due to the inconsistent amplification coefficients of the conditioning circuits, the amplitudes of the long-wave infrared radiation data and the mid-wave infrared radiation data are inconsistent. All collected infrared radiation data are normalized to unify the dual-band infrared radiation data into the (-1,1) interval.

[0016] S203. Clean the normalized dataset, delete unusable data, and ensure the reliability of the dataset.

[0017] Preferably, in step S3, in the CBGTE model, the BGT neural network is used to compensate for random errors in the dual-band infrared attitude measurement process, and the extended Kalman filter is used to eliminate geometric errors in the dual-band infrared attitude measurement process.

[0018] The BGT neural network, which integrates Bi-GRU and Transformer Encoder networks, is used to compensate for random errors in dual-band infrared radiation information. The BGT neural network consists of an encoder and a decoder. The encoder contains Bi-GRU modules and Transformer Encoder modules, and the decoder consists of Bi-GRU and fully connected layers.

[0019] An extended Kalman filter is used to derive a nonlinear prediction equation from the observation equation and run it in tangent approximation. The first-order Taylor expansion around the average state is used to eliminate geometric errors in the infrared radiation information compensated by the BGT neural network.

[0020] Preferably, in step S3, data augmentation is performed before training the CBGTE by sampling infrared data, with each 10-second data point serving as a training sample. The loss function is the mean squared error function, the optimizer is set to Adamw, the learning rate is 0.0005, the epoch is set to 200, and the weight decay is set to 0.01. The attitude data curve is obtained based on a deep learning-assisted extended Kalman algorithm.

[0021] The present invention also provides: an electronic device, comprising:

[0022] One or more processors;

[0023] A storage device on which one or more programs are stored;

[0024] When the one or more programs are executed by the one or more processors, the one or more processors implement any of the deep learning-assisted extended Kalman dual-band infrared radiation attitude estimation methods described above.

[0025] The present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps in any of the above-mentioned deep learning-assisted extended Kalman dual-band infrared radiation attitude estimation methods.

[0026] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0027] 1. Compared with existing methods, the dual-band infrared radiation attitude estimation method of this invention has demonstrated superior performance through semi-physical experiments, with the calculated roll angle error being ±0.4° and the pitch angle error being ±0.2°.

[0028] 2. The dual-band infrared radiation attitude estimation method of the present invention uses extended Kalman filtering to eliminate geometric errors in the dual-band infrared attitude measurement process and BGT neural network to compensate for random errors in the dual-band infrared attitude measurement process.

[0029] 3. This invention uses a BGT architecture. Unlike the traditional Bi-GRU neural network, the dual-band infrared radiation attitude estimation method of this invention introduces a multi-head attention mechanism to capture key information features in infrared radiation data, thereby realizing a high-precision BGT neural network framework structure. Attached Figure Description

[0030] Figure 1 This is a flowchart of the dual-band infrared radiation attitude estimation method of deep learning-assisted extended Kalman scattering in this invention;

[0031] Figure 2 This is a schematic diagram of the semi-physical experimental data of the present invention;

[0032] Figure 3 This is a structural diagram of the CBGTE model of the present invention;

[0033] Figure 4 This is a schematic diagram of the error values ​​calculated by the CBGTE model in this invention. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the application will be further described in detail below with reference to the accompanying drawings. The described embodiments are only a part of the embodiments involved in this invention. All non-innovative embodiments based on these embodiments by other researchers in the art are within the protection scope of this invention. Furthermore, the step numbers in the embodiments of this invention are only set for ease of explanation and do not limit the order of the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0035] In one embodiment of the present invention, a deep learning-assisted extended Kalman scattering algorithm for dual-band infrared radiation attitude estimation is provided, such as... Figure 1 As shown, it includes the following steps:

[0036] S1. Analyze the impact of geometric errors and random errors in dual-band infrared radiation measurement signals on the reliability of attitude estimation, including: a) analyzing the causes of geometric errors and their impact on attitude reliability, b) analyzing the causes of random errors and their impact on attitude reliability.

[0037] Among them, a) geometric errors are often caused by factors such as sensor measurement noise, assembly position, and motor interference. Geometric errors are characterized by randomness in value, limited amplitude, and non-temporal sequence. The impact on dual-wave infrared radiation attitude information is that a random noise with limited amplitude is superimposed on the true attitude value.

