Equipment instrument quality interpretation device based on AI vision

Abstract

The invention discloses an equipment instrument quality interpretation device based on AI vision, and relates to the field of computer vision and artificial intelligence. Comprising an optical acquisition system, an image processing unit, an intelligent detection unit and a data management unit, the optical acquisition system acquires image or video information of equipment and instruments through a camera or an image sensor, is matched with a corresponding light source to enable the acquired image to have no shadow, and is connected with the image processing unit through a network; the image processing unit preprocesses the acquired image; the intelligent detection unit extracts features related to target quality from the preprocessed image by using a deep learning algorithm; and the data management unit stores the data and sends an interpretation result to a user through the feedback module according to the quality interpretation result, the user performs corresponding operation according to the interpretation result, and real-time quality detection is realized by utilizing efficient image processing and AI interpretation algorithms, so that the detection efficiency is remarkably improved, and the requirements of a high-rhythm production line are met.

Description

Technical Field

[0001] The present application relates to the fields of computer vision and artificial intelligence, and specifically to an equipment and instrument quality judgment device based on AI vision. Background Art

[0002] Traditionally, the quality judgment of equipment and instruments is mainly carried out through manual inspection, which is not only inefficient but also prone to misjudgment due to human factors. In recent years, with the development of computer vision and artificial intelligence technology, more and more researchers have tried to use these technologies to automatically judge the quality of equipment and instruments. However, the existing quality judgment systems based on AI vision generally have problems such as low accuracy and poor stability, which cannot meet the needs of practical applications.

[0003] The results of the present invention are intended to solve the problems of low efficiency, poor accuracy and interference from human factors in traditional equipment and instrument quality inspection methods. By introducing an automated quality judgment device based on AI vision, the present invention significantly improves the accuracy and real-time performance of quality inspection, reduces the possibility of human misjudgment, and is applicable to various types of equipment and instruments, enhancing the versatility and adaptability of the system. In terms of technical effects, the present invention achieves efficient and reliable quality inspection, and improves the automation level of the production line and the product quality control capability. Summary of the invention

[0004] This application aims at the current existing technology that the equipment and instrument quality inspection method mainly relies on manual inspection and traditional machine vision technology, which has the following technical problems:

[0005] Low detection efficiency: Manual inspection is time-consuming and labor-intensive, and it is difficult to meet the real-time detection needs of high-speed production lines.

[0006] Insufficient accuracy: Manual inspection results are limited by the experience and status of the inspectors, and are prone to missed detections and misjudgments.

[0007] Interference from human factors: During the manual testing process, factors such as the tester’s emotions and fatigue will affect the reliability of the test results.

[0008] Poor adaptability: Traditional detection methods are designed for specific types of equipment and instruments, lack versatility, and are difficult to adapt to the detection needs of various types of equipment.

[0009] Poor real-time performance: Traditional methods make it difficult to achieve real-time testing and feedback, and there are delays in problem discovery and resolution.

[0010] Difficulty in data recording and analysis: Manual inspection is difficult to achieve automatic recording and analysis, which limits the historical data tracing and problem location of quality control.

[0011] The present invention proposes an AI vision-based equipment and instrument quality judgment device, which aims to solve the above problems. The device includes an optical acquisition system, an image processing unit, an intelligent detection unit, and a data management unit.

[0012] Optical acquisition system: responsible for collecting images or video information from equipment and instruments, using high-resolution cameras or image sensors. It mainly uses high-resolution industrial cameras and appropriate light sources to ensure that the collected images are clear and without shadows. The camera is connected to the processing unit through the network.

[0013] Image processing unit: pre-process the collected images, including image cropping, grayscale, image denoising, binarization and opening operation. Through image cropping and grayscale, the amount of data processing can be reduced, and the unnecessary background area can be removed while retaining the integrity of the image features of the area of ​​interest. Then filter denoising is performed to reduce the noise caused by surface roughness and shooting environment. Binarization can separate parts and areas of interest from the background. However, there are burrs and holes in the binary image that affect the quality of edge extraction, so most of the burrs and holes need to be removed through opening operation.

