Intelligent algorithm air test base with multi-modal interface

Abstract

The application provides a smart algorithm over-the-air test base with a multi-modal interface, the base comprising: a standard 3U-VPX plug-in chassis, internally integrated with a VPX backplane; a main control unit plugged on a corresponding slot of the VPX backplane through a VPX connector, serving as the central management core of the whole machine; an interface module arranged on the VPX backplane, the interface module being configured as multiple different types of physical interfaces, constituting a multi-modal interface system facing over-the-air test scenarios; a supporting software deployed on the main control unit; multiple smart models can realize plug-and-play through the supporting software without the need to write any underlying code. The application realizes plug-and-play verification of smart models by sinking heterogeneous computing capability and real-time data processing capability into a modular, plug-in test base architecture, and internally embedding a general smart algorithm framework.

Description

Technical Field

[0001] This application belongs to the field of avionics and artificial intelligence testing and verification technology, and in particular relates to an airborne test base for intelligent algorithms with a multimodal interface. Background Technology

[0002] With the deep penetration of artificial intelligence technology into the field of aviation, deep learning-based intelligent algorithms are gradually becoming core functional modules of airborne systems for unmanned and manned aircraft. However, compared with ground-based unmanned platforms, the operating environment faced by airborne vehicles is highly dynamic, latency-sensitive, and uncertain, placing more stringent requirements on the real-time performance, robustness, and reliability of airborne intelligent algorithms.

[0003] Currently, the testing and verification of intelligent algorithms mainly involves three levels of technical approaches: First, pure digital simulation verification. This method simulates the operation of intelligent algorithms in a computer-built virtual environment without requiring any real hardware peripherals. Its advantages are low cost and fast iteration, but there are significant differences between the virtual environment and the real physical world. A high simulation pass rate does not equal a high experimental pass rate, and the algorithm often experiences a significant performance drop after being physically ported, a problem known as the "porting gap."

[0004] Second, hardware-in-the-loop (HIL) simulation verification. This method integrates some real hardware into the simulation system to form a semi-physical simulation environment, which to some extent bridges the gap between pure digital simulation and physical testing. However, existing HIL simulation platforms are generally designed for ground-based application scenarios, have poor scalability, and cannot effectively simulate the unique dynamic flight environment of aircraft and the real data characteristics of airborne sensors.

[0005] Third, real-world flight testing. This method is the most direct way to verify intelligent algorithms, which involves testing the algorithm in real airspace to obtain the most reliable test data. However, real-world flight testing has significant drawbacks, including extremely high testing costs, long testing cycles, difficulty in airspace coordination, poor reproducibility, and high safety risks.

[0006] Furthermore, current technologies lack an airborne testing platform capable of supporting plug-and-play deployment of any intelligent model. Most existing testing platforms are hard-coded for specific algorithms, requiring rewriting of data interface code and underlying drivers each time an algorithm is changed, resulting in extremely low algorithm iteration efficiency. Simultaneously, existing platforms have limited computing power in their intelligent computing modules and do not support hybrid deployment of mainstream AI frameworks, making it difficult to adapt to the rapid evolution of deep learning algorithms. Summary of the Invention

[0007] The purpose of this invention is to propose an aerial test base for intelligent algorithms with multimodal interfaces. By sinking heterogeneous computing capabilities and real-time data processing capabilities into a modular, pluggable test base architecture, and incorporating a general intelligent algorithm framework, it enables plug-and-play verification of intelligent models.

[0008] This application provides an aerial testing base for intelligent algorithms with a multimodal interface, the base comprising: Standard 3U-VPX card-type chassis with integrated VPX backplate; The main control unit is plugged into the corresponding slot on the VPX backplane via a VPX connector, serving as the central management core of the entire unit; An interface module is installed on the VPX backplane. The interface module is configured as a variety of different types of physical interfaces to form a multimodal interface system for airborne test scenarios. The accompanying software is deployed on the main control unit; various intelligent models can be used plug-and-play through the accompanying software without writing any underlying code.

[0009] Preferably, the supporting software includes a deep learning and reinforcement learning algorithm library SDK, which is a general intelligent algorithm framework.

