Unmanned aerial vehicle target detection method in 5G environment
Technical Field
The invention relates to a 5G technology, an unmanned aerial vehicle, a target detection and edge calculation technology, in particular to a target detection method of the unmanned aerial vehicle in a 5G environment.
Background
An unmanned aircraft, abbreviated as "drone", and abbreviated in english as "UAV", is an unmanned aircraft that is operated by a radio remote control device and a self-contained program control device, or is operated autonomously, either completely or intermittently, by an onboard computer. Along with the development of unmanned aerial vehicle technique and the continuous decline of key spare part cost, greatly reduced the threshold of developing unmanned aerial vehicle system, this also makes unmanned aerial vehicle obtain extensive application, and unmanned aerial vehicle's volume and weight also are constantly diminishing in order to adapt to wider application scene, and unmanned aerial vehicle also becomes more intelligent simultaneously, and intelligent functions such as automatic obstacle avoidance, automatic tracking target all begin to popularize.
With the coming of a new information revolution represented by technologies and businesses such as cloud computing and mobile internet, revolutionary transition from product provision to service provision is brought, and the information industry is a key element for promoting the upgrading of the information industry and promoting the integration of the vertical industry and the internet. The cloud center aggregates a large amount of physical hardware resources, realizes unified allocation, scheduling and management of heterogeneous network computing resources by adopting a virtualization technology, and greatly reduces computing and storage costs by intensively building a data center. Along with the increasing huge cloud data volume, the transmission rate is reduced, and even a large network delay sometimes exists, and the network delay becomes a large factor restricting the application. The current situation of 5G is technically changed, 5G has the characteristics of ultrahigh bandwidth, low delay and large capacity, and the end-to-end network slicing capability can flexibly and dynamically allocate required network resources to different requirements, so that a customized network is provided for users aiming at the vertical industry. On the other hand, with the development of artificial intelligence technology in recent years, especially the algorithm evolution taking deep learning as a core, the large-scale convolutional neural network similar to the human brain structure is obtained through training and construction with the support of big data and the super-strong evolutionary capability, and landing is realized in different fields. Under the 5G environment, along with the continuous surge of the demand of the edge side, the capability of artificial intelligence gradually sinks to the edge side from the cloud, the artificial intelligence is closer to the demand side of the equipment, and the edge computing node represented by the 5G base station brings new possibility to the whole edge side application industry.
Utilize unmanned aerial vehicle to carry out target detection is unmanned aerial vehicle's important application, nevertheless receive unmanned aerial vehicle size, load, computational resource's restriction, unmanned aerial vehicle is more as an acquisition device, and the data of collection also are mostly to shoot with the angle of overlooking perpendicularly, and target detection's rate of accuracy has certain restriction. Under the circumstances, how to effectively utilize the 5G network and the edge computing capability and improve the target detection accuracy of the unmanned aerial vehicle become problems which need to be solved urgently.
Disclosure of Invention
In order to solve the technical problems, the invention provides a method for detecting the target of the unmanned aerial vehicle in the 5G environment, which effectively utilizes a 5G high-bandwidth high-speed network and a 5G base station edge intelligent service, combines the computing capability of the terminal side of the unmanned aerial vehicle, detects the image obtained by aerial shooting of the unmanned aerial vehicle in real time, transmits the undetermined target back to the 5G base station side through 5G for analysis, feeds the result back to the unmanned aerial vehicle in real time, changes the advancing mode of the unmanned aerial vehicle, and changes the angle or adjusts sensing equipment to complete the target detection task. In addition, edge intelligent computing nodes are managed in a unified mode through a cloud center, the edge nodes are subjected to personalized distribution of intelligent models and services according to unmanned aerial vehicle target detection tasks, and meanwhile images collected by the unmanned aerial vehicles are collected and used for continuous optimization of the cloud target detection models.
According to the invention, a large amount of computing resources are gathered by the cloud center, model training is carried out on a target detection task executed by the unmanned aerial vehicle, multi-dimensional cutting optimization is carried out, and services and models are respectively pushed to the 5G edge base station terminal and the unmanned aerial vehicle terminal by combining business logic. The unmanned aerial vehicle interacts with the edge computing node in real time through a 5G network, and real-time target detection of images collected by the unmanned aerial vehicle is achieved jointly. In addition, the edge base station provides the storage function, can have the edge base station with the image that unmanned aerial vehicle gathered temporarily to unified timesharing is uploaded to cloud center and is used for future model optimization, further promotes the recognition accuracy, has also promoted unmanned aerial vehicle's duration to a certain extent.