[0014] The intelligent detection unit uses a deep learning algorithm to extract features related to the target quality, such as shape, size, color, texture, etc. from the preprocessed image. The feature extraction module of the present invention is implemented based on a convolutional neural network, and the specific steps include a convolution layer, a pooling layer, and a fully connected layer. The convolution layer refers to extracting low-level and high-level features of an image through multiple convolution layers; the pooling layer refers to reducing the dimension of the feature map through the pooling layer to reduce the amount of calculation; the fully connected layer refers to inputting the extracted feature information into the fully connected layer and outputting a feature vector. Then, based on the extracted features, the quality of the equipment and instrument is automatically judged using a trained quality judgment model. The model uses a supervised learning algorithm and is trained with a large amount of labeled data to achieve a high-precision and high-stability judgment effect.

[0015] Data management unit: Based on the quality judgment results, send feedback signals to users or related systems for subsequent maintenance, improvement, etc.

[0016] This application has the following beneficial effects:

[0017] 1. Using efficient image processing and AI interpretation algorithms, real-time quality inspection can be achieved, which significantly improves inspection efficiency and meets the needs of high-paced production lines;

[0018] 2. The feature extraction and interpretation algorithm based on deep learning reduces the possibility of human misjudgment and missed detection, and greatly improves the accuracy and reliability of the detection results;

[0019] 3. The present invention has a high degree of automation, and the detection process no longer relies on manual operation, which greatly reduces the interference of human factors and reduces the risk of operational errors and production accidents;

[0020] 4. The present invention is applicable to various types of equipment and instrument quality inspection, has strong versatility and adaptability, and can be flexibly applied to different production scenarios;

[0021] 5. The present invention can realize real-time detection and feedback, timely discover and solve quality problems, and reduce the cost and risk of the production process;

[0022] 6. The present invention can automatically record the test data, which is convenient for subsequent analysis and tracing, and improves the management level of quality control and the traceability of historical data. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0024] Figure 1 A schematic diagram of the structure of an equipment instrument quality judgment device based on AI vision provided in an embodiment of the present application.

[0025] Figure 2 This is a hardware diagram of an optical acquisition system provided in an embodiment of the present application.

[0026] Figure 3 This is a workflow diagram of an image processing unit provided in an embodiment of the present application.

[0027] Figure 4 A workflow diagram of an intelligent detection unit provided in an embodiment of the present application.

[0028] Figure 5 An embodiment of the present application provides a workflow diagram of a graph data management unit. DETAILED DESCRIPTION

[0029] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0030] The scheme of the present application is further described below in conjunction with the accompanying drawings. In practical applications, users can automatically judge the quality of equipment and instruments through the device of the present invention. First, the optical acquisition system collects image information of the equipment and instruments; then, the image processing unit preprocesses the image; then, the intelligent detection unit extracts features related to the target quality from the preprocessed image, and judges the quality based on the extracted features, and sends the judgment results to the user through the feedback module. The user can perform corresponding operations based on the judgment results, such as repair, replacement, etc.

[0031] The optical acquisition system mainly refers to the hardware platform for image acquisition, such as Figure 1 Schematic diagram of the structure of the interpretation device in . The image acquisition module is the core of the entire hardware device. It must not only meet the resolution of image capture, but also ensure the width of the field of view of image capture. It involves specific hardware equipment, mainly industrial cameras, lenses, auxiliary light sources and high-performance deep learning PCs. Industrial cameras are the core components of image acquisition hardware. Their essential function is to convert optical signals into orderly electrical signals, and then convert the signals into analog-to-digital signals and send them to the processor to complete image processing, analysis and recognition. Selecting a suitable industrial camera is an important part of the design of a machine vision system. The type of industrial camera not only directly determines the resolution and image quality of the captured images, but is also directly related to the operating mode of the entire system. The lens is an optical component used by industrial cameras to generate images. The auxiliary light source is to solve the problem of insufficient light during image acquisition and shooting. The light source can be placed in front of the camera to provide lighting for the object to be photographed during shooting. High-performance deep learning PCs can improve the model training rate and model accuracy after image acquisition and reduce the time of interpretation.

