An image recognition method for online detection sensor calibration of titanium alloys

Abstract

The invention discloses an image recognition method for calibration of a titanium alloy online detection sensor, which relates to the technical field of sensor calibration. An image acquisition device is deployed on a titanium alloy production line to acquire titanium alloy surface images in real time, image processing is performed on the titanium alloy surface images, and feature extraction is performed after converting them into standard images to be analyzed, and corresponding surface defect characteristic elements are extracted. The surface defect characteristic elements are screened to obtain key element information, and a surface defect recognition model for surface detection of titanium alloy is constructed according to the key element information. An online detection sensor is deployed on the titanium alloy production line, and the surface defect recognition model is associated with the online detection sensor for application. For the same titanium alloy workpiece on the production line, the actual detection result of the online detection sensor is compared with the recognition result of the surface defect recognition model, and it is determined whether to perform accuracy calibration on the corresponding online detection sensor according to the comparison result.

Description

Technical Field

[0001] The invention relates to the technical field of sensor calibration, in particular to an image recognition method for calibrating a titanium alloy online detection sensor. Background Art

[0002] Online detection sensors are sensors used to monitor products in production in real time. For example, they are devices that monitor and detect titanium alloy materials or components in real time. Such sensors can be integrated into titanium alloy production lines and can quickly detect the surface quality, size, shape and defects of titanium alloy workpieces without stopping the machine.

[0003] Since online detection sensors are often accompanied by accuracy abnormalities during use, and accuracy imbalance will cause false detection and missed detection of surface defects of titanium alloy workpieces on the production line, it has become an urgent task to screen out online detection sensors with accuracy abnormalities on the titanium alloy production line, calibrate their accuracy in time, reduce missed detection and false detection of surface defects of titanium alloy workpieces, and ensure the efficient and stable operation of the titanium alloy production line. Summary of the invention

[0004] In order to solve the above problems, the object of the present invention is to provide an image recognition method for calibrating a titanium alloy online detection sensor.

[0005] The purpose of the present invention can be achieved by the following technical solution: An image recognition method for calibrating a titanium alloy online detection sensor comprises the following steps:

[0006] Step S1: deploying an image acquisition device on the titanium alloy production line, acquiring a titanium alloy surface image in real time through the image acquisition device, and performing image processing on the titanium alloy surface image, thereby converting the titanium alloy surface image into a corresponding standard image to be analyzed;

[0007] Step S2: extracting features from the standard image to be analyzed, and then extracting corresponding surface defect feature elements, screening the surface defect feature elements through a machine learning algorithm, and then obtaining key element information;

[0008] Step S3: constructing a surface defect recognition model for surface detection of titanium alloy according to the key element information, deploying online detection sensors on the titanium alloy production line, and associating the surface defect recognition model with the online detection sensors;

[0009] Step S4: For the same titanium alloy workpiece on the production line, compare the actual detection result of the online detection sensor with the recognition result of the surface defect recognition model, and decide whether to perform accuracy calibration on the corresponding online detection sensor according to the comparison result.

[0010] Furthermore, an image acquisition device is deployed on the titanium alloy production line. The process of real-time acquisition of titanium alloy surface images by the image acquisition device includes:

[0011] Selecting several equipment deployment nodes on the titanium alloy production line, deploying image acquisition equipment at each equipment deployment node, obtaining several processing position coordinates of titanium alloy workpieces on the titanium alloy production line, and inputting them into several image acquisition devices, the image acquisition devices adjust their own shooting lenses, and thus make the shooting lenses focus on the titanium alloy workpieces corresponding to the processing position coordinates;

[0012] The image acquisition device is used to test shoot the titanium alloy workpiece, obtain the test image of the workpiece, and determine whether the test image of the workpiece meets the imaging quality standard. If so, the titanium alloy surface image of the titanium alloy workpiece is directly captured in real time by the image acquisition device. If not, the image acquisition device is repaired.

