A comprehensive display device based on FPGA and dual ARM processor architecture

Abstract

This application provides an integrated display device based on an FPGA and dual ARM processor architecture, relating to the field of avionics. It first acquires flight attitude data of the target aircraft and transmits it to a first ARM processor via a MIPI interface. The first ARM processor preprocesses the data and stores it in an input buffer of dual-port shared memory. Next, the FPGA chip reads the preprocessed data from the input buffer to generate a flight attitude indication image and writes it to an output buffer. The second ARM processor reads the attitude indication image from the output buffer and performs image synthesis based on semantic feature interaction with the navigation image to obtain a synthesized flight display image, which is then stored in a display frame buffer. Finally, the synthesized image is read from the display frame buffer and transmitted to a cockpit multi-function display screen via an LVDS interface for display. This enables more efficient, accurate, and reliable flight information display.

Description

Technical Field

[0001] This application relates to the field of avionics, and more specifically, to an integrated display device based on an FPGA and dual ARM processor architecture. Background Technology

[0002] In modern avionics systems, the real-time and accurate display of flight attitude and navigation information is crucial for pilot safety. With advancements in aviation technology and increasing complexity of operating environments, standards for flight control systems are rising, which in turn drives increasingly stringent requirements for the performance of display devices.

[0003] However, traditional display devices typically employ a single or relatively simple processor architecture, which struggles to meet the real-time processing demands of modern aircraft's massive amounts of flight attitude and navigation data. For instance, in highly dynamic flight scenarios or complex weather conditions, when a flood of sensor data arrives, traditional devices may experience processing delays, failing to update displayed information in a timely manner. This results in delayed flight data acquisition for pilots, impacting their decision-making. Furthermore, traditional display devices may use older transmission interface technologies with lower data transmission rates, making it impossible to quickly transmit large amounts of data such as high-resolution flight attitude indication images and navigation images.

[0004] Therefore, a comprehensive display solution based on FPGA and dual ARM processor architecture is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide an integrated display device based on an FPGA and dual ARM processor architecture.

[0006] According to one aspect of this application, a comprehensive display device based on an FPGA and dual ARM processor architecture is provided, comprising:

[0007] The data input module is used to collect the time series of flight attitude data of the target aircraft and input it into the first ARM processor through the MIPI interface.

[0008] The data preprocessing module is used to preprocess the time series set of the flight attitude data through the first ARM processor and write it into the input buffer of the dual-port shared memory;

[0009] The flight attitude indication image generation module is used to read the timing set of the preprocessed flight attitude data from the input buffer through the FPGA chip, and write the generated flight attitude indication image into the output buffer of the dual-port shared memory.

[0010] The flight image synthesis module is used to read the flight attitude indication image from the output buffer through the second ARM processor, and synthesize the flight attitude indication image and the navigation image based on key semantic feature interaction to obtain the synthesized flight display image, and write it into the display frame buffer.

[0011] The display module is used to read the composite flight display image from the display frame buffer and send the composite flight display image to the multi-function display screen in the cockpit via the LVDS interface for display.

[0012] Compared with existing technologies, the integrated display device based on FPGA and dual ARM processor architecture provided in this application first acquires the flight attitude data of the target aircraft and transmits it to the first ARM processor via a MIPI interface. The first ARM processor preprocesses the data and stores it in an input buffer of dual-port shared memory. Then, the FPGA chip reads the preprocessed data from the input buffer to generate a flight attitude indication image and writes it to the output buffer. The second ARM processor reads the attitude indication image from the output buffer and performs image synthesis based on semantic feature interaction with the navigation image to obtain a synthesized flight display image, which is then stored in the display frame buffer. Finally, the synthesized image is read from the display frame buffer and transmitted to the cockpit multi-function display screen via an LVDS interface for display. This enables more efficient, accurate, and reliable flight information display. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0014] Figure 1 This is a block diagram of an integrated display device based on an FPGA and dual ARM processor architecture according to an embodiment of this application.

[0015] Figure 2 This is a block diagram of a flight image synthesis module in an integrated display device based on an FPGA and dual ARM processor architecture according to an embodiment of this application.

[0016] Figure 3 This is a schematic diagram of the data flow in a flight image synthesis module of an integrated display device based on an FPGA and dual ARM processor architecture according to an embodiment of this application.

[0017] Figure 4This is a block diagram of the flight attitude-navigation state feature interaction unit in an integrated display device based on an FPGA and dual ARM processor architecture according to an embodiment of this application. Detailed Implementation

[0018] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0019] In modern avionics systems, the real-time and accurate display of flight attitude and navigation information is crucial for pilot safety. With the continuous advancement of aviation technology and the increasing complexity of flight environments, the technical standards of flight control systems are constantly rising, a trend that also places more stringent demands on the performance of display devices.

[0020] However, traditional display devices typically employ a single or relatively simple processor architecture, which often struggles to achieve efficient real-time processing when dealing with the massive amounts of flight attitude data and navigation information from modern aircraft. For example, in highly dynamic flight environments or complex weather conditions, a large amount of sensor data is input simultaneously. Traditional devices may experience delays due to limited processing capabilities, resulting in untimely updates to displayed information. This leads to a lag in the flight data acquired by the pilot, thus affecting their decision-making. Furthermore, many traditional display devices still rely on relatively outdated transmission interface technologies, resulting in low data transmission rates and making it difficult to quickly process and present high-resolution flight attitude indication images and navigation information.

[0021] As can be understood, an FPGA (Field-Programmable Gate Array) is a highly programmable integrated circuit that allows users to configure its hardware logic to achieve specific functions. FPGAs are characterized by high parallel processing capabilities, low latency, and high reliability, making them suitable for applications requiring real-time data processing. A dual-ARM processor, on the other hand, refers to a processor containing two ARM architecture processing cores. This architecture is typically used for high-performance computing tasks, offering low power consumption, high efficiency, and multitasking capabilities, making it suitable for applications requiring parallel computing and real-time response. In flight display systems, the combination of an FPGA and a dual-ARM processor significantly improves system performance and reliability. Specifically, the FPGA's high-speed parallel processing capabilities enable real-time processing of sensor data, video signals, and graphics rendering, ensuring rapid updates and accurate display of flight information. The dual-ARM processor, however, runs the operating system, executes flight management software, and handles human-machine interaction tasks, providing flexible software control capabilities. The two complement each other, ensuring high computing performance while maintaining low power consumption and high stability. Simultaneously, the FPGA executes tasks directly through hardware logic, reducing the possibility of software failures and improving system reliability, while the dual-ARM processor efficiently runs multiple tasks, further enhancing system flexibility. Furthermore, the programmability of FPGAs allows for system upgrades as needed, while ARM processors support a rich software ecosystem, facilitating future functional expansion. In summary, the combined application of FPGAs and dual ARM processors enables flight display systems to operate more efficiently and reliably, improving flight safety and the pilot's operating experience.