[0032] The purpose of the image processing unit preprocessing the image is to improve the image quality, enhance the image features and reduce noise in the further image analysis and processing process, and improve the accuracy and stability of the subsequent recognition and analysis results. Figure 3 , using image preprocessing technology, the original image is cropped, grayed, denoised, binarized and opened. Image cropping refers to cropping the size of the captured image to make the size of the image meet the subsequent algorithm.

[0033] The grayscale process is to calculate the grayscale value of each pixel of the RGB image according to a certain ratio of the three-channel components. Compared with other processing methods, this method can improve the contrast of the image and make the image clearer.

[0034] Gamma correction calculates grayscale:

[0035]

[0036] Psychological Gooch algorithm calculates grayscale:

[0037] Gray=R×0.299+G×0.587+B×0.114

[0038] Average grayscale calculation:

[0039] Gray=(R+G+B) / 3

[0040] The specific process of image denoising is as follows: select the corresponding filter template according to the density of noise pollution. The template size consists of m×n pixels. The gray value of the pixel point f(x, y) to be processed is replaced by the gray value of all pixels in the template covering the original image area. The formula is as follows:

[0041]

[0042] Where f(x,y) is the gray value of the pixel to be processed in the image, and g(x,y) is the gray value of the pixel after processing. There are two commonly used filter templates: 3×3 and 5×5. Its advantages are fast calculation speed and short time.

[0043] Binarization threshold segmentation is a method of image segmentation. By setting an appropriate threshold, an image with multiple gray levels can be segmented into only two gray levels, 0 and 255. By comparing the pixels in the image, those greater than the set threshold are converted to 255 (white), and those less than the set threshold are converted to 0 (black), thereby converting an image with multiple gray levels into a binary image. The formula is:

[0044]

[0045] Where g(x,y) is the image after segmentation, f(x,y) is the image before segmentation, and e is the threshold.

[0046] The mathematical morphology of opening operation is an image processing method based on set theory and lattice theory. Its goal is to quantitatively describe the geometric structure of the image. The opening operation first uses the set B to corrode the original image A, and then expands the result. The result of the opening operation is represented by A○B and is defined as:

[0047]

[0048] The intelligent detection unit is based on Figure 5 As shown in the figure, the intelligent detection unit mainly adopts the convolutional neural network in the deep learning algorithm. By training normal samples, a discrimination model is constructed to detect and discriminate the real-time collected images. It includes an input layer, a convolution layer, a pooling layer, and a fully connected layer.

[0049] The input layer is the beginning of the data flow of the convolutional neural network. The input layer of the convolutional neural network can receive the raw data input of 1D signals or 2D graphs, and pass the data to the following convolutional layer and pooling layer for feature extraction. The size of the input layer of the classic LeNet-5 model of the convolutional neural network is 32×32×1, 32×32 represents the width and height of the input layer, and 1 represents the depth of the input layer (when the input is an RGB image, it is 3).

[0050] The convolution layer is composed of a set of trainable filters. The convolution operation extracts features from the input signal through a set of filters. These filter groups are called convolution kernels. The formula represents the convolution operation:

[0051]

[0052] where w m,n Represents the weight of the convolution kernel, i.e., the filter parameter, x i+m,j+n is the value in the local receptive field of the convolution kernel, b represents the bias value, and y i,j Represents the output of the convolution operation.

[0053] The pooling layer is often placed after the convolution layer. It further reduces the dimension by removing unimportant components in the feature map to achieve the purpose of feature extraction. Adding a pooling layer can reduce the number of parameters of the convolutional neural network, thereby suppressing overfitting and improving the robustness of the model.