[0038] b) Random errors are unique to dual-band infrared radiation attitude measurement systems. The core principle of this system is to extract navigation information from infrared radiation fields in different regions using infrared radiation sensors. However, if the infrared radiation in space changes, the output data of the infrared radiation sensors will also change accordingly.

[0039] Experiments show that variations in radiation interference in space are mainly caused by solar and cloud interference. Mid-wave sensors are more sensitive to solar interference, while long-wave sensors are more sensitive to cloud interference. These interferences introduce random errors into infrared radiation data. These random errors cause excessive deviations in attitude angle values ​​during infrared radiation attitude calculation, significantly affecting the infrared radiation data within the corresponding range and rendering the attitude angle information unreliable.

[0040] Preferably, in step S1 of this embodiment, from a time series perspective, geometric errors are often random in value and independent of time, with amplitudes within the range of [-0.02, 0.02]V during the measurement process on the hardware-in-the-loop platform. From a time series perspective, random errors are often limited by a time interval, affecting an area within approximately half a time period, and have no significant limitation on amplitude.

[0041] S2. Collect dual-band infrared radiation data through a hardware-in-the-loop simulation platform, and then process the data as follows: a) Classify and organize the dataset according to the type of interference; b) Normalize and clean the dataset through the preprocessing module.

[0042] Among them, a) dual-band infrared radiation data under different radiation backgrounds were collected through a hardware-in-the-loop simulation platform, and the dataset was divided into three types according to the type of interference: no interference, solar interference, and cloud interference.

[0043] b) Due to factors such as inconsistent amplification coefficients in the conditioning circuit, the amplitudes of long-wave infrared radiation data and mid-wave infrared radiation data are inconsistent. Therefore, it is necessary to normalize all infrared radiation data and clean the normalized dataset to ensure its reliability.

[0044] Preferably, in step S2 of this embodiment, as follows: Figure 2 As shown, the hardware-in-the-loop simulation platform was set to a sampling rate of 10kHz and a sampling time of 10s to collect infrared radiation data. Data sets were collected and preprocessed under different atmospheric radiation backgrounds over a year.

[0045] First, normalization is performed to unify the dual-band infrared radiation data into the (-1, 1) interval, as shown in the following formula:

[0046]

[0047] In the formula, X i To filter the input data, Xmin and X max and represent the minimum and maximum values ​​in the input matrix, respectively, and X represents the normalized output data.

[0048] Next, the dataset is cleaned by removing unusable data.

[0049] S3. Construct the CBGTE model by inputting the collected dual-band infrared radiation dataset into the CBGTE model, training the CBGTE algorithm, and saving the optimal weights of the model. Specifically, the CBGTE model consists of two main modules: a) a BGT neural network model and b) an extended Kalman filter.

[0050] In this embodiment, a) the BGT neural network is a fusion variant of Bi-GRU and Transformer Encoder, mainly consisting of two parts: an encoder and a decoder.

[0051] The BGT neural network combines the advantages of both architectures: Bi-GRU effectively captures contextual information in a sequence by processing it bidirectionally, thus considering past and future states and enhancing the understanding of temporal dependencies. The Transformer employs a self-attention mechanism, which allows for parallel processing of the entire sequence and captures global contextual information, thereby increasing the weight of key information.

[0052] The main function of the BGT neural network is to compensate for random errors in dual-band infrared radiation information. The encoder mainly consists of a Bi-GRU and a Transformer Encoder, while the decoder is composed of a Bi-GRU and fully connected layers.

[0053] First, Bi-GRU is used to extract time series information from two directions. The bidirectional extracted feature information is then superimposed according to weights, which can maximize the data utilization rate.

[0054] Next, the extracted feature information is processed through the Transformer Encoder module, which can effectively extract features of key information.

[0055] Then, the feature information encoded by the encoder is decoded by the decoder. First, it is decoded through the Bi-GRU layer, and then the decoded features are mapped through the fully connected layer.

[0056] Finally, BGT can output dual-band infrared radiation information after compensating for random errors.