[0010] Preferably, the general intelligent algorithm framework provides a unified model loading interface, a unified data preprocessing interface, a unified inference scheduling interface, and a unified result output interface; the intelligent model does not require any modification to the underlying hardware of the base and can be loaded, run, and verified in a plug-and-play manner under the general intelligent algorithm framework.

[0011] Preferably, the general intelligent algorithm framework is divided into four layers: The first layer is the model loading layer, which provides a unified model loading interface and supports the mixed deployment of multiple model formats. The framework automatically identifies the model file format and calls the corresponding parser to extract the model structure, weights, and input / output tensor information. The model container adopts a pluggable design, and new model formats can be dynamically extended through the plugin mechanism. The second layer is the data preprocessing layer, which provides a unified data preprocessing interface. The framework automatically extracts the corresponding data from the multimodal data stream collected by the interface module according to the specifications of the model input tensor, and completes the preprocessing operations of size transformation, normalization, and format conversion without the need for manual adaptation code. The third layer is the inference scheduling layer, which has a built-in heterogeneous computing resource automatic scheduler. The scheduler manages the concurrent execution of multiple instances of the model at the same time, realizing the time-division multiplexing of computing resources. The fourth layer is the result output layer, which provides a unified result output interface. It encapsulates the model inference results into a standardized data structure and distributes them to the test monitoring software for display and storage through the main control unit, or feeds them back to the test machine platform through the bus interface. Through the above four-layer design, the intelligent models can all be plug-and-play without writing any underlying code.

[0012] Preferably, multiple model formats include ONNX, TensorRT, OpenVINO, and TFLite.

[0013] Preferably, the multimodal interface system can simultaneously acquire airborne sensor data in a real air environment from multiple data sources, thus forming a complete airborne multimodal data acquisition capability.

[0014] Preferably, the multimodal interface system includes: Four 100 / 1000 Mbps Ethernet communication interfaces for high-speed data exchange and video streaming; Dual redundant AFDX network interfaces for interconnection with avionics systems compliant with ARINC 664 standards; Four RS422 interfaces are used for communication with inertial navigation and GPS serial sensors; Dual redundant 1553B communication interfaces are used for reliable bus communication with the airborne mission computer and flight control computer. ARINC429 interface: 4 transmit, 4 receive, used for data exchange with avionics equipment.

[0015] Preferably, the intelligent model includes a target tracking model, an intelligent recognition model, an intelligent decision-making model, a path planning model, and a classification model.

[0016] The beneficial technical effects of this application are as follows: The base can be statically deployed on the ground or carried by an experimental aircraft platform for in-flight testing. It can simultaneously access multiple heterogeneous data streams through data links, providing multimodal data input consistent with the real flight environment for any intelligent model running on it. The base has a built-in general intelligent algorithm framework that supports plug-and-play loading, automatic adaptation, and heterogeneous accelerated inference for various intelligent models without requiring modification of model code or underlying drivers. The base can comprehensively test and evaluate the functional correctness, real-time performance, and robustness of intelligent models under real flight conditions, thereby filling the gap in in-flight physical-in-the-loop verification functionality between ground simulation and actual flight. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 Diagram of the base architecture for aerial testing of intelligent algorithms; Figure 2 Overall design concept diagram of the aerial test base for intelligent algorithms; Figure 3 Diagram of the software development environment for the aerial test base of intelligent algorithms. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The features and illustrative embodiments of various aspects of the present invention will now be described in detail. Numerous specific details are set forth in the following detailed description to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention may be practiced without requiring some of these specific details. The following description of embodiments is merely intended to provide a better understanding of the invention by illustrating examples of the invention. The invention is by no means limited to any specific setups and methods set forth below, but covers any improvements, substitutions, and modifications to structures, methods, and devices without departing from the spirit of the invention. Well-known structures and techniques are not shown in the drawings and the following description to avoid unnecessarily obscuring the invention.

[0021] In the description of this invention, it should be noted that the directions or positional relationships indicated by terms such as "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer" are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing and simplifying the invention, and should not be construed as limiting the invention. Furthermore, the use of ordinal numbers (e.g., "first and second," etc.) is for distinguishing objects and is not limited to this order, and should not be construed as indicating or implying relative importance.

