Image target detection method and system

Abstract

The invention relates to the technical field of image detection, and discloses an image target detection method and system.The method comprises the steps that an edge end executes an image target detection algorithm to detect a set target after collecting image data, and when the set target is detected, the edge end transmits the image data and a detection result of the set target to a cloud end; the detection result comprises a set target bounding box, a category label and confidence. The problems of low detection efficiency and the like in the prior art are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of image detection, and in particular to an image target detection method and system. Background Art

[0002] Traditional cloud-edge collaborative image target detection mainly relies on centralized data processing, that is, all image processing and computing tasks are completed by cloud servers. However, with the surge in data volume and high requirements for real-time performance, the traditional model may cause a large amount of data transmission delays, network bandwidth pressure and storage costs. Transmitting a large amount of image data from edge devices to the cloud for processing can easily lead to excessive use of network bandwidth, especially in the case of large-scale equipment or high-frequency data collection, which may cause long delays and affect real-time requirements. At the same time, the cloud must process a large amount of irrelevant or redundant data, resulting in low processing efficiency. In scenarios that require rapid response, effective warnings and decisions cannot be made in a timely manner. Summary of the invention

[0003] In order to overcome the deficiencies of the prior art, the present invention provides an image target detection method and system to solve the problems of low detection efficiency and the like in the prior art.

[0004] The technical solution adopted by the present invention to solve the above problems is: An image target detection method, after collecting image data, the edge end executes an image target detection algorithm to detect a set target. When the set target is detected, the edge end transmits the image data and detection result of the set target to the cloud. The detection result includes a set target boundary box, a category label and a confidence level.

[0005] As a preferred technical solution, the following steps are included: A. Deploy cloud components: deploy the cloud core on the cloud and the cloud container that communicates with the cloud core; the cloud core is used to communicate with the edge, allocate computing tasks and manage devices; the cloud container is used to receive the set target image data and detection results transmitted by the edge, and issue an early warning; B. Deploy edge components: deploy edge cores at the edge and edge containers that communicate with the edge cores; the edge cores are used to: after collecting image data, control edge containers to execute image target detection algorithms to detect set targets, communicate with the cloud, upload target detection results to the cloud, and receive computing tasks from the cloud; edge containers are used to: execute image target detection algorithms and transmit target detection results to the edge cores; C, execute the image object detection algorithm to detect the set target.

[0006] As a preferred technical solution, step C comprises the following steps: C1, edge end collects image data; C2, the cloud controls the edge to execute the image target detection algorithm. The edge marks the target in the image according to the detection result of the target detection algorithm and generates a detection result. When the edge detects the set target, it enters step C3; C3, when the set target is detected, the edge uploads the original image and its target detection result annotation map to the cloud through the MQTT protocol.

[0007] As a preferred technical solution, in step C2, when marking the target in the image, a bounding box is drawn and the target category is marked.

[0008] As a preferred technical solution, in step C3, the application programs to be executed on the cloud and the edge are mirrored, and then the image target detection algorithm is executed.

[0009] As a preferred technical solution, the method further comprises the following steps: D. After receiving the test results, the cloud issues an early warning based on the test results.

[0010] As a preferred technical solution, the image target detection algorithm is the YOLOv5 algorithm.

[0011] As a preferred technical solution, the cloud and the edge communicate via the MQTT communication protocol.

[0012] As a preferred technical solution, the cloud and edge are built based on the KubeEdge platform.

[0013] An image target detection system is used to implement an image target detection method, comprising a cloud and an edge end communicating with each other, wherein the cloud includes a cloud core and a cloud container communicating with the cloud core, and the edge end includes an edge core and an edge container communicating with the edge core, wherein the cloud core is communicatively connected with the edge core; wherein the cloud core is used to communicate with the edge end, allocate computing tasks and manage equipment; the cloud container is used to receive set target image data and detection results transmitted by the edge end, and issue an early warning; the edge core is used to control the edge container to execute an image target detection algorithm to detect the set target after collecting the image data, and communicate with the cloud, upload the target detection results to the cloud, and receive computing tasks from the cloud; the edge container is used to execute the image target detection algorithm and transmit the target detection results to the edge core.