Wherein,
the unmanned aerial vehicle is provided with an image acquisition device and has certain calculation and storage capacities, a target detection model specially aiming at the unmanned aerial vehicle can be loaded after the cloud center training, cutting and optimizing, real-time target detection reasoning is carried out at the unmanned aerial vehicle end, interaction with a 5G base station and the cloud center is completed through a 5G communication module, and a cooperative task is completed.
In addition, unmanned aerial vehicle possesses the data compression function, further reduces storage and network bandwidth.
The 5G base station is a node with edge computing capability, comprises the capabilities of computing, storing and networking, provides 5G network service externally, is upwards connected with a core network in a butt joint mode and is connected with a cloud center, can realize the services of target detection, data temporary storage, safety certification, unmanned aerial vehicle interaction control, data compression, time-sharing uploading and the like, realizes two-way message transmission and action interaction with the unmanned aerial vehicle, carries out two-way communication with the cloud center, and timely updates and manages the node assembly service.
The cloud center realize the distribution of target detection task and the multidimension degree of target detection model and tailor the optimization to with service and model, carry out the propelling movement with and the business demand according to 5G basic station and unmanned aerial vehicle's actual specification, realize the management to 5G basic station and unmanned aerial vehicle, the high in the clouds carries out the model training through the image that comes from marginal basic station and unmanned aerial vehicle collection that continues to collect and optimizes, continues to promote the discernment precision.
The invention is used for an unmanned aerial vehicle to execute image acquisition and target detection tasks, and comprises the following steps:
101, the cloud center gathers a large amount of hardware resources, and performs model training by using actually collected images to form corresponding detection models for target detection tasks of pictures collected by the unmanned aerial vehicle;
102, establishing an image acquisition task for the unmanned aerial vehicle through the cloud center, and determining the target detection requirement and the general driving route of the unmanned aerial vehicle;
103, cutting and optimizing a corresponding model according to the actual specification of the unmanned aerial vehicle and the requirement of a target detection task to form a detection model which is most suitable for the unmanned aerial vehicle, and downloading the detection model into the unmanned aerial vehicle through a safety channel;
104, selecting a model which is most suitable for the 5G base station to operate according to the target detection task requirement of the unmanned aerial vehicle, and directionally pushing a detection model aiming at the task, a picture quality evaluation model and related services interacted with the unmanned aerial vehicle to the 5G base station through a safety channel according to a driving route, so that the 5G base station can be conveniently and quickly loaded in the future;
105, the unmanned aerial vehicle starts to execute an image acquisition task, flies above a relevant area, acquires images for ground targets, performs real-time target detection and analysis by using a local model of the unmanned aerial vehicle, determines target objects and takes pictures for multiple times;
106, the unmanned aerial vehicle quickly transmits the acquired image to the 5G base station through a 5G network, and simultaneously, the image with low certainty factor is specially marked, so that the 5G base station is requested to provide service, and further image processing is completed;
step 107, the 5G base station receives the image sent by the unmanned aerial vehicle, detects the image by adopting a target detection model of the base station local to the unmanned aerial vehicle image acquisition task, and determines a detection result;
step 108, the 5G base station analyzes and detects the image sequence from the unmanned aerial vehicle, synthesizes multiple factors such as confidence coefficient, picture quality and the like, plans an optimal shooting place and feeds back the optimal shooting place to the unmanned aerial vehicle;
step 109, the unmanned aerial vehicle adjusts the flight line of the unmanned aerial vehicle in real time according to the feedback from the 5G base station, changes the position and the angle of the unmanned aerial vehicle and shoots;
and step 110, repeating the steps 105 to 109, and continuously improving the quality of the acquired image to meet the acquisition task.
111, temporarily storing the images acquired by the unmanned aerial vehicle in a local place by the 5G base station, classifying and compressing the images according to actual acquisition task requirements, and uploading the images to the cloud center;
step 112, (optional) when the timeliness of the acquisition task is not strong, the 5G base station can select the relative idle time of the network, and the network is compressed and uploaded to the cloud center;
step 113, the unmanned aerial vehicle executes an image acquisition task;
and step 114, the cloud center further processes the image according to the task execution condition, and uses the image collected this time for training the task type target detection model, so as to continuously improve the detection precision.
