Method for enhancing runway data with aircraft information
By designing a runway data enhancement method with aircraft information, and using generative adversarial models and loss functions, the problems of small sample size of runway database and gradient disappearance and explosion in intelligent take-off and landing research were solved, and the generated sample quality was improved, which promoted the progress of intelligent take-off and landing research.
Patent Information
- Application Number
- CN202411670427.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-21
- Publication Date
- 2025-03-07
AI Technical Summary
In the study of intelligent take-off and landing, the runway database sample size is small and difficult to obtain, which affects the training of deep learning algorithms. The existing sample generation algorithm is prone to gradient vanishing and gradient explosion during the training process.
Design a runway data enhancement method with aircraft information. By designing activation functions and introducing a generative adversarial model, designing the loss function of the generation network and the discriminative network, and constructing training data with aircraft information. Use these data to separately train the generation network and the discriminative network to generate enhanced runway data.
Through this method, the runway database with aircraft information was expanded, the data shortage in the research on intelligent take-off and landing was improved, and the problems of gradient vanishing and gradient explosion were solved. The generated samples were less noise, the images were clearer, and the details were more detailed, which promoted the progress of research on intelligent take-off and landing.
Smart Images
Figure CN119578466A_ABST
Abstract
Description
Technical Field
[0001] The disclosed embodiments relate to the technical field of sample generation, and in particular to a runway data enhancement method with aircraft information. Background Art
[0002] Intelligent take-off and landing research and design of various deep learning algorithms require a large amount of runway data. In the deep learning training process, not only runway data is required, but also the corresponding aircraft information, such as aircraft pitch, roll, heading deviation, DME distance, radio altitude, etc. The airport runway database has the problem of small sample size and difficulty in obtaining samples, which affects the training of deep learning algorithms. Use sample generation algorithms to generate reliable runway images, expand the runway database, improve the training of deep learning algorithms, and promote the research of intelligent take-off and landing.
[0003] The sample generation algorithm mainly uses the oversampling method to process the data set and generate new sample data. The simplest oversampling is simple random oversampling, which randomly copies small samples, but the data information is not increased, which may cause overfitting and waste training time. The oversampling method based on K nearest neighbor divides the iterative data through the K nearest neighbor algorithm, reduces the noise in the data, and improves the generation quality. The SMOTE algorithm based on random oversampling generates new data between small samples and their K nearest neighbor samples through the sampling formula, which effectively increases the diversity of samples. Generative adversarial networks (GAN) are an emerging method that can learn data distribution, generate small samples, and generate high-quality sample data.
[0004] The Generative Adversarial Network consists of a generative network and a discriminative network, and its purpose is to generate data that is very similar to real data. The generator G (Generator) receives a random variable Z and outputs the generated data G (Z), so that the distribution of the generated data G (Z) is as similar as possible to that of the real data X. The discriminator D (Discriminator) should distinguish the real data X and the generated data G (Z) as much as possible. It inputs the real data X and the generated data G (Z) and outputs a probability value, indicating the probability that the generated data G (Z) is the real data X. At the same time, the output of the discriminator D will also be fed back to the generator G to cooperate with the training of the generator G. During the training, the generator G and the discriminator D compete with each other, and the generation and discrimination capabilities are continuously strengthened. In an ideal case, after the training is completed, the output of the discriminator D is 1 / 2, and it is unable to distinguish the type of input data. Deep Convolutional Generative Adversarial Networks (DCGAN) adds a deep convolutional network structure based on GAN. The generator G uses the deconvolution operation to upsample the input of each layer to generate an image. The original DCGAN model is unstable in training and prone to gradient vanishing and gradient exploding problems, and the quality of generated samples needs to be improved.
[0005] Therefore, it is necessary to improve one or more problems existing in the above-mentioned related technical solutions.