[0022] Based on this, this application proposes an integrated display device based on FPGA and dual ARM processor architecture. By adopting high-speed data interface, parallel processing architecture, advanced machine vision technology and intelligent image synthesis method, it provides a more efficient, accurate and reliable flight information display solution, which significantly improves flight safety and ease of operation. Figure 1 This is a block diagram of an integrated display device based on an FPGA and dual ARM processor architecture according to an embodiment of this application. Figure 1As shown, the integrated display device 100 based on an FPGA and dual ARM processor architecture includes: a data input module 110, used to acquire a time series set of flight attitude data of a target aircraft object and input it to a first ARM processor via a MIPI interface; a data preprocessing module 120, used to preprocess the time series set of flight attitude data via the first ARM processor and write it into an input buffer of a dual-port shared memory; a flight attitude indication image generation module 130, used to read the preprocessed time series set of flight attitude data from the input buffer via an FPGA chip, and write it into an output buffer of the dual-port shared memory after generating a flight attitude indication image; a flight image synthesis module 140, used to read the flight attitude indication image from the output buffer via a second ARM processor, and synthesize the flight attitude indication image and navigation image based on key semantic feature interaction to obtain a synthesized flight display image, and write it into a display frame buffer; and a display module 150, used to read the synthesized flight display image from the display frame buffer and send the synthesized flight display image to a multi-function display screen in the cockpit via an LVDS interface for display.

[0023] In this embodiment, the data input module 110 is used to collect a time-series set of flight attitude data of the target aircraft and input it into the first ARM processor via the MIPI interface. Specifically, the flight attitude data includes pitch angle, roll angle, and yaw angle. It should be understood that the time-series set of flight attitude data contains information such as the pitch angle, roll angle, and yaw angle of the target aircraft at different times. This data changes dynamically in real time and needs to be collected and transmitted at a high frequency to ensure that the pilot can obtain the aircraft's attitude information in a timely and accurate manner. The MIPI interface has high-speed data transmission and strong anti-interference capabilities. Transmitting data through the MIPI interface ensures that a large amount of flight attitude data is accurately transmitted to the first ARM processor in a short time. This ensures that in complex aviation electromagnetic environments, the flight attitude data is not subject to electromagnetic interference during transmission, making the data quality transmitted to the first ARM processor reliable. Furthermore, the processor can process the data immediately, avoiding processing delays caused by data transmission latency, thereby ensuring that the entire device can respond to changes in flight attitude in real time, and thus providing the pilot with timely and accurate information.

[0024] In this embodiment, the data preprocessing module 120 is used to preprocess the time series set of flight attitude data by the first ARM processor and write it into the input buffer of the dual-port shared memory. Accordingly, considering that flight attitude data may be affected by various factors during acquisition and transmission, such as sensor errors and transmission noise, resulting in noise and errors in the data, the first ARM processor preprocesses the time series set of flight attitude data, performing operations such as cleaning and filtering. These preprocessing steps improve data quality and provide a reliable foundation for subsequent processing and analysis. Furthermore, since subsequent data processing requires efficient access and processing of this data, storing the preprocessed data in the input buffer of the dual-port shared memory allows different hardware components (such as FPGA and ARM processor) to access this data quickly and in parallel, thereby accelerating the overall device response speed.

[0025] The following is a detailed explanation of a specific implementation process of "preprocessing the timing set of the flight attitude data by the first ARM processor and writing it into the input buffer of the dual-port shared memory":

[0026] Upon receiving the timing set of flight attitude data, the ARM1 processor immediately begins data parsing. It extracts key valid fields, such as pitch and roll angles, from each ARINC429 standard frame. During this parsing process, the ARM1 processor meticulously analyzes and breaks down each frame of data according to preset protocols and rules, ensuring that no important information is missed.

[0027] Next, the ARM1 processor performs rigorous verification on the parsed data. It verifies the integrity and accuracy of the data by checking the CRC (Cyclic Redundancy Check) bits. Specifically, CRC technology calculates a checksum based on a specific algorithm on the received data and compares it with the CRC checksum bits attached to the data. If the two do not match, it means the data "encountered an accident" during transmission and contains errors. In this case, the ARM1 processor will decisively discard this erroneous data to prevent it from interfering with subsequent processing. This data verification mechanism not only improves the robustness of the device but also ensures the reliability of the final displayed information. By rigorously verifying the data, the ARM1 processor can effectively filter out noise and outliers, ensuring that only high-quality data can proceed to the next processing step.

[0028] The parsed and verified data already possesses high accuracy and usability; the next step is data caching. The first ARM processor establishes a connection with the dual-port shared memory through the Memory Management Unit (MMU), successfully acquiring access to the input buffer. At this point, the processor writes the parsed and verified attitude data sequentially to the designated storage location in the input buffer according to a predefined data storage format. During this process, the processor ensures the correctness of the data writing order and format so that subsequent data can be accurately read for further processing.

