AI intelligent visual inspection system based on cloud computing

Abstract

The invention relates to an AI intelligent visual detection system based on cloud computing, and the system comprises an image collection module which employs a high-definition camera to shoot and obtain a detection image of a to-be-detected object; the image preprocessing module is used for carrying out optimization processing on the detection image, removing image noise and improving the image contrast; the image segmentation module is used for segmenting the detection image into a plurality of small image blocks, generating a unique identifier ID for each small image block and marking the coordinate of each small image block in the original detection image; the cloud computing scheduling center is composed of a task generation unit and a task issuing unit, and is used for generating a detection task from the small image blocks segmented by the image segmentation module and issuing the detection task; the distributed computing cluster is composed of a plurality of cloud computing processing units, and each cloud computing processing unit receives and processes one detection task; and the global result fusion module is used for receiving and summarizing the detection results returned by each cloud computing processing unit, mapping the detection results to the original detection image according to the coordinates of the corresponding small image blocks and outputting the detection results.

Description

Technical Field

[0001] This invention relates to the field of visual inspection technology, specifically an AI-powered intelligent visual inspection system based on cloud computing. Background Technology

[0002] With the rapid development of industrial automation and intelligence, visual inspection technology, as a non-contact inspection method, has been widely used in various production and daily life scenarios. Traditional visual inspection systems mostly adopt a local computing mode, using a single processor to process and analyze the acquired images. However, this mode has obvious drawbacks: when the amount of image data to be inspected is large and the inspection task is complex, insufficient local computing power will lead to low inspection efficiency and cannot meet the needs of large-scale batch inspection. Summary of the Invention

[0003] To address the above problems, this invention proposes an AI-powered intelligent visual inspection system based on cloud computing. A cloud-based AI intelligent visual inspection system includes: Image acquisition module: Uses a high-definition camera to capture and acquire images of the items to be inspected; Image preprocessing module: Optimizes the detected image, removes image noise, and improves image contrast; Image segmentation module: Divides the detection image into several small patches, generates a unique identifier ID for each patch, and marks the coordinates of each patch in the original detection image; The cloud computing scheduling center consists of a task generation unit and a task publishing unit, which generates detection tasks from the small image patches divided by the image segmentation module and publishes them. Distributed computing cluster: Composed of multiple cloud computing processing units, each of which accepts and processes a detection task; Global result fusion module: Receives and aggregates the detection results returned by each cloud computing processing unit, maps them to the original detection image according to the coordinates of the corresponding small image patches, and outputs the detection results; Network communication module: Configures a protocol adaptation unit, a data encryption unit, a transmission control unit, and a status monitoring unit, used to realize bidirectional data interaction between the image acquisition module, image preprocessing module, image segmentation module, cloud computing scheduling center, distributed computing cluster, and global result fusion module.

[0004] Furthermore, the task generation unit encapsulates each small patch into an independent detection task. The task information of the detection task includes the small patch image data, identifier ID, coordinate information, and detection requirements, and is sent to the distributed computing cluster by the task publishing unit.

[0005] Furthermore, the task publishing unit collects the load status of each cloud computing processing unit in real time and uses a round-robin load balancing strategy to distribute and publish detection tasks.

[0006] Furthermore, each cloud computing processing unit in the distributed computing cluster is an independent virtualized node. The cloud computing processing unit has a built-in Python runtime environment and PyTorch deep learning framework, and deploys a pre-trained AI visual detection model.

[0007] Furthermore, the image segmentation module has a built-in image segmentation model and uses semantic segmentation to identify potential target regions in the detected image and segment the potential target regions into independent small patches.

[0008] Furthermore, after the image segmentation module segments and generates small image patches, it generates an identifier ID for each small image patch and records the coordinates of the top-left vertex of the small image patch in the original detection image. The global result fusion module maps the detection results of the small image patches to the original detection image based on the identifier ID and coordinate relationship.

[0009] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: the large-size detection image is divided into multiple small image patches by the image segmentation module, and the patches are processed in parallel by multiple cloud computing processing units of the distributed computing cluster, which greatly improves the processing speed of the detection task. Compared with the traditional local computing system, the processing efficiency is greatly improved, which can meet the needs of large-scale batch detection. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of the structure of the present invention.

[0012] Figure 2 This is a schematic diagram of the structure of the cloud computing scheduling center in this invention.