[0006] It should be noted that this section is intended to provide background or context for the technical solutions of the present disclosure stated in the claims. The description herein is not admitted to be prior art by virtue of being included in this section. Summary of the invention
[0007] The purpose of the embodiments of the present disclosure is to provide a runway data enhancement method with aircraft information, thereby overcoming one or more problems caused by limitations and defects of related technologies at least to a certain extent.
[0008] According to an embodiment of the present disclosure, a method for enhancing runway data with aircraft information is provided, the method comprising: Designing an activation function and introducing the activation function into a generative adversarial model; wherein the generative adversarial model includes a generative network and a discriminative network; Designing a generator loss function of the generative network; Designing a discriminator loss function of the discriminant network; Constructing training data with aircraft information, and using the training data to separately train the generative network and the discriminative network to obtain a trained generative adversarial model; The runway data to be enhanced is input into the trained generative adversarial model to obtain enhanced runway data.
[0009] Furthermore, the activation function of the generative adversarial model is an S_ReLU function, and the expression of the S_ReLU function is:
[0010] in, is the first coefficient term, is the second coefficient term, x For input data.
[0011] Furthermore, the generation network includes: The first input layer, the first fully connected layer, several deconvolution layers, the first normalization layer, the first S_ReLU activation layer and the first output layer are connected in sequence; wherein the last layer in the first S_ReLU activation layer uses a tanh activation layer, the first input layer inputs noise, and feature data is generated through the first fully connected layer, the deconvolution layer, the first normalization layer and the first S_ReLU activation layer, and finally the first output layer outputs the generated data.
[0012] Furthermore, the expression of the generator loss function is:
[0013] in, is the generator loss function, is the loss function of the original generator, is the total variation loss function, For the discriminant network, To generate samples, for The (i,j)th pixel of .
[0014] Furthermore, the discriminant network includes: The second input layer, several convolutional layers, the second normalization layer, the second S_ReLU activation layer, the second fully connected layer and the second output layer; wherein, The second input layer inputs the generated data and training data, extracts data features through the second convolutional layer, the second normalization layer and the second S_ReLU activation layer, and outputs the probability that the training data is real data through the second fully connected layer and the second output layer.
[0015] Furthermore, the expression of the discriminator loss function is:
[0016] in, is the discriminator loss function, is the loss function of the original discriminator, is the gradient penalty term, is the third coefficient term, is the fourth coefficient term.
[0017] Furthermore, the step of constructing training data with aircraft information includes: The binary numbers of the aircraft pitch, roll, heading deviation, DME distance and radio altitude corresponding to the runway image in the original training data are padded with 0 to the same width as the runway image; The binary number padded with 0 is added to the tail column of the runway image to obtain the training data with aircraft information.
[0018] Furthermore, the step of separately training the generative network and the discriminative network using the training data to obtain a trained generative adversarial model includes: Randomly select m noise samples from the noise distribution ; Select m training data ; Train the generative network G, input the noise sample into the generative network G, obtain the generated sample, minimize the generator loss function, fix the discriminant network D, and calculate the generation model parameters of the generator loss function Gradient ,use Update Parameters ;in,
[0019] Train the discriminant network D, input the training data and the generated sample into the discriminant network D, maximize the discriminator loss function, fix the generated network G, and calculate the discriminant network parameters of the discriminator loss function Gradient ,use Update Parameters ;in, .
[0020] Furthermore, the step of inputting the to-be-enhanced runway data into the trained generative adversarial model to obtain the enhanced runway data includes: Randomly select the runway data to be enhanced from the noise distribution; The runway data to be enhanced is input into the trained generative adversarial model, and the enhanced runway data is output.
