A portable industrial vision inspection device and its inspection method

Abstract

This invention relates to a portable industrial vision inspection device and method, belonging to the field of industrial vision inspection. The device adopts a handheld, integrated design, including a main frame with a touchscreen, camera, light source, and processing unit, as well as a detachable handle, solving the problems of large size, high cost, and inflexible deployment associated with fixed systems. Its core lies in its built-in "interactive fusion" vision processing architecture, which deeply integrates the efficient preprocessing and localization capabilities of traditional vision algorithms with the complex recognition capabilities of artificial intelligence vision models. Through collaborative work in modes such as "decision fusion" or "preprocessing guidance," it significantly improves the detection accuracy and robustness on embedded platforms. The device communicates with a server via a wireless network to achieve task distribution and result management. This invention is particularly suitable for high-precision, flexible inspection in areas with limited space and where robotic arms are difficult to operate.

Description

Technical Field

[0001] This invention belongs to the fields of industrial vision inspection, image processing technology and embedded technology, and specifically relates to a miniaturized, handheld industrial vision inspection device based on an embedded system and its inspection method. Background Technology

[0002] With the continuous development of industrial automation and intelligent manufacturing technologies, the requirements for product quality, production efficiency, and consistency in the production process are increasing. Inspection methods that rely on manual visual inspection or contact measurement generally suffer from problems such as high labor intensity, low inspection efficiency, strong subjectivity, and difficulty in ensuring consistency, and are no longer able to meet the needs of modern industrial large-scale, high-precision, and continuous production.

[0003] Industrial vision inspection technology is a non-contact inspection technology that uses optical imaging devices to acquire image information of the target object and analyzes and judges the appearance, size, position, or defect features of the target through image processing and pattern recognition algorithms. This technology typically consists of a light source system, an imaging system, an image acquisition unit, a processing and analysis module, and a control and communication interface. It has advantages such as high inspection speed, high repeatability, and adaptability to automated production lines, and has been widely used in fields such as electronics manufacturing, machining, semiconductor packaging, automotive parts, and packaging printing.

[0004] However, current mainstream industrial vision inspection systems typically consist of fixed industrial cameras, industrial computers, dedicated light sources, and complex mechanical mounting structures (such as robotic arms and XYZ slides). While these systems offer high inspection accuracy, they have significant limitations: they are bulky, costly to deploy, complex to install and debug, and impose stringent requirements on the space and rigidity of the production line. For small and medium-sized production lines, flexible manufacturing units, or sampling and inspection scenarios requiring manual intervention, such fixed systems are cumbersome and uneconomical.

[0005] In addition, existing visual inspection solutions tend to be simple in their processing logic: either they rely entirely on rule-based algorithms based on image processing, which have the advantages of being fast and highly deterministic, but are difficult to deal with defects such as complex appearance and varied features (e.g., irregular scratches and stains); or they rely entirely on artificial intelligence models based on deep learning, which have strong generalization ability, but require high computing resources, are difficult to run in real time on embedded devices, and have "black box" uncertainties.

[0006] Therefore, there is an urgent need for an industrial vision inspection device that can balance portability, intelligent inspection, and cost control. It should be handheld by operators, adaptable to complex field environments, and its internal vision processing system should intelligently integrate the advantages of different technological approaches to handle diverse industrial inspection tasks. Summary of the Invention

[0007] The purpose of this invention is to overcome the aforementioned shortcomings of the prior art and provide a portable industrial vision inspection device and method. This device, through its unique integrated hardware and software design, particularly its internal "interactive fusion" vision processing architecture, achieves the goal of highly robust and high-precision intelligent inspection on a portable embedded platform.

[0008] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a portable industrial vision inspection device, the device comprising: A handheld mechanical structure includes a main frame with a built-in touch screen, industrial camera, light source and processing unit, and handles detachably connected to both sides of the main frame. The software system includes device-side visual inspection software running on the processing unit, and a host computer service program running on a remote server and communicating with the device-side visual inspection software. A visual inspection algorithm module, integrated into the software system, is used to analyze and process images acquired by the industrial camera; The device-side visual inspection software is used to drive the industrial camera to acquire images, receive user operations through the touch screen or buttons on the grip, run the visual inspection algorithm module, and interact with the host computer service program via a wireless network. The host computer service program is used to manage the visual inspection template library, configure and distribute inspection algorithms, receive and statistically analyze inspection results, and run the visual inspection algorithm module.