[0057] Preferably, in step S3 of this embodiment, as follows: Figure 3 As shown, the BGT neural network consists of an encoder and a decoder, and the specific processing is as follows:

[0058] The preprocessed infrared radiation information is input into the encoder. First, bidirectional information extraction is performed using a Bi-GRU, which is an improvement on the GRU network. The principle of GRU is shown in the following formula:

[0059]

[0060] In the formula, σ represents the sigmoid activation function, and W r W z W and X represent the weights of the forget gate, update gate, and hidden layer, respectively. k h represents the input of the model at time k. k-1 Represents the hidden layer state at time k-1. h represents the sum of the past and current hidden layer states at time k. k This represents the output of the hidden layer at time k.

[0061] Bi-GRU adds bidirectional feature extraction to GRU, as shown in the following formula:

[0062]

[0063] In the formula, GRU(·) represents the GRU operation. and For the forward hidden sequence and the backward hidden sequence, W a and W b This represents the weights of the forward and backward hidden sequences.

[0064] Next, the feature information extracted by Bi-GRU is used to extract key information features through the Transformer Encoder module, which mainly consists of two key parts: multilayer perceptron (MLP) and multi-head self-attention mechanism.

[0065] The Multilayer Perceptron (MLP) consists of two fully connected linear layers, one activation function, and two dropout layers. Unlike the traditional Transformer Encoder, changing the activation function from ReLU to SiLU significantly improves accuracy. The MLP calculation is shown in the following formula:

[0066]

[0067] In the formula, w1 and b1 represent the weights and biases of the first fully connected layer, and w2 and b2 represent the weights and biases of the second fully connected layer.

[0068] Multi-head self-attention mechanism is an improvement on self-attention mechanism. Self-attention mechanism enhances the correlation of internal features by minimizing external interference. It transforms the input vector into query matrix Q, key matrix K, and value matrix V, weights the input matrices and passes them through the softmax(·) function, and finally multiplies them by the value matrix. The detailed formula is shown below:

[0069]

[0070] In the formula, d k This represents the dimension of the key matrix. T Indicates transpose;

[0071] The multi-head self-attention mechanism maps the self-attention mechanism into a multi-dimensional space and concatenates the features extracted from each space using the Concat(·) function, as shown in the following formula:

[0072]

[0073] In the formula, W i Q W i K and W i V These are the corresponding spatial mapping matrices of the Q, K, and V matrices, respectively. i It is the head matrix in the i-th dimension space. The Attention(·) function is a self-attention mechanism, and the MHAT(·) function is a multi-head attention mechanism.

[0074] Therefore, the entire Transformer Encoder module can be expressed by the following formula:

[0075] X o =MLP(LN(MHAT(LN(X) i ))+X i ))+MHAT(LN(X i ))+X i (7)

[0076] In the formula, X i and X o It is the input and output of the Transformer Encoder layer, and the LN(·) function is the LayerNormalization function;

[0077] The overall flow of the BGT neural network is shown in the following formula:

[0078]

[0079] In the formula, X is the input of the neural network, Bi-GRU(·) function is the Bi-GRU operation, TransformerEncoder(·) function is the Transformer Encoder operation, FC(·) function is the fully connected operation, h0 is the bidirectional information feature passed through Bi-GRU for the first time, h1 is the key feature extracted by the Transformer Encoder module, h2 is the decoded information feature passed through Bi-GRU, and Y is the infrared information feature after compensation RE.

[0080] Furthermore, in this embodiment, b) geometric errors pose a challenge to the dual-band infrared radiation attitude system in practical applications. During attitude measurement, various factors can lead to geometric errors, including sensor measurement noise, assembly position changes, and motor interference, which reduce the reliability of attitude data.

[0081] In this embodiment, an extended Kalman filter is used to address this challenge during attitude measurement. The infrared radiation information compensated by the BGT neural network is then processed by the extended Kalman filter to eliminate geometric errors.

[0082] The extended Kalman filter is a special type of Kalman filter that treats nonlinear systems by linearizing them into local approximations. It operates by deriving nonlinear prediction equations from the observation equations and using tangent approximations. Its core principle is based on a first-order Taylor expansion around the average state.

[0083] The state matrix of the extended Kalman filter is set as X = [γ ω θ] T γ is the roll angle, ω is the rotational speed, and θ is the pitch angle.