[0014] Compared with the prior art, the present invention has the following beneficial effects: (1) Edge image target detection: By deploying the target detection algorithm at the edge, the present invention can quickly complete image analysis locally, significantly reducing the delay of data transmission and processing, and achieving near real-time target recognition; (2) Reduce the amount of data transmission: Only relevant data is uploaded to the cloud when specific image targets such as aircraft or drones are detected, reducing the transmission of irrelevant data and improving transmission efficiency; (3) Intelligent warning and decision-making: After the cloud receives the detection from the edge, it can conduct further analysis and intelligent decision-making based on the data, such as issuing real-time warnings. (4) Platform synergy advantages: By utilizing the distributed computing capabilities of the KubeEdge platform, it is possible to efficiently manage the data flow and task scheduling between the edge and the cloud, making the entire system more flexible, scalable, and fault-tolerant, ensuring stability and efficiency in large-scale deployments. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a structural block diagram of an image target detection system according to the present invention; Figure 2 Code diagram for building a KubeEdge cluster. DETAILED DESCRIPTION

[0016] The present invention will be further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0017] Example 1 like Figure 1 to Figure 2 As shown, the present invention aims to combine the edge computing capabilities of the KubeEdge platform to offload image target detection tasks to edge devices, and upload relevant images and detection results to the cloud only when specific targets are detected, greatly improving the efficiency and intelligence of the system.

[0018] More specifically, as follows: 1. System model: like Figure 1 The image target detection system shown in the figure includes a cloud and N edge terminals (N≥1 and N is an integer). The cloud control node manages and schedules all resources and workloads in the cluster. The entire system cluster is built on the KubeEdge edge computing platform, and image target detection and intelligent warning are achieved by deploying container applications on the cloud and edge terminals. In this system, the image target detection task is performed by the edge terminal. After collecting the image data, the edge terminal immediately executes the image target detection algorithm in the local container to detect specific targets. When a specific target is detected, the edge terminal transmits the original image data and detection results of the specific target to the cloud through the MQTT communication protocol. After receiving the detection results, the cloud sends a real-time warning based on the image detection results.

[0019] 2. KubeEdge platform construction: In order to realize the image target recognition and intelligent early warning functions at the edge, the present invention builds an efficient edge computing architecture based on the KubeEdge platform. KubeEdge is an open source platform based on Kubernetes, which aims to provide containerized management and scheduling for edge devices. The KubeEdge platform consists of two parts: cloud components and edge components. The cloud component is responsible for the management and task scheduling of the overall system, while the edge component is responsible for performing image target detection tasks on local nodes.

[0020] The KubeEdge platform construction part of the present invention mainly includes the following key steps: (1) Deploy cloud components. Build a Kubernetes cluster in the cloud environment and deploy KubeEdge's cloud component CloudCore. CloudCore is responsible for establishing connections with each edge terminal, allocating computing tasks, and collecting data. Configure cloud resource scheduling and management strategies to ensure that tasks can be reasonably scheduled based on computing power and that data transmission is stable and real-time.

[0021] (2) Deploy edge components. Install KubeEdge's edge component EdgeCore on the edge. Configure the edge's local computing resources, including CPU, memory, and storage. Each edge independently runs and manages local applications such as image target detection algorithms. The edge will communicate with the cloud regularly through KubeEdge, upload target detection results to the cloud, and receive instructions and updates from the cloud.

[0022] (3) Application image management. The main purpose of Dockerizing the application to be executed on the cloud and edge and compiling it into a Docker image file is to package the application and all its dependencies, configurations, and environments into a single executable, portable, and reusable file. The image file ensures that the application runs in the same way on different machines and environments. Whether in the cloud or on the edge, the container image can ensure the consistency of the same operating system, libraries, dependencies, and application code in all environments, which can eliminate problems caused by environmental differences. At the same time, the image file provides a standardized and automated deployment method. The image is the basic unit for KubeEdge to deploy applications.

[0023] It is worth noting that regarding the order of deploying cloud components and edge components: you can deploy cloud components first and then edge components, or you can deploy edge components first and then cloud components, or you can do both simultaneously or alternately.