[0054] The fully connected layer is often located after the convolutional layer or the pooling layer, which can further nonlinearize the features extracted by the previous layer.

[0055] The data management unit outputs the detection results of the intelligent detection unit and stores the data for subsequent query and result tracing. The specific design steps are as follows.

[0056] Design database model: Design the database model, including data tables and fields, based on requirements and data structure.

[0057] Create a database table: In Microsoft Access software, create a blank database as a machine vision database. The basis of Access database is table. The first step in database development is to create a table. Create a new table in the database and add fields to the table to record, such as user information, inspection time, qualification and type. Set the part number field as the primary key and set the properties of each field. The user number, picture number type, etc. are text types, the picture is an OLE object type, and the inspection date is a date type.

[0058] Image table field design: including image ID (a unique identifier used to distinguish each record of image data), image name (a name or identifier used to describe the image), image path (the storage path or relative path of the image file), category (the category or label to which the image belongs), and other metadata fields (add other image-related fields such as shooting date, employee, etc. as needed).

[0059] Insert data: When an image is judged to be correct, insert its related information into the database table. Create an insert statement to insert values ​​such as image ID, name, path, and category into the database table.

[0060] Query and retrieve data: Access query is a tool for processing and analyzing data. By setting the query conditions, you can filter and classify the information. Use the created table as the data source, set the three fields of quality inspector and test date in the data table as the query conditions, and use these three fields to search and filter the data to be queried. For example, if you enter the number of any picture in the query, you can find the interpretation result of the corresponding picture. Similarly, you can also query the interpretation result of a certain quality inspector or a certain time.

[0061] Update operation: When you need to update the data in the database, you can perform the corresponding update operation. Write update statements to regularly check and update the model to adapt to new image classification requirements and improve performance.

[0062] Deletion operation: According to the results of the model interpretation, for incorrect images, you can perform a deletion operation to delete the corresponding image record from the database. Write a deletion statement to delete the incorrect image from the database based on the interpretation results.

[0063] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that an article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such article or device. In the absence of more restrictions, the elements defined by the sentence "includes..." do not exclude the existence of other identical elements in the article or device including the elements.

[0064] The above is a detailed introduction to the provided AI vision-based equipment and instrument quality judgment device. This article uses specific examples to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the AI ​​vision-based equipment and instrument quality judgment device of the present application and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A device for judging the quality of equipment and instruments based on AI vision, characterized in that: It includes optical acquisition system, image processing unit, intelligent detection unit and data management unit; The optical acquisition system acquires images or video information of equipment and instruments through a camera or an image sensor, and cooperates with a corresponding light source to make the acquired images shadow-free. The optical acquisition system is connected to the image processing unit through a network; The image processing unit performs preprocessing on the collected image, including image cropping, grayscale, image denoising, binarization and opening operation processing; The intelligent detection unit uses a deep learning algorithm to extract features related to target quality from the preprocessed image, including shape, size, color, and texture; The data management unit saves the data and sends the judgment result to the user through the feedback module according to the quality judgment result, and the user performs corresponding operations according to the judgment result.

2. According to the AI ​​vision-based equipment and instrument quality judgment device of claim 1, it is characterized in that: In the image processing unit, the image cropping and graying are to retain the image feature integrity of the region of interest through image cropping and graying, while removing unnecessary background areas; the image denoising is to reduce the noise caused by surface processing roughness and shooting environment through filtering denoising; the binarization process separates the parts and the region of interest from the background; The opening operation removes burrs and holes that affect the edge extraction quality in the binary image after the binarization process.

3. The device for judging the quality of equipment and instruments based on AI vision according to claim 2, characterized in that: The grayscale process is a process of obtaining the grayscale value of the corresponding pixel by calculating the components of the three channels of each pixel of the RGB image in proportion, specifically including: Gamma correction calculates grayscale: Psychological Gooch algorithm calculates grayscale: Gray=R×0.299+G×0.587+B×0.114 Average grayscale calculation: Gray=(R+G+B) / 3.