[0009] Furthermore, the detection process performed by the visual detection algorithm module of the present invention includes: Template configuration process: Collect standard template images and upload them to the server. After configuring the detection algorithm in the host computer service program, publish them to the device-side visual inspection software. On-site inspection process: Real-time images are acquired through the industrial camera, and then coarse matching and positioning, target identification, fine matching and positioning, target area extraction, inspection, result determination, and the determination result is uploaded to the server.

[0010] Furthermore, the core improvement of the present invention lies in the architecture of the visual detection algorithm module, which is configured as a processing architecture that interactively fuses visual algorithms with artificial intelligence visual models. The visual algorithm is used to preprocess and enhance the image. The preprocessing includes at least filtering and histogram equalization to suppress noise and uneven illumination. The feature enhancement includes at least highlighting target features related to the recognition of the artificial intelligence visual model through morphological operations or specific filters, aiming to provide better quality input for subsequent recognition. The artificial intelligence vision model receives images or image regions that have been preprocessed and enhanced by the vision algorithm and then identifies and classifies them. The interactive fusion is not a simple chain of algorithms, but rather manifests as at least one of the following deep collaborative modes: The interactive fusion is manifested in at least one of the following ways: Preprocessing guidance: During the recognition process, the artificial intelligence visual model can generate attention maps or region suggestions for specific features of the image. Based on this information, the visual algorithm dynamically adjusts the enhancement strategy of subsequent frames or adjacent regions to focus on key features. Decision fusion: For the same detection item, the visual algorithm outputs a first judgment result and confidence level based on the rule model, and the artificial intelligence visual model outputs a second judgment result and confidence level. The final detection result is determined by the first and second judgment results and their confidence levels. The device-side visual inspection software is configured to dynamically allocate the workflow of the visual algorithm and the artificial intelligence visual model for the same inspection task, so that the input of the artificial intelligence visual model depends on the preprocessing result of the visual algorithm, while the enhancement strategy of the visual algorithm is adapted according to the feature type that the artificial intelligence visual model needs to identify.

[0011] Furthermore, the interactive fusion described in this invention is specifically implemented through a decision fusion method: For rule defects that can be determined by geometric features, the visual algorithm analyzes them based on edge, contour, or brightness features and outputs a first determination result and its confidence level. For irregular or complex defects that need to be determined by semantic features, the visual algorithm performs image preprocessing to extract candidate regions containing suspected defects and optimize their quality. The optimized regions are then input into the artificial intelligence visual model for analysis, and a second determination result and its confidence level are output. The device-side visual inspection software is configured to generate a final defect judgment conclusion based on the first judgment result and the second judgment result and their respective confidence levels, according to preset decision rules.

[0012] Furthermore, the visual algorithm described in this invention is used to perform one or more of the following: image scaling, matching and positioning, filtering, QR code recognition, color difference recognition, structural defect recognition, or orientation recognition; The artificial intelligence vision model is used to perform one or more of optical character recognition, crack and scratch detection, or target detection and classification.

[0013] Furthermore, when used for character recognition, the visual algorithm is configured to perform perspective correction and binarization processing to provide the artificial intelligence visual model with a regular character region image with a clean background; the artificial intelligence visual model is an optical character recognition model.

[0014] Furthermore, the device-side visual inspection software of the present invention also includes an image preprocessing unit for performing anti-shake processing on the acquired image sequence. The anti-shake processing is achieved by extracting ORB feature points of consecutive frame images for matching, calculating the inter-frame transformation matrix, and performing affine transformation correction on the image.

[0015] Furthermore, the device of the present invention also includes a gyroscope sensor for real-time sensing of the device's attitude information; the visual detection algorithm module is configured to combine the attitude information provided by the gyroscope sensor with a visual feature matching algorithm to compensate and correct image distortion or positional deviation caused by device movement or tilt.

[0016] Furthermore, the grip of the present invention is provided with a battery compartment, which is connected to the grip using a quick-release structure.

[0017] Based on the same inventive concept, the present invention also provides an industrial visual inspection method, applied to a portable industrial visual inspection device as described in any of the above descriptions, the method comprising: The host computer service program creates and configures detection templates and algorithms for specific targets to be inspected, and then sends them to the portable industrial vision inspection device. At the inspection site, the device is held in hand and the industrial camera is aimed at the target to be inspected to acquire images; The device-side visual inspection software calls the visual inspection algorithm module to process and analyze the acquired real-time images and output the inspection results. The detection results are uploaded to the host computer service program via wireless network for recording and statistics.