[0029] In this embodiment, the flight attitude indication image generation module 130 is used to read the preprocessed timing set of flight attitude data from the input buffer via the FPGA chip, and write it to the output buffer of the dual-port shared memory after generating the flight attitude indication image. Specifically, the preprocessed timing set of flight attitude data is read from the input buffer of the dual-port shared memory via a DMA controller. By quickly and accurately acquiring the timing set of preprocessed flight attitude data through the DMA controller, the FPGA chip can perform timely calculations and processing based on the latest flight attitude data, generating an indication image that accurately reflects the aircraft's current attitude, and writing it to the output buffer of the dual-port shared memory, so that subsequent data processing steps can flexibly obtain the required data. For example, based on data such as pitch angle, roll angle, and yaw angle, the FPGA can accurately draw the aircraft's attitude graph in three-dimensional space, providing the pilot with intuitive flight attitude information. It is worth mentioning that the DMA controller can directly access memory, avoiding the additional overhead caused by the processor participating in data transfer. This allows for rapid reading of data from the input buffer into the FPGA when processing large amounts of flight attitude data timing sets, meeting the FPGA's need for rapid data processing. Specifically, in a specific example of this application, the implementation steps of the FPGA chip generating a flight attitude indication image include: after reading the flight attitude data, the FPGA chip first performs a coordinate transformation operation on it. The flight attitude data collected by the aircraft's sensors is based on the aircraft's own body coordinate system. However, to intuitively present the flight attitude on the multi-function display screen in the cockpit, this data must be converted into the screen coordinate system. Taking the aircraft's pitch angle data as an example, the FPGA will accurately map it to the Y-axis of the screen according to certain mapping rules; the roll angle data will be mapped to other relevant dimensions of the screen. This coordinate transformation process transforms the abstract flight attitude data into coordinate information that can be intuitively represented on the screen, laying the foundation for the accurate generation of subsequent graphics. Based on the results of the coordinate transformation, the FPGA chip begins to generate the graphic elements of the attitude indicator (ADI). The attitude indicator is an important visual tool for pilots to understand the aircraft's flight attitude. The FPGA utilizes its powerful parallel processing capabilities to quickly and accurately draw various graphic elements based on the flight attitude data. For example, the line representing the horizontal plane is precisely drawn, allowing the pilot to clearly determine whether the aircraft is flying horizontally or tilted by observing the relative position of this line and the aircraft icon. The aircraft icon is also accurately displayed according to the actual flight attitude, showing the aircraft's specific position and direction in the current attitude. After generating the graphical elements of the attitude indicator, the FPGA chip needs to overlay them with a background map loaded from the second ARM processor. The background map provides the pilot with geographical context information about the aircraft's flight, which is crucial for the pilot to understand the aircraft's position and surrounding environment.The FPGA cleverly overlays the attitude indicator graphic elements onto the background map to generate the final flight attitude indication image, allowing the pilot to obtain both flight attitude and geographical background information in one view.

[0030] In this embodiment, the flight image synthesis module 140 is used to read the flight attitude indication image from the output buffer via a second ARM processor, and to synthesize the flight attitude indication image and the navigation image based on key semantic feature interaction to obtain a synthesized flight display image, which is then written into the display frame buffer. It should be understood that synthesizing the flight attitude indication image and the navigation image provides the pilot with more comprehensive and intuitive flight information. The flight attitude indication image displays the aircraft's current attitude (such as pitch, roll, heading, etc.), while the navigation image displays the aircraft's flight path, geographical location, surrounding terrain, and other information. The synthesized flight display image integrates these two types of information, allowing the pilot to simultaneously obtain the aircraft's attitude and navigation information in one frame, thus more accurately grasping the flight status. During flight, the flight attitude and navigation conditions constantly change, and the image content becomes very complex. However, traditional rule-based image synthesis often uses relatively simple and fixed feature extraction rules, which are difficult to adapt to such complex and ever-changing scenarios and cannot extract features that match the current flight state in a timely and accurate manner. For example, when encountering severe weather or special flight missions, the features in the image may change significantly, and traditional methods may not be able to effectively extract key features, resulting in poor image synthesis results.

[0031] Based on this, the technical concept of this application is to use machine vision-based image analysis and processing technology to enhance the image resolution of the flight attitude indication image and navigation image. Then, feature extraction is performed on the preprocessed flight attitude indication image and preprocessed navigation image. Based on the key semantic interaction representation between the extracted flight attitude features and navigation state features, a synthesized flight display image is automatically generated. This application can dynamically adjust the feature extraction strategy under different environmental conditions. Whether in clear weather or inclement weather (such as heavy rain, dense fog, etc.), the system can accurately extract key features matching the current flight state to ensure the accuracy and reliability of image synthesis.

[0032] Figure 2 This is a block diagram of a flight image synthesis module in an integrated display device based on an FPGA and dual ARM processor architecture according to an embodiment of this application. Figure 3 This is a schematic diagram of the data flow in a flight image synthesis module of an integrated display device based on an FPGA and dual ARM processor architecture according to an embodiment of this application. Figure 2 and Figure 3As shown, the flight image synthesis module 140 includes: an image preprocessing unit 141, used to enhance the image resolution of the flight attitude indication image and the navigation image to obtain a preprocessed flight attitude indication image and a preprocessed navigation image; a flight attitude feature extraction unit 142, used to extract flight attitude features from the preprocessed flight attitude indication image to obtain a flight attitude feature vector; a navigation state feature extraction unit 143, used to extract navigation state features from the preprocessed navigation image to obtain a navigation state feature vector; a flight attitude-navigation state feature interaction unit 144, used to perform rapid interaction analysis of key semantics between flight attitude and navigation state on the flight attitude feature vector and the navigation state feature vector to obtain a flight attitude-navigation state feature interaction encoding vector; and a flight image synthesis unit 145, used to obtain the synthesized flight display image based on the flight attitude-navigation state feature interaction encoding vector.

[0033] In this embodiment, the image preprocessing unit 141 is used to enhance the image resolution of the flight attitude indication image and the navigation image to obtain preprocessed flight attitude indication images and preprocessed navigation images. Correspondingly, considering that image acquisition devices are limited by factors such as cost, size, and technical specifications, the resolution of the acquired flight attitude indication images and navigation images often falls short of ideal levels. For example, when sensors on an aircraft acquire images, issues such as lighting conditions and equipment precision may lead to blurry original images or loss of detail. Furthermore, the multi-function display screen in the cockpit has specific resolution and display accuracy requirements. If the original image resolution is too low, problems such as image stretching and pixelation will occur when displayed on the screen, severely affecting the pilot's ability to read information. Therefore, in the technical solution of this application, the image resolution of the flight attitude indication image and the navigation image is enhanced to obtain preprocessed flight attitude indication images and preprocessed navigation images. In this way, the preprocessed high-resolution image can provide more accurate data for subsequent feature extraction, thereby improving the accuracy of the extracted flight attitude features and navigation status features.