[0084] The state transition matrix is Where Δt is the time step.

[0085] The prediction noise covariance matrix can be obtained by the following formula:

[0086] P(k|k-1)=FP(k-1)F T +Q (9)

[0087] In the formula, P(k|k-1) is the prediction noise covariance matrix, Q is the system noise covariance matrix, and P(k-1) is the covariance matrix at time k-1.

[0088] The observation matrix is ​​represented as Y = [L1 L2 M1 M2] T Based on the dual-band infrared attitude measurement model, the linearized output matrix can be obtained as follows:

[0089] The update step involves predicting the output matrix H and the noise covariance matrix P(k|k-1) to calculate the Kalman gain K, as shown in the following formula:

[0090] K(k)=P(k|k-1)H(k) T [H(k)P(k|k-1)H(k) T +R] -1 (10)

[0091] In the formula, R represents the observation noise variance matrix.

[0092] The gain matrix K adjusts the estimated attitude parameters and updates the state matrix X, as shown in the following formula:

[0093] X(k)=X(k-1)+K(k)[Y(k)-H(k)] (11)

[0094] In the formula, Y(k) represents the actual measurement data of the sensor at time k, and H(k) represents the estimated data of the sensor at time k.

[0095] Finally, the noise covariance matrix P(k) of the filter is updated by the following formula:

[0096] P(k)=[IK(k)H(k)]P(k|k-1) (12)

[0097] It is particularly important to note that in step S3, data augmentation is required before training the CBGTE model. Infrared data is sampled, with each 10-second data point serving as a training sample. The loss function is the mean squared error function, the optimizer is set to Adamw, the learning rate is 0.0005, the epoch is set to 200, and the weight decay is set to 0.01. Finally, the pose data curve is obtained based on the deep learning-assisted extended Kalman algorithm.

[0098] Specifically, the optimal model weights for BGT are saved as follows: The optimal weights of the BGT model are saved, and the optimal loss (loss_best) is set to 100 before training begins. After each training round, the validation set is input into the BGT neural network to obtain the loss value on the validation set. If the current loss is less than loss_best, the loss is assigned to loss_best, and the model weights from this training round are saved as the optimal weights, continuing to the next training round. If the loss is greater than loss_best, no saving is required, and the next training round proceeds directly. The final optimal model weights are saved and validated on the device.

[0099] S4. Put the optimal weights of the model into the FPGA (ZYNQ7020) device to verify the feasibility of the algorithm and estimate the dual-band infrared radiation attitude of the spin-type UAV.

[0100] Specifically, based on the optimized weights of the CBGTE model obtained through optimization, model weight transfer is performed, and the dual-band infrared radiation attitude measurement data collected by the device is input into the FPGA (ZYNQ7020) device for model prediction output.

[0101] In summary, this invention designs a deep learning-assisted extended Kalman spectroscopy dual-band infrared radiation attitude estimation algorithm. Its superior performance is verified through semi-physical experiments, with the calculated roll angle error at steady state being ±0.2° and the pitch angle error at ±0.5°. Figure 4 As shown.

[0102] In this embodiment of the invention, an electronic device is also provided, comprising: one or more processors; a storage device storing one or more programs thereon; when the one or more programs are executed by the one or more processors, the one or more processors implement the deep learning-assisted extended Kalman dual-band infrared radiation attitude estimation method described in any of the above embodiments.

[0103] In this embodiment of the invention, a computer-readable storage medium is also provided, on which a computer program is stored. When the program is executed by a processor, it implements the steps in any of the deep learning-assisted extended Kalman dual-band infrared radiation attitude estimation methods described in the above embodiments.