[0024] Based on the above platform building steps, the present invention preliminarily built an image target detection system based on KubeEdge. The code is as follows Figure 2 shown.

[0025] 3. Edge image target detection: In order to make full use of the computing resources of edge devices, the present invention deploys containerized applications to the edge through the KubeEdge platform to execute image target detection algorithms to reduce data transmission delays and improve real-time performance. In the present invention, the edge calls the YOLOv5 algorithm for image target recognition. YOLOv5 is an efficient deep learning target detection algorithm that can simultaneously predict the category and position of the target in a single forward propagation. Compared with traditional target detection methods, YOLOv5 has higher accuracy and computational efficiency, and is particularly suitable for real-time image processing tasks. When the edge detects specific targets such as airplanes, the edge transmits the original image data and detection results to the cloud through the MQTT communication protocol. MQTT is a lightweight publish / subscribe message protocol that is mainly used to achieve communication between devices through limited bandwidth and unreliable network environments. Its design goal is to achieve low power consumption, low network occupancy and real-time and reliable message delivery, which is particularly suitable for the Internet of Things and edge computing scenarios. The process of deploying edge image target detection and transmission tasks through the KubeEdge platform is as follows: (1) Deploy edge devices near the data source and collect image data through camera devices.

[0026] (2) The cloud control node deploys the compiled image target detection algorithm image to the edge container through the KubeEdge platform for execution. The image integrates the pre-trained YOLOv5 model and can identify specific categories of targets, such as airplanes, vehicles, pedestrians, etc. According to the detection results of YOLOv5, the edge marks the targets in the image, draws the bounding box and marks the target category, and generates the detection results. The present invention specifies the specific category of target as an airplane. When the edge detects the airplane result, it executes the next step of the image transmission task.

[0027] (3) Only when a specific target aircraft is detected, the edge uploads the original image and its target detection results to the cloud through the MQTT protocol. This method minimizes the amount of data uploaded by the edge, avoids the transmission of irrelevant image data, and thus improves bandwidth utilization efficiency.

[0028] 4. Cloud-based intelligent warning: In the present invention, when the cloud system receives specific target data (such as aircraft image data and detection results) sent by the edge, it will issue an intelligent warning in time to notify relevant personnel. The process of deploying the cloud intelligent warning application through the KubeEdge platform is as follows: (1) The cloud decodes the received raw image data and extracts key information from the target detection results, including the target category (such as airplane), the location of the bounding box (such as coordinates), and the confidence level of the target (such as 95%).

[0029] (2) Target verification and identification are performed in the cloud to verify whether the target category is a preset key monitoring target aircraft and to confirm whether the target detection result meets the set threshold (for example, the confidence level is higher than 60%).

[0030] (3) If the target meets the preset key monitoring targets and threshold requirements, the cloud determines that the target is a potential threat, issues an early warning, and notifies relevant personnel to take relevant measures.

[0031] Example 2 like Figure 1 to Figure 2 As shown, based on Example 1, this example provides a more detailed implementation method.

[0032] Based on the above system model, the present invention simulates the edge image target detection process in the built KubeEdge cluster. We deployed the image target detection application to the edge container through the KubeEdge platform on the cloud control node, and performed the target recognition task on 100 local JPEG image data on the edge. Among these 100 image data, only 1 is the aircraft image data that meets the preset specific target requirements.

[0033] Table 1 shows the comparison of the transmission performance between the present invention and the traditional cloud-edge collaborative image target detection system. Compared with the traditional cloud-edge collaborative image target detection system, the present invention changes the image target detection task from the cloud to the edge detection, and only transmits the original image data and detection results of the target to the cloud when a specific target aircraft is detected, which greatly reduces the amount of transmitted data and the transmission time. As can be seen from Table 1, taking 100 JPEG format image data with a total local size of 16.1MB on the edge as an example, under the condition of a transmission rate of 1.315Mbps (obtained according to the experimental system test), the present invention can reduce the transmission time from 100.1s to 1.17s, significantly reducing the data transmission time, and providing more efficient and real-time target detection and intelligent early warning capabilities.

