Satellite-borne intelligent computing device applied to ecological environment industry

Abstract

The invention provides satellite-borne intelligent computing equipment for ecological environment industry application, and relates to the technical field of data processing. The satellite-borne intelligent computing device applied to the ecological environment industry comprises a hardware acceleration layer, a self-adaptive power management layer and an ecological environment algorithm application layer which are connected in sequence to form a three-layer architecture of heterogeneous computing, self-adaptive power consumption management and algorithm hardening, and deep coupling design of an ecological environment application algorithm and a satellite-borne hardware platform can be achieved; the method has the advantages of high computing power, low power consumption and radiation resistance, dynamic reconfiguration is perceived by using power consumption, algorithm load is pre-judged through a task scheduling engine, a computing resource allocation strategy is adjusted in advance, the optimal energy efficiency is realized, a complex space environment can be met, and related algorithm reasoning calculation can be stably carried out on a satellite.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a satellite-borne intelligent computing device for ecological environment industry applications. BACKGROUND

[0002] With increasing global attention to climate change, environmental protection, and sustainable development, large-scale, long-period, and high-frequency ecological environment monitoring using satellite remote sensing technology has become an indispensable means. Current monitoring tasks are increasingly complex, covering multiple dimensions such as atmospheric composition and pollutant monitoring (e.g., greenhouse gases, aerosols, etc.), water quality and eutrophication assessment (e.g., chlorophyll concentration, algal blooms, etc.), vegetation coverage and ecosystem health assessment (e.g., vegetation index, pest and disease, etc.), and soil key parameter monitoring (e.g., humidity, temperature, salinity, etc.). In addition, in recent years, the development of satellite computing technology has enabled satellites to process data on orbit. Satellite artificial intelligence technology has made a leap from passive compliance with preset instructions to active intelligent analysis and decision-making. Satellite artificial intelligence technology has achieved remarkable results in fields such as Earth remote sensing, astronomical observation, and deep space exploration, and has become the core of satellite intelligence development.

[0003] Traditional satellite data processing modes usually involve downlinking raw remote sensing data to ground stations, which are then processed and analyzed by ground supercomputing centers. This mode has the following significant bottlenecks: 1. High data transmission delay: The limitations of satellite overflight and data downlink bandwidth result in a delay of several hours or even days between data collection and obtaining analysis results, which cannot meet the needs of high-time-sensitive applications such as disaster warning and response to sudden pollution incidents.

[0004] 2. Heavy burden on satellite-ground link: The data volume generated by modern high-resolution and high-spectral sensors grows exponentially, putting enormous pressure on limited satellite-ground communication bandwidth, resulting in a large amount of raw data that cannot be timely downlinked or must be compressed.

[0005] 3. Insufficient on-orbit computing capacity: Although existing radiation-resistant processors are highly reliable, their computing performance is usually far behind that of commercial processors, with low computing density and power efficiency, making it difficult to deploy and run complex and state-of-the-art artificial intelligence models on orbit.

[0006] 4. Severe space environment: The radiation environment in space (including total ionizing dose effect and single event effect) poses a fatal threat to commercially available high-performance computing chips, easily leading to data errors, function interruptions, and even permanent damage. At the same time, the heat dissipation problem in the vacuum environment also limits the upper limit of the power consumption of high-performance processors.

[0007] In order to break through the above bottleneck, the edge computing and artificial intelligence technology are currently introduced into on-board processing, i.e. on-orbit intelligent processing. However, to design a spaceborne computer capable of meeting the complex ecological environment monitoring task requirements, while taking into account the high-performance AI (Artificial Intelligence) inference ability and strict on-board power budget, and can be stably operated in the harsh space environment for a long time, it faces difficulties in high-performance computing, radiation-resistant reinforcement, low-power management, efficient thermal control, and system reliability. SUMMARY