[0022] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly, encompassing both direct connection and indirect connection via an intermediate medium. Those skilled in the art can understand the specific meaning of these terms in this invention based on the specific circumstances.

[0023] It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other, and the various embodiments can be referenced and cited in each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0024] The following describes the embodiments and appendices. Figure 1 - Appendix Figure 3 The present invention will be described in further detail, but the embodiments of the present invention are not limited thereto.

[0025] This invention provides an airborne test platform for intelligent algorithms with multimodal interfaces. The platform includes a power module, a main control unit, an intelligent computing module, an interface module, debugging and development tools, supporting software, and connectors. The airborne test platform adopts an integrated design using a standard 3U VPS board-based plug-in chassis. Each functional module is installed on the VPS backplane via plug-in, enabling modular and reconfigurable deployment. The platform can be statically deployed on the ground or carried by a test aircraft platform for airborne testing. It simultaneously accesses various heterogeneous data streams via data links, providing multimodal data input consistent with the real flight environment for any intelligent model running on it. The platform incorporates a general-purpose intelligent algorithm framework, supporting plug-and-play loading, automatic adaptation, and heterogeneous accelerated inference for various intelligent models without requiring modification of model code or underlying drivers. The platform can comprehensively test and evaluate the functional correctness, real-time performance, and robustness of intelligent models under real flight conditions, thus filling the gap in airborne physical-in-the-loop verification functionality between ground simulation and actual flight.

[0026] This invention provides an aerial testing base for intelligent algorithms with a multimodal interface, comprising: The intelligent algorithm aerial test base includes a power module, a main control unit, an intelligent computing module, an interface module, debugging and development tools, and supporting software. The aerial test base adopts an integrated design with a standard 3U-VPX board plug-in chassis. Each functional module is installed on the VPX backplane in a plug-in manner, realizing modular and reconfigurable deployment. The interface module is configured to simultaneously deploy sensor data acquisition interfaces, standardized data communication interfaces, and extended custom interfaces, forming a multimodal interface system to simultaneously access heterogeneous data streams from multiple airborne sensors.

[0027] Furthermore, the power module supports 28VDC airborne power input, has a wide voltage input range and overvoltage, undervoltage and overcurrent protection functions, and is suitable for complex power supply environments in the field.

[0028] Furthermore, the main control unit adopts a Phytium D2000 processor and X100 chipset, is equipped with 16GB of onboard memory and a 2TB solid-state drive, runs the Ubuntu 18.04 operating system and loads the PREEMPT_RT real-time kernel to ensure the real-time performance of data processing.

[0029] Furthermore, the intelligent computing module adopts a fully domestically produced AI processor, with the Ascend 310B (HI1910) as its core chip. This processor includes one DaVinci V300 AI Core, four 64-bit Taishan V200M processor cores (ARMv8-A architecture), an integrated 96-bit LPDDR4X memory interface, 12GB of memory, and provides 10 TFLOPS of FP16 computing power and 20 TOPS of INT8 computing power. It supports AI frameworks and inference engines such as PyTorch, TensorFlow, Caffe, and ONNX, and supports CNN, RNN neural network algorithms and reinforcement learning algorithms. This module exchanges data at high speed with the VPX backplane and main control unit through a PCIe Gen4.0 x8 lane interface.

[0030] Furthermore, the multimodal interface system deployed in the interface module specifically includes: four 100 / 1000 Mbps Ethernet communication interfaces (electrical ports), dual redundant AFDX network interfaces, four RS422 interfaces, dual redundant 1553B communication interfaces, and four ARINC429 interfaces for transmitting and receiving. These interfaces can simultaneously connect to visual sensors, radar sensors, inertial navigation sensors, and avionics bus data links, meeting the synchronous acquisition requirements of multi-source heterogeneous data in airborne testing scenarios.

[0031] Furthermore, the debugging and development tools include a testing instrument capable of detecting four Ethernet interfaces, dual redundant AFDX interfaces, dual redundant 1553B interfaces, four transmit / four receive ARINC429 interfaces, and four RS422 interfaces, and providing testing software to automatically test all interfaces; the testing instrument software can drive the intelligent algorithm prototype (i.e., the test base) to perform functional and performance tests; the intelligent algorithm prototype is pre-installed with PyTorch, TensorFlow, Caffe, and ONNX development environments, as well as neural network algorithm libraries and reinforcement learning algorithm libraries, facilitating algorithm development.

