CN115561790A - Satellite-ground cloud edge collaborative ground target identification method and device based on deep network - Google Patents
Satellite-ground cloud edge collaborative ground target identification method and device based on deep network Download PDFInfo
- Publication number
- CN115561790A CN115561790A CN202211295375.6A CN202211295375A CN115561790A CN 115561790 A CN115561790 A CN 115561790A CN 202211295375 A CN202211295375 A CN 202211295375A CN 115561790 A CN115561790 A CN 115561790A
- Authority
- CN
- China
- Prior art keywords
- target
- satellite
- ground
- model
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S19/00—Satellite radio beacon positioning systems; Determining position, velocity or attitude using signals transmitted by such systems
- G01S19/01—Satellite radio beacon positioning systems transmitting time-stamped messages, e.g. GPS [Global Positioning System], GLONASS [Global Orbiting Navigation Satellite System] or GALILEO
- G01S19/13—Receivers
- G01S19/35—Constructional details or hardware or software details of the signal processing chain
- G01S19/37—Hardware or software details of the signal processing chain
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S19/00—Satellite radio beacon positioning systems; Determining position, velocity or attitude using signals transmitted by such systems
- G01S19/01—Satellite radio beacon positioning systems transmitting time-stamped messages, e.g. GPS [Global Positioning System], GLONASS [Global Orbiting Navigation Satellite System] or GALILEO
- G01S19/13—Receivers
- G01S19/24—Acquisition or tracking or demodulation of signals transmitted by the system
Landscapes
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a satellite-ground cloud edge collaborative ground target identification method and device based on a deep network. The device comprises a ground cloud service node and a satellite side node, wherein the satellite side node comprises a calculation control module, an on-satellite communication module, a satellite data acquisition module, a data preprocessing module, a data labeling module and an on-satellite identification processing module. The method comprises the following steps: firstly, constructing a task support environment for identifying a ground target, loading an existing target recognition task model, and executing a recognition task if a current model matches the requirement of the recognition task; if not, collecting target data of various identification tasks as standard training samples; then, a target recognition model is built, and a standard training sample is adopted to train the model; then sending the trained target recognition model to the satellite edge node; and finally, the satellite edge nodes execute the target identification task and return the target identification result to the ground cloud service node. The invention improves the timeliness and the accuracy of the satellite to ground target identification.
Description
Technical Field
The invention belongs to the technical field of satellite intelligent application, and particularly relates to a satellite-ground cloud edge collaborative ground target identification method and device based on a deep network.
Background
With the development of artificial intelligence, computing power moves from the center cloud to the edge and the terminal, and the coordination of cloud computing power, edge computing power and terminal computing power is more and more important. The cooperative application of cloud edges to spatial information systems is a great trend. The cloud edge cooperation develops towards the cloud edge-side three-level computing power demand-following scheduling direction, the cooperation aims to improve the computing power resource utilization rate, the network is switched from cloud network integration to computing network integration, and a credible and efficient demand-following network is provided for computing services.
The existing satellite earth target identification mostly belongs to a storage and return ground type, and due to the limitation of the satellite operation orbit, the image information shot outside the country or outside the measurement and control station is only transmitted to the ground station in the transit period and then is processed by the ground station. The ground station has perfect computing equipment and rich target model base, can quickly and comprehensively identify the target, but has the biggest defect that the time effectiveness of the target is poor, only fixed targets can be identified well, and the timeliness of the information of the moving target is poor; meanwhile, the original identification method is limited in identification capability, and the target identification accuracy and rate can be improved by using the deep network.
Disclosure of Invention
The invention aims to provide a satellite-ground cloud edge collaborative ground target identification method and device based on a deep network, which are high in timeliness and accuracy.