[0008] In view of the above problems, the present application is proposed to provide a spaceborne intelligent computing device for ecological environment industry application to overcome the above problems or at least partially solve the above problems. The technical solution is as follows: The present application provides a spaceborne intelligent computing device for ecological environment industry application, comprising a hardware acceleration layer, an adaptive power management layer and an ecological environment algorithm application layer connected in sequence, constituting a three-layer architecture of heterogeneous computing, adaptive power management and algorithm hardening, wherein: The hardware acceleration layer comprises a radiation-resistant basic platform, a reconfigurable FPGA, a multi-core CPU, an artificial intelligence AI acceleration core and a neuromorphic coprocessor, constituting a radiation-hardened hybrid computing array, for providing a variety of computing power; The adaptive power management layer comprises a dynamic voltage and frequency adjustment DVFS controller, a power gating network and a task-aware scheduler, for dynamically regulating the voltage, frequency and power supply state of each computing unit in the hardware acceleration layer according to the computing task, to realize intelligent power management; The ecological environment algorithm application layer is a software-defined layer, comprising a task scheduling engine and a plurality of ecological monitoring algorithm modules scheduled by the task scheduling engine; the task scheduling engine is used to intelligently distribute the computing tasks in the plurality of ecological monitoring algorithm modules to the matching computing units in the hardware acceleration layer according to the on-orbit task requirements.

[0009] In one possible implementation, the hardware acceleration layer adopts a master-slave-monitor heterogeneous architecture, comprising: The master processing module integrates an AI acceleration core, a multi-core CPU and a neuromorphic coprocessor, for providing AI inference and general-purpose computing; and in order to adapt to space applications, the master processing module is packaged and shielded based on a radiation-resistant basic platform; The collaborative processing module integrates a reconfigurable FPGA, for receiving raw data streams from a spectral imager and executing preprocessing algorithms, while serving as a health monitor of the master processing module; A monitoring module running a real-time operating system is responsible for power-on, reset and mode switching of the entire satellite intelligent computing device, executes watchdog timer and system heartbeat monitoring, and executes redundancy switching instructions when serious faults are detected.

[0010] In a possible implementation, a task-aware scheduler is configured to generate power consumption control instructions according to satellite task planning and current imaging area characteristics. A DVFS controller is configured to dynamically adjust the operating frequency and core voltage of the AI acceleration core and / or the multi-core CPU in the main processing module according to the power consumption control instructions. A power gating network is configured to perform fine-grained power supply on-off control on each computing unit in the hardware acceleration layer.

[0011] In a possible implementation, the adaptive power management layer is configured to support at least three preset power consumption modes, and the task-aware scheduler dynamically switches modes according to the satellite flight area and task planning, the preset power consumption modes including: A regular cruise mode, when the satellite flies over a non-key monitoring area, the AI acceleration core operates in a low-frequency state and performs light abnormality detection; A standard monitoring mode, when the satellite flies over land or a regular monitoring area, the AI acceleration core operates at a standard frequency and performs a complete set of cloud detection, land classification and vegetation index calculation; A fine analysis mode, when a specific target is detected, the AI acceleration core operates at a short-time overclocking frequency within a thermal management margin and executes a physical inversion or water component analysis model.

[0012] In a possible implementation, the ecological monitoring algorithms corresponding to the plurality of ecological monitoring algorithm modules in the ecological environment algorithm application layer are deployed in the form of hardware IP cores on different computing units in the hardware acceleration layer.

[0013] In a possible implementation, the plurality of ecological monitoring algorithm modules include a pixel-level index calculation module, a land classification coverage calculation module, a physical inversion calculation module and a deep learning calculation module.

[0014] In a possible implementation, the pixel-level index calculation algorithm corresponding to the pixel-level index calculation module is fixed and deployed in a reconfigurable FPGA to implement fixed-point pipeline calculation.

[0015] In a possible implementation, the land classification coverage calculation algorithm corresponding to the land classification coverage calculation module is fixed and deployed in a multi-core CPU to implement parallel processing of feature extraction and classification.

[0016] In a possible implementation, the pre-computed results of the physical inversion calculation module are stored in the anti-radiation non-volatile memory as a lookup table, and the acceleration query and inversion are performed by a special hardware lookup module.