[0018] The beneficial effects of this invention are as follows: Low cost and high flexibility: The integrated and handheld design of the equipment eliminates the need for expensive industrial robotic arms, fixed brackets and large industrial control computers. A single device can be used on multiple production lines or different workstations, which significantly reduces the initial investment and deployment costs.

[0019] High spatial adaptability: Its compact body can directly penetrate into areas with limited space, such as inside vehicles, where it is difficult for robotic arms to move, to perform inspections, breaking through the application boundaries of fixed systems.

[0020] Balancing detection accuracy and efficiency: Through an "interactive fusion" architecture, simple tasks are intelligently assigned to efficient traditional algorithms and complex tasks to powerful AI models, even with limited embedded computing power. Decision fusion enhances the robustness of the final judgment, achieving detection accuracy close to that of high-end fixed systems while maintaining the detection pace of handheld operations.

[0021] The system is easy to use and manage: The "server configuration - terminal execution" mode centralizes the complex algorithm settings and model training work in the backend, and the front-end operator only needs to press a button to take a picture, which greatly reduces the difficulty of use. Moreover, all detection data can be centrally stored and analyzed, improving the level of quality management. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the handheld mechanical structure described in this invention; Figure 2 This is a schematic diagram of the structure of the balancing support leg described in this invention; Figure 3 This is a schematic diagram of the main frame structure described in this invention; Figure 4 This is an interface diagram of one operation mode of the device-side visual inspection software described in this invention; Figure 5 This is an interface diagram of another operation mode of the device-side visual inspection software described in this invention. Figure 6 This is a diagram of the identification and display interface of the host computer service program described in this invention; Figure 7 This is a template configuration interface diagram for a certain part described in this invention; Figure 8 This is a diagram of the defect identification interface for a certain part described in this invention. Detailed Implementation

[0024] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. The following examples will help those skilled in the art to further understand the present invention, but do not limit the present invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention.

[0025] Example 1: This example provides a portable industrial vision inspection device, which aims to solve the problems of high cost, large size, and strict space requirements of existing industrial vision inspection devices.

[0026] The portable industrial vision inspection device described in this embodiment includes: A handheld mechanical structure includes a main frame with a built-in touch screen, industrial camera, light source and processing unit, and handles detachably connected to both sides of the main frame. The software system includes device-side visual inspection software running on the processing unit, and a host computer service program running on a remote server and communicating with the device-side visual inspection software. A visual inspection algorithm module, integrated into the software system, is used to analyze and process images acquired by the industrial camera; The device-side visual inspection software is used to drive the industrial camera to acquire images, receive user operations through the touch screen or buttons on the grip, run the visual inspection algorithm module, and interact with the host computer service program via a wireless network. The host computer service program is used to manage the visual inspection template library, configure and distribute inspection algorithms, receive and statistically analyze inspection results, and run more complex visual inspection algorithm modules.

[0027] Furthermore, the detection process of the visual detection algorithm module is designed to include two stages: offline template configuration and online on-site detection. The template configuration stage completes algorithm training and parameter setting on the server side; the on-site detection stage performs efficient real-time analysis on the device side, and ensures positioning accuracy under hand-held shaking through a strategy of "coarse matching positioning + fine matching positioning".

[0028] Specifically: The process involves coarse matching followed by fine matching. Coarse matching extracts ORB feature points from both the real-time and template images. Once a sufficient number of matches are found, position correction is performed. In the corrected image, images are extracted from the configured feature ROI regions. After coarse registration, key detection regions are extracted from the corrected image based on the pre-configured feature ROI regions. Fine matching is then performed within these ROI regions to extract SIFT feature points that are more robust to scale and rotation changes. A high-precision feature matching algorithm is then used for fine position matching, further calculating precise geometric transformation relationships and performing fine correction on the ROI regions. This approach, combining rapid coarse localization with high-precision fine correction, significantly improves image registration accuracy while maintaining computational efficiency.