[0034] In this embodiment, the flight attitude feature extraction unit 142 is used to extract flight attitude features from the preprocessed flight attitude indication image to obtain a flight attitude feature vector. Specifically, in this embodiment, the flight attitude feature extraction unit is used to: pass the preprocessed flight attitude indication image through a flight attitude feature extractor based on the MobileNet model to obtain the flight attitude feature vector. Accordingly, considering that the preprocessed flight attitude indication image contains key features related to flight attitude, such as the representation characteristics of aircraft pitch, roll, and heading in the image, in order to reflect the actual attitude of the aircraft, this application obtains the flight attitude feature vector by passing the preprocessed flight attitude indication image through a flight attitude feature extractor based on the MobileNet model. It should be understood that MobileNet is a lightweight convolutional neural network model, whose design philosophy is to minimize the number of model parameters and computational load while ensuring a certain level of accuracy. Compared to some large convolutional neural networks (such as ResNet and VGG), MobileNet requires less computational resources when processing images, enabling it to run quickly on ARM processors without placing an excessive burden on the processor, thus ensuring the system's real-time performance. Furthermore, the MobileNet model, trained on a large amount of image data, possesses powerful feature extraction capabilities, able to extract key latent features related to flight attitude from images, transforming image information into a compact and representative set of flight attitude feature vectors.

[0035] In this embodiment, the navigation state feature extraction unit 143 is used to extract navigation state features from the preprocessed navigation image to obtain a navigation state feature vector. Specifically, in this embodiment, the navigation state feature extraction unit is used to: pass the preprocessed navigation image through a ViT model-based navigation state feature extractor to obtain the navigation state feature vector. It should be understood that navigation images contain rich information such as aircraft flight routes, geographical locations, and surrounding terrain, and these information have complex long-sequence dependencies. For example, the continuity of flight routes, the correlation between geographical locations and surrounding terrain, etc. At the same time, the information in the navigation image may involve different scales, such as macroscopic flight area maps and microscopic details of specific landmarks. Based on this, in the technical solution of this application, the preprocessed navigation image is passed through a ViT model-based navigation state feature extractor to quantify and represent key information in the navigation image through its Transformer architecture's self-attention mechanism, thereby obtaining a navigation state feature vector. Specifically, the ViT model, based on the Transformer architecture, is well-suited for handling long-sequence dependencies. It can more effectively capture long-term dependencies and contextual information between different regions and elements in navigation images, thus providing a more comprehensive understanding of the navigation image's content. Furthermore, through its self-attention mechanism, the ViT model can automatically model information at different scales, adaptively focusing on important regions and features in the image, regardless of whether these features are at a global or local scale. These features include information such as flight path direction, distance, geographic coordinates, and terrain features, helping pilots and flight control systems accurately grasp the aircraft's navigation status and ensure that the flight proceeds safely and accurately along the predetermined route.

[0036] The following is a detailed explanation of a specific implementation process for "obtaining the navigation state feature vector by passing the preprocessed navigation image through a navigation state feature extractor based on the ViT model":

[0037] First, the preprocessed navigation image undergoes block segmentation and positional encoding. The preprocessed navigation image contains rich information such as aircraft flight paths, geographical locations, and surrounding terrain. These information exhibits complex long-sequence dependencies and involves different scales. The navigation image is divided into multiple fixed-size image blocks, each becoming the basic unit for subsequent processing. Since the positional information of each image block within the original image is crucial for understanding the navigation image's content—such as the continuity of flight paths and the correlation between geographical locations and surrounding terrain—positional encoding is added to each image block. Positional encoding is generated using a specific algorithm that integrates positional information numerically into the image block features. This allows the model to perceive the spatial relationships between different image blocks, thus laying the foundation for accurately understanding various types of information in the navigation image.

[0038] After image segmentation and location encoding, linear embedding and feature fusion operations are performed. For the segmented and location-encoded image patches, linear embedding is performed using a linear projection layer. This operation transforms the originally high-dimensional pixel values ​​of each image patch into low-dimensional feature vectors, reducing data dimensionality while preserving key information to the greatest extent possible and reducing the computational burden of subsequent processing. These low-dimensional feature vectors are then concatenated with learnable class tokens. Class tokens are specially designed vectors that aggregate global features to represent the comprehensive information of the entire image. This concatenation method achieves the fusion of local and global features from each image patch, providing a more comprehensive and richer information foundation for subsequent feature extraction, enabling the model to understand navigation images from both macroscopic and microscopic perspectives.

[0039] Then, the fused feature vector sequence enters the Transformer encoder for deep processing. The Transformer encoder consists of multiple identical stacked layers, each primarily containing a multi-head self-attention mechanism and a multilayer perceptron (MLP). In the multi-head self-attention mechanism, the model performs parallel attention calculations on the feature vector sequence through different heads. The unique aspect of this mechanism is that it allows the model to focus on the relationships between image patches from multiple different perspectives, adaptively capturing long-term dependencies and contextual information between different regions and elements in the navigation image. For example, when focusing on flight routes, the model can simultaneously consider the influence of geographical location and surrounding terrain on the flight route; when processing geographical location information, it can also relate it to flight route planning and direction. The output of the multi-head self-attention mechanism, after layer normalization and residual connections, is input into the multilayer perceptron. The multilayer perceptron further transforms and maps the features, enhancing the expressive power of the features through nonlinear activation functions, and mining deeper and more representative feature information.

[0040] After multiple layers of deep processing by the Transformer encoder, a feature representation containing rich navigation information is obtained. From these complex feature representations, key features closely related to the navigation state are extracted. These key features cover multiple important aspects such as flight path direction, distance, geographic coordinates, and terrain features. To integrate and compress these key features into a compact vector that facilitates subsequent processing, specific mapping or pooling operations are employed. Through these operations, key features scattered across different dimensions are effectively integrated, redundant information is removed, and a navigation state feature vector that can comprehensively quantify and represent the key information of the navigation image is ultimately generated.

[0041] In this embodiment of the application, the flight attitude-navigation state feature interaction unit 144 is used to perform rapid interaction analysis of key semantics between flight attitude and navigation state on the flight attitude feature vector and the navigation state feature vector to obtain a flight attitude-navigation state feature interaction encoding vector. Specifically, Figure 4 This is a block diagram of the flight attitude-navigation state feature interaction unit in an integrated display device based on an FPGA and dual ARM processor architecture according to an embodiment of this application. Figure 4 As shown, the flight attitude-navigation state feature interaction unit 144 includes: a feature implicit encoding subunit 1441, used to perform principal component implicit encoding and implicit feature extraction on the flight attitude feature vector and the navigation state feature vector respectively to obtain a set of flight attitude feature principal component implicit encoding vectors and a navigation state feature encoding vector; a flight attitude-navigation state implicit feature anchoring subunit 1442, used to perform flight attitude-navigation state implicit key clue anchoring on the set of navigation state feature encoding vectors and flight attitude feature principal component implicit encoding vectors to obtain a feature pair of {navigation state implicit feature encoding vector, anchored flight attitude principal component implicit feature encoding vector}; and a flight attitude-navigation state feature fine-grained interaction subunit 1443, used to perform flight attitude-navigation state feature fine-grained interaction on the flight attitude feature vector and the navigation state feature vector based on the feature pair of {navigation state implicit feature encoding vector, anchored flight attitude principal component implicit feature encoding vector} to obtain the flight attitude-navigation state feature interaction encoding vector.