[0017] In a possible implementation, the deep learning algorithm corresponding to the deep learning calculation module is deployed in the AI acceleration core and the neuromorphic coprocessor, wherein the neuromorphic coprocessor is configured with a special convolution engine and an attention mechanism engine.

[0018] By means of the above technical solutions, the spaceborne intelligent computing device for ecological environment industry application provided by the present application can realize deep coupling design of ecological environment application algorithms and spaceborne hardware platforms, has high computing power, low power consumption, and radiation resistance, utilizes power consumption perception dynamic reconfiguration, predicts algorithm load through a task scheduling engine, and adjusts computing resource allocation strategies in advance, so as to realize energy efficiency optimization, meet complex space environment, and stably perform related algorithm inference calculation on the satellite.

[0019] Moreover, the spaceborne intelligent computing device solidifies ecological monitoring algorithms into a hardware acceleration layer, can flexibly realize water bloom identification, chlorophyll concentration inversion, vegetation change, and other applications on the satellite in orbit, and rapidly downloads and distributes generated products to corresponding users for use. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the description of the embodiments of the present application will be briefly introduced.

[0021] Figure 1 An architecture diagram of the spaceborne intelligent computing device for ecological environment industry application provided by the embodiments of the present application is shown; Figure 2 Another architecture diagram of the spaceborne intelligent computing device for ecological environment industry application provided by the embodiments of the present application is shown. DETAILED DESCRIPTION

[0022] Exemplary embodiments of the present application will be described in detail below with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided so that the present application can be more thoroughly understood and the scope of the present application can be fully conveyed to those skilled in the art.

[0023] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and in the above drawings are used to distinguish similar objects, and do not necessarily have to describe a specific order or sequence. It should be understood that such use can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the term "comprising" and its variants are to be interpreted as meaning "including but not limited to" an open term.

[0024] To solve the above technical problems, the embodiments of the present application provide a spaceborne intelligent computing device for ecological environment industry application, which is designed as a core data processing payload of an imaging satellite running in space, as shown in Figure 1 The spaceborne intelligent computing device includes a hardware acceleration layer, an adaptive power management layer and an ecological environment algorithm application layer connected in sequence, forming a three-layer architecture of heterogeneous computing, adaptive power consumption management and algorithm hardening, wherein: The hardware acceleration layer includes a radiation-hardened basic platform, a reconfigurable FPGA, a multi-core CPU, an artificial intelligence AI acceleration core and a neuromorphic coprocessor, forming a radiation-hardened hybrid computing array for providing various computing power; here, FPGA stands for Field Programmable Gate Array, which is a programmable chip at the hardware level; CPU stands for Central Processing Unit, which is a central processing unit; The adaptive power management layer includes a dynamic voltage and frequency scaling DVFS controller, a power gating network and a task-aware scheduler, which is used to dynamically control the voltage, frequency and power supply state of each computing unit in the hardware acceleration layer according to the computing task, and realizes intelligent power consumption management; here, DVFS stands for Dynamic Voltage and Frequency Scaling, which is dynamic voltage and frequency scaling; The ecological environment algorithm application layer is a software-defined layer, including a task scheduling engine and a plurality of ecological monitoring algorithm modules scheduled by the task scheduling engine; the task scheduling engine is used to intelligently distribute the computing tasks in the plurality of ecological monitoring algorithm modules to the matching computing units in the hardware acceleration layer according to the on-orbit task requirements.

[0025] The spaceborne intelligent computing device for ecological environment industry application provided by the embodiments can realize deep coupling design of ecological environment application algorithm and spaceborne hardware platform, has high computing power, low power consumption and radiation resistance, utilizes power consumption-aware dynamic reconfiguration, predicts algorithm load through a task scheduling engine, adjusts computing resource allocation strategy in advance, realizes energy efficiency optimization, and can meet complex space environment to stably perform related algorithm inference calculation on the satellite.