The technical solution for realizing the purpose of the invention is as follows: a satellite-ground cloud edge collaborative ground target identification method based on a deep network comprises the following steps:
step 1, constructing a satellite-ground cloud edge collaborative ground target identification task support environment, and loading an existing target identification task model;
step 2, the ground cloud service node confirms the recognition task model according to the requirement of the recognition task, and if the current recognition task model is matched, the step 7 is carried out to execute the recognition task; if the current recognition task model cannot meet the requirements of the recognition task, entering step 3;
step 3, collecting target data of various identification tasks, wherein the identification tasks comprise ground large-scale equipment identification tasks and marine large-scale ship identification tasks, storing the identification tasks into a target database, preprocessing and marking the target data, and generating standard training samples;
step 4, building a target recognition model according to the on-satellite resource constraints, and configuring training parameters of the target recognition model;
step 5, training a target recognition model by the ground cloud service node, and testing the prediction accuracy of the trained target recognition model;
step 6, the ground cloud service node sends the trained target recognition model to the satellite edge node through the satellite-ground communication link;
7, the satellite edge nodes execute target identification tasks and return target identification results and target real-time states to the ground cloud service nodes;
and 8, controlling the satellite edge nodes to collect target area data by the ground cloud service nodes, and updating the task database.
A satellite-ground cloud edge collaborative ground target identification device based on a deep network comprises a ground cloud service node and a satellite edge node, wherein the satellite edge node comprises a calculation control module, an on-satellite communication module, a satellite data acquisition module, a data preprocessing module, a data labeling module and an on-satellite identification processing module;
the ground cloud service node adopts a high-performance server cluster and is used for business process control, data receiving and transmitting control, model training optimization and model database support;
the computing control module adopts a heterogeneous computing platform based on a CPU + GPU and is used for satellite-borne service flow control, data transceiving control and image processing control;
the satellite communication module is used for receiving communication signals from other satellites, demodulating and unframing the signals and transmitting the signals to the calculation control module; meanwhile, data from the calculation control module is received, framing and modulation of signals are achieved, and then the signals are sent to a ground cloud service node or other satellite side nodes;
the satellite data acquisition module is used for acquiring a satellite image to be identified;
the data preprocessing module is used for preprocessing the satellite image to be identified to generate a standard image for labeling;
the data marking module is used for marking the target content in the preprocessed satellite image for training of the model training module;
and the on-satellite identification processing module is used for inputting the preprocessed standard satellite image into the uploaded target identification model and outputting an identification result.
Compared with the prior art, the invention has the following remarkable advantages: (1) The ground target identification is carried out based on satellite-ground cloud edge cooperation, and data of the target are acquired in real time through satellite edge nodes, so that the timeliness of target identification is improved; (2) The target identification and acquisition tasks of the target area are cooperatively performed through the ground cloud service node and the satellite edge node, so that the accuracy of target identification is improved.
Drawings
Fig. 1 is a schematic flow chart of a satellite-ground cloud edge collaborative ground-target identification method based on a deep network according to the present invention.
FIG. 2 is a schematic flow chart of the process of constructing the object recognition model according to the present invention.
Fig. 3 is a block diagram of a structure of a satellite-ground cloud edge collaborative ground-target identification device based on a deep network.
Detailed Description
The invention is described in further detail below with reference to the following figures and specific examples.
With reference to fig. 1, a satellite-ground cloud edge collaborative ground-target identification method based on a deep network includes the following steps:
step 1, constructing a satellite-ground cloud edge collaborative ground target identification task support environment, and loading an existing target identification task model;
step 2, the ground cloud service node confirms the recognition task model according to the requirement of the recognition task, and if the current recognition task model is matched, the step 7 is carried out to execute the recognition task; if the current recognition task model cannot meet the requirements of the recognition task, entering step 3;
step 3, collecting target data of various identification tasks, wherein the identification tasks comprise ground large-scale equipment identification tasks and marine large-scale ship identification tasks, storing the identification tasks into a target database, preprocessing and marking the target data, and generating standard training samples;
step 4, building a target recognition model according to the on-satellite resource constraints, and configuring training parameters of the target recognition model;
step 5, training a target recognition model by the ground cloud service node, and testing the prediction accuracy of the trained target recognition model;
step 6, the ground cloud service node sends the trained target recognition model to the satellite edge node through the satellite-ground communication link;
step 7, the satellite edge nodes execute the target identification task and return the target identification result and the target real-time state to the ground cloud service node;
and 8, controlling the satellite edge nodes to collect target area data by the ground cloud service nodes, and updating the task database.