[0032] Furthermore, the accompanying software features power-on self-testing, flight self-testing, and maintenance self-testing functions; each module provides drivers, libraries, and example code; the main control unit runs the Ubuntu 18.04 desktop operating system and loads a real-time kernel; it pre-installs deep learning and reinforcement learning algorithm libraries (SDKs); and it pre-installs an automated deployment software suite. The accompanying software incorporates a general-purpose intelligent algorithm framework, which provides a unified model loading interface, a unified data preprocessing interface, a unified inference scheduling interface, and a unified result output interface. All intelligent models can be plugged in and run without code modification. The framework automatically parses the model's input and output structure and automatically matches it with multimodal sensor data. It also has a built-in heterogeneous computing resource automatic scheduler that intelligently selects NPU, GPU, or FPGA execution units based on the model's computational characteristics.

[0033] The introduction of the aforementioned general intelligent algorithm framework solves the technical challenge of unifying and adapting heterogeneous models, heterogeneous buses, and heterogeneous hardware, enabling plug-and-play functionality for any intelligent model. In existing technologies, airborne avionics bus interfaces (1553B, AFDX, RS422, ARINC429, etc.) have complex data types, intelligent models come from diverse sources (PyTorch, TensorFlow, Caffe, ONNX, etc.), and the underlying hardware includes heterogeneous units such as NPUs, GPUs, and FPGAs. Typically, a large amount of adaptation code needs to be written for each "model + bus + hardware" combination, resulting in a long algorithm porting cycle.

[0034] In other embodiments of this application: Overall structure of the aerial test base like Figure 2 This embodiment provides an intelligent algorithm aerial test base with a multimodal interface. The base uses a standard 3U-VPX plug-in chassis as the physical carrier. The chassis integrates a VPX backplane, and each functional module is installed on the VPX backplane in a plug-in manner.

[0035] Specifically, the aerial test base includes: The power module provides a stable power supply to the entire base. It supports 28VDC onboard power input, features a wide voltage input design, and is equipped with input overvoltage protection, undervoltage protection, and output overcurrent protection circuits. The power module is electrically connected to the main control unit, intelligent computing module, interface module, and other power-consuming modules via the power supply bus on the VPX backplane, enabling single-point power supply and overall system distribution.

[0036] The main control unit, connected to the corresponding slot on the VPX backplane via a VPX connector, serves as the central management core of the entire system. The main control unit uses a Phytium D2000 processor and X100 chipset, configured with 16GB of onboard memory and a 2TB solid-state drive, running the Ubuntu 18.04 operating system with the PREEMPT_RT real-time kernel. In actual testing, the main control unit is responsible for managing the buffering, distribution, and synchronization of multimodal data streams received by the interface modules, as well as monitoring the status of each module and diagnosing faults during the testing process. The main control unit provides front-panel interfaces: one RJ45 Ethernet port (for debugging), one HDMI port, and two USB ports, all of which are standard interfaces.

[0037] The intelligent computing module exchanges data with the main control unit via a PCIe Gen4.0 x8 lane high-speed bus. This module is responsible for running the target intelligent model.

[0038] The interface module provides a rear-mounted aviation plug interface and a front-mounted debugging interface on its external interface panel.

[0039] The debugging and development tools include a testing instrument capable of detecting four Ethernet interfaces, dual-redundant AFDX interfaces, dual-redundant 1553B interfaces, four-transmit and four-receive ARINC429 interfaces, and four RS422 interfaces. It also provides testing software to automatically test all interfaces. The testing instrument software can drive the intelligent algorithm prototype to perform functional and performance tests. The intelligent algorithm prototype comes pre-installed with PyTorch, TensorFlow, Caffe, and ONNX development environments, as well as neural network algorithm libraries and reinforcement learning algorithm libraries.

[0040] The accompanying software includes the Ubuntu 18.04 operating system, drivers, libraries and routine code deployed on the main control unit, as well as self-testing software, a pre-installed deep learning and reinforcement learning algorithm library SDK, and a pre-installed automated deployment software suite.