[0029] The core improvement of this invention lies in the architecture of the visual detection algorithm module. This module is configured as a processing architecture that interactively integrates visual algorithms and artificial intelligence visual models. The visual algorithms are responsible for efficient preprocessing and feature enhancement of the image, such as filtering, histogram equalization, and morphological operations, aiming to provide higher-quality input for subsequent recognition. The AI ​​visual model is responsible for complex pattern recognition and classification of the processed image regions.

[0030] The so-called "interactive fusion" is not a simple chain of algorithms, but rather manifests as at least one of the following deep collaborative modes: 1. Preprocessing guidance: During the recognition process, the artificial intelligence vision model can generate attention maps or region suggestions for specific features of the image. Based on this feedback, the device software can dynamically guide the vision algorithm to adjust the enhancement strategy for subsequent image frames or adjacent regions (such as changing filter parameters or focusing on specific regions), thereby achieving adaptive optimization of the processing.

[0031] 2. Decision Fusion: For the same detection item, the system analyzes the data in parallel or sequentially using visual algorithms (based on rule models) and artificial intelligence visual models, with each outputting a judgment result with confidence levels. The final detection conclusion is not determined by a single model, but by the device software fusing the two results according to preset decision rules (such as weighted averaging, confidence comparison, etc.), thereby significantly improving the reliability and robustness of the judgment.

[0032] Furthermore, in the typical application of defect detection, the aforementioned decision fusion model is specifically applied: for regular defects, the algorithm directly analyzes and outputs the results; for irregular and complex defects, the algorithm extracts and optimizes candidate regions, which are then analyzed by an artificial intelligence visual model. The results and confidence levels of both are aggregated for a comprehensive decision.

[0033] Furthermore, in character recognition applications, the algorithm is dedicated to providing "preprocessing services" for artificial intelligence vision models, such as performing perspective correction and binarization to eliminate perspective and background interference, creating regular and clean input conditions for OCR models. This reflects the deep integration of the algorithm at the task flow level.

[0034] Furthermore, to address the inherent jitter of handheld operation, the device software integrates a vision-based real-time image stabilization algorithm that corrects images through feature point matching and affine transformation. Optionally, the device can integrate a gyroscope sensor, whose attitude data can be combined with the vision algorithm to achieve joint correction of image distortion caused by significant device movement and tilt.

[0035] Through the implementation of this embodiment, the present invention successfully solves the problems pointed out in the background art and achieves the following positive effects: Low cost and high flexibility: The integrated and handheld design of the equipment eliminates the need for expensive industrial robotic arms, fixed brackets and large industrial control computers. A single device can be used on multiple production lines or different workstations, which significantly reduces the initial investment and deployment costs.

[0036] High spatial adaptability: Its compact body can directly penetrate into areas with limited space, such as inside vehicles, where it is difficult for robotic arms to move, to perform inspections, breaking through the application boundaries of fixed systems.

[0037] Balancing detection accuracy and efficiency: Through an "interactive fusion" architecture, simple tasks are intelligently assigned to efficient traditional algorithms and complex tasks to powerful AI models, even with limited embedded computing power. Decision fusion enhances the robustness of the final judgment, achieving detection accuracy close to that of high-end fixed systems while maintaining the detection pace of handheld operations.

[0038] The system is easy to use and manage: The "server configuration - terminal execution" mode centralizes the complex algorithm settings and model training work in the backend, and the front-end operator only needs to press a button to take a picture, which greatly reduces the difficulty of use. Moreover, all detection data can be centrally stored and analyzed, improving the level of quality management.

[0039] Example 2, in conjunction with the following Figures 1 to 8 This invention describes the specific implementation of a portable industrial vision inspection device.

[0040] Portable industrial vision inspection equipment includes: Handheld mechanical structure, its structure is as follows Figure 1 As shown, it includes a detachable left and right hand grip structure with mechanical buttons, a quick-release battery structure, and a main frame structure with a touchscreen, camera, and light source. The main frame structure is as follows... Figure 3 As shown; in practical applications, the handheld mechanical structure is equipped with, for example... Figure 2 The balancing support leg structure shown is used to support the equipment in an upright and balanced position.

[0041] The device-side visual inspection software is a Linux-based visual inspection software that supports touch operation, button operation, drives the camera to capture images, configures templates, and communicates with the server-side host computer via a wireless network.

[0042] The server-side host computer statistics service program can manage the detection library, view detection information in real time, view the statistics board, create algorithm templates, and perform calculations for complex detection programs.