As a specific implementation manner, the constructing of the satellite-ground cloud edge collaborative ground target identification task support environment in step 1 specifically includes the following steps:
step 1.1, constructing a satellite-ground cloud edge collaborative ground target identification task support environment, including constructing a task operation environment and constructing a cloud edge collaborative communication link;
step 1.2, constructing a task running environment, and deploying the task running environment in a ground cloud service node and a satellite side node;
the task operation environment of the ground cloud service node is composed of a server cluster, is used for performing business process control, data transceiving control, model training optimization and model database support, and is provided with a task model library and a target database, wherein the task model library is used for managing the existing target identification model which can be directly used, the target database is used for storing various target data, including target data of ground large equipment and target data of marine large ships, and the stored target data is used for retraining or adjusting the target identification model at a later stage;
the task operation environment of the satellite edge node is constructed based on a software virtualization technology and is used for performing cross-platform deployment and migration of a task system, on-satellite task resource management, control model transceiving, task starting and stopping, message simulation, and task execution effect and observation parameters presentation;
step 1.3, a cloud edge cooperative communication link is constructed, wherein the cloud edge cooperative communication link comprises an uplink transmitted from a ground cloud service node to a satellite edge node, a downlink transmitted from the satellite edge node to the ground cloud service node and a parallel link transmitted between the satellite edge nodes, and is used for satellite-ground control instruction transmission, identification model transmission, target data transmission and state information transmission, and relay transmission of the ground cloud service node and a non-direct-connection satellite edge node.
As a specific implementation manner, the ground cloud service node in step 2 confirms an identification task model according to the requirement of an identification task, and if the current identification task model is matched, the step 7 is performed to execute the identification task; if the current recognition task model fails to meet the requirements of the recognition task, the method proceeds to step 3, specifically as follows:
after receiving the target identification task, the ground cloud service node extracts an identification target identifier, searches the identifier in a cloud service task model library, stores a target identification model which can be directly used, and can execute the target identification task after being annotated with satellite edge nodes. In the searching process, if the target model matched with the identifier is successfully searched, the satellite side nodes of the corresponding area are annotated on the model, and the side nodes are controlled to execute an identification task to obtain an identification result; if the target model matching the identifier cannot be retrieved, model construction is performed.
As a specific implementation manner, the step 3 collects target data of various types of recognition tasks, stores the target data in a target database, preprocesses and labels the target data, and generates a standard training sample, which specifically includes the following steps, with reference to fig. 2:
step 3.1, collecting target data of various identification tasks and storing the target data into a target database;
the following two ways are adopted to collect the target data of various identification tasks:
in the first mode, data of a target to be identified is collected through channels of public network information collection, professional organization service and historical data accumulation of the system and stored in a target database of a ground cloud service node;
in the second mode, the ground cloud service node sends an instruction to a satellite side node in a nearby set range according to the code of a target area to be identified, collects data of the target to be identified in real time, and stores the data into a target database of the ground cloud service node after the data is collected;
3.2, the ground cloud service node acquires sensor parameters of the satellite edge node, determines a preprocessing index of target data to be recognized according to the sensor parameters, and performs preprocessing operation on the target data to be recognized, wherein the preprocessing index is as follows:
the method comprises the steps that a ground cloud service node obtains sensor parameters of a satellite edge node, the sensor parameters comprise optical sensor resolution, picture and sampling frequency, preprocessing indexes of target data are determined according to the sensor parameters, the preprocessing indexes comprise resolution and picture, normalization operation is conducted on the target data to be recognized, and the outstanding texture characteristics of pictures are sharpened.
And 3.3, sending the preprocessed target data to be recognized into a target data labeling link, performing frame selection and labeling on all targets in the target data to be recognized, storing the target data to be recognized into a target database as a model training sample after labeling is completed, and training a target recognition model in the next step.
As a specific implementation manner, step 4, building a target recognition model according to the on-satellite resource constraints, and configuring training parameters of the target recognition model, specifically as follows:
step 4.1, the ground cloud service node acquires software and hardware resource parameters of the satellite edge node, and constructs a target identification model which is suitable for hardware constraint of the satellite edge node and comprises a data preprocessing layer, an internal hiding layer and an output layer; the target identification model has the characteristics of small calculation amount, low frame rate requirement, small model scale and high precision;
the target identification model adopts a MobileNet SSD model as a basic model;
step 4.2, designing a network structure and parameters of the target identification model, wherein the network structure and parameters comprise model layer number design, internal hidden layer design, unit number design of each layer, network layer number iteration number design, optimizer design, evaluation function design, activation function selection and Dropout layer;
and 4.3, confirming the network updating frequency, the training iteration times and the evaluation function of the target recognition model according to the operation rate and the sample number of the ground cloud service nodes.