[0013] Furthermore, the process of performing image processing on the titanium alloy surface image and then converting the titanium alloy surface image into a corresponding standard image to be analyzed includes:

[0014] The image processing performed on the titanium alloy surface image includes image denoising, image enhancement and format calibration. The titanium alloy surface image is converted into a first operation image through image denoising, the first operation image is enhanced, and then the first operation image is converted into a second operation image, and the second operation image is format calibrated to generate a final standard image to be analyzed.

[0015] Furthermore, the process of extracting features from the standard image to be analyzed and then extracting corresponding surface defect feature elements includes:

[0016] The standard image to be analyzed is converted into a corresponding grayscale image, and a grayscale co-occurrence matrix corresponding to the grayscale image is constructed. Based on the grayscale co-occurrence matrix, the texture feature information corresponding to each standard image to be analyzed is extracted, and contour analysis is performed on the grayscale image to extract the contour of the defect area corresponding to the titanium alloy workpiece. The edge detection algorithm is used to detect the edge position of the contour of the defect area corresponding to the titanium alloy workpiece under the grayscale image to obtain the perimeter, area and rectangularity of the defect area, which are integrated as shape feature information, and all geometric structure features of the defect area corresponding to the titanium alloy workpiece are extracted as geometric structure feature information; the texture feature information, shape feature information and geometric structure feature information extracted from each standard image to be analyzed are integrated as the surface defect feature elements of the standard image to be analyzed.

[0017] Furthermore, the process of screening the characteristic elements of surface defects through machine learning algorithms and obtaining key element information includes:

[0018] The titanium alloy surface images taken at historical time points are analyzed using a machine learning algorithm to obtain a texture feature mapping information set, a shape feature mapping information set, and a geometric structure feature mapping information set for determining the surface defect types of the titanium alloy workpiece;

[0019] The texture feature mapping information set is used to characterize several types of texture feature information corresponding to the titanium alloy workpiece when there are surface defects;

[0020] The shape feature mapping information set is used to characterize several types of shape feature information corresponding to the titanium alloy workpiece when there are surface defects;

[0021] The geometric structure feature mapping information set is used to characterize several types of geometric structure feature information corresponding to the titanium alloy workpiece when there are surface defects;

[0022] Determine in sequence whether the surface defect feature elements of each titanium alloy workpiece are successfully and synchronously mapped to the texture feature mapping information set, the shape feature mapping information set, and the geometric structure feature mapping information set;

[0023] If yes, the corresponding surface defect characteristic elements are screened as key element information;

[0024] If not, the corresponding surface defect characteristic elements are eliminated.

[0025] Furthermore, the process of determining whether the surface defect characteristic elements of each titanium alloy workpiece are successfully synchronously mapped includes:

[0026] Decomposing the surface defect characteristic elements of the current titanium alloy workpiece into corresponding texture characteristic information, shape characteristic information and geometric structure characteristic information;

[0027] If the texture feature information belongs to a certain category in the texture feature mapping information set, it is determined that the texture feature information mapping is successful, otherwise, it is determined that the texture feature information mapping fails;

[0028] If the shape feature information belongs to a certain category in the shape feature mapping information set, it is determined that the shape feature information mapping is successful; otherwise, it is determined that the shape feature information mapping is unsuccessful;

[0029] If the geometric structure feature information belongs to a certain category in the geometric structure feature mapping information set, it is determined that the mapping of the geometric structure feature information is successful; otherwise, it is determined that the mapping of the geometric structure feature information is unsuccessful;

[0030] When the texture feature information, the shape feature information, and the geometric structure feature information are respectively mapped successfully, it is determined that the surface defect feature elements are successfully mapped synchronously; otherwise, it is determined that the surface defect feature elements are failed to be mapped synchronously.