[0021] The technical solution provided by the embodiments of the present disclosure may have the following beneficial effects: In the embodiments of the present disclosure, through the above-mentioned runway data enhancement method with aircraft information, on the one hand, an activation function is designed and introduced into the generative adversarial model; a generator loss function of the generative network is designed; a discriminator loss function of the discriminant network is designed; training data with aircraft information is constructed, and the generative network and the discriminative network are trained separately using the training data to obtain a trained generative adversarial model; the runway data to be enhanced is input into the trained generative adversarial model to obtain enhanced runway data. On the other hand, the runway database with aircraft information is expanded through sample generation, which improves the problem of insufficient number and difficulty in obtaining runway databases in the research of intelligent take-off and landing. The problem of gradient vanishing and gradient exploding during the training process of the sample generation algorithm is solved, and samples with less noise, clearer images, and more detailed image details can be generated, further promoting the research of intelligent take-off and landing. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.
[0023] Figure 1 A step diagram showing a method for enhancing runway data with aircraft information in an exemplary embodiment of the present disclosure; Figure 2 A structural diagram showing a generated network in an exemplary embodiment of the present disclosure; Figure 3 A structural diagram showing a discriminant network in an exemplary embodiment of the present disclosure; Figure 4 A specific flow chart of a method for enhancing runway data with aircraft information according to an exemplary embodiment of the present disclosure is shown. DETAILED DESCRIPTION
[0024] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
[0025] In addition, the accompanying drawings are only schematic illustrations of the embodiments of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated descriptions will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities.
[0026] This example embodiment provides a runway data enhancement method with aircraft information. Figure 1 As shown in , the runway data enhancement method with aircraft information may include: steps S101 to S105.
[0027] Step S101: designing an activation function, and introducing the activation function into a generative adversarial model; wherein the generative adversarial model includes a generative network and a discriminative network; Step S102: designing a generator loss function of the generator network; Step S103: designing a discriminator loss function of the discriminant network; Step S104: constructing training data with aircraft information, and using the training data to separately train the generative network and the discriminative network to obtain a trained generative adversarial model; Step S105: inputting the runway data to be enhanced into the trained generative adversarial model to obtain enhanced runway data.
[0028] Through the above-mentioned runway data enhancement method with aircraft information, on the one hand, an activation function is designed and introduced into the generative adversarial model; the generator loss function of the generative network is designed; the discriminator loss function of the discriminant network is designed; training data with aircraft information is constructed, and the generative network and the discriminative network are trained separately using the training data to obtain a trained generative adversarial model; the runway data to be enhanced is input into the trained generative adversarial model to obtain enhanced runway data. On the other hand, the runway database with aircraft information is expanded through sample generation, which improves the problem of insufficient number and difficulty in obtaining runway databases in the research of intelligent take-off and landing. The problem of gradient vanishing and gradient exploding during the training process of the sample generation algorithm is solved, and samples with less noise, clearer images, and more detailed image details can be generated, which further promotes the research of intelligent take-off and landing.
[0029] Next, we will refer to Figures 1 to 4 Each step of the above-mentioned runway data enhancement method with aircraft information in this example implementation is described in more detail.
[0030] In step S101, an activation function is designed, and the S_ReLU function shown in the following formula is designed as the activation function of the discriminant network D.
[0031]
[0032] in, and is the coefficient term. For the runway image, it is preferred to take and .
[0033] In step S102, a generator network G is constructed, a generator loss function is designed as shown in the following formula, and Figure 2 The structure of the generative network G shown in the figure is that the input layer inputs noise Z, and generates feature data through the fully connected layer, deconvolution layer, normalization layer and S_ReLU activation layer. The last layer uses a tanh activation layer, and finally the output layer outputs the generated data G(Z) (i.e., generated samples).
[0034]
[0035] in, represents the loss function of the generative model, represents the loss function of the original generative model, represents the simplified total variation loss function.
[0036] Furthermore, the loss function of the original generative model is It enables the generative model to learn the data distribution of real data as much as possible, thereby generating data that conforms to the distribution of real data.
[0037]
[0038] Simplified total variation loss function It can not only reduce the block effect in the generated data but also reduce the data noise, making the generated image clearer and ensuring the detailed description of the image.