[0013] The system includes: image acquisition module 1, image preprocessing module 2, image segmentation module 3, cloud computing scheduling center 4, distributed computing cluster 5, global result fusion module 6, task generation unit 41, and task publishing unit 42. Detailed Implementation

[0014] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0015] Please see Figures 1-2 A cloud-based AI intelligent visual inspection system includes: Image acquisition module 1: It adopts a high-definition camera with a resolution of no less than 1920×1080. The lens focal length can be adjusted according to the detection scene (range 5-50mm). It acquires continuous frame detection images of the object to be detected in real time. The image format is JPEG or PNG, and the frame rate can reach 30fps, ensuring the clarity and real-time performance of image acquisition and providing high-quality raw data for subsequent processing.

[0016] Image preprocessing module 2 optimizes the detection image, removing image noise and improving image contrast. After receiving the detection image transmitted from image acquisition module 1, it sequentially performs grayscale conversion, Gaussian filtering for noise reduction, and histogram equalization for contrast enhancement. Specifically, grayscale conversion converts the color image into a single-channel grayscale image, reducing the amount of data; Gaussian filtering uses a 5×5 convolution kernel with a standard deviation set to 0.8-1.2, effectively removing Gaussian noise and salt-and-pepper noise from the image; histogram equalization improves the overall image contrast by adjusting the image's grayscale distribution, making potential defect areas easier to identify.

[0017] Image Segmentation Module 3: This module incorporates an improved U-Net-based image segmentation model. It employs semantic segmentation to analyze the preprocessed detection image, automatically identifying potential target regions (such as product defect candidate regions and regions to be detected). It then segments the potential target regions and background regions into several uniformly sized small patches (recommended size 256×256 pixels, adjustable according to actual detection needs). After segmentation, a unique 16-bit identifier ID (format "TB-XXXXXXXXXXXXXXXX") is generated for each small patch. Simultaneously, the pixel coordinates (X, Y) of the top-left corner vertex of each small patch in the original detection image are recorded. The origin of the coordinates is the top-left corner vertex of the original image, with the X-axis pointing horizontally to the right and the Y-axis pointing vertically downwards, ensuring the unique and traceable positional information of each small patch.

[0018] Cloud Computing Scheduling Center 4: As the core scheduling unit of the system, it consists of a task generation unit 41 and a task publishing unit 42. The task generation unit 41 receives small image patch data (including image data, identifier ID, and coordinate information) transmitted by the image segmentation module 3. Combining this with preset detection requirements (such as defect type, detection accuracy threshold, and detection time limit), it encapsulates each small image patch into an independent detection task. The task data format is JSON, containing the fields: task_id (unique task identifier), tile_id (small image patch identifier ID), tile_data (binary data of the small image patch), coordinate (coordinate information X, Y), and detection requirements (detection requirement parameters). The task publishing unit 42 collects the load status (including CPU utilization, memory usage, and current task queue length) of each cloud computing processing unit in the distributed computing cluster in real time through the network communication module. It uses a round-robin load balancing strategy to evenly distribute detection tasks to each cloud computing processing unit, ensuring balanced load across units and avoiding single-point overload.

[0019] Distributed computing cluster 5 consists of N (N≥2, elastically scalable) independent cloud computing processing units. Each cloud computing processing unit is a virtualized node based on KVM virtualization technology. Each cloud computing processing unit has a built-in Python 3.8+ runtime environment and PyTorch 1.10+ deep learning framework, and deploys a pre-trained AI vision detection model. When a detection task is received, the cloud computing processing unit calls the AI ​​vision detection model to detect small patches, identify the defect type, location coordinates (local coordinates relative to the small patch), confidence level, and other information of the target area, and generates detection result data.

[0020] Global Result Fusion Module 6: Receives detection result data returned by each cloud computing processing unit via the network communication module. First, it performs deduplication and integrity verification based on the identifier ID of each small image patch to ensure that all detection results for each small image patch have been received and are unique. Then, based on the global coordinates (X, Y) of the small image patch and the local coordinates (X1, Y1) in the detection results, it calculates the global actual coordinates (X+X1, Y+Y1) of the target region in the original detection image. It then maps the detection results of each small image patch to the original detection image according to their coordinate positions, generating a complete global detection report. The report includes information such as the original image, defect location markers, defect type, confidence level, and detection time. Network communication module: Configures a protocol adaptation unit, a data encryption unit, a transmission control unit, and a status monitoring unit, used to realize bidirectional data interaction between image acquisition module 1, image preprocessing module 2, image segmentation module 3, cloud computing scheduling center 4, distributed computing cluster 5, and global result fusion module 6.