[0031] Furthermore, a surface defect recognition model for surface detection of titanium alloy is constructed based on key element information, an online detection sensor is deployed on a titanium alloy production line, and the process of associating the surface defect recognition model with the online detection sensor for application includes:

[0032] Divide all key element information into modeling data and model iteration data;

[0033] Divide the model iteration data into training set, test set and validation set;

[0034] The modeling data is processed through a deep learning algorithm to build an initial surface defect recognition model. The surface defect recognition model is used to perform surface inspection on titanium alloys, and then the surface defect types of titanium alloy workpieces are obtained. The surface defect types include crack defects, hole defects, and scratch defects.

[0035] The initial surface defect recognition model is trained through the training set, and the output results after each model training are obtained through the test set test. The output results are verified through the validation set, and the model prediction accuracy of the surface defect recognition model is obtained through real-time verification;

[0036] Set the passing threshold for the model’s prediction accuracy;

[0037] If the model prediction accuracy exceeds the passing threshold, the final surface defect recognition model is constructed; otherwise, the surface defect recognition model continues to be trained;

[0038] Deploy online detection sensors on the titanium alloy production line, configure the data ports of the online detection sensors, and associate the surface defect recognition model with the corresponding online detection sensors through the data port application.

[0039] Furthermore, the process of comparing the actual detection result of the online detection sensor with the recognition result of the surface defect recognition model and determining whether to perform accuracy calibration on the corresponding online detection sensor according to the comparison result includes:

[0040] The actual detection results of each online detection sensor on the production line on the titanium alloy workpiece are obtained in sequence. For the same titanium alloy workpiece, the recognition result of the surface defect recognition model on the current titanium alloy workpiece is obtained, and the actual detection result of the online detection sensor detecting the titanium alloy workpiece is compared with the recognition result of the surface defect recognition model;

[0041] If the actual detection result of the online detection sensor corresponds to the recognition result of the surface defect recognition model and the surface defect type of the titanium alloy workpiece is the same, it means that the accuracy of the online detection sensor is normal, and the accuracy calibration of the current online detection sensor is not performed;

[0042] If the actual detection result of the online detection sensor corresponds to the recognition result of the surface defect recognition model and the surface defect type of the titanium alloy workpiece is not the same, it means that the accuracy of the online detection sensor is abnormal, and the accuracy of the current online detection sensor is calibrated.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows: by deploying image acquisition equipment on the titanium alloy production line to collect real-time titanium alloy surface images, converting the images into standard images to be analyzed after image processing, performing feature extraction on the standard images to be analyzed to extract surface defect characteristic elements, and obtaining key element information after screening, constructing a surface defect recognition model for surface detection of titanium alloys according to the key element information, deploying online detection sensors on the titanium alloy production line, comparing the actual detection results of the online detection sensors with the recognition results of the surface defect recognition model for the same titanium alloy workpiece, identifying the online detection sensors with inconsistent comparison results as having accuracy abnormalities, achieving timely discovery of abnormal online detection sensors, and performing accuracy calibration on the online detection sensors with accuracy abnormalities, reducing missed detection and false detection of surface defects of titanium alloy workpieces, and effectively ensuring the efficient and stable operation of the titanium alloy production line. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0045] like Figure 1 As shown, an image recognition method for calibrating a titanium alloy online detection sensor comprises the following steps:

[0046] Step S1: deploying an image acquisition device on the titanium alloy production line, acquiring a titanium alloy surface image in real time through the image acquisition device, and performing image processing on the titanium alloy surface image, thereby converting the titanium alloy surface image into a corresponding standard image to be analyzed;

[0047] Step S2: extracting features from the standard image to be analyzed, and then extracting corresponding surface defect feature elements, screening the surface defect feature elements through a machine learning algorithm, and then obtaining key element information;

[0048] Step S3: constructing a surface defect recognition model for surface detection of titanium alloy according to the key element information, deploying online detection sensors on the titanium alloy production line, and associating the surface defect recognition model with the online detection sensors;

[0049] Step S4: For the same titanium alloy workpiece on the production line, compare the actual detection result of the online detection sensor with the recognition result of the surface defect recognition model, and decide whether to perform accuracy calibration on the corresponding online detection sensor according to the comparison result.