As a specific implementation manner, the ground cloud service node in step 5 trains a target recognition model, and tests the prediction accuracy of the trained target recognition model, specifically as follows:
step 5.1, training the target recognition model built in the step 4 by using the target training sample preprocessed in the step 3 to obtain a trained target recognition model;
and 5.2, testing the prediction accuracy of the trained target recognition model.
As a specific implementation manner, the satellite edge node in step 7 executes a target recognition task, and returns a target recognition result and a target real-time state to the ground cloud service node, which is specifically as follows:
establishing a connection between the ground cloud service node and a satellite side node, determining a satellite side node identifier for executing a target recognition task, transmitting a trained target recognition model to the corresponding satellite side node, executing the target recognition task by the satellite side node, dynamically adjusting sensor parameters based on different orbit heights, and returning a target recognition result and a target real-time state according to the requirements of the ground cloud service node; and the ground cloud service node changes the target recognition model requirement and completes the target recognition and acquisition tasks of the target area through the cloud edge cooperative communication link.
With reference to fig. 3, the satellite-to-ground cloud edge collaborative ground-to-target identification device based on the deep network comprises a ground cloud service node and a satellite edge node, wherein the satellite edge node comprises a calculation control module, an on-satellite communication module, a satellite data acquisition module, a data preprocessing module, a data labeling module and an on-satellite identification processing module;
the ground cloud service node adopts a high-performance server cluster and is used for business process control, data transceiving control, model training optimization and model database support;
the computing control module adopts a heterogeneous computing platform based on a CPU + GPU and is used for satellite-borne business process control, data receiving and sending control and image processing control;
the satellite communication module comprises an antenna, a radio frequency module, a baseband processing module and an information processing module, and is used for receiving communication signals from other satellites, demodulating and unframing the signals and transmitting the signals to the calculation control module; meanwhile, data from the computing control module is received, framing and modulation of signals are achieved, and then the signals are sent to a ground cloud service node or other satellite side nodes; the module and the calculation control module carry out data interaction through LVDS and RapidIO interfaces, and the type of interaction data comprises instructions, states, services, a target and images;
the satellite data acquisition module is used for acquiring a satellite image to be identified;
the data preprocessing module is used for preprocessing the satellite image to be identified and generating a standard image for marking;
the data marking module is used for marking the target content in the preprocessed satellite image for training of the model training module;
and the satellite recognition processing module is used for inputting the preprocessed standard satellite image into the uploaded target recognition model and outputting a recognition result.
Example 1
In this embodiment, optical images including a large transport vehicle, a large ship, and a large transport airplane are used as target data to be recognized, and the image data are stored in a target database in a ground cloud service node. And the ground cloud service node retrieves the satellite edge node sensor parameters and the calculation controller parameters of the task execution region, obtains the optical sensor resolution of 4000 x 3000, the model storage space capacity P and calculates the controller memory parameters. And then, according to the sensor resolution and the model storage space, executing preprocessing operation on the optical image data, normalizing all the image resolutions of the transport vehicle to 800 × 600, simultaneously performing gray processing on the target image, and cutting the frame into the sensor picture size. And finally, dividing the preprocessed image data into two groups A and B according to the proportion of 8.
In the embodiment, a MobileNet-based SSD model is adopted as a target identification basic model, the model has the characteristics of high identification speed, small model size and high calculation efficiency, and the model internally comprises a data preprocessing layer, an internal hiding layer and an output layer. The MobileNet is responsible for 28 layers of feature extraction, the SSD is responsible for sample classification, the preprocessing layer adopts a structure of 800 × 600 × 3, the internal hidden layer adopts a convolution network structure, the convolution kernel is 3 × 3, and Softmax is used as an elicitation function.
In this embodiment, after the target recognition basic model is imported, according to the obtained sensor data, the sensor data in this embodiment is optical sensor resolution 600 × 800, the computational power of the training calculation node is 2.4GHz, the training samples are trained in 3 classes by using a group of data in the target database, the training step length is 6 times, and 1000 rounds of training are performed.