[0026] In an embodiment of the present application, a possible implementation is provided, and the hardware acceleration layer can adopt a master-slave-monitor heterogeneous architecture, which includes: The master processing module integrates an AI acceleration core, a multi-core CPU and a neuromorphic coprocessor, and is used to provide AI inference and general computing; and in order to adapt to space applications, the master processing module is packaged and shielded based on an anti-radiation basic platform; The cooperative processing module integrates a reconfigurable FPGA, and is used to receive raw data streams from a spectral imager and execute pre-processing algorithms, and simultaneously acts as a health monitor of the master processing module; The monitoring module runs a real-time operating system, is responsible for power-on, reset and mode switching of the entire on-board intelligent computing device, executes a watchdog timer and system heartbeat monitoring, and executes a redundancy switching instruction when a serious fault is detected.

[0027] In the embodiment, the master-slave-monitor heterogeneous architecture is adopted, the master processing module focuses on high-performance computing, the cooperative processing module is responsible for high-throughput preprocessing, and the monitoring module ensures the absolute reliability of the operating system at the basic control level. Through division of labor, cooperation and monitoring, an intelligent computing system with high performance, high reliability and high autonomy for harsh environments such as space is constructed.

[0028] In an embodiment of the present application, a possible implementation is provided, and a task-aware scheduler is used to generate power consumption control instructions according to satellite task planning and current imaging area characteristics; The DVFS controller is used to dynamically adjust the operating frequency and core voltage of the AI acceleration core and / or the multi-core CPU in the master processing module according to the power consumption control instructions; The power gate network is used to perform fine-grained power supply on-off control on each computing unit in the hardware acceleration layer.

[0029] In the embodiment, the task-aware scheduler predicts the computing power demand according to the satellite task planning and the current imaging area characteristics, generates power consumption control instructions, avoids idle or overload of computing resources, and formulates an optimal energy efficiency strategy from the system level; and the DVFS controller dynamically adjusts the operating frequency and core voltage of the AI acceleration core and / or the multi-core CPU in the master processing module according to the power consumption control instructions, greatly reduces the dynamic power consumption of the chip during task execution, and realizes on-demand power supply. In addition, the power gate network performs fine-grained power supply on-off control on each computing unit in the hardware acceleration layer, eliminates the static power consumption of idle units, realizes zero leakage, and is especially suitable for intermittent systems.

[0030] In an embodiment of the present application, a possible implementation is provided, the adaptive power management layer is configured to support at least three preset power consumption modes, and a task-aware scheduler dynamically switches modes according to a satellite flight area and a task plan, the preset power consumption modes include: A conventional cruise mode, when the satellite flies through a non-key monitoring area, the AI acceleration core operates in a low-frequency state and performs light abnormality detection; A standard monitoring mode, when the satellite flies through land or a conventional monitoring area, the AI acceleration core operates at a standard frequency and performs a complete set of cloud detection, ground object classification, and vegetation index calculation; A fine analysis mode, when a specific target is detected, the AI acceleration core operates at a short-time overclocking frequency within a thermal management margin and performs a physical inversion or water body composition analysis model.

[0031] In the embodiment, through the deep integration of task awareness and dynamic power consumption mode switching, the on-board AI computing resources are accurately and autonomously balanced between the province and the strength, thereby comprehensively improving the on-orbit efficiency of the satellite.

[0032] In an embodiment of the present application, a possible implementation is provided, the ecological environment algorithm application layer includes a plurality of ecological monitoring algorithm modules corresponding to ecological monitoring algorithms, and the ecological monitoring algorithms are deployed in different computing units of the hardware acceleration layer in the form of hardware IP cores. Here, IP stands for Intellectual Property.

[0033] The hardware IP core of the embodiment is solidified for a specific algorithm, can realize highly parallel computing, avoids the instruction scheduling overhead of a general-purpose processor, and can realize order-of-magnitude improved throughput and significantly reduced unit computing power consumption when processing massive data such as remote sensing images and video streams.

[0034] In an embodiment of the present application, a possible implementation is provided, the plurality of ecological monitoring algorithm modules include a pixel-level index calculation module, a land classification and coverage calculation module, a physical inversion calculation module, and a deep learning calculation module.