[0042] It should be understood that the flight attitude feature vector mainly reflects the aircraft's attitude information, such as pitch, roll, and heading; the navigation state feature vector focuses on the aircraft's navigation information, such as flight path and geographical location. Although they each have their own emphasis, they are closely related in actual flight. Changes in the aircraft's attitude affect its navigation path, and navigation information determines the direction of flight attitude adjustments. Therefore, in order to integrate the information they carry, achieve complementarity, and comprehensively represent the aircraft's flight state, this application performs rapid interactive analysis of key semantics between the flight attitude feature vector and the navigation state feature vector to obtain a flight attitude-navigation state feature interactive encoding vector. That is, in actual flight, the environment is complex and changeable, and situations such as severe weather, terrain limitations, and air traffic control may occur, causing frequent changes in flight attitude and navigation state. Interactive analysis of the feature vectors enables the system to better adapt to such complex dynamic changes, comprehensively capture the impact of various key factors on flight, and provide richer and more accurate feature representations for subsequent processing.

[0043] Specifically, in this embodiment, the feature implicit coding subunit is used to: perform feature principal component analysis on the flight attitude feature vectors to obtain a set of flight attitude feature principal component coding vectors, which can be expressed by the following formula:

[0044]

[0045] Where V2 is the flight attitude feature vector, PCA(·) is the principal component analysis operation, C2 is the covariance matrix of the flight attitude feature samples calculated by V2, and U2 is the orthogonal matrix of the principal components of the flight attitude features. 21 v 22 v 2i and v 2m Let be the 1st, 2nd, ith, and mth principal component encoding vectors of flight attitude features, respectively, and let Λ2 be the diagonal matrix of flight attitude features. 21 , λ 2m v 21 and v 2m The corresponding eigenvalue, U2 T Let U be the transpose of U2;

[0046] The set of flight attitude feature principal component encoding vectors is processed by performing flight attitude point convolution latent feature extraction on each flight attitude feature principal component encoding vector in the set of flight attitude feature principal component encoding vectors. This process can be expressed by the following formula:

[0047]

[0048] Among them, Conv 1×1 It is a point-matrix convolutional encoding, sigmoid is the sigmoid function, h 21 h 22 h 2i and h 2m These are the 1st, 2nd, 1st, and 1mth flight attitude feature principal component latent coding vectors in the set of flight attitude feature principal component latent coding vectors, respectively, and H2 is the set of flight attitude feature principal component latent coding vectors.

[0049] The navigation state feature vector is subjected to navigation state point convolution latent feature extraction to obtain the navigation state feature encoding vector. This process can be expressed by the following formula:

[0050] H1 = sigmoid[Conv] 1×1 (V1)]

[0051] Among them, Conv 1×1It is a point convolutional encoding, sigmoid is the sigmoid function, V1 is the navigation state feature vector, and H1 is the navigation state feature encoding vector.

[0052] It's understandable that flight attitude feature vectors encompass information across multiple dimensions, such as pitch, roll, and heading. While this information is rich, it also results in high dimensionality. When processing large amounts of flight attitude data, this high dimensionality presents computational challenges, increasing computation time and potentially leading to excessive resource consumption. By performing principal component analysis (PCA) on the flight attitude feature vectors, mapping the original features to a new low-dimensional feature space, we can reduce both dimensionality and computational complexity. Specifically, this low-dimensional feature space construction process can accurately capture the variance distribution in the data, allowing data analysis to focus on directions with higher variance—that is, emphasizing information-rich features. In this way, the most critical parts of the original flight attitude features can be preserved. In other words, the resulting set of principal component encoded vectors for flight attitude features expresses most of the key information from the original high-dimensional feature vectors with fewer dimensions, enabling the entire system to operate more efficiently.

[0053] Accordingly, although the set of principal component encoding vectors for flight attitude features has reduced dimensionality through principal component analysis, complex interactions may still exist between the feature space channels represented by each vector. These interactions contain deeper and more fundamental information about flight attitude, helping the model to accurately understand and describe it. Therefore, this application requires performing flight attitude point convolution latent feature extraction processing on each flight attitude feature principal component encoding vector in the set of flight attitude feature principal component encoding vectors to deeply explore the interactions between the feature space channels of each vector, thereby discovering the potential patterns hidden behind the data. Specifically, point convolution integrates information at the feature channel level through cross-dimensional weight mapping, enabling the fusion of information from different feature channels. This helps eliminate redundancy between features while retaining important information. For example, in flight attitude analysis, different sensors may provide related but not identical information; through point convolution information integration, this information can be effectively fused to form a more accurate and comprehensive description of flight attitude. Furthermore, by combining it with a nonlinear activation function, the feature representation capability can be significantly improved. That is, the set of hidden encoding vectors of the principal components of flight attitude features obtained after processing can express flight attitude information more accurately and richly.

[0054] It can be understood that point convolution does not rely on the locality of the original feature space, but directly performs cross-dimensional weight mapping at the feature channel level. This characteristic allows for the effective integration of information from different dimensions in the navigation state feature vector. Taking the aircraft's geographical location information and flight route information as examples, these belong to different dimensions of features, and traditional methods may find it difficult to quickly and accurately establish the connection between them. However, point convolution, through weight mapping, can directly correlate geographical location information with flight route information, allowing the system to quickly understand the specific situation of the aircraft's current position within its planned flight route, such as determining whether the aircraft has deviated from the predetermined route, and the degree and direction of the deviation. This cross-dimensional information integration greatly improves the efficiency and accuracy of processing navigation state information. Moreover, by introducing nonlinear activation functions during the processing of navigation state feature vectors, the expressive power of navigation state features can be further enhanced. Specifically, the navigation state of an aircraft exhibits complex and variable characteristics in actual flight, and nonlinear activation functions allow the model to learn these complex feature patterns. When encountering temporary changes in flight routes due to air traffic control or irregular changes in flight speed and direction caused by airflow, nonlinear activation functions can help the model capture these nonlinear changes, enabling the navigation state feature encoding vector to more accurately and richly represent the actual navigation state. In summary, the navigation state feature encoding vector obtained after extracting latent features through navigation state point convolution integrates rich navigation information, uncovers the intrinsic relationships between features, enhances the expressive power of complex navigation states, and lays a solid foundation for subsequent interactive analysis with flight attitude features.