[0041] In other embodiments of this application: Multimodal interface configuration The multimodal interface system of the interface module specifically includes: Four 100 / 1000 Mbps Ethernet communication interfaces: used for high-speed data exchange and video streaming; Dual redundant AFDX network interfaces: used for interconnection with avionics systems compliant with ARINC 664 standards; 4 RS422 interfaces: used for communication with serial sensors such as inertial navigation and GPS; Dual-redundant 1553B communication interface: used for reliable bus communication with the airborne mission computer and flight control computer; ARINC429 interface: 4 transmit, 4 receive, used for data exchange with avionics equipment.

[0042] The aforementioned interfaces are diverse and can simultaneously acquire airborne sensor data in real-world aerial environments from multiple data sources, forming a complete airborne multimodal data acquisition capability.

[0043] In other embodiments of this application: Hardware design of the intelligent computing module The intelligent computing module in this embodiment adopts a fully domestically produced design, and its core component is a computing board built based on the Ascend 310B AI processor.

[0044] The main technical features of the Ascend 310B (model HI1910) AI processor are as follows: AI computing core: 1 DaVinci V300 AI Core, designed specifically for accelerating deep neural network inference.

[0045] CPU cores: 4 64-bit Taishan V200M processor cores, using the ARMv8-A architecture, for general computing and task scheduling.

[0046] Memory interface: 96-bit LPDDR4X interface, supporting high-bandwidth and low-power memory access.

[0047] High-speed interface: 8-lane SerDes interface, supporting multiplexing of PCIe, SATA and USB3.0.

[0048] Low-speed interfaces: UART, I2C, SPI, CAN, PWM, eMMC.

[0049] Network interfaces: RGMII, SGMII.

[0050] Audio and video interfaces: MIPI CSI, MIPI DSI, HDMI, I2S.

[0051] The intelligent computing module is equipped with 12GB of LPDDR4X memory, providing 10 TFLOPS of FP16 computing power and 20 TOPS of INT8 computing power. This module supports mainstream AI frameworks and inference engines such as PyTorch, TensorFlow, Caffe, and ONNX, as well as neural network algorithms such as CNN and RNN, and reinforcement learning algorithms.

[0052] The intelligent computing module exchanges high-speed data with the VPX backplane 102 and the main control unit 12 via a PCIe Gen4.0 x8 lane interface. This module can run a Linux system and works in conjunction with the main control unit 12 to complete the loading, inference calculation, and result output of the intelligent model.

[0053] Other embodiments in this application: General intelligent algorithm framework The accompanying software comes pre-installed with a deep learning and reinforcement learning algorithm library SDK, which is divided into four layers: The first layer is the model loading layer. It provides a unified model loading interface, supporting mixed deployment of multiple model formats such as ONNX, TensorRT, OpenVINO, and TFLite. The framework automatically identifies the model file format and calls the corresponding parser to extract the model structure, weights, and input / output tensor information. The model container adopts a pluggable design; new model formats can be dynamically extended through a plugin mechanism.

[0054] The second layer is the data preprocessing layer. It provides a unified data preprocessing interface. Based on the specifications of the model input tensor, the framework automatically extracts the corresponding data from the multimodal data stream collected by the interface module and completes preprocessing operations such as size transformation, normalization, and format conversion, without the need for manual adaptation code.

[0055] The third layer: Inference scheduling layer. It includes a built-in automatic scheduler for heterogeneous computing resources. The scheduler simultaneously manages the concurrent execution of multiple instances of the model, achieving time-division multiplexing of computing resources.

[0056] The fourth layer is the result output layer. It provides a unified result output interface, encapsulates the model inference results into a standardized data structure, and distributes them to the test monitoring software for display and storage through the main control unit, or feeds them back to the test machine platform through the bus interface.

[0057] Through the above four-layer design, the intelligent models can all be plug-and-play without writing any underlying code.

[0058] In other embodiments of this application: the test platform carries out in-flight verification. The intelligent algorithm aerial test base of this embodiment can be mounted on a test machine platform for aerial physical verification.