[0034] Table 1 Transmission performance comparison table

[0035] In summary, the present invention proposes an image target detection system based on KubeEdge, and realizes intelligent early warning in the cloud. The system significantly improves the real-time and computational efficiency of target detection by deploying image target detection tasks on edge devices. Only when a specific target is detected, the edge uploads the specific target detection results and original image data to the cloud for intelligent decision-making and early warning, avoiding the transmission of a large amount of image data and reducing the burden of network bandwidth, which is particularly suitable for limited bandwidth or remote environments. At the same time, the present invention utilizes the containerized management capabilities of the KubeEdge platform to support flexible deployment of image target detection algorithms on different edge devices, and unified scheduling and management through KubeEdge. This approach enables the system to dynamically adjust computing resources, support deployment of different scales, and facilitate system maintenance and expansion.

[0036] As described above, the present invention can be preferably implemented.

[0037] All features disclosed in all embodiments in this specification, or steps in all methods or processes implicitly disclosed, except for mutually exclusive features and / or steps, can be combined and / or expanded or replaced in any manner.

[0038] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. According to the technical essence of the present invention, within the spirit and principles of the present invention, any simple modification, equivalent replacement and improvement made to the above embodiment still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for detecting an image target, characterized in that: After collecting the image data, the edge executes the image target detection algorithm to detect the set target. When the set target is detected, the edge transmits the image data and detection results of the set target to the cloud. The detection results include the set target bounding box, category label and confidence.

2. The image target detection method according to claim 1, characterized in that: The following steps are involved: A. Deploy cloud components: deploy the cloud core on the cloud and the cloud container that communicates with the cloud core; the cloud core is used to communicate with the edge, allocate computing tasks and manage devices; the cloud container is used to receive the image data and detection results of the set target transmitted by the edge, and issue an early warning; B. Deploy edge components: deploy edge cores at the edge and edge containers that communicate with the edge cores; the edge cores are used to: after collecting image data, control edge containers to execute image target detection algorithms to detect set targets, communicate with the cloud, upload target detection results to the cloud, and receive computing tasks from the cloud; edge containers are used to: execute image target detection algorithms and transmit target detection results to the edge cores; C, execute the image object detection algorithm to detect the set target.

3. The image target detection method according to claim 2, characterized in that: Step C includes the following steps: C1, edge end collects image data; C2, the cloud controls the edge to execute the image target detection algorithm. The edge labels the target in the image according to the detection result calculated by the target detection algorithm, and generates a detection result labeling map. When the edge detects the set target, it enters step C3; C3, when the set target is detected, the edge uploads the original image and its target detection result annotation map to the cloud through the MQTT protocol.

4. The image target detection method according to claim 3, characterized in that: In step C2, when annotating the object in the image, a bounding box is drawn and the object category is annotated.

5. The image target detection method according to claim 3, characterized in that: In step C3, after the applications to be executed on the cloud and edge are mirrored, the image object detection algorithm is executed.

6. The image target detection method according to any one of claims 2 to 5, characterized in that: The following steps are also included: After receiving the test results, the cloud will issue an early warning based on the test results.

7. The image target detection method according to claim 1, characterized in that: The image target detection algorithm is the YOLOv5 algorithm.

8. The image target detection method according to claim 1, characterized in that: The cloud and the edge communicate via the MQTT protocol.

9. The image target detection method according to claim 1, characterized in that: The cloud and edge are built on the KubeEdge platform.

10. An image target detection system, characterized in that: A method for implementing an image target detection method as described in any one of claims 1 to 8, comprising a cloud and an edge end communicating with each other, wherein the cloud includes a cloud core and a cloud container communicating with the cloud core, and the edge end includes an edge core and an edge container communicating with the edge core, and the cloud core is communicatively connected to the edge core; wherein the cloud core is used to: communicate with the edge end, allocate computing tasks and manage equipment; the cloud container is used to: receive set target image data and detection results transmitted by the edge end, and issue an early warning; the edge core is used to: control the edge container to execute an image target detection algorithm to detect the set target after collecting the image data, and communicate with the cloud, upload the target detection results to the cloud, and receive computing tasks from the cloud; the edge container is used to: execute the image target detection algorithm and transmit the target detection results to the edge core.

Images