[0035] In this embodiment, the pixel-level index calculation module performs fast calculation on each pixel point to generate basic quantitative information such as vegetation index and water body index, and realizes basic information extraction; the land classification and coverage calculation module identifies and divides different land types (such as forest, farmland, water area, and construction land) based on pixel index to generate a land use / coverage map, and realizes feature recognition and mapping; the physical inversion calculation module inverses key ecological environment parameters (such as leaf area index, evapotranspiration, and water turbidity) that are difficult to directly measure from surface observation data based on physical laws (such as energy balance and radiation transfer); and the deep learning calculation module processes high-dimensional and complex nonlinear relationships to realize more intelligent tasks such as precise identification of specific targets, change detection, and prediction of future ecological trends.

[0036] In this embodiment, a possible implementation is provided, and a pixel-level index calculation algorithm corresponding to the pixel-level index calculation module is solidified and deployed in a reconfigurable FPGA to realize fixed-point pipeline calculation. In this embodiment, a large number of pixel points can be processed simultaneously by using the hardware parallel computing capability of the FPGA, thereby avoiding the bottleneck of sequential execution of a traditional CPU; in combination with pipeline design, data is continuously processed by various processing units like a pipeline, thereby realizing ultra-high data throughput and meeting the real-time processing requirements of satellites for massive remote sensing data.

[0037] In this embodiment, a possible implementation is provided, and a land classification and coverage calculation algorithm corresponding to the land classification and coverage calculation module is solidified and deployed in a multi-core CPU to realize parallel processing of feature extraction and classification. In this embodiment, the parallel computing capability of the multi-core CPU can be used to decompose the feature extraction and classification tasks into multiple subtasks, which are processed simultaneously by multiple computing cores; for example, feature calculation can be performed on different blocks of an image simultaneously, or classification tasks of multiple different regions can be processed in parallel, thereby greatly shortening the time from imaging to outputting land classification results, and meeting the requirements of satellites for fast analysis of massive remote sensing data.

[0038] In this embodiment, a possible implementation is provided, and the precalculation result of the physical inversion calculation module is stored in a radiation-resistant nonvolatile memory as a lookup table, and is accelerated and queried by a special hardware lookup module. In this embodiment, complex calculation is completed in advance, and only one or several memory accesses are required during inversion, thereby realizing fast response at the microsecond or even nanosecond level; in addition, the calculation process is simplified to low-power memory access, which greatly reduces dynamic power consumption and is particularly suitable for space tasks with limited energy; the radiation-resistant nonvolatile memory can permanently maintain data integrity in extreme environments, ensures the safety and reliability of the lookup table, and provides a certain and error-free calculation basis for the system.

[0039] In the embodiments of the present application, a possible implementation is provided, a deep learning algorithm corresponding to a deep learning computing module, and an optimized algorithm model is deployed in an AI acceleration core and a neuromorphic coprocessor, wherein the neuromorphic coprocessor is configured with a special convolution engine and an attention mechanism engine. In the embodiments, the AI acceleration core provides stable and powerful peak computing power, ensuring fast and high-precision processing of regular data; the neuromorphic coprocessor realizes continuous perception with ultra-low power consumption and efficient time series analysis, and is particularly suitable for processing streaming data.

[0040] The above introduces Figure 1 Various implementation manners of each link of the embodiments shown in the above are described below. The satellite-borne intelligent computing device for ecological environment industry application in the embodiments of the present application will be further described through specific embodiments.

[0041] In specific embodiments, as Figure 2 shown, the satellite-borne intelligent computing device for ecological environment industry application can include a hardware acceleration layer, an adaptive power management layer and an ecological environment algorithm application layer connected in sequence, forming a three-layer architecture of heterogeneous computing, adaptive power consumption management and algorithm hardening, wherein: The hardware acceleration layer includes a radiation-resistant basic platform, a reconfigurable FPGA, a multi-core RISC-V CPU, an AI acceleration core and a neuromorphic coprocessor, forming a radiation-hardened hybrid computing array, for providing various computing power; here, RISC-V stands for Reduced Instruction Set Computing-V, which is the fifth generation of a free and open processor instruction set architecture; the AI acceleration core has a >275TOPS, where TOPS stands for Tera Operations Per Second, which is a unit of processor computing power, and 1TOPS represents one trillion basic operation operations per second of the processor; The adaptive power management layer includes a DVFS controller, a power gate network and a task-aware scheduler, for dynamically regulating the voltage, frequency and power supply state of each computing unit in the hardware acceleration layer according to the computing task, to realize intelligent power consumption management; The ecological environment algorithm application layer is a software-defined layer, including a task scheduling engine and a plurality of ecological monitoring algorithm modules scheduled by the task scheduling engine; the task scheduling engine is used to intelligently distribute the computing tasks in the plurality of ecological monitoring algorithm modules to the matching computing units in the hardware acceleration layer according to the on-orbit task requirements.