In this embodiment, after the training of the target recognition model is completed, the target recognition model is run on the group B data, in this example, 98% is used as the qualified recognition rate, 4 groups of tests are performed, if the target recognition result accuracy reaches 98% after averaging, the model is considered to be qualified, and if the target recognition result accuracy cannot reach the requirement, the training is performed again. And when the model test is qualified, uploading the model to a satellite boundary point.
In the embodiment, in the process of executing the target identification task, firstly, a link is established between the ground cloud service node and the satellite side node executing the task, and other satellite side nodes can be used as relays. After uploading the target model to the satellite side node, the ground cloud service node sends a target identification instruction to the satellite side node, wherein the target identification instruction comprises target area sampling and returning, target area uninterrupted identification and the like. After receiving the instruction, the task satellite side node starts a virtualized computing environment, calls an identification sensor to execute a shooting task on a target area, and transmits shooting data back to a ground cloud server through a satellite link; and the ground cloud service node sends an identification task instruction to the satellite side node, and the satellite side node executes an identification task in real time and returns an identification image and identification result information.
Claims (10)
1. A satellite-ground cloud edge collaborative ground target identification method based on a deep network is characterized by comprising the following steps:
step 1, constructing a satellite-ground cloud edge collaborative earth target identification task support environment, and loading an existing target identification task model;
step 2, the ground cloud service node confirms an identification task model according to the requirement of the identification task, and if the current identification task model is matched, the ground cloud service node enters step 7 to execute the identification task; if the current recognition task model cannot meet the requirements of the recognition task, entering step 3;
step 3, collecting target data of various identification tasks, wherein the identification tasks comprise ground large-scale equipment identification tasks and marine large-scale ship identification tasks, storing the identification tasks into a target database, preprocessing and marking the target data, and generating a standard training sample;
step 4, building a target recognition model according to the satellite resource constraints, and configuring training parameters of the target recognition model;
step 5, training a target recognition model by the ground cloud service node, and testing the prediction accuracy of the trained target recognition model;
step 6, the ground cloud service node sends the trained target recognition model to the satellite edge node through the satellite-ground communication link;
step 7, the satellite edge nodes execute the target identification task and return the target identification result and the target real-time state to the ground cloud service node;
and 8, controlling the satellite edge nodes to collect target area data by the ground cloud service nodes, and updating the task database.
2. The method for identifying a satellite-ground cloud edge collaborative ground target identification based on a deep network according to claim 1, wherein the step 1 of constructing a satellite-ground cloud edge collaborative ground target identification task support environment specifically comprises the following steps:
step 1.1, constructing a satellite-ground cloud edge collaborative ground target identification task support environment, including constructing a task operation environment and constructing a cloud edge collaborative communication link;
step 1.2, constructing a task operation environment, and deploying the task operation environment in a ground cloud service node and a satellite side node;
the task running environment of the ground cloud service node consists of a server cluster, is used for performing business process control, data transceiving control, model training optimization and model database support, and is provided with a task model library and a target database, wherein the task model library is used for managing the existing target recognition model which can be directly used, the target database is used for storing various target data, including target data of ground large-scale equipment and target data of marine large-scale ships, and the stored target data is used for retraining or adjusting the target recognition model at a later stage;
the task operation environment of the satellite edge node is constructed based on a software virtualization technology and is used for performing cross-platform deployment and migration of a task system, on-satellite task resource management, control model transceiving, task starting and stopping, message simulation, and task execution effect and observation parameters presentation;
step 1.3, a cloud edge cooperative communication link is constructed, wherein the cloud edge cooperative communication link comprises an uplink transmitted from a ground cloud service node to a satellite edge node, a downlink transmitted from the satellite edge node to the ground cloud service node and a parallel link transmitted between the satellite edge nodes, and is used for satellite-ground control instruction transmission, identification model transmission, target data transmission and state information transmission, and relay transmission of the ground cloud service node and a non-direct-connection satellite edge node.
3. The method for cooperatively identifying the ground target based on the satellite-ground cloud edge of the deep network as claimed in claim 1, wherein the step 3 of collecting target data of various types of recognition tasks, storing the target data into a target database, preprocessing and labeling the target data, and generating a standard training sample comprises the following specific steps:
step 3.1, collecting target data of various identification tasks and storing the target data into a target database;
step 3.2, the ground cloud service node acquires sensor parameters of the satellite edge nodes, determines preprocessing indexes of target data to be recognized according to the sensor parameters, and performs preprocessing operation on the target data to be recognized;
and 3.3, sending the preprocessed target data to be recognized into a target data labeling link, performing frame selection and labeling on all targets in the target data to be recognized, taking the target data to be recognized as a model training sample after the labeling is finished, storing the model training sample into a target database, and training a target recognition model in the next step.