[0043] The visual inspection algorithm module is a processing architecture that interactively integrates visual algorithms and artificial intelligence visual models. It can intelligently identify inspection targets, automatically register, and intelligently correct them to ensure accuracy for handheld use. It features modular algorithms, rapid barcode / QR code recognition, and configurable production line inspection wizards.

[0044] The complete visual workflow is as follows: (1) Template configuration: Collect template images → Upload to server → Configure algorithm → Publish to handheld devices; (2) Testing process: Image acquisition → coarse matching and localization → target identification and detection → fine matching and localization → target extraction → detection → result judgment → upload to server.

[0045] In practical applications, visual algorithms are responsible for the basic processing of image information, including scaling, matching and localization, filtering, material filters, and some basic detection algorithms, including QR code recognition, color difference recognition, structural missing recognition, and component orientation recognition.

[0046] Artificial intelligence vision models are responsible for complex detection scenarios, such as OCR character recognition, crack and scratch recognition, and target detection and classification.

[0047] Since handheld devices have limited computing power and storage, many computational requirements are not complex and can be met by visual algorithms. Using artificial intelligence visual models for computation may introduce uncertainty and cause over-detection or under-detection.

[0048] However, for detection tasks such as minor scratches, bumps, stains, and target identification and localization, which cannot be obtained through traditional calculations, a method of image cleaning processing using visual algorithms combined with detection and recognition using artificial intelligence visual models is adopted.

[0049] In practical applications, a dual protection system, encompassing both hardware and software layers, is designed to address noise, lighting variations, and vibrations in industrial environments. Hardware-level preprocessing: In terms of hardware design, environmental interference is reduced by stabilizing imaging conditions, and fixed-focal-length industrial lenses are used to avoid focal length drift caused by autofocus, ensuring consistent imaging scale for each acquired image.

[0050] Turn off the automatic exposure and automatic gain functions in the camera manufacturer's driver, and set a fixed exposure time and gain value to stabilize the image brightness.

[0051] Supplemental lighting: The handheld device has a built-in supplemental light source, and additional industrial light sources can be used when necessary to ensure uniform lighting in the target area and reduce shadows and highlight reflections.

[0052] Software-level preprocessing: At the software level, image enhancement and stabilization are performed on the acquired real-time video stream. Histogram equalization and block-based automatic equalization methods are used to perform illumination compensation processing on the template and real-time images.

[0053] Based on OpenCV, ORB feature points are extracted from the image and then quickly matched. The inter-frame transformation matrix is ​​calculated in real time, and the image is corrected by affine transformation to achieve the image stabilization effect and reduce the impact of hand shakiness on the shooting and image acquisition.

[0054] At the same time, filters are designed and adjusted for different testing environments and materials to highlight the features to be tested.

[0055] In practical applications, to address the issue of handheld instability, this invention employs a real-time anti-shake algorithm and real-time feature matching, automatically initiating registration and correction upon target detection. Simultaneously, to ensure high-precision detection even under movement, tilt, and distance changes, this invention can optionally incorporate an IMU (Integrated Measurement Unit) sensor to acquire real-time device attitude information (such as pitch, roll, and yaw angles) for position-sensitive applications. When the detection device tilts or moves, the sensor data provides a preliminary estimate of the image attitude change. Furthermore, a visual template matching algorithm is used to perform feature point matching and geometric transformation calculations between the real-time acquired image and the template image. Affine transformation or homography matrix is ​​then used for fine-tuning the image's position and perspective correction, compensating for image distortion or positional deviations caused by device attitude changes. Moreover, scale-invariant feature matching and dynamic ROI (Region of Interest) localization enable the system to adapt to image scale changes caused by varying shooting distance. Through the combined approach of IMU attitude information-assisted correction and visual feature registration algorithms, stable positioning of the detection area can be maintained even under device movement, tilt, or distance changes.

[0056] In practical applications, for defects with strong regularity (such as gaps, scratches, missing parts, positional deviations, etc.), image processing algorithms are mainly used for detection, including image filtering and denoising, edge detection (Sobel / Canny), binarization segmentation, adaptive threshold segmentation, and morphological processing (erosion, dilation, opening and closing operations). The target contour, area, shape, or grayscale features are analyzed to identify abnormal areas.