[0042] 1) Hardware acceleration layer architecture and implementation, the hardware acceleration layer adopts a master-slave-monitor heterogeneous architecture.

[0043] (1) Main processing module (including AI acceleration core, multi-core CPU and neuromorphic coprocessor): industrial-grade chips are selected to provide high-power AI inference performance. To adapt to space applications, the main processing module is packaged and shielded. The core chips of the original module are re-packaged in an airtight ceramic package, and a lightweight tantalum alloy local shielding box is designed for the entire module to protect the GPU (Graphic Processing Unit) and DDR (Double Data Rate) memory and other radiation-sensitive areas. In addition, a dedicated anti-radiation ruggedized power module is designed for the main processing module to provide stable multi-channel low-voltage and high-current power supply.

[0044] (2) Cooperative processing module: This module is based on a single board design and integrates FPGA, which has programmable logic, DSP (Digital Signal Processing) units and AI engines. It is responsible for receiving raw data streams from the spectral imager (through a high-speed serial interface). It performs on-orbit radiation correction, geometric correction and other preprocessing algorithms. As a health monitor for the main processing module, it monitors its running status through a dedicated interface (such as PCIe). It implements dynamic reconfiguration logic for fault recovery. PCIe stands for Peripheral Component Interconnect Express, which is a high-speed computer expansion bus standard.

[0045] (3) Monitoring module: As the highest authority system brain, it runs a high-reliability real-time operating system, which is responsible for the power-on, reset and mode switching of the entire machine, and executes watchdog timers and system heartbeat monitoring. When a serious fault is detected, it executes redundancy switching instructions, such as disconnecting the power supply of the main processing module and enabling the backup module.

[0046] 2) Task-aware adaptive ultra-low power management implementation.

[0047] To control the total power consumption of each module within a certain range, a fine-grained power management strategy needs to be implemented. At the hardware level, high-efficiency gallium nitride DC-DC converters are used, and fine-grained power gating and clock gating are implemented for AI chips, FPGAs and memory, etc. Here, DC stands for Direct Current.

[0048] At the driver level, the self-reconfigurable hybrid architecture idea is adopted, and the dynamic voltage and frequency adjustment technology based on task load prediction is introduced. The monitoring unit dynamically adjusts the operating frequency and core voltage of the main AI processing unit according to the complexity of the current ecological monitoring task and the characteristics of the sensor data stream, and implements the preset power consumption mode according to the satellite task planning and the current imaging area.

[0049] (1) Normal cruise mode (low power consumption): When the satellite is flying over non-priority monitoring areas (such as oceans), the main AI processor is in low-frequency mode, running only a lightweight anomaly detection algorithm.

[0050] (2) Standard monitoring mode (steady power consumption): When the satellite is flying over land or regular monitoring areas, the main AI processor runs at a standard frequency, performing a full set of cloud detection, ground feature classification, and vegetation index calculation.

[0051] (3) Detailed analysis mode (short-time high power consumption): When a specific target (such as a suspected water bloom in a lake) is detected, the main AI processor is overclocked to the highest performance, running high-precision physical inversion or water component analysis models. This mode is limited by the buffering capacity of the thermal management system and phase change material, and the duration does not exceed a few minutes.

[0052] 3) Realization of deployable modular structure and multi-level composite thermal management.