[0050] It should be further explained that, in a specific implementation, an image acquisition device is deployed on a titanium alloy production line, and the process of real-time acquisition of titanium alloy surface images by the image acquisition device includes:

[0051] Select several equipment deployment nodes on the titanium alloy production line, and number the several equipment deployment nodes, record the number as i, then i = 1, 2, 3, ..., n, where n is a natural number greater than 0, and deploy the image acquisition device at each equipment deployment node;

[0052] Determine whether the real-time working parameters of each image acquisition device meet the preset standard parameters;

[0053] If so, no action is taken;

[0054] If not, debug the image acquisition device whose real-time working parameters do not meet the standard parameters;

[0055] Acquire several processing position coordinates of titanium alloy workpieces on the titanium alloy production line, and input the processing position coordinates into the image acquisition devices corresponding to several equipment deployment nodes, so that the image acquisition devices adjust their own shooting lenses, thereby focusing the shooting lenses and the titanium alloy workpieces corresponding to the processing position coordinates;

[0056] The image acquisition device takes a test image of the titanium alloy workpiece to obtain the test image of the workpiece, and determines whether the test image of the workpiece meets the imaging quality standard. The imaging quality standard includes the judgment of the imaging clarity and the imaging completeness;

[0057] Set the clarity threshold, denoted as Qx;

[0058] Set the completeness threshold, denoted as Wz;

[0059] The values ​​of the imaging clarity and imaging integrity of the workpiece test image are recorded as d1 and d2 respectively. When d1≥Qx, the imaging clarity of the workpiece test image meets the requirements. When d1<Qx, the imaging clarity does not meet the requirements. When d2≥Wz, the imaging integrity of the workpiece test image meets the requirements. When d2<Wz, the imaging integrity does not meet the requirements.

[0060] When the image clarity or image integrity of the workpiece test image does not meet the requirements, it is judged that the workpiece test image does not meet the imaging quality standard, and the corresponding image acquisition equipment is repaired. After the repair is completed, the titanium alloy surface image of the titanium alloy workpiece is collected in real time through the image acquisition equipment;

[0061] Otherwise, the titanium alloy surface image of the titanium alloy workpiece is directly acquired in real time through the image acquisition device.

[0062] It should be further explained that, in a specific implementation, the process of performing image processing on the titanium alloy surface image and then converting the titanium alloy surface image into a corresponding standard image to be analyzed includes:

[0063] The image processing performed on the titanium alloy surface image includes image denoising, image enhancement and format calibration. The titanium alloy surface image is converted into a first operation image by image denoising, the first operation image is enhanced, and then the first operation image is converted into a second operation image. The second operation image is format calibrated to generate a final standard image to be analyzed.

[0064] The contents of the image denoising are as follows:

[0065] A unit area image is set, and the titanium alloy surface image is divided into a number of sub-denoising areas according to the image size of the unit area image. For each sub-denoising area, it is divided into a number of corresponding noise filtering points according to the pixel cell size. The sub-denoising area is a square, and the noise average value of the noise filtering point corresponding to the center point of the square is selected as the target noise value;

[0066] The noise values ​​of the noise-filtered points other than the center point in the sub-denoising area are replaced with the target noise value, and the above operation is repeated to complete the image denoising of each sub-denoising area, and the several sub-denoising areas that have completed the image denoising are integrated into a first operation image;

[0067] The contents of the image enhancement are as follows:

[0068] The first operation image is composed of several pixel channels. The color moment of each pixel channel is counted. The color moment includes the first-order color moment, the second-order color moment and the third-order color moment. The formulas of the first-order color moment, the second-order color moment and the third-order color moment are respectively recorded as , as well as ;

[0069] Then there are as follows:

[0070] ;