[0059] During the test preparation phase, the intelligent algorithm airborne test base was integrated and installed in the equipment bay of the test aircraft platform, and the interface module of the airborne test base was interconnected with the flight control computer of the test aircraft platform.

[0060] During the test execution phase, the test platform and aerial test base are powered on and started. The supporting software automatically performs power-on self-tests. The general intelligent algorithm framework automatically loads the pre-set target intelligent model.

[0061] During the test, the debugging and development tools can perform online monitoring and fault diagnosis of each interface. After the test, the ground control terminal exports all recorded data, including raw data, intermediate model outputs, final outputs, timestamps, flight status parameters, etc., for offline analysis.

[0062] The above detailed embodiments are a description of the present invention. It should not be considered that the specific embodiments of the present invention are limited to these descriptions. For those skilled in the art, several simple deductions and substitutions can be made without departing from the concept of the present invention, and all of these should be considered to fall within the protection scope of the present invention.

Claims

1. An aerial test base for intelligent algorithms with a multimodal interface, characterized in that, The base includes: Standard 3U-VPX card-type chassis with integrated VPX backplate; The main control unit is plugged into the corresponding slot on the VPX backplane via a VPX connector, serving as the central management core of the entire unit; An interface module is installed on the VPX backplane. The interface module is configured as a variety of different types of physical interfaces to form a multimodal interface system for airborne test scenarios. The accompanying software is deployed on the main control unit; various intelligent models can be used plug-and-play through the accompanying software without writing any underlying code.

2. The base according to claim 1, characterized in that, The supporting software includes a deep learning and reinforcement learning algorithm library SDK, which is a general intelligent algorithm framework.

3. The base according to claim 2, characterized in that, The general intelligent algorithm framework provides a unified model loading interface, a unified data preprocessing interface, a unified inference scheduling interface, and a unified result output interface. The intelligent model does not require any modification to the underlying hardware of the base and can be loaded, run, and verified in a plug-and-play manner under the general intelligent algorithm framework.

4. The base according to claim 3, characterized in that, The general intelligent algorithm framework is divided into four layers: The first layer is the model loading layer, which provides a unified model loading interface and supports the mixed deployment of multiple model formats. The framework automatically identifies the model file format and calls the corresponding parser to extract the model structure, weights, and input / output tensor information. The model container adopts a pluggable design, and new model formats can be dynamically extended through the plugin mechanism. The second layer is the data preprocessing layer, which provides a unified data preprocessing interface. The framework automatically extracts the corresponding data from the multimodal data stream collected by the interface module according to the specifications of the model input tensor, and completes the preprocessing operations of size transformation, normalization, and format conversion without the need for manual adaptation code. The third layer is the inference scheduling layer, which has a built-in heterogeneous computing resource automatic scheduler. The scheduler manages the concurrent execution of multiple instances of the model at the same time, realizing the time-division multiplexing of computing resources. The fourth layer is the result output layer, which provides a unified result output interface. It encapsulates the model inference results into a standardized data structure and distributes them to the test monitoring software for display and storage through the main control unit, or feeds them back to the test machine platform through the bus interface. Through the above four-layer design, the intelligent models can all be plug-and-play without writing any underlying code.

5. The base according to claim 4, characterized in that, Multiple model formats are available, including ONNX, TensorRT, OpenVINO, and TFLite.

6. The base according to claim 1, characterized in that, The multimodal interface system can simultaneously acquire airborne sensor data in real-world aerial environments from multiple data sources, thus forming a complete airborne multimodal data acquisition capability.

7. The base according to claim 6, characterized in that, The multimodal interface system includes: Four 100 / 1000 Mbps Ethernet communication interfaces for high-speed data exchange and video streaming; Dual redundant AFDX network interfaces for interconnection with avionics systems compliant with ARINC 664 standards; Four RS422 interfaces are used for communication with inertial navigation and GPS serial sensors; Dual redundant 1553B communication interfaces are used for reliable bus communication with the airborne mission computer and flight control computer. ARINC429 interface: 4 transmit, 4 receive, used for data exchange with avionics equipment.

8. The base according to claim 1, characterized in that, The intelligent model includes a target tracking model, an intelligent recognition model, an intelligent decision-making model, a path planning model, and a classification model.

Images