[0053] In order to realize the heat dissipation of the equipment in the vacuum environment, by adopting the improved high-density modular design, the computing, storage, power supply, interface and other functional units are designed as standardized, independently replaceable and upgradeable modules. Inverted soldering ceramic packaging technology is adopted for key heat generating chips (such as AI processors), combined with high-performance thermal interface material, to maximize the thermal resistance from the chip core to the package shell. In the PCB board inside or back of each computing module, micro flat heat pipes are embedded to quickly and evenly conduct the concentrated heat generated by the chip to the entire module metal frame and edge. At the system level, the guide rails of the modular case are in close contact with the module edges, and the heat is collected to the case bottom plate through efficient heat conduction. The case bottom plate integrates the evaporation end of the loop heat pipe, and the loop heat pipe efficiently and long-distance transfers the heat to the external radiator panel of the satellite platform for heat dissipation. In addition, phase change materials are integrated inside the case to absorb and buffer the transient peak heat generated by the AI processor when running at full load, ensuring the stability of the core temperature.

[0054] 4) Realization of ecological environment algorithm hardware.

[0055] In order to realize the efficient application of ecological environment algorithm on the intelligent computing device on satellite, the core ecological algorithm is hardened as IP core to form a reconfigurable ecological environment algorithm library, which solves the efficiency loss problem caused by the separation of algorithm and hardware of on-board computer. Pixel-level exponential hardware acceleration engine design is adopted, and parallel multi-channel spectral data processing pipeline is used to realize pixel-level fixed-point pipeline in FPGA. For commonly used land classification algorithms, the algorithm logic is fixed in CPU to realize parallel processing of feature extraction and classification. Based on the physical inversion remote sensing monitoring demand, a look-up table (LUT, Look Up Table) hardware acceleration module is constructed, and the pre-computed results of radiation transfer model are stored in radiation-resistant MRAM to realize parallel inversion of related applications. Based on the demand of deep learning computing module, a special convolution engine and attention mechanism engine are built on the neuromorphic coprocessor, and storage optimization is realized by using loop blocking and cache hierarchy. Model lightweight is realized by pruning, quantization and other methods at the model level.

[0056] Those skilled in the art can clearly understand the specific working process of the above-described system, device and module, and can refer to the corresponding process in the foregoing method embodiments. For brevity, no further description is given here.

[0057] Those skilled in the art can understand that the technical solutions of the present application can be embodied in the form of a software product in essence or in whole or part of the technical solutions, which is stored in a storage medium and includes a plurality of program instructions for causing an electronic device (such as a personal computer, a server, or a network device) to execute all or part of the steps of the method described in the embodiments of the present application when the program instructions are executed. The storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various storage media that can store program codes.

[0058] Alternatively, all or part of the steps of the foregoing method embodiments can be completed by program instruction related hardware (such as an electronic device of a personal computer, a server, or a network device), which can be stored in a computer readable storage medium. When the program instructions are executed by the processor of the electronic device, the electronic device executes all or part of the steps of the method described in the embodiments of the present application.

[0059] The above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that within the spirit and principles of the present application, the technical solutions described in the foregoing embodiments can still be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not cause the corresponding technical solutions to deviate from the protection scope of the present application.

Claims

1. A spaceborne intelligent computing device for applications in the ecological and environmental industry, characterized in that, It comprises a hardware acceleration layer, an adaptive power management layer, and an ecosystem algorithm application layer, connected in sequence, forming a three-layer architecture of heterogeneous computing, adaptive power management, and algorithm hardening, wherein: The hardware acceleration layer, including a radiation-hardened base platform, a reconfigurable FPGA, a multi-core CPU, an AI acceleration core, and a neuromorphic coprocessor, constitutes a radiation-hardened hybrid computing array to provide various computing powers. The adaptive power management layer, including the Dynamic Voltage and Frequency Adjustment (DVFS) controller, power gating network, and task-aware scheduler, is used to dynamically adjust the voltage, frequency, and power supply status of each computing unit in the hardware acceleration layer according to the computing task, thereby achieving intelligent power consumption management. The ecological environment algorithm application layer is a software-defined layer, which includes a task scheduling engine and multiple ecological monitoring algorithm modules scheduled by the task scheduling engine. The task scheduling engine is used to intelligently allocate the computing tasks in the multiple ecological monitoring algorithm modules to the matching computing units in the hardware acceleration layer for execution according to the needs of the on-orbit mission.