[0071] For each pixel channel, the average value of its pixel value is calculated to obtain the first-order color moment, which can reflect the brightness information of the image. The pixel channels are numbered as j, and the numbering is j=1, 2, 3, ..., N, where N is an integer greater than 0 and N represents the total number of pixel channels. Represents the pixel value of the jth pixel channel;

[0072] ;

[0073] Among them, the pixel value variance of each pixel channel is calculated to obtain the second-order color moment, which is used to reflect the discrete degree of image color;

[0074] ;

[0075] Among them, the pixel value skewness of each pixel channel is calculated to obtain the third-order color moment, which is used to describe the symmetry of color distribution;

[0076] If a pixel channel of the first operation image has abnormal brightness, abnormal color discreteness, and asymmetric color distribution, the corresponding pixel channel is adjusted until the brightness, color discreteness, and color distribution of the current pixel channel meet the requirements, otherwise, no operation is performed; when the brightness, color discreteness, and color distribution of all pixel channels of the first operation image meet the requirements, the first operation image is converted into the second operation image;

[0077] The content of the format calibration is: comparing whether the image format and image size of the current second operation image meet the respective preset standards, if so, converting the second operation image into the final standard image to be analyzed, if not, adjusting the image format and image size of the second operation image until the final standard image to be analyzed is generated.

[0078] It should be further explained that, in a specific implementation manner, the process of extracting features from the standard image to be analyzed and then extracting corresponding surface defect feature elements includes:

[0079] Convert the standard image to be analyzed into a corresponding grayscale image, construct a grayscale co-occurrence matrix corresponding to the grayscale image, extract the texture feature information corresponding to each standard image to be analyzed based on the grayscale co-occurrence matrix, and use the grayscale image to perform contour analysis to extract the contour of the defect area corresponding to the titanium alloy workpiece, and use the edge detection algorithm to detect the edge position of the contour of the defect area corresponding to the titanium alloy workpiece under the grayscale image to obtain the perimeter, area and rectangularity of the defect area, and integrate them as shape feature information to extract all geometric structure features of the defect area corresponding to the titanium alloy workpiece and use them as geometric structure feature information. The geometric structure feature information is used to represent all lines, angles formed by lines and intersections formed by lines in the defect area corresponding to the titanium alloy workpiece;

[0080] The texture feature information, shape feature information and geometric structure feature information extracted from each standard image to be analyzed are integrated as the surface defect feature elements of the standard image to be analyzed.

[0081] It should be further explained that, in a specific implementation, the process of screening the characteristic elements of surface defects by a machine learning algorithm to obtain key element information includes:

[0082] The titanium alloy surface images taken at historical time points are analyzed using a machine learning algorithm to obtain a texture feature mapping information set, a shape feature mapping information set, and a geometric structure feature mapping information set for determining the surface defect types of the titanium alloy workpiece;

[0083] The texture feature mapping information set is used to characterize several types of texture feature information corresponding to the titanium alloy workpiece when there are surface defects;

[0084] The shape feature mapping information set is used to characterize several types of shape feature information corresponding to the titanium alloy workpiece when there are surface defects;

[0085] The geometric structure feature mapping information set is used to characterize several types of geometric structure feature information corresponding to the titanium alloy workpiece when there are surface defects;

[0086] Determine in sequence whether the surface defect feature elements of each titanium alloy workpiece are successfully and synchronously mapped to the texture feature mapping information set, the shape feature mapping information set, and the geometric structure feature mapping information set;

[0087] If yes, the corresponding surface defect characteristic elements are screened as key element information;

[0088] If not, the corresponding surface defect characteristic elements are eliminated.