[0055] Next, the set of navigation state feature encoding vectors and flight attitude feature principal component latent encoding vectors is anchored using flight attitude-navigation state latent key clues to obtain feature pairs of {navigation state latent feature encoding vector, anchored flight attitude principal component latent feature encoding vector}. The above process can be expressed by the following formula:

[0056]

[0057] F bestpair ={H1; h 2k}

[0058] Among them, h 2i H1 is the i-th latent coding vector of the principal components of flight attitude features in the set of latent coding vectors of principal components of flight attitude features. H1 is the coding vector of navigation state features. <·> denotes the inner product, ‖·‖ is the Euclidean norm for calculating the vector, ε is the modulation coefficient, and argmax is the maximum value. j (·) represents the j value corresponding to the maximum value, k is the position of the maximum approximate match value in the set of latent encoding vectors of the principal components of flight attitude features, and h2k It is the anchored flight attitude principal component latent feature encoding vector, F bestpair It is a feature pair of {navigation state latent feature encoding vector, anchored flight attitude principal component latent feature encoding vector}.

[0059] It is understandable that flight attitude and navigation status are two different but closely related aspects of aircraft flight. Their feature encoding vectors originate from different sources and have different semantics and feature distributions. Navigation status feature encoding vectors primarily reflect navigation information such as the aircraft's flight path and geographical location, while the set of implicit encoding vectors for the principal components of flight attitude features focuses on attitude information such as pitch, roll, and heading. In actual flight scenarios, although these two types of information influence each other, they exhibit semantic inconsistencies due to differences in their sources and representations. By using flight attitude-navigation status implicit key cue anchoring, the features from these two different sources can be semantically aligned. It establishes a unified semantic standard for different features in a multi-dimensional feature space, enabling navigation status features and flight attitude features to be understood and analyzed within the same semantic framework. For example, when an aircraft performs a turn, its flight attitude changes (e.g., roll), and the flight path in the navigation status also changes accordingly. Through semantic alignment, the model can clearly identify the intrinsic relationship between these two changes, accurately corresponding the attitude change with the adjustment of the navigation path. In practical implementation, the intrinsic relationship between navigation state feature encoding vectors and the set of latent encoding vectors of flight attitude feature principal components can be identified by calculating the correlation between them. Based on this correlation, the most valuable feature pairs are extracted, such as the flight attitude features corresponding to a specific navigation position. These feature pairs provide clear correlations for subsequent feature interactions, enabling the system to better understand the mutual influence between flight attitude and navigation state, thereby achieving more accurate flight state analysis and prediction.

[0060] Finally, based on the feature pair {navigation state latent feature encoding vector, anchored flight attitude principal component latent feature encoding vector}, a fine-grained interaction of flight attitude and navigation state features is performed on the flight attitude feature vector and the navigation state feature vector to obtain the flight attitude-navigation state feature interaction encoding vector. The above process can be expressed by the following formula:

[0061]

[0062] Wherein, V2 is the flight attitude feature vector, V1 is the navigation state feature vector, H1 is the navigation state feature encoding vector, and h 2k It is the anchored flight attitude principal component latent feature encoding vector, h 2k T It is h 2kThe transpose of , softmax is the activation function, and S is the h 2k Length, It is a matrix multiplication, where α and β are weighted hyperparameters, and V f It is the flight attitude-navigation state feature interactive encoding vector.

[0063] It is understandable that after achieving semantic alignment by anchoring implicit key cues between flight attitude and navigation state, fine-grained interaction of flight attitude and navigation state features can uncover deeper and more complex nonlinear relationships between these features. Specifically, aircraft flight is a complex dynamic system, and changes in flight attitude (such as pitch, roll, and yaw) and navigation state (such as flight path and geographical location) are not simply linear relationships. For example, when flying in mountainous areas, the aircraft needs to constantly adjust its flight attitude to avoid peaks, while the navigation system also plans new flight paths in real time based on terrain and target location. In this case, the interaction between flight attitude and navigation state exhibits a high degree of nonlinearity. Fine-grained interaction can capture these complex relationships, enabling the system to have a deeper and more accurate understanding of the flight state. Multi-head self-attention mechanisms are one of the core technologies for achieving fine-grained interaction of flight attitude and navigation state features. It can simultaneously focus on different parts of the input features, capturing long-distance dependencies and multi-dimensional semantic interactions. In the interaction of flight attitude and navigation state features, multi-head self-attention mechanisms can analyze and fuse features from multiple perspectives. For example, one "head" can focus on the impact of pitch angle changes in flight attitude features on the navigation route, while another "head" can focus on how changes in geographical location in the navigation state prompt adjustments to flight attitude. This approach provides greater flexibility for feature fusion, enabling it to adapt to the complex and ever-changing relationship between flight attitude and navigation state in different flight scenarios. The resulting flight attitude-navigation state feature interaction encoding vector integrates key information from both flight attitude and navigation state, and fully considers the complex nonlinear relationship between them, providing a solid feature data foundation for the subsequent generation of high-quality synthesized flight display images.

[0064] Preferably, in another embodiment of this application, the fine-grained interaction subunit for flight attitude-navigation state features is used to: perform weakly blurred semantic correction based on eigenvalues ​​on the feature pair {navigation state latent feature encoding vector, anchored flight attitude principal component latent feature encoding vector} to obtain a corrected feature pair {navigation state latent feature encoding vector, anchored flight attitude principal component latent feature encoding vector}; and based on the corrected feature pair {navigation state latent feature encoding vector, anchored flight attitude principal component latent feature encoding vector}, perform fine-grained interaction of flight attitude feature vector and navigation state feature vector to obtain the flight attitude-navigation state feature interaction encoding vector. The above process is expressed by the following formula:

[0065]

[0066] Among them, H 1i h is the eigenvalue at the i-th position in H1. 2ki It is h 2k The feature value at position i, H 1i ′ is H 1i The corrected eigenvalue, h 2ki ′ is h 2ki The corrected feature values, H1′ is the corrected navigation state latent feature encoding vector, h 2k ′ is the modified anchored principal component latent feature encoding vector of flight attitude, V f It is the flight attitude-navigation state feature interactive encoding vector.