[0039]
[0040] Among them, i and j represent the pixel coordinate positions of the generated data.
[0041] In step S103, a discriminant network D is constructed, a discriminator loss function is designed as shown in the following formula, and Figure 3 The discriminant network D structure shown in the figure has an input layer that inputs the generated data G(Z) and the real data X, and extracts the data features through the convolution layer, the normalization layer and the S_ReLU activation layer. The output layer outputs the probability that the input is the real data through the fully connected layer.
[0042]
[0043] in, represents the loss function of the discriminant network, is the loss function of the original discriminant network, is the gradient penalty term, and For the runway image, as a preference, and .
[0044] Furthermore, the loss function of the original discriminant network is It enables the discriminant network to accurately distinguish generated data from real data. In adversarial training, the generative model will also learn the data distribution of real data as much as possible and generate data that can deceive the discriminator.
[0045]
[0046] Gradient Penalty It can alleviate the problems of gradient explosion and gradient vanishing that may occur in the discriminant network during training, and plays an important role in improving the stability of network training.
[0047]
[0048] in, is the coefficient term, represents the linear uniform sampling between the real data sample X and the generated data sample G(Z), .
[0049] In step S104, training data with aircraft information is constructed, and the aircraft pitch, roll, heading deviation, DME distance, and radio altitude corresponding to the runway image in the original training data are visualized, and the binary numbers of the above data are padded with 0 to the same width as the runway image. Then, the binary numbers of the aircraft pitch, roll, heading deviation, DME distance, and radio altitude are added to the tail column of the runway image to become part of the runway image, thereby constructing training data with aircraft information.
[0050] Train the network. During the training process, the generative model and the discriminative network are trained separately. Fix the generative network G and train the discriminative network D, fix the discriminative network D and train the generative network G alternately. It includes the following specific steps: Randomly select 500 noise samples from the noise distribution ; Select 500 constructed runway images as training data ; Train the discriminant network D, input the training data X and the generated samples into the discriminant network D. Maximize the discriminator loss function, fix the generated network G, and calculate the discriminator loss function with respect to the discriminant network parameters Gradient , and then use Update Parameters ;
[0051] Train the generative network G, input the noise sample into the generative network G, and obtain the generated sample. Minimize the generator loss function, fix the discriminant network D, and calculate the generator loss function with respect to the generation model parameters Gradient , and then use Update Parameters ;
[0052] In step S105, the runway data to be enhanced is randomly selected from the noise distribution; the runway data to be enhanced is input into the trained generative adversarial model, and the enhanced runway data is output.
[0053] like Figure 4 As shown, it is a specific flow chart of the runway data enhancement method with aircraft information.
[0054] Through the above-mentioned runway data enhancement method with aircraft information, on the one hand, an activation function is designed and introduced into the generative adversarial model; the generator loss function of the generative network is designed; the discriminator loss function of the discriminant network is designed; training data with aircraft information is constructed, and the generative network and the discriminative network are trained separately using the training data to obtain a trained generative adversarial model; the runway data to be enhanced is input into the trained generative adversarial model to obtain enhanced runway data. On the other hand, the runway database with aircraft information is expanded through sample generation, which improves the problem of insufficient number and difficulty in obtaining runway databases in the research of intelligent take-off and landing. The problem of gradient vanishing and gradient exploding during the training process of the sample generation algorithm is solved, and samples with less noise, clearer images, and more detailed image details can be generated, which further promotes the research of intelligent take-off and landing.
[0055] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.
[0056] In the embodiments of the present disclosure, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present disclosure can be understood according to specific circumstances.
[0057] In the embodiments of the present disclosure, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may include the first and second features being in direct contact, or may include the first and second features not being in direct contact but being in contact through another feature between them. Moreover, a first feature being "above", "above" and "above" a second feature includes the first feature being directly above and obliquely above the second feature, or simply indicates that the first feature is higher in level than the second feature. A first feature being "below", "below" and "below" a second feature includes the first feature being directly below and obliquely below the second feature, or simply indicates that the first feature is lower in level than the second feature.