[0089] The contents of judging whether the surface defect feature elements are successfully mapped to the texture feature mapping information set, the shape feature mapping information set, and the geometric structure feature mapping information set are as follows:

[0090] Decomposing the surface defect characteristic elements of the current titanium alloy workpiece into corresponding texture characteristic information, shape characteristic information and geometric structure characteristic information;

[0091] If the texture feature information belongs to a certain category in the texture feature mapping information set, it is determined that the texture feature information mapping is successful, otherwise, it is determined that the texture feature information mapping fails;

[0092] If the shape feature information belongs to a certain category in the shape feature mapping information set, it is determined that the shape feature information mapping is successful; otherwise, it is determined that the shape feature information mapping is unsuccessful;

[0093] If the geometric structure feature information belongs to a certain category in the geometric structure feature mapping information set, it is determined that the mapping of the geometric structure feature information is successful; otherwise, it is determined that the mapping of the geometric structure feature information is unsuccessful;

[0094] When the texture feature information, the shape feature information, and the geometric structure feature information are respectively mapped successfully, it is determined that the surface defect feature elements are successfully mapped synchronously; otherwise, it is determined that the surface defect feature elements are failed to be mapped synchronously.

[0095] It should be further explained that, in a specific implementation, a surface defect recognition model for surface detection of titanium alloy is constructed according to key element information, an online detection sensor is deployed on a titanium alloy production line, and the process of associating the surface defect recognition model with the online detection sensor for application includes:

[0096] Divide all key element information into modeling data and model iteration data;

[0097] Divide the model iteration data into training set, test set and validation set;

[0098] The modeling data is processed by a deep learning algorithm to construct an initial surface defect recognition model. The surface defect recognition model is used to perform surface inspection on titanium alloys to obtain the surface defect types of titanium alloy workpieces, including crack defects, hole defects, and scratch defects.

[0099] The initial surface defect recognition model is trained through the training set, and the output results after each model training are obtained through the test set test. The output results are verified through the validation set, and the model prediction accuracy of the surface defect recognition model is obtained through real-time verification;

[0100] Set the passing threshold for the model’s prediction accuracy;

[0101] If the model prediction accuracy exceeds the passing threshold, the final surface defect recognition model is constructed; otherwise, the surface defect recognition model continues to be trained.

[0102] Deploy online detection sensors on the titanium alloy production line, and debug the online detection sensors until the working parameters of the online detection sensors meet the preset sensor specification working parameters, and detect the titanium alloy workpieces on the titanium alloy production line through the online detection sensors;

[0103] Configure the data port of the online detection sensor, associate the surface defect recognition model used for surface detection of titanium alloy workpieces with the corresponding online detection sensor through the data port application, and test whether the data communication of the data port is normal;

[0104] If so, no action is taken;

[0105] If not, reconfigure the data port until the data communication of the data port is normal.

[0106] It should be further explained that, in a specific implementation manner, for the same titanium alloy workpiece on the production line, the process of comparing the actual detection result of the online detection sensor with the recognition result of the surface defect recognition model and determining whether to perform accuracy calibration on the corresponding online detection sensor according to the comparison result includes:

[0107] The actual detection results of each online detection sensor on the production line on the titanium alloy workpiece are obtained in sequence. For the same titanium alloy workpiece, the recognition result of the surface defect recognition model on the current titanium alloy workpiece is obtained, and the actual detection result of the online detection sensor detecting the titanium alloy workpiece is compared with the recognition result of the surface defect recognition model;

[0108] If the actual detection result of the online detection sensor corresponds to the recognition result of the surface defect recognition model and the surface defect type of the titanium alloy workpiece is the same, it means that the accuracy of the online detection sensor is normal, and the accuracy calibration of the current online detection sensor is not performed;

[0109] If the actual detection result of the online detection sensor corresponds to the recognition result of the surface defect recognition model and the surface defect type of the titanium alloy workpiece is different, it means that the accuracy of the online detection sensor is abnormal, and the accuracy of the current online detection sensor is calibrated;