[0067] Furthermore, to address the alignment ambiguity between the implicit features of navigation state and the implicit features of the anchored flight attitude principal components caused by source uncertainty, a weakening and blurring mechanism is introduced to perform weakly blurred power-law extension. Specifically, a 3 / 8 exponent is used as a precursor prior extension to perform power-law a priori responsive ambiguity convergence on the alignment conditions of the navigation state implicit feature parameters, while a 1 / 4 exponent is used as the main alignment extension to strictly constrain the alignment attenuation relaxation of the power-law distribution of the anchored flight attitude principal component implicit features. Thus, in the case where the alignment boundary condition constraints within the effective association range are unclear, the value relevance mechanism avoids the prior ambiguity of system behavior under a single mechanism, corrects the semantic distribution consistency ambiguity within the alignment interval, and improves the intuitiveness of mining the implicit fine-grained interaction associations of the anchored features.

[0068] In this embodiment, the flight image synthesis unit 145 is used to obtain the synthesized flight display image based on the flight attitude-navigation state feature interaction encoding vector. Specifically, in this embodiment, the flight image synthesis unit is used to: pass the flight attitude-navigation state feature interaction encoding vector through an AIGC-based flight display synthesizer to obtain the synthesized flight display image. That is, the flight attitude-navigation state feature interaction encoding vector obtained by performing key semantic interaction between the flight attitude feature vector and the navigation state feature vector is generated. This is then used in conjunction with an AIGC-based flight display synthesizer to utilize deep learning algorithms to understand the complex semantic information in the flight attitude-navigation state feature interaction encoding vector. Based on this information, the compositional patterns of images under different flight states are learned from a large amount of training data, thereby generating a high-quality synthesized flight display image that conforms to the actual flight situation. The synthesized flight display image contains detailed flight attitude indication information and navigation information, providing the pilot with an intuitive and comprehensive view. In this image, an icon representing the aircraft clearly shows its current attitude, including its roll and pitch angles. The horizon indicator indicates whether the aircraft is ascending, descending, or flying horizontally, and the heading indicator displays the aircraft's current heading angle. Furthermore, the image includes map information of the current location, destination, preset flight path, an overview of the surrounding terrain, and waypoint markings.

[0069] The following is a detailed explanation of a specific implementation process for "obtaining the synthesized flight display image by passing the flight attitude-navigation state feature interactive encoding vector through an AIGC-based flight display synthesizer":

[0070] The model training phase is the foundation and key to the entire synthesis process, determining the flight display synthesizer's understanding of flight data and its image generation capabilities. First, it requires the extensive collection of rich data across various flight scenarios, including flight attitude data, navigation data, and corresponding actual flight display images in different weather conditions (e.g., clear skies, heavy rain, dense fog), flight phases (takeoff, cruise, landing), and geographical environments (plains, mountains, urban areas, etc.). The diversity and comprehensiveness of this data are crucial, enabling the flight display synthesizer to learn flight information features under various complex conditions. After collection, data preprocessing is performed to clean the data, removing outliers and noise to ensure accuracy and reliability. Flight attitude and navigation data are then precisely correlated and labeled with actual flight display images, constructing a high-quality training dataset for the model to learn from.

[0071] Subsequently, a suitable deep learning architecture was selected to build a flight display synthesizer based on AIGC. Taking a Generative Adversarial Network (GAN) as an example, it includes two important components: a generator and a discriminator. The generator is responsible for generating synthetic images based on the input flight attitude-navigation state feature interaction encoding vector, while the discriminator is used to judge the differences between the generated images and the real flight display images. The two cooperate and compete with each other. During training, preprocessed training data is continuously input into the synthesizer, and the parameters of the generator and discriminator are continuously adjusted using the backpropagation algorithm. The generator strives to generate high-quality images that can deceive the discriminator, while the discriminator strives to accurately distinguish between real and generated images. After a large number of training iterations, the flight display synthesizer gradually stabilizes and reaches a good performance state, capable of deeply understanding the complex semantic information in the flight attitude-navigation state feature interaction encoding vector.

[0072] Once the model training is complete and reaches the expected performance, the image generation stage begins. At this point, the flight attitude-navigation state feature interaction encoding vector, obtained through the flight attitude-navigation state feature interaction unit, is input into the generator of the pre-trained flight display synthesizer. The generator decodes the encoding vector based on the image composition patterns learned during training for different flight states. It systematically constructs the synthesized image from the rich feature patterns extracted from the training data. For example, based on the attitude information in the encoding vector, it accurately determines the attitude of the icon representing the aircraft in the image, including tilt and pitch angles; and generates map information of the current location, destination, preset route, surrounding terrain overview, and waypoint markers based on navigation information. These elements are logically and spatially combined in the image to initially form a synthesized image containing flight attitude indication and navigation information.

[0073] However, the generated preliminary synthesized image still needs to undergo rigorous evaluation by a discriminator (in a GAN architecture). The discriminator comprehensively judges the generated image from various aspects, such as the accuracy of elements, the rationality of the layout, and the degree of matching with the real flight scene. If the generated image meets the preset quality standards, it is output as the final synthesized flight display image; if it does not meet the requirements, the generator will further adjust its parameters based on the information from the discriminator and regenerate the image. This process is repeated until the generated image meets the high-quality requirements. The final output synthesized flight display image contains detailed and accurate flight attitude indication information and navigation information, providing the pilot with an intuitive and comprehensive view of the flight status.

[0074] In summary, the flight image synthesis module 140 is clearly described. It uses machine vision-based image analysis and processing technology to enhance the image resolution of the flight attitude indication image and the navigation image. Then, it extracts features from the preprocessed flight attitude indication image and the preprocessed navigation image, thereby automatically generating the synthesized flight display image based on the key semantic interaction representation between the extracted flight attitude features and navigation state features. This allows for dynamic adjustment of the feature extraction strategy under different environmental conditions, thus ensuring the accuracy and reliability of the synthesized flight display image.