The invention has the advantages that
The unmanned aerial vehicle tasks are managed in a unified mode through the cloud center, targeted model training is performed according to target detection requirements generated by different tasks executed by the unmanned aerial vehicle and by combining business logic, multi-dimensional model cutting optimization is performed according to the specification of the unmanned aerial vehicle and the specification of the tasks, the unmanned aerial vehicle is enabled to have the capability of real-time analysis and processing, and the problem of high energy consumption calculation caused by the power consumption and volume limitation of the unmanned aerial vehicle is solved; simultaneously according to the task that unmanned aerial vehicle executed, the optimization back relevant detection model that will relate to, evaluation model and relevant service, push to 5G edge base station end, make full use of the high-speed transmission ability of 5G network and the marginal intelligent ability of 5G basic station, through the interdynamic between edge calculation node and the unmanned aerial vehicle, alleviate the not enough of unmanned aerial vehicle resource, promote unmanned aerial vehicle's real-time target detection rate of accuracy simultaneously, and according to the feedback of 5G edge calculation node, carry out image acquisition route planning, promote the execution efficiency who gathers the image task, and then promote the image quality who finally gathers the image. In addition, images acquired by the unmanned aerial vehicle temporarily have edge base stations, the cruising ability of the unmanned aerial vehicle is improved to a certain extent, the images are uploaded to the cloud center in a unified time-sharing mode for future model optimization, the network bandwidth and the computing resources are used more efficiently, and meanwhile the target detection and identification accuracy is also continuously improved.
Drawings
Fig. 1 is a schematic diagram of a node where an unmanned aerial vehicle performs a task to participate.
Fig. 2 is a flowchart of the unmanned aerial vehicle performing image acquisition and target detection tasks.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer and more complete, the technical solutions in the embodiments of the present invention will be described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention, and based on the embodiments of the present invention, all other embodiments obtained by a person of ordinary skill in the art without creative efforts belong to the scope of the present invention.
According to the invention, the tasks of the unmanned aerial vehicle are uniformly managed through the cloud center, targeted model training is carried out according to target detection requirements generated by different tasks executed by the unmanned aerial vehicle and by combining business logic, multi-dimensional model cutting optimization is carried out according to the specification and the task specification of the unmanned aerial vehicle, the unmanned aerial vehicle is enabled to have the capability of real-time analysis and processing, and the problem of high energy consumption calculation caused by the power consumption and volume limitation of the unmanned aerial vehicle is solved; simultaneously according to the task that unmanned aerial vehicle executed, the optimization back relevant detection model that will relate to, evaluation model and relevant service, push to 5G edge base station end, make full use of the high-speed transmission ability of 5G network and the marginal intelligent ability of 5G basic station, through the interdynamic between edge calculation node and the unmanned aerial vehicle, alleviate the not enough of unmanned aerial vehicle resource, promote unmanned aerial vehicle's real-time target detection rate of accuracy simultaneously, and according to the feedback of 5G edge calculation node, carry out image acquisition route planning, promote the execution efficiency who gathers the image task, and then promote the image quality who finally gathers the image. In addition, images acquired by the unmanned aerial vehicle temporarily have edge base stations, the cruising ability of the unmanned aerial vehicle is improved to a certain extent, the images are uploaded to the cloud center in a unified time-sharing mode for future model optimization, the network bandwidth and the computing resources are used more efficiently, and meanwhile the target detection and identification accuracy is also continuously improved.
As shown in fig. 1, the cloud center aggregates a large amount of computing resources, performs model training on a target detection task executed by the unmanned aerial vehicle, performs multidimensional cutting optimization, and pushes services and models to a 5G edge base station end and an unmanned aerial vehicle terminal respectively in combination with business logic. The unmanned aerial vehicle interacts with the edge computing node in real time through a 5G network, and real-time target detection of images collected by the unmanned aerial vehicle is achieved jointly. In addition, the edge base station provides the storage function, can have the edge base station with the image that unmanned aerial vehicle gathered temporarily to unified timesharing is uploaded to cloud center and is used for future model optimization, further promotes the recognition accuracy, has also promoted unmanned aerial vehicle's duration to a certain extent.