[0104] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A deep learning-assisted extended Kalman spectroscopy dual-band infrared radiation attitude estimation method, characterized in that, The steps include: S1. Analyze the causes of geometric and random errors in the dual-band infrared radiation measurement signals of a spin-operated UAV and their impact on the reliability of attitude estimation. S2. Collect dual-band infrared radiation data of the spin-type UAV through a hardware-in-the-loop simulation platform, and preprocess the collected dataset. S3. Construct a CBGTE model, including a BGT neural network and an extended Kalman filter. Input the collected dataset into the CBGTE model, train the CBGTE model, and obtain the optimal weights of the CBGTE model. In the CBGTE model, the BGT neural network is used to compensate for random errors in the dual-band infrared attitude measurement process, and the extended Kalman filter is used to eliminate geometric errors in the dual-band infrared attitude measurement process. The BGT neural network integrates Bi-GRU and Transformer Encoder networks to compensate for random errors in dual-band infrared radiation information. The BGT neural network consists of an encoder and a decoder. The encoder includes a Bi-GRU module and a Transformer Encoder module, and the decoder consists of a Bi-GRU and a fully connected layer. The extended Kalman filter derives a nonlinear prediction equation from the observation equation and runs it in tangent approximation. It is expanded in first-order Taylor around the average state and is used to eliminate geometric errors in the infrared radiation information compensated by the BGT neural network. S4. The optimal weights of the CBGTE model are put into the test equipment to verify the feasibility of the algorithm and to estimate the attitude of the dual-band infrared radiation of the spin-type UAV.

2. The deep learning-assisted extended Kalman dual-band infrared radiation attitude estimation method according to claim 1, characterized in that, In step S1, the geometric error is caused by sensor measurement noise, assembly position and motor interference; from a time series perspective, the geometric error value is random and independent of time, and the amplitude is within the range of [-0.02, 0.02]V during the measurement process on the semi-physical platform. The random error is caused by solar interference and cloud interference. From a time series perspective, the random error is limited by a time interval, and its influence range is within half a time period. There is no limit to the amplitude.

3. The deep learning-assisted extended Kalman dual-band infrared radiation attitude estimation method according to claim 1, characterized in that, In step S2, the sampling rate and sampling time of the hardware-in-the-loop simulation platform are set, dual-band infrared radiation data are collected under different atmospheric radiation backgrounds, and the collected dataset is preprocessed as follows: S201. Classify and organize the dataset according to the type of interference, dividing the dataset into: no interference, solar interference, and cloud interference. S202. Due to the inconsistent amplification coefficients of the conditioning circuits, the amplitudes of the long-wave infrared radiation data and the mid-wave infrared radiation data are inconsistent. All collected infrared radiation data are normalized to unify the dual-band infrared radiation data into the [-1,1] interval. S203. Clean the normalized dataset, delete unusable data, and ensure the reliability of the dataset.

4. The deep learning-assisted extended Kalman dual-band infrared radiation attitude estimation method according to claim 3, characterized in that, In step S202, all infrared radiation data are normalized using the following formula: In the formula, X i To filter the input data, X min and X max and represent the minimum and maximum values ​​in the input matrix, respectively, and X represents the normalized output data.