[0075] In this embodiment, the display module 150 is used to read the composite flight display image from the display frame buffer and send the composite flight display image to the multi-function display screen in the cockpit via the LVDS interface for display. It should be understood that the display frame buffer acts as a data relay station, pre-storing the composite flight display image. Reading images from the buffer ensures data stability and continuity. During flight, the processing of flight data and the generation of images is a continuous dynamic process. If images are directly obtained from the data generation end for display, delays and fluctuations in data processing may cause display stuttering, flickering, or even data loss. The existence of the buffer allows the display module to read image data at a stable rate, ensuring smooth display. The composite flight display image contains rich flight attitude and navigation information, resulting in a large data volume. To transmit this image data completely to the multi-function display screen in a short time, a high-speed data transmission interface is required. The LVDS interface has high-speed transmission characteristics, and its data transmission rate can meet the requirements of real-time flight image display, ensuring that the pilot can obtain the latest flight information in a timely manner. By observing the images on the display screen, pilots can intuitively understand key information such as the aircraft's flight attitude, flight path, and geographical location, so as to make timely and correct flight decisions and ensure flight safety.

[0076] In summary, the integrated display device 100 based on the FPGA and dual ARM processor architecture of this application embodiment is described. It first acquires flight attitude data of the target aircraft and transmits it to the first ARM processor via a MIPI interface. The first ARM processor preprocesses the data and stores it in an input buffer of dual-port shared memory. Next, the FPGA chip reads the preprocessed data from the input buffer to generate a flight attitude indication image and writes it to the output buffer. The second ARM processor reads the attitude indication image from the output buffer and performs image synthesis based on semantic feature interaction with the navigation image to obtain a synthesized flight display image, which is then stored in a display frame buffer. Finally, the synthesized image is read from the display frame buffer and transmitted to the cockpit multi-function display screen via an LVDS interface for display. This enables more efficient, accurate, and reliable flight information display.

Claims

1. A comprehensive display device based on FPGA and dual ARM processor architecture, characterized in that, include: The data input module is used to collect the time series of flight attitude data of the target aircraft and input it into the first ARM processor through the MIPI interface. The data preprocessing module is used to preprocess the time series set of the flight attitude data through the first ARM processor and write it into the input buffer of the dual-port shared memory; The flight attitude indication image generation module is used to read the timing set of the preprocessed flight attitude data from the input buffer through the FPGA chip, and write the generated flight attitude indication image into the output buffer of the dual-port shared memory. The flight image synthesis module is used to read the flight attitude indication image from the output buffer through the second ARM processor, and synthesize the flight attitude indication image and the navigation image based on key semantic feature interaction to obtain the synthesized flight display image, and write it into the display frame buffer. The display module is used to read the composite flight display image from the display frame buffer and send the composite flight display image to the multi-function display screen in the cockpit via the LVDS interface for display. The flight image synthesis module includes: An image preprocessing unit is used to enhance the image resolution of the flight attitude indication image and the navigation image to obtain a preprocessed flight attitude indication image and a preprocessed navigation image. A flight attitude feature extraction unit is used to extract flight attitude features from the preprocessed flight attitude indication image to obtain a flight attitude feature vector. A navigation state feature extraction unit is used to extract navigation state features from the preprocessed navigation image to obtain a navigation state feature vector. The flight attitude-navigation state feature interaction unit is used to perform rapid interaction analysis of key semantics between flight attitude-navigation state on the flight attitude feature vector and the navigation state feature vector to obtain the flight attitude-navigation state feature interaction encoding vector. The flight image synthesis unit is used to obtain the synthesized flight display image based on the flight attitude-navigation state feature interactive encoding vector; The flight attitude-navigation state feature interaction unit includes: The feature implicit coding subunit is used to perform principal component implicit coding and implicit feature extraction on the flight attitude feature vector and the navigation state feature vector respectively to obtain the set of flight attitude feature principal component implicit coding vectors and the navigation state feature coding vector; The flight attitude navigation state implicit feature anchoring subunit is used to anchor the set of navigation state feature encoding vector and flight attitude feature principal component implicit encoding vector to obtain the feature pair {navigation state implicit feature encoding vector, anchored flight attitude principal component implicit feature encoding vector}. The flight attitude-navigation state feature fine-grained interaction subunit is used to perform fine-grained interaction on the flight attitude feature vector and the navigation state feature vector based on the feature pair {navigation state latent feature encoding vector, anchored flight attitude principal component latent feature encoding vector} to obtain the flight attitude-navigation state feature interaction encoding vector.

2. The integrated display device based on FPGA and dual ARM processor architecture according to claim 1, characterized in that, The flight attitude data includes pitch angle, roll angle, and heading angle.

3. The integrated display device based on FPGA and dual ARM processor architecture according to claim 1, characterized in that, The flight attitude feature extraction unit is used to: pass the preprocessed flight attitude indication image through a flight attitude feature extractor based on the MobileNet model to obtain the flight attitude feature vector.

4. The integrated display device based on FPGA and dual ARM processor architecture according to claim 3, characterized in that, The navigation state feature extraction unit is used to: pass the preprocessed navigation image through a navigation state feature extractor based on the ViT model to obtain the navigation state feature vector.

5. The integrated display device based on FPGA and dual ARM processor architecture according to claim 1, characterized in that, The feature hidden coding subunit is used for: Perform principal component analysis on the flight attitude feature vectors to obtain a set of flight attitude feature principal component encoded vectors; The flight attitude feature principal component encoding vectors in the set of flight attitude feature principal component encoding vectors are subjected to flight attitude point convolution latent feature extraction to obtain the set of flight attitude feature principal component latent encoding vectors; The navigation state feature vector is subjected to navigation state point convolution latent feature extraction to obtain the navigation state feature encoding vector.

6. The integrated display device based on FPGA and dual ARM processor architecture according to claim 5, characterized in that, The fine-grained interaction subunit for flight attitude-navigation state features is used for: The feature pair {navigation state latent feature encoding vector, anchored flight attitude principal component latent feature encoding vector} is subjected to weak fuzzy semantic correction based on feature values ​​to obtain the corrected feature pair {navigation state latent feature encoding vector, anchored flight attitude principal component latent feature encoding vector}. Based on the modified {navigation state latent feature encoding vector, anchored flight attitude principal component latent feature encoding vector} feature pair, a fine-grained interaction of flight attitude and navigation state features is performed on the flight attitude feature vector and the navigation state feature vector to obtain the flight attitude and navigation state feature interaction encoding vector.

7. The integrated display device based on FPGA and dual ARM processor architecture according to claim 6, characterized in that, The flight image synthesis unit is used to: pass the flight attitude-navigation state feature interactive encoding vector through an AIGC-based flight display synthesizer to obtain the synthesized flight display image.

Images