2. The spaceborne intelligent computing device for ecological and environmental industry applications according to claim 1, characterized in that, The hardware acceleration layer adopts a master-slave-monitor heterogeneous architecture, including: The main processing module integrates an AI acceleration core, a multi-core CPU, and a neuromorphic coprocessor to provide AI inference and general computing. Furthermore, to adapt to space applications, the main processing module is encapsulated and shielded based on a radiation-resistant platform. The collaborative processing module, which integrates a reconfigurable FPGA, is used to receive the raw data stream from the spectral imager and execute preprocessing algorithms, while also serving as a health monitor for the main processing module. The monitoring module runs a real-time operating system and is responsible for powering on, resetting, and switching modes of the entire onboard intelligent computing device. It executes watchdog timers and system heartbeat monitoring, and executes redundancy switching instructions when a serious fault is detected.

3. The spaceborne intelligent computing device for ecological and environmental industry applications according to claim 2, characterized in that, The mission-aware scheduler is used to generate power control commands based on satellite mission planning and current imaging area characteristics; The DVFS controller is used to dynamically adjust the operating frequency and core voltage of the AI ​​acceleration core and / or multi-core CPU in the main processing module according to power consumption control instructions. Power gating networks are used to perform fine-grained power on / off control on each computing unit in the hardware acceleration layer.

4. The spaceborne intelligent computing device for ecological and environmental industry applications according to claim 3, characterized in that, The adaptive power management layer is configured to support at least three preset power consumption modes. The mission-aware scheduler dynamically switches modes based on the satellite flight area and mission planning. The preset power consumption modes include: In normal cruise mode, when the satellite flies over non-key monitoring areas, the AI ​​acceleration core operates at a low frequency to perform lightweight anomaly detection; In standard monitoring mode, when the satellite flies over land or a regular monitoring area, the AI ​​acceleration core operates at a standard frequency, performing a full set of cloud detection, land cover classification, and vegetation index calculations. In the fine analysis mode, when a specific target is detected, the AI ​​acceleration core operates at a short-term overclock within the thermal management margin to execute physical inversion or water composition analysis models.

5. The spaceborne intelligent computing device for ecological and environmental industry applications according to claim 1, characterized in that, The ecological monitoring algorithms corresponding to multiple ecological monitoring algorithm modules in the ecological environment algorithm application layer are deployed in the form of hardware IP cores in different computing units of the hardware acceleration layer.

6. The spaceborne intelligent computing device for ecological and environmental industry applications according to claim 5, characterized in that, Multiple ecological monitoring algorithm modules include a pixel-level index calculation module, a land classification and coverage calculation module, a physical inversion calculation module, and a deep learning calculation module.

7. The spaceborne intelligent computing device for ecological and environmental industry applications according to claim 6, characterized in that, The pixel-level exponent calculation algorithm corresponding to the pixel-level exponent calculation module is embedded and deployed in a reconfigurable FPGA to realize fixed-point pipelined calculation.

8. The spaceborne intelligent computing device for ecological and environmental industry applications according to claim 6, characterized in that, The land classification and coverage calculation algorithm corresponding to the land classification and coverage calculation module is fixed and deployed on a multi-core CPU to achieve parallel processing of feature extraction and classification.

9. The spaceborne intelligent computing device for ecological and environmental industry applications according to claim 6, characterized in that, The pre-calculation results of the physical inversion calculation module are stored as a lookup table in a radiation-resistant non-volatile memory, and the lookup and inversion are accelerated by a dedicated hardware lookup module.

10. The spaceborne intelligent computing device for ecological and environmental industry applications according to claim 6, characterized in that, The deep learning computing module contains deep learning algorithms, and the optimized algorithm models are deployed in the AI ​​acceleration core and the neuromorphic coprocessor. The neuromorphic coprocessor is equipped with a dedicated convolution engine and an attention mechanism engine.

Images