Wherein,
the unmanned aerial vehicle is provided with an image acquisition device, has certain calculation and storage capacities, can load a target detection model specially aiming at the unmanned aerial vehicle after training, cutting and optimizing of a cloud center, carries out real-time target detection reasoning at an unmanned aerial vehicle end, completes interaction with a 5G base station and the cloud center through a 5G communication module, and completes a cooperative task, and in addition, the unmanned aerial vehicle has a data compression function, so that the storage and network bandwidth are further reduced;
the 5G base station is a node with edge computing capability, comprises the capabilities of computing, storing and networking, provides 5G network service to the outside, is upwards butted with a core network and is connected with a cloud center, can realize the services of target detection, data temporary storage, security authentication, unmanned aerial vehicle interaction control, data compression, time-sharing uploading and the like, realizes bidirectional message transmission and action interaction with the unmanned aerial vehicle, performs bidirectional communication with the cloud center, and timely updates and manages the node assembly service;
the cloud center realize the distribution of target detection task and the multidimension degree of target detection model and tailor the optimization to with service and model, carry out the propelling movement with and the business demand according to 5G basic station and unmanned aerial vehicle's actual specification, realize the management to 5G basic station and unmanned aerial vehicle, the high in the clouds carries out the model training through the image that comes from marginal basic station and unmanned aerial vehicle collection that continues to collect and optimizes, continues to promote the discernment precision.
For clarity of description, the unmanned aerial vehicle mentioned below is provided with an image acquisition module, a data storage module, a calculation analysis module, a position positioning module, a security authentication module and the like in addition to a 5G communication module, wherein a handshake protocol can be adopted between the unmanned aerial vehicle and an edge node of the 5G base station for mutual authentication to form a security channel, and data is transmitted through encryption to ensure the security of the unmanned aerial vehicle; the target detection algorithm adopted by the unmanned aerial vehicle can be an SSD algorithm, a YOLO algorithm and the like. Those skilled in the art will appreciate that the configurations according to embodiments of the present invention can be applied on other scenarios in addition to using the above terminals and algorithms.
The method provided by the invention will be described in detail with reference to specific examples.
Unmanned aerial vehicle executes image acquisition and target detection tasks
Referring to fig. 2, the unmanned aerial vehicle performing image acquisition and target detection tasks includes the following steps:
101, the cloud center gathers a large amount of hardware resources, and performs model training by using actually collected images to form corresponding detection models for target detection tasks of pictures collected by the unmanned aerial vehicle;
102, establishing an image acquisition task for the unmanned aerial vehicle through the cloud center, and determining the target detection requirement and the general driving route of the unmanned aerial vehicle;
103, cutting and optimizing a corresponding model according to the actual specification of the unmanned aerial vehicle and the requirement of a target detection task to form a detection model which is most suitable for the unmanned aerial vehicle, and downloading the detection model into the unmanned aerial vehicle through a safety channel;
104, selecting a model which is most suitable for the 5G base station to operate according to the target detection task requirement of the unmanned aerial vehicle, and directionally pushing a detection model aiming at the task, a picture quality evaluation model and related services interacted with the unmanned aerial vehicle to the 5G base station through a safety channel according to a driving route, so that the 5G base station can be conveniently and quickly loaded in the future;
105, the unmanned aerial vehicle starts to execute an image acquisition task, flies above a relevant area, acquires images for ground targets, performs real-time target detection and analysis by using a local model of the unmanned aerial vehicle, determines target objects and takes pictures for multiple times;
106, the unmanned aerial vehicle quickly transmits the acquired image to the 5G base station through a 5G network, and simultaneously, the image with low certainty factor is specially marked, so that the 5G base station is requested to provide service, and further image processing is completed;
step 107, the 5G base station receives the image sent by the unmanned aerial vehicle, detects the image by adopting a target detection model of the base station local to the unmanned aerial vehicle image acquisition task, and determines a detection result;
step 108, the 5G base station analyzes and detects the image sequence from the unmanned aerial vehicle, synthesizes multiple factors such as confidence coefficient, picture quality and the like, plans an optimal shooting place and feeds back the optimal shooting place to the unmanned aerial vehicle;
step 109, the unmanned aerial vehicle adjusts the flight line of the unmanned aerial vehicle in real time according to the feedback from the 5G base station, changes the position and the angle of the unmanned aerial vehicle and shoots;
and step 110, repeating the steps 105 to 109, and continuously improving the quality of the acquired image to meet the acquisition task.
111, temporarily storing the images acquired by the unmanned aerial vehicle in a local place by the 5G base station, classifying and compressing the images according to actual acquisition task requirements, and uploading the images to the cloud center;
step 112, (optional) when the timeliness of the acquisition task is not strong, the 5G base station can select the relative idle time of the network, and the network is compressed and uploaded to the cloud center;
step 113, the unmanned aerial vehicle executes an image acquisition task;
and step 114, the cloud center further processes the image according to the task execution condition, and uses the image collected this time for training the task type target detection model, so as to continuously improve the detection precision.
The above description is only a preferred embodiment of the present invention, and is only used to illustrate the technical solutions of the present invention, and not to limit the protection scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention shall fall within the protection scope of the present invention.