[0021] In this embodiment, the task generation unit 41 encapsulates each small image patch into an independent detection task. The task information of the detection task includes the small image patch image data, identifier ID, coordinate information and detection requirements, and is sent to the distributed computing cluster 5 by the task publishing unit 42.

[0022] In this embodiment, the task publishing unit 42 collects the load status of each cloud computing processing unit in real time and uses a round-robin load balancing strategy to allocate and publish detection tasks.

[0023] In this embodiment, each cloud computing processing unit in the distributed computing cluster 5 is an independent virtualized node. The cloud computing processing unit has a built-in Python runtime environment and PyTorch deep learning framework, and deploys a pre-trained AI visual detection model.

[0024] In this embodiment, the image segmentation module 3 has a built-in image segmentation model and uses semantic segmentation to identify potential target regions in the detection image and segment the potential target regions into independent small patches.

[0025] In this embodiment, after the image segmentation module 3 segments and generates small image patches, it generates an identifier ID for the small image patches and records the coordinates of the top left corner vertex of the small image patch in the original detection image. The global result fusion module 6 maps the detection result of the small image patch to the original detection image according to the identifier ID and coordinate relationship.

[0026] The working principle of this invention is as follows: the image acquisition module 1 captures images of the object to be detected in real time, generates a detection image, and transmits it to the image preprocessing module 2; Image preprocessing module 2 performs grayscale conversion, noise reduction, and contrast enhancement on the detected image, and outputs an optimized image; Image segmentation module 3 uses semantic segmentation to segment the optimized image and generates small image patches with identifiers (IDs) and coordinates. The task generation unit 41 of the cloud computing scheduling center 4 encapsulates the detection task, and the task publishing unit 42 distributes the task to the distributed computing cluster 5 through a round-robin load balancing strategy. The cloud computing processing units of the distributed computing cluster 5 process the detection task in parallel and output the small image patch detection results; The global result fusion module 6 summarizes the detection results, maps them to the original detection image, and generates a global detection report.

[0027] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A cloud-based AI intelligent visual inspection system, characterized in that, include: Image acquisition module: Uses a high-definition camera to capture and acquire images of the items to be inspected; Image preprocessing module: Optimizes the detected image, removes image noise, and improves image contrast; Image segmentation module: Divides the detection image into several small patches, generates a unique identifier ID for each patch, and marks the coordinates of each patch in the original detection image; The cloud computing scheduling center consists of a task generation unit and a task publishing unit, which generates detection tasks from the small image patches divided by the image segmentation module and publishes them. Distributed computing cluster: Composed of multiple cloud computing processing units, each of which accepts and processes a detection task; Global result fusion module: Receives and aggregates the detection results returned by each cloud computing processing unit, maps them to the original detection image according to the coordinates of the corresponding small image patches, and outputs the detection results; Network communication module: Configures a protocol adaptation unit, a data encryption unit, a transmission control unit, and a status monitoring unit, used to realize bidirectional data interaction between the image acquisition module, image preprocessing module, image segmentation module, cloud computing scheduling center, distributed computing cluster, and global result fusion module.

2. The AI ​​intelligent visual inspection system based on cloud computing according to claim 1, characterized in that, The task generation unit encapsulates each small patch into an independent detection task. The task information of the detection task includes the small patch image data, identifier ID, coordinate information and detection requirements, and is sent to the distributed computing cluster by the task publishing unit.

3. The AI ​​intelligent visual inspection system based on cloud computing according to claim 1, characterized in that, The task publishing unit collects the load information of each cloud computing processing unit in real time and uses a round-robin load balancing strategy to distribute and publish detection tasks.

4. The AI ​​intelligent visual inspection system based on cloud computing according to claim 1, characterized in that, In a distributed computing cluster, each cloud computing processing unit is an independent virtualized node. The cloud computing processing unit has a built-in Python runtime environment and PyTorch deep learning framework, and deploys a pre-trained AI visual detection model.

5. The AI ​​intelligent visual inspection system based on cloud computing according to claim 1, characterized in that, The image segmentation module has a built-in image segmentation model and uses semantic segmentation to identify potential target regions in the detected image and segment the potential target regions into independent small patches.

6. The AI ​​intelligent visual inspection system based on cloud computing according to claim 1, characterized in that, After the image segmentation module segments and generates small image patches, it generates an identifier ID for each small image patch and records the coordinates of the top-left vertex of the small image patch in the original detection image. The global result fusion module maps the detection results of the small image patches to the original detection image based on the identifier ID and coordinate relationship.

Images