[0057] For complex or irregular defects (such as surface stains, material damage, printing defects, etc.), deep learning target detection models, such as the YOLO series and lightweight CNN networks, can be introduced for feature learning and defect classification. A defect sample database can be collected to train model weights, improving recognition accuracy in complex scenarios. Artificial intelligence vision models can utilize ResNet, MobileNet, and lightweight convolutional neural network (CNN) structures. By training on sample data, they learn the feature distributions of different target categories, achieving automatic classification of product categories, states, or types.

[0058] In practical applications, when recognizing barcodes / QR codes, the paper, plastic, cloth, and metal surfaces are analyzed, and different filters are used to highlight the QR code features. After the image is captured, it is converted into a grayscale image and then input into the ZBAR QR code recognition library for recognition.

[0059] It also supports adaptive detection of multiple types, sizes, and poses of targets, specifically designed as follows: Feature point matching and image registration: Image registration is performed using feature point matching algorithms such as ORB / SIFT to achieve automatic adaptation to target translation, rotation and scale changes.

[0060] ROI Dynamic Localization: Automatically locates the detection region based on template features, avoiding detection region shift due to changes in target position. Scale-Invariant Feature Detection: Uses a feature algorithm with scale and rotation invariance (SIFT) to improve the recognition stability of targets of different sizes and poses.

[0061] Deep learning data augmentation training: Adding data augmentation such as rotation, scaling, and brightness changes during the model training phase improves the model's robustness to different poses and environmental conditions.

[0062] The following describes the operation method for a portable industrial vision inspection device equipped with a portable embedded hardware and software system, 4GB RAM + 64GB ROM, an octa-core CPU with four A72 cores and four A53 cores, a powerful Arm high-performance GPU, and a built-in 6T computing power NPU. Connect the device to the industrial control computer via Wi-Fi, and with high-performance industrial software, automatically upload task logs, modify task configurations, configure detection templates, and view detection statistics.

[0063] The server can deploy high-performance, complex model detection algorithms to meet complex detection needs.

[0064] For vehicle interior inspection solutions: Step 1: Determine the testing content and use the present invention to extract images from the standard template; Step 2: Perform template annotation on the server's host computer, configuring the algorithm to detect color difference and errors / omissions; Step 3: After configuration, take a picture of the target vehicle while holding the present invention; Step 4: The algorithm automatically identifies and detects targets from real-time images and outputs the detection results; Step 5: The handheld device automatically uploads the test results to the server database, and the server views the test statistics.

[0065] For workpiece character recognition solutions: Step 1: Use this invention to capture an image of the text area of ​​a standard target workpiece; Step 2: Label the text region on the server's host computer, and configure the algorithm for character recognition; Step 3: After configuration, hold the present invention over the target workpiece; Step 4: The algorithm automatically identifies and detects targets from real-time images and outputs the detection results; Step 5: The handheld device automatically uploads the test results to the server database, and the server views the test statistics.

[0066] Figures 4 to 8 The diagrams show the detection interface of the portable industrial vision inspection device. It can be seen that the portable industrial vision inspection device proposed in this invention can effectively complete the detection of areas with limited space and where robotic arms have difficulty moving, breaking through the application boundaries of fixed systems.

[0067] The above description of the technical solution provided by the present invention through several specific embodiments is intended to highlight the advantages and benefits of the technical solution provided by the present invention. However, the above-described specific embodiments are not intended to limit the present invention. Any reasonable modifications and improvements to the present invention, reasonable combinations of implementation methods and equivalent substitutions based on the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A portable industrial vision inspection device, characterized in that the device... include: A handheld mechanical structure includes a main frame with a built-in touch screen, industrial camera, light source and processing unit, and handles detachably connected to both sides of the main frame. The software system includes device-side visual inspection software running on the processing unit, and a host computer service program running on a remote server and communicating with the device-side visual inspection software. A visual inspection algorithm module, integrated into the software system, is used to analyze and process images acquired by the industrial camera; The device-side visual inspection software is used to drive the industrial camera to acquire images, receive user operations through the touch screen or buttons on the grip, run the visual inspection algorithm module, and interact with the host computer service program via a wireless network. The host computer service program is used to manage the visual inspection template library, configure and distribute inspection algorithms, receive and statistically analyze inspection results, and run the visual inspection algorithm module.