[0058] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.
[0059] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the appended claims.
Claims
1. A runway data enhancement method with aircraft information, characterized in that: The method includes: Designing an activation function and introducing the activation function into a generative adversarial model; wherein the generative adversarial model includes a generative network and a discriminative network; Designing a generator loss function of the generative network; Designing a discriminator loss function of the discriminant network; Constructing training data with aircraft information, and using the training data to separately train the generative network and the discriminative network to obtain a trained generative adversarial model; The runway data to be enhanced is input into the trained generative adversarial model to obtain enhanced runway data.
2. The runway data enhancement method with aircraft information according to claim 1, characterized in that: The activation function of the generative adversarial model is the S_ReLU function, and the expression of the S_ReLU function is: in, is the first coefficient term, is the second coefficient term, x For input data.
3. The runway data enhancement method with aircraft information according to claim 2, characterized in that: The generation network comprises: The first input layer, the first fully connected layer, several deconvolution layers, the first normalization layer, the first S_ReLU activation layer and the first output layer are connected in sequence; wherein the last layer in the first S_ReLU activation layer uses a tanh activation layer, the first input layer inputs noise, and feature data is generated through the first fully connected layer, the deconvolution layer, the first normalization layer and the first S_ReLU activation layer, and finally the first output layer outputs the generated data.
4. The runway data enhancement method with aircraft information according to claim 3, characterized in that: The expression of the generator loss function is: in, is the generator loss function, is the loss function of the original generator, is the total variation loss function, For the discriminant network, To generate samples, for The (i,j)th pixel of .
5. The runway data enhancement method with aircraft information according to claim 4, characterized in that: The discriminant network comprises: The second input layer, several convolutional layers, the second normalization layer, the second S_ReLU activation layer, the second fully connected layer and the second output layer; wherein, The second input layer inputs the generated data and training data, extracts data features through the second convolutional layer, the second normalization layer and the second S_ReLU activation layer, and outputs the probability that the training data is real data through the second fully connected layer and the second output layer.
6. The runway data enhancement method with aircraft information according to claim 5, characterized in that: The expression of the discriminator loss function is: in, is the discriminator loss function, is the loss function of the original discriminator, is the gradient penalty term, is the third coefficient term, is the fourth coefficient term.
7. The runway data enhancement method with aircraft information according to claim 1, characterized in that: The steps to construct training data with aircraft information include: The binary numbers of the aircraft pitch, roll, heading deviation, DME distance and radio altitude corresponding to the runway image in the original training data are padded with 0 to the same width as the runway image; The binary number padded with 0 is added to the tail column of the runway image to obtain the training data with aircraft information.
8. The runway data enhancement method with aircraft information according to claim 7, characterized in that: The step of separately training the generative network and the discriminative network using the training data to obtain a trained generative adversarial model includes: Randomly select m noise samples from the noise distribution ; Select m training data ; Train the generative network G, input the noise sample into the generative network G, obtain the generated sample, minimize the generator loss function, fix the discriminant network D, and calculate the generation model parameters of the generator loss function Gradient ,use Update Parameters ;in, Train the discriminant network D, input the training data and the generated sample into the discriminant network D, maximize the discriminator loss function, fix the generated network G, and calculate the discriminant network parameters of the discriminator loss function Gradient ,use Update Parameters ;in, 。 9. The runway data enhancement method with aircraft information according to claim 7, characterized in that: The step of inputting the to-be-enhanced runway data into the trained generative adversarial model to obtain the enhanced runway data comprises: Randomly select the runway data to be enhanced from the noise distribution; The runway data to be enhanced is input into the trained generative adversarial model, and the enhanced runway data is output.