5. The deep learning-assisted extended Kalman dual-band infrared radiation attitude estimation method according to claim 3, characterized in that, In step S3, the collected dataset is input into the CBGTE model to train and learn the model, obtaining the optimal weights. The method is as follows: S301. Input the preprocessed infrared radiation information into the encoder, extract bidirectional information through the Bi-GRU module, and superimpose the bidirectional extracted feature information according to weights to maximize data utilization. The Bi-GRU module is based on an improved GRU network, and the formula is as follows: In the formula, σ represents the sigmoid activation function, and W r W z W and X represent the weights of the forget gate, update gate, and hidden layer, respectively. k h represents the input of the model at time k. k-1 This represents the hidden layer state at time k-1. h represents the sum of the past and current hidden layer states at time k. k This represents the output of the hidden layer at time k; Bidirectional feature extraction is performed using the Bi-GRU module, as shown in the following formula: In the formula, GRU(·) represents the GRU operation. and For the forward hidden sequence and the backward hidden sequence, W a and W b Represents the weights of the forward and backward hidden sequences; S302. Input the bidirectional feature information extracted by the Bi-GRU module into the Transformer Encoder module to extract the key information features; The Transformer Encoder module includes: a multilayer perceptron and a multi-head self-attention mechanism; The multilayer perceptron consists of two fully connected linear layers, one activation function, and two dropout layers. The activation function is the SiLu function, and the MLP is calculated as follows: In the formula, w1 and b1 represent the weights and biases of the first fully connected layer, and w2 and b2 represent the weights and biases of the second fully connected layer. The multi-head self-attention mechanism is an improvement on the self-attention mechanism. It transforms the input vector into a query matrix Q, a key matrix K, and a value matrix V. The input matrices are weighted and then multiplied by the value matrix using the softmax(·) function, as shown in the following formula: In the formula, d k The dimension of the key matrix is ​​represented by T, which represents the transpose. The self-attention mechanism is mapped to a multi-dimensional space through a multi-head self-attention mechanism, and the features extracted from each space are concatenated using the Concat(·) function, as shown in the following formula: In the formula, W i Q W i K and W i V These are the corresponding spatial mapping matrices of the Q, K, and V matrices, respectively. i It is the head matrix in the i-th dimension space. The Attention(·) function is a self-attention mechanism, and the MHAT(·) function is a multi-head attention mechanism. The complete Transformer Encoder module can be represented by the following formula: X o =MLP(LN(MHAT(LN(X i ))+X i ))+MHAT(LN(X i ))+X i ; In the formula, X i and X o It is the input and output of the Transformer Encoder layer, and the LN(·) function is the LayerNormalization function; S303. The feature information encoded by the encoder is decoded by the decoder. First, the decoded features are obtained by decoding through a Bi-GRU network. Then, the decoded features are mapped through a fully connected layer to output the dual-band infrared radiation information after the overall BGT neural network compensates for random errors, as shown below: In the formula, X is the input of the neural network, Bi-GRU(·) function is the Bi-GRU operation, TransformerEncoder(·) function is the Transformer Encoder operation, FC(·) function is the fully connected operation, h0 is the bidirectional information feature passed through Bi-GRU for the first time, h1 is the key feature extracted by the Transformer Encoder module, h2 is the decoded information feature passed through Bi-GRU, and Y is the infrared information feature after compensation RE.

6. The deep learning-assisted extended Kalman dual-band infrared radiation attitude estimation method according to claim 5, characterized in that, In step S3, the extended Kalman filter has the following state matrix: X=[goth] T ; The state transition matrix is: Where γ is the roll angle, ω is the rotational speed, θ is the pitch angle, and Δt is the time step. The prediction noise covariance matrix is ​​obtained by the following formula: P(k|k-1)=FP(k-1)F T +Q; In the formula, P(k|k-1) is the prediction noise covariance matrix, Q is the system noise covariance matrix, and P(k-1) is the covariance matrix at time k-1. The observation matrix is ​​represented as: Y=[L1 L2 M1 M2] T ; In the formula, L1, L2, M1, and M2 represent the output data of long wave 1, long wave 2, medium wave 1, and medium wave 2, respectively. Based on the dual-band infrared attitude measurement model, the linearized output matrix is ​​obtained as follows: The update step includes predicting the output matrix H and the noise covariance matrix P(k|k-1) to calculate the Kalman gain K, as follows: K(k)=P(k|k-1)H(k) T [H(k)P(k|k-1)H(k) T +R] -1 ; In the formula, R represents the observation noise variance matrix; The estimated attitude parameters are adjusted using the gain matrix K, and the state matrix X is updated, as shown in the following formula: X(k)=X(k-1)+K(k)[Y(k)-H(k)]; In the formula, Y(k) represents the actual measurement data of the sensor at time k, and H(k) represents the estimated data of the sensor at time k; The noise covariance matrix P(k) of the extended Kalman filter is updated by the following formula: P(k)=[IK(k)H(k)]P(k|k-1).

7. The deep learning-assisted extended Kalman dual-band infrared radiation attitude estimation method according to claim 1, characterized in that, In step S3, data augmentation is performed before training the CBGTE model. Infrared data is sampled, with each 10 seconds of data serving as a training sample. The loss function is set to mean squared error, the optimizer to Adamw, the learning rate to 0.0005, the epoch to 200, and the weight decay to 0.

01. The attitude data curve is obtained based on the deep learning-assisted extended Kalman algorithm.

8. An electronic device, characterized in that, include: one or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the deep learning-assisted extended Kalman dual-band infrared radiation attitude estimation method as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the deep learning-assisted extended Kalman dual-band infrared radiation attitude estimation method according to any one of claims 1 to 7.

Images