[0110] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An image recognition method for calibrating a titanium alloy online detection sensor, characterized in that: The following steps are involved: Step S1: deploying an image acquisition device on the titanium alloy production line, acquiring a titanium alloy surface image in real time through the image acquisition device, and performing image processing on the titanium alloy surface image, thereby converting the titanium alloy surface image into a corresponding standard image to be analyzed; Step S2: extracting features from the standard image to be analyzed, and then extracting corresponding surface defect feature elements, screening the surface defect feature elements through a machine learning algorithm, and then obtaining key element information; Step S3: constructing a surface defect recognition model for surface detection of titanium alloy according to the key element information, deploying online detection sensors on the titanium alloy production line, and associating the surface defect recognition model with the online detection sensors; Step S4: for the same titanium alloy workpiece on the production line, compare the actual detection result of the online detection sensor with the recognition result of the surface defect recognition model, and decide whether to perform accuracy calibration on the corresponding online detection sensor according to the comparison result; The process of constructing a surface defect recognition model for surface inspection of titanium alloy based on key element information, deploying online inspection sensors on the titanium alloy production line, and associating the surface defect recognition model with the online inspection sensors includes: Divide all key element information into modeling data and model iteration data; Divide the model iteration data into training set, test set and validation set; The modeling data is processed through a deep learning algorithm to construct an initial surface defect recognition model. The surface defect recognition model is used to perform surface inspection on titanium alloys, and then the surface defect types of titanium alloy workpieces are obtained. The surface defect types include crack defects, hole defects, and scratch defects. The initial surface defect recognition model is trained through the training set, and the output results after each model training are obtained through the test set test. The output results are verified through the validation set, and the model prediction accuracy of the surface defect recognition model is obtained through real-time verification; Set the passing threshold for the model’s prediction accuracy; If the model prediction accuracy exceeds the passing threshold, the final surface defect recognition model is constructed; otherwise, the surface defect recognition model continues to be trained; Deploy online detection sensors on the titanium alloy production line, configure the data ports of the online detection sensors, and associate the surface defect recognition model with the corresponding online detection sensors through the data port application.

2. The image recognition method for calibrating a titanium alloy online detection sensor according to claim 1 is characterized in that: The image acquisition equipment is deployed on the titanium alloy production line. The process of real-time acquisition of titanium alloy surface images by the image acquisition equipment includes: Selecting several equipment deployment nodes on the titanium alloy production line, deploying image acquisition equipment at each equipment deployment node, obtaining several processing position coordinates of titanium alloy workpieces on the titanium alloy production line, and inputting them into several image acquisition devices, the image acquisition devices adjust their own shooting lenses, and thus make the shooting lenses focus on the titanium alloy workpieces corresponding to the processing position coordinates; The image acquisition device is used to test shoot the titanium alloy workpiece, obtain the test image of the workpiece, and determine whether the test image of the workpiece meets the imaging quality standard. If so, the titanium alloy surface image of the titanium alloy workpiece is directly captured in real time by the image acquisition device. If not, the image acquisition device is repaired.

3. The image recognition method for calibrating a titanium alloy online detection sensor according to claim 2 is characterized in that: The process of performing image processing on the titanium alloy surface image and then converting the titanium alloy surface image into a corresponding standard image to be analyzed includes: The image processing performed on the titanium alloy surface image includes image denoising, image enhancement and format calibration. The titanium alloy surface image is converted into a first operation image through image denoising, the first operation image is enhanced, and then the first operation image is converted into a second operation image, and the second operation image is format calibrated to generate a final standard image to be analyzed.

4. The image recognition method for calibrating a titanium alloy online detection sensor according to claim 3 is characterized in that: The process of extracting features from the standard image to be analyzed and then extracting the corresponding surface defect feature elements includes: The standard image to be analyzed is converted into a corresponding grayscale image, and a grayscale co-occurrence matrix corresponding to the grayscale image is constructed. Based on the grayscale co-occurrence matrix, the texture feature information corresponding to each standard image to be analyzed is extracted, and contour analysis is performed on the grayscale image to extract the contour of the defect area corresponding to the titanium alloy workpiece. The edge detection algorithm is used to detect the edge position of the contour of the defect area corresponding to the titanium alloy workpiece under the grayscale image to obtain the perimeter, area and rectangularity of the defect area, which are integrated as shape feature information, and all geometric structure features of the defect area corresponding to the titanium alloy workpiece are extracted as geometric structure feature information; the texture feature information, shape feature information and geometric structure feature information extracted from each standard image to be analyzed are integrated as the surface defect feature elements of the standard image to be analyzed.

5. The image recognition method for calibrating a titanium alloy online detection sensor according to claim 4 is characterized in that: The process of screening the characteristic elements of surface defects through machine learning algorithms and obtaining key element information includes: The data of titanium alloy surface images taken at historical time nodes are analyzed by machine learning algorithms, thereby obtaining a texture feature mapping information set, a shape feature mapping information set, and a geometric structure feature mapping information set for determining the surface defect types of titanium alloy workpieces; The texture feature mapping information set is used to characterize several types of texture feature information corresponding to the titanium alloy workpiece when there are surface defects; The shape feature mapping information set is used to characterize several types of shape feature information corresponding to the titanium alloy workpiece when there are surface defects; The geometric structure feature mapping information set is used to characterize several types of geometric structure feature information corresponding to the titanium alloy workpiece when there are surface defects; Determine in sequence whether the surface defect feature elements of each titanium alloy workpiece are successfully and synchronously mapped to the texture feature mapping information set, the shape feature mapping information set, and the geometric structure feature mapping information set; If yes, the corresponding surface defect characteristic elements are screened as key element information; If not, the corresponding surface defect characteristic elements are eliminated.

6. The image recognition method for calibrating a titanium alloy online detection sensor according to claim 5, characterized in that: The process of judging whether the surface defect characteristic elements of each titanium alloy workpiece are successfully synchronously mapped includes: Decomposing the surface defect characteristic elements of the current titanium alloy workpiece into corresponding texture characteristic information, shape characteristic information and geometric structure characteristic information; If the texture feature information belongs to a certain category in the texture feature mapping information set, it is determined that the texture feature information mapping is successful, otherwise, it is determined that the texture feature information mapping fails; If the shape feature information belongs to a certain category in the shape feature mapping information set, it is determined that the shape feature information mapping is successful; otherwise, it is determined that the shape feature information mapping is unsuccessful; If the geometric structure feature information belongs to a certain category in the geometric structure feature mapping information set, it is determined that the mapping of the geometric structure feature information is successful; otherwise, it is determined that the mapping of the geometric structure feature information is unsuccessful; When the texture feature information, the shape feature information, and the geometric structure feature information are respectively mapped successfully, it is determined that the surface defect feature elements are successfully mapped synchronously; otherwise, it is determined that the surface defect feature elements are failed to be mapped synchronously.

7. The image recognition method for calibrating a titanium alloy online detection sensor according to claim 1, characterized in that: The process of comparing the actual detection results of the online detection sensor with the recognition results of the surface defect recognition model and determining whether to perform accuracy calibration on the corresponding online detection sensor according to the comparison results includes: The actual detection results of each online detection sensor on the production line on the titanium alloy workpiece are obtained in sequence. For the same titanium alloy workpiece, the recognition result of the surface defect recognition model on the current titanium alloy workpiece is obtained, and the actual detection result of the online detection sensor detecting the titanium alloy workpiece is compared with the recognition result of the surface defect recognition model; If the actual detection result of the online detection sensor corresponds to the recognition result of the surface defect recognition model and the surface defect type of the titanium alloy workpiece is the same, it means that the accuracy of the online detection sensor is normal, and the accuracy calibration of the current online detection sensor is not performed; If the actual detection result of the online detection sensor corresponds to the recognition result of the surface defect recognition model and the surface defect type of the titanium alloy workpiece is not the same, it means that the accuracy of the online detection sensor is abnormal, and the accuracy of the current online detection sensor is calibrated.

Images