4. The satellite-ground cloud edge collaborative ground-target identification method based on the deep network as claimed in claim 3, wherein the step 3.1 collects target data of various types of identification tasks, and stores the target data into a target database, specifically as follows:
the following two ways are adopted to collect the target data of various identification tasks:
in the first mode, data of a target to be identified is collected through public network information collection, professional organization service and historical data accumulation of the system and stored in a target database of a ground cloud service node;
in the second mode, the ground cloud service node sends an instruction to a satellite side node in a nearby set range according to the code of the target area to be identified, collects data of the target to be identified in real time, and stores the data into a target database of the ground cloud service node after the data is collected.
5. The satellite-ground cloud edge collaborative ground target identification method based on the deep network as claimed in claim 3, wherein the ground cloud service node in step 3.2 obtains sensor parameters of the satellite edge node, determines a preprocessing index of target data to be identified according to the sensor parameters, and performs preprocessing operation on the target data to be identified, specifically as follows:
the method comprises the steps that a ground cloud service node obtains sensor parameters of a satellite edge node, the sensor parameters comprise optical sensor resolution, picture frame and sampling frequency, preprocessing indexes of target data are determined according to the sensor parameters, the preprocessing indexes comprise resolution and picture frame, normalization operation is conducted on the target data to be recognized, and the outstanding texture characteristics of pictures are sharpened.
6. The method for identifying the satellite-ground cloud edge collaborative ground target identification based on the deep network according to claim 1, wherein in step 4, a target identification model is built according to the on-satellite resource constraints, and training parameters of the target identification model are configured as follows:
step 4.1, the ground cloud service node acquires software and hardware resource parameters of the satellite edge node, and constructs a target identification model which is suitable for hardware constraint of the satellite edge node and comprises a data preprocessing layer, an internal hiding layer and an output layer;
step 4.2, designing a network structure and parameters of the target recognition model, wherein the network structure and parameters comprise a model layer number, an internal hidden layer, a layer unit number, a network layer number iteration number, an optimizer, an evaluation function, an activation function and a Dropout layer;
and 4.3, confirming the network updating frequency, the training iteration times and the evaluation function of the target recognition model according to the operation rate and the sample number of the ground cloud service nodes.
7. The method for cooperatively identifying a target to ground based on a satellite-ground cloud edge of a deep network according to claim 6, wherein the target identification model in step 4.1 adopts a MobileNet SSD model as a basic model.
8. The satellite-ground cloud edge collaborative ground target identification method based on the deep network as claimed in claim 1, wherein the ground cloud service node in step 5 trains a target recognition model and tests the prediction accuracy of the trained target recognition model, specifically as follows:
step 5.1, training the target recognition model set up in the step 4 by using the target training sample preprocessed in the step 3 to obtain a trained target recognition model;
and 5.2, testing the prediction accuracy of the trained target recognition model.
9. The satellite-ground cloud edge collaborative ground target identification method based on the deep network as claimed in claim 1, wherein the satellite edge node in step 7 executes a target identification task and returns a target identification result and a target real-time state to the ground cloud service node, specifically as follows:
establishing a connection between the ground cloud service node and a satellite side node, determining a satellite side node identifier for executing a target recognition task, transmitting a trained target recognition model to the corresponding satellite side node, executing the target recognition task by the satellite side node, dynamically adjusting sensor parameters based on different orbit heights, and returning a target recognition result and a target real-time state according to the requirements of the ground cloud service node; and the ground cloud service node changes the target identification model requirement and completes the target identification and acquisition tasks of the target area through the cloud edge cooperative communication link.