4. The device for judging the quality of equipment and instruments based on AI vision according to claim 3 is characterized in that: The image denoising specifically includes: The corresponding filtering template is selected according to the density of noise pollution. The template size consists of m×n pixels. The grayscale value of the pixel point f(x, y) to be processed in the image is replaced by the grayscale average value of all pixels in the template covering the original image area. The formula is as follows: Where f(x, y) is the gray value of the pixel to be processed in the image, g(x, y) is the gray value of the pixel after processing, and the filtering templates include 3×3 and 5×5 templates.

5. The device for judging the quality of equipment and instruments based on AI vision according to claim 4 is characterized in that: The binarization divides the image with multiple gray levels into only two gray levels of 0 and 255 by setting a threshold; by comparing the pixels in the image, the pixels greater than the set threshold are converted to 255, corresponding to white, and the pixels less than the set threshold are converted to 0, corresponding to black, and the image with multiple gray levels is converted into a binary image. The formula is: Where g(x,y) is the image after segmentation, f(x,y) is the image before segmentation, and e is the threshold.

6. The device for judging the quality of equipment and instruments based on AI vision according to claim 1, characterized in that: The intelligent detection unit adopts the convolutional neural network in the deep learning algorithm, builds a discrimination model by training normal samples, and detects and discriminates the real-time collected images, specifically including the input layer, convolution layer, pooling layer, and fully connected layer; The input layer is the beginning of the convolutional neural network data flow, which receives the original data input of 1D signal or 2D spectrum and passes the data to the subsequent convolutional layer and pooling layer for feature extraction; The convolution layer is composed of a group of trainable filters. The convolution operation is to extract features from the input signal through a group of filters. The filter group is the convolution kernel. The convolution operation is represented as: where w m,n is the weight of the convolution kernel, i.e. the parameter of the filter, x i+m,j+n is the value in the local receptive field of the convolution kernel, b is the bias value, y i,j is the output of the convolution operation; The pooling layer is after the convolution layer, and reduces the dimension by removing unimportant parts of the feature map to achieve the purpose of feature extraction; the fully connected layer is located after the convolution layer or the pooling layer, and makes the extracted features nonlinear.

7. The device for judging the quality of equipment and instruments based on AI vision according to claim 6, characterized in that: The data management unit outputs the detection results of the intelligent detection unit, stores the data, and provides subsequent query and result tracing. The specific steps include: Design database model: Design the database model, including data tables and fields, based on requirements and data structure; Create a database table: Create a blank database as a machine vision database, add fields to be recorded in the table, including user information, inspection time, qualified and type fields, set the part number field as the primary key, and set the properties of each field. The user number and picture number type are text types, the picture is an OLE object type, and the inspection date is a date type; Image table field design: including image ID, image name, image path, category and other metadata fields; wherein the image ID is specifically a unique identifier used to distinguish each image data record, the image name is used to describe the name or identifier of the image, the image path is the storage path or relative path of the image file, the category is the category or label to which the image belongs, and the other metadata fields are other image-related fields added according to needs, including shooting date and employee; Insert data: When the image is judged to be correct, the information is inserted into the machine vision database; create an insert statement to insert the values ​​in the image table field design into the machine vision database respectively. Query and retrieve data: Filter and classify information by setting query conditions. Use the created table as the data source and set the three fields of quality inspector and test date in the data table as query conditions. The data to be queried can be searched and filtered through these three fields. Update operation: Write update statements, check and update the model regularly, and execute corresponding update operations when data in the database needs to be updated to adapt to new image classification requirements; Deletion operation: Write a deletion statement and execute the deletion operation for incorrect images based on the results of the model interpretation to delete the incorrect images from the database.

Images