2. The portable industrial vision inspection device according to claim 1, characterized in that, The detection process performed by the visual detection algorithm module includes: Template configuration process: Collect standard template images and upload them to the server. After configuring the detection algorithm in the host computer service program, publish them to the device-side visual inspection software. On-site inspection process: Real-time images are acquired through the industrial camera, and then coarse matching and positioning, target identification, fine matching and positioning, target area extraction, inspection, result determination, and the determination result is uploaded to the server.

3. The portable industrial vision inspection device according to claim 1 or 2, characterized in that, The visual detection algorithm module is configured as a processing architecture that interactively integrates visual algorithms and artificial intelligence visual models. The visual algorithm is used to preprocess and enhance the image. The preprocessing includes at least filtering and histogram equalization to suppress noise and uneven illumination. The feature enhancement includes at least highlighting target features related to the recognition of the artificial intelligence visual model through morphological operations or specific filters. The artificial intelligence vision model receives images or image regions that have been preprocessed and enhanced by the vision algorithm and then identifies and classifies them. The interactive fusion is manifested in at least one of the following ways: Preprocessing guidance: During the recognition process, the artificial intelligence visual model can generate attention maps or region suggestions for specific features of the image. Based on this information, the visual algorithm dynamically adjusts the enhancement strategy of subsequent frames or adjacent regions to focus on key features. Decision fusion: For the same detection item, the visual algorithm outputs a first judgment result and confidence level based on the rule model, and the artificial intelligence visual model outputs a second judgment result and confidence level. The final detection result is determined by the first and second judgment results and their confidence levels. The device-side visual inspection software is configured to dynamically allocate the workflow of the visual algorithm and the artificial intelligence visual model for the same inspection task, so that the input of the artificial intelligence visual model depends on the preprocessing result of the visual algorithm, while the enhancement strategy of the visual algorithm is adapted according to the feature type that the artificial intelligence visual model needs to identify.

4. The portable industrial vision inspection device according to claim 3, characterized in that, The interactive fusion is specifically achieved through a decision fusion approach: For rule defects that can be determined by geometric features, the visual algorithm analyzes them based on edge, contour, or brightness features and outputs a first determination result and its confidence level. For irregular or complex defects that need to be determined by semantic features, the visual algorithm performs image preprocessing to extract candidate regions containing suspected defects and optimize their quality. The optimized regions are then input into the artificial intelligence visual model for analysis, and a second determination result and its confidence level are output. The device-side visual inspection software is configured to generate a final defect judgment conclusion based on the first judgment result and the second judgment result and their respective confidence levels, according to preset decision rules.

5. The portable industrial vision inspection device according to claim 4, characterized in that, The visual algorithm is used to perform one or more of the following: image scaling, matching and localization, filtering, QR code recognition, color difference recognition, structural missing recognition, or orientation recognition. The artificial intelligence vision model is used to perform one or more of optical character recognition, crack and scratch detection, or target detection and classification.

6. The portable industrial vision inspection device according to claim 5, characterized in that, When used for character recognition, the visual algorithm is configured to perform perspective correction and binarization processing to provide the artificial intelligence visual model with a regular character region image with a clean background; the artificial intelligence visual model is an optical character recognition model.

7. The portable industrial vision inspection device according to claim 1, characterized in that, The device-side visual inspection software also includes an image preprocessing unit, which is used to perform anti-shake processing on the acquired image sequence. The anti-shake processing is achieved by extracting ORB feature points of consecutive frame images for matching, calculating the inter-frame transformation matrix, and performing affine transformation correction on the image.

8. The portable industrial vision inspection device according to claim 1, characterized in that, The device also includes a gyroscope sensor for real-time sensing of the device's attitude information; the visual detection algorithm module is configured to combine the attitude information provided by the gyroscope sensor with a visual feature matching algorithm to compensate and correct image distortion or positional deviation caused by device movement or tilt.

9. The portable industrial vision inspection device according to claim 1, characterized in that, The grip has a battery compartment, which is connected to the grip via a quick-release mechanism.

10. An industrial vision inspection method, applied to a portable industrial vision inspection device as described in any one of claims 1-9, characterized in that, include: The host computer service program creates and configures detection templates and algorithms for specific targets to be inspected, and then sends them to the portable industrial vision inspection device. At the inspection site, the device is held in hand and the industrial camera is aimed at the target to be inspected to acquire images; The device-side visual inspection software calls the visual inspection algorithm module to process and analyze the acquired real-time images and output the inspection results. The detection results are uploaded to the host computer service program via wireless network for recording and statistics.

Images