10. A satellite-ground cloud edge collaborative ground target identification device based on a deep network is characterized by comprising a ground cloud service node and a satellite edge node, wherein the satellite edge node comprises a calculation control module, an on-satellite communication module, a satellite data acquisition module, a data preprocessing module, a data labeling module and an on-satellite identification processing module;
the ground cloud service node adopts a high-performance server cluster and is used for business process control, data transceiving control, model training optimization and model database support;
the computing control module adopts a heterogeneous computing platform based on a CPU + GPU and is used for satellite-borne business process control, data receiving and sending control and image processing control;
the satellite communication module is used for receiving communication signals from other satellites, demodulating and unframing the signals and transmitting the signals to the calculation control module; meanwhile, data from the computing control module is received, framing and modulation of signals are achieved, and then the signals are sent to a ground cloud service node or other satellite side nodes;
the satellite data acquisition module is used for acquiring a satellite image to be identified;
the data preprocessing module is used for preprocessing the satellite image to be identified to generate a standard image for labeling;
the data marking module is used for marking the target content in the preprocessed satellite image for training of the model training module;
and the satellite recognition processing module is used for inputting the preprocessed standard satellite image into the uploaded target recognition model and outputting a recognition result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211295375.6A CN115561790A (en) | 2022-10-21 | 2022-10-21 | Satellite-ground cloud edge collaborative ground target identification method and device based on deep network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211295375.6A CN115561790A (en) | 2022-10-21 | 2022-10-21 | Satellite-ground cloud edge collaborative ground target identification method and device based on deep network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115561790A true CN115561790A (en) | 2023-01-03 |
Family
ID=84746959
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211295375.6A Pending CN115561790A (en) | 2022-10-21 | 2022-10-21 | Satellite-ground cloud edge collaborative ground target identification method and device based on deep network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115561790A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117278100A (en) * | 2023-09-21 | 2023-12-22 | 之江实验室 | Service execution method, device and storage medium based on space-based information system |
-
2022
- 2022-10-21 CN CN202211295375.6A patent/CN115561790A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117278100A (en) * | 2023-09-21 | 2023-12-22 | 之江实验室 | Service execution method, device and storage medium based on space-based information system |
| CN117278100B (en) * | 2023-09-21 | 2024-04-26 | 之江实验室 | Service execution method, device and storage medium based on space-based information system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110335507B (en) | Flight operation situation rule analysis method based on air traffic control track big data | |
| US20120029812A1 (en) | Method and system for automatically planning and scheduling a remote sensing satellite mission | |
| CN111447251A (en) | System and method for collecting, monitoring and analyzing vehicle data from multiple vehicles using edge calculations | |
| CN110377593B (en) | Method for processing and screening reconnaissance target by low-orbit multifunctional satellite | |
| CN116310689B (en) | Unmanned intelligent multi-mode information fusion and target perception system and operation method | |
| CN115561790A (en) | Satellite-ground cloud edge collaborative ground target identification method and device based on deep network | |
| CN113420939A (en) | Cloud picture forecasting method, equipment and storage medium | |
| CN102263800A (en) | Geographic information internet of things system constructed based on cloud computing, and application | |
| CN113326826A (en) | Network model training method and device, electronic equipment and storage medium | |
| CN115294476A (en) | Edge calculation intelligent detection method and device for unmanned aerial vehicle power inspection | |
| CN115453541A (en) | Precipitation amount prediction method and device, electronic device and storage medium | |
| CN117173527A (en) | Automatic driving multi-mode data updating and fusing method, device, equipment and medium | |
| US11263453B2 (en) | Method and system for tracking and displaying object trajectory | |
| CN114513805B (en) | Wireless modeling method and system | |
| CN113114399B (en) | Three-dimensional spectrum situation complementing method and device based on generation countermeasure network | |
| CN112422214B (en) | Communication effect demonstration verification system suitable for aviation channel | |
| CN110991378B (en) | Individual identification method and device for power amplifier | |
| CN113780371B (en) | Insulator state edge identification method based on edge calculation and deep learning | |
| CN112486667B (en) | Method and device for accurately processing data based on edge calculation | |
| CN115497052A (en) | Saline soil identification monitoring method and system and electronic equipment | |
| CN112732446B (en) | Task processing method and device and storage medium | |
| CN114140289A (en) | Smart city planning system based on satellite remote sensing technology | |
| CN115482277A (en) | Social distance risk early warning method and device | |
| CN115077539A (en) | Map generation method, device, equipment and storage medium | |
| CN110019608A (en) | A kind of information collecting method, device and system and storage equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |