Horizontal continuous casting cutting screening method, device and system and storage medium

Abstract

The invention relates to the technical field of motor testing, and provides a horizontal continuous casting cutting screening method, device and system and a storage medium. The method comprises the steps that when a bar is detected, an image of the bar is captured according to a preset time interval; performing fitting splicing on the images sorted according to the time interval by adopting an RANSAC algorithm, and performing shadow preprocessing; obtaining a preprocessed image construction data set, and generating adversarial network fitting; acquiring an image which is successfully fitted by the adversarial network, performing edge extraction and measurement on the image, comparing edge data of the image with a system database, and judging whether the image is a bar section which does not meet the standard or not; and cutting and removing the bar sections which do not meet the standard, and cutting and collecting the bar sections which meet the standard according to the specified length. By means of the technical scheme, the problems that in the prior art, the automation degree is low, manufacturing difficulty is large, and defective products cannot be recognized are solved.

Description

Technical Field

[0001] The present invention relates to the field of motor testing technology, and in particular to a horizontal continuous casting cutting and screening method, device, system and storage medium. Background Art

[0002] The horizontal continuous casting cutting and screening device is a device used to improve the speed and efficiency of bar production. The device has the advantages of high production efficiency, strong finished product screening ability, and automatic adjustment of material size, and has shown broad application prospects in metallurgy, manufacturing and other fields. However, the existing sawing and screening systems still have many limitations in the application of bar production and cannot meet the needs of efficient and automated production.

[0003] At present, most companies still rely on the traditional method of manual sawing and visual screening for bar processing and quality inspection. This method is not only inefficient, but also easily affected by human factors, resulting in frequent problems such as missed inspections and misjudgments. In addition, during the manual screening process, workers are in a dangerous working environment for a long time, which poses a high risk of occupational safety hazards. Especially when operating in high temperature or dangerous areas, workers face greater health and safety risks.

[0004] Another key issue is that the sampling and screening method used in traditional production cannot fully detect every bar, which may lead to unqualified products entering the market and reduce the company's product quality reputation. In addition, the manual setting of material dimensions not only has errors, but also requires frequent adjustments, which is difficult to adapt to the flexible needs of modern production for multi-specification products. Summary of the invention

[0005] The present invention proposes a more efficient, accurate and safe horizontal continuous casting cutting and screening method, device, system and storage medium, which improves production efficiency, significantly reduces labor costs and safety risks, and ensures the stability and consistency of product quality.

[0006] The technical solution of the present invention is as follows:

[0007] A horizontal continuous casting cutting and screening method, comprising:

[0008] When the bar is detected, an image of the bar is captured at each time interval according to a preset time interval;

[0009] The RANSAC algorithm is used to fit and splice the images sorted by time intervals, and shadow preprocessing is performed on the spliced ​​images.

[0010] Obtain the preprocessed images to construct a data set, generate an adversarial network fitting based on the image data in the data set, and refit the image data where the adversarial network fitting fails until the adversarial network fitting succeeds;

[0011] Obtain an image that has been successfully fitted by the adversarial network, extract and measure the edge of the image, compare the edge data of the image with the system database, and determine whether it is a bar segment that does not meet the standards;

[0012] The bar segments that do not meet the standards are cut and removed, and the bar segments that meet the standards are cut into specified lengths and collected.

[0013] Furthermore, the step of fitting and splicing the images sorted according to the time interval using the RANSAC algorithm includes:

[0014] Randomly select feature points as internal points and determine the transformation matrix model through calculation;

[0015] Repeat step 1 continuously, traverse all feature points, and establish multiple transformation matrices;

[0016] The number of feature points in the transformation matrix is ​​counted, and the matrix with the largest number of feature points is the optimal transformation matrix.

[0017] Furthermore, the transformation matrix of the transformation matrix model determined by calculation is a homography matrix, and the matrix parameters are obtained through 4 sets of feature points and 8 linear equations. The specific expressions are as follows:

[0018]

[0019] Assume that the probability of a point selected from the image feature point set being an internal point is P, then the probability formula is:

[0020] P=λ n ,

[0021]

[0022] In the formula, N represents the total number of inliers, M represents the total number of feature points, λ represents the probability that all randomly selected points are inliers, and λ n It means iterating n times. The number of iterations n needs to take the logarithm of the formula. The derived formula is:

[0023]

[0024] The RANSAC algorithm is used to determine the internal points, the feature point set is traversed, the optimal transformation is calculated through the transformation matrix model, and finally the stitching of multiple images is completed.

[0025] Furthermore, the edge data of the image is compared with the system database to determine whether it is a bar segment that does not meet the standards based on the structural similarity of the compared image and the system database image and the peak signal-to-noise ratio of the compared image and the system database image.

[0026] Furthermore, the calculation formula of the structural similarity is:

[0027]

[0028] In the formula, μ is the value, which represents the brightness estimation; σ is the variance, which represents the contrast estimation; σ xy is the covariance, indicating the structural similarity estimation, C1=(k1L) 2 , C2=(k2L) 2 , k1 and k2 are 0.01 and 0.02 respectively, and L is the dynamic value range of the pixel.

[0029] Furthermore, the calculation formula of the peak signal-to-noise ratio is:

[0030]

[0031] Where, MAE is the noise energy signal value, It is the square of the maximum value of the pixel color.

[0032] A horizontal continuous casting cutting and screening device, comprising:

[0033] The image acquisition module, when detecting the bar material, captures the image of the bar material at each time interval according to a preset time interval;

[0034] The image stitching module uses the RANSAC algorithm to fit and stitch images sorted by time intervals, and performs shadow preprocessing on the stitched images;

[0035] The adversarial network fitting module obtains the preprocessed image to construct a data set, generates an adversarial network fitting based on the image data in the data set, and re-fits the image data that fails to fit the adversarial network until the adversarial network fitting succeeds;

[0036] The edge detection module obtains the image that has been successfully fitted by the adversarial network, extracts and measures the edge of the image, compares the edge data of the image with the system database, and determines whether it is a bar segment that does not meet the standards;

[0037] The cutting and separation module cuts and removes the bar segments that do not meet the standards, and cuts and collects the bar segments that meet the standards into specified lengths.

[0038] Furthermore, the image acquisition module includes a laser sensor and a high-definition camera, wherein the laser sensor is used to detect the position and length of the bar material, and the high-definition camera is used to acquire the image of the bar material;

[0039] The cutting and separation module includes a robotic arm, a cutting box, a front clamping device, and a rear clamping device. The robotic arm is used to pick up bars that meet the standards or do not meet the standards and place them in corresponding positions. The cutting box includes a tool moving device and a cutting tool. The tool moving device is used to drive the cutting tool to move. The cutting tool is used to cut the bars. The front clamping device and the rear clamping device are used to clamp the bars so that the cutting tool can perform stable cutting.

[0040] A horizontal continuous casting cutting and screening system, the system comprising:

[0041] one or more memories for storing instructions; and

[0042] One or more processors, configured to call and execute the instructions from the memory to execute the method described in any one of the preceding embodiments.

[0043] A computer-readable storage medium, the computer-readable storage medium comprising:

[0044] The program, when the program is executed by a processor, the method as described in any one of the above embodiments is executed.

[0045] The working principle and beneficial effects of the present invention are:

[0046] 1. The present invention realizes a fully automatic bar material detection and screening process through image capture, algorithm splicing, adversarial network fitting and other technologies, which greatly reduces the intervention of manual operation and reduces labor costs. At the same time, it reduces the error and missed detection rate caused by human judgment, and improves production efficiency and finished product qualification rate.

[0047] 2. The present invention uses the RANSAC algorithm for image stitching and shadow preprocessing, and combines it with adversarial network fitting to accurately detect subtle defects and irregular shapes on the surface of the bar. Compared with traditional visual screening and sampling screening, the present invention can conduct a comprehensive inspection of each bar, effectively improve the screening accuracy, and prevent unqualified products from entering the market.

[0048] 3. When the adversarial network fails to fit, the system will automatically refit until the fitting is successful. This adaptive intelligent adjustment process ensures the accuracy of image quality and detection results. Compared with traditional methods, the present invention can more flexibly cope with various complex situations in production and reduce false detection.

[0049] 4. The present invention can accurately obtain the edge data of the bar material by extracting and measuring the edge of the image, and compare it with the standard data in the system database in real time. The traditional method usually relies on manual setting and measurement, which has large errors. The present invention can greatly improve the accuracy and consistency of dimensional measurement through automated image processing technology.

[0050] 5. The present invention can automatically judge the bar segments that do not meet the standards and cut and remove them immediately, and cut and collect the bar segments that meet the standards according to the set specifications. This fully automated cutting and screening process greatly improves production efficiency, reduces the risks and errors of manual cutting, and ensures high consistency and high quality of the finished bar products. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0052] Figure 1 is a flow chart of the present invention;

[0053] Figure 2 It is a structural schematic diagram of a horizontal continuous casting cutting and screening device in the present invention;

[0054] Figure 3 for Figure 2 Schematic diagram of the internal structure of the cutting box;

[0055] In the figure: 1. Robotic arm; 2. Front clamping device; 3. Cutting box; 4. Rear clamping device; 5. High-definition camera; 6. Bar material; 7. Laser sensor; 8. Tool moving device; 9. Cutting tool; 10. Feeding device; 11. Feeding port. DETAILED DESCRIPTION

[0056] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] like Figure 1 As shown, this embodiment proposes a horizontal continuous casting cutting and screening method, comprising:

[0058] When the bar is detected, the image of the bar is captured at each time interval according to the preset time interval, realizing continuous monitoring of the bar and ensuring that the image data capture covers the entire length and process of the bar. At the same time, by capturing images at fixed time intervals, the integrity and timing of the image data are guaranteed, providing a reliable basis for subsequent image stitching and processing.

[0059] The RANSAC algorithm is used to fit and stitch images sorted by time intervals, and shadow preprocessing is performed on the stitched images; the RANSAC algorithm can deal with noise and abnormal data when processing images to ensure the accuracy of image stitching. The RANSAC algorithm stitches images collected at multiple time points into a complete bar image, which can avoid the problem that a single image cannot cover the entire bar. At the same time, shadow preprocessing removes light interference in the image, improves the quality of subsequent image analysis, and effectively reduces the misjudgment rate.

[0060] Obtain the preprocessed images to construct a data set, generate an adversarial network fitting based on the image data in the data set, and refit the image data where the adversarial network fitting fails until the adversarial network fitting succeeds; by performing image fitting through generative adversarial networks (GANs), details in the image can be identified and optimized at a higher level. The adversarial network can be continuously adjusted and optimized through self-learning to ensure that the generated image data is as close to the real data as possible. This embodiment allows automatic refitting when fitting fails, ensuring that the quality of the image finally used for analysis meets the requirements, thereby avoiding misjudgment due to image quality problems.

[0061] Get the image that the adversarial network has successfully fitted, extract and measure the edge of the image, compare the edge data of the image with the system database, and determine whether it is a bar segment that does not meet the standards; through edge extraction technology, accurately obtain the shape and size data of the bar to ensure fine measurement of the bar. Compare the edge data with the standard value in the system database to accurately determine whether the quality of the bar meets the requirements. This process reduces the error of human detection, while improving the accuracy and consistency of measurement, ensuring the effective identification of unqualified bar segments.

[0062] Cut and remove the bar segments that do not meet the standards, and cut and collect the bar segments that meet the standards to a specified length; through automated cutting and removal, unqualified bar segments can be processed immediately to prevent them from entering the downstream process. This function makes the entire production process more efficient and reduces the scrap rate. In addition, precise length cutting and automatic collection of bar segments that meet the standards ensure product consistency and efficient output, further improving the automation and efficiency of the production line.

[0063] In this embodiment, the RANSAC (Random Sample Consensus) algorithm is an iterative algorithm for correctly estimating the parameters of a mathematical model from a set of data containing outliers. Outliers are image noise, which refers to data that does not conform to the model; inliers are data that constitute the model parameters. The RANSAC algorithm does not fit the line with the minimum distance between feature points, but counts the number of inliers, and the fitted line with the most inliers is the optimal line. The steps of fitting and splicing images sorted by time intervals using the RANSAC algorithm include:

[0064] Randomly select feature points as internal points and determine the transformation matrix model through calculation;

[0065] Repeat step 1 continuously, traverse all feature points, and establish multiple transformation matrices;

[0066] The number of feature points in the transformation matrix is ​​counted, and the matrix with the largest number of feature points is the optimal transformation matrix.

[0067] In this embodiment, the transformation matrix of the transformation matrix model is determined by calculation to be a homography matrix, and the matrix parameters are obtained through 4 sets of feature points and 8 linear equations. The specific expressions are as follows:

[0068]

[0069] Assume that the probability of a point selected from the image feature point set being an internal point is P, then the probability formula is:

[0070] P=λ n ,

[0071]

[0072] In the formula, N represents the total number of inliers, M represents the total number of feature points, λ represents the probability that all randomly selected points are inliers, and λ n It means iterating n times. The number of iterations n needs to take the logarithm of the formula. The derived formula is:

[0073]

[0074] The RANSAC algorithm is used to determine the internal points, the feature point set is traversed, the optimal transformation is calculated through the transformation matrix model, and finally the stitching of multiple images is completed.

[0075] In this embodiment, image preprocessing refers to preprocessing the image by image noise reduction, image filtering, etc., so as to improve the shadow removal effect and make the established data set samples objective and robust.

[0076] In this embodiment, constructing a data set means actively collecting clear defective images for training to generate adversarial grids. In order to prevent overfitting, the data set should include various scenes of the target object as much as possible, so that the trained network model can have good generalization performance in various test scenes.

[0077] In this embodiment, the generative adversarial network model is divided into a generative network and a discriminative network. The generative network and the discriminative network train the model in an adversarial game, the generative network tries to generate new data, and the discriminative network tries to determine whether the output is false data or real data. The generative network is implemented in a coding and decoding manner, and is composed of multiple convolutional layers, deconvolutional layers, and fully connected layers. The generative network inputs a row of normally distributed random numbers Z, outputs a fake picture similar to the real image, and sends the real image and the generated fake image together to the discriminative network, which discriminates them. The discriminative network is implemented by a convolutional neural network of coding and decoding. The discriminative network inputs an image Y and outputs a judgment on whether the input image is true or false. The goal of the generative network is to make the generated image as close to the real image as possible by training the model, that is, to minimize the difference between the real image and the generated image, so that the discriminative network judges the generated image as a real image. In the training iteration, the generative network and the discriminative network will continue to evolve and confront each other until they reach a state of equilibrium. In the equilibrium state, the discriminative network can no longer recognize the synthetic false data, thereby restoring the original features of the photo as much as possible and reducing the shadows caused by taking pictures.

[0078] In this embodiment, the images processed by the technical means of image preprocessing, data set construction and generative adversarial network model guarantee the authenticity of the image to the greatest extent, provide more accurate image data for edge extraction and measurement, improve the accuracy of measurement, and thus reduce the device's misjudgment of waste, ultimately improving the working efficiency of the horizontal continuous casting sawing and screening device and reducing the probability of defective products flowing out.

[0079] In this embodiment, the edge data of the image is compared with the system database, and the basis for determining whether it is a bar segment that does not meet the standards is the structural similarity between the compared image and the system database image and the peak signal-to-noise ratio between the compared image and the system database image.

[0080] In this embodiment, the structural similarity (SSIM) can measure the degree of image distortion and the similarity between two images. SSIM is a perceptual model. This indicator conforms to the intuitive perception of the human eye. It mainly considers three key features of the image: brightness, contrast, and structure. The calculation formula of structural similarity is:

[0081]

[0082] In the formula, μ is the value, which represents the brightness estimation; σ is the variance, which represents the contrast estimation; σ xyis the covariance, indicating the structural similarity estimation, C1=(k1L) 2 , C2=(k2L) 2 , k1 and k2 are 0.01 and 0.02 respectively, and L is the dynamic value range of the pixel.

[0083] In this embodiment, the peak signal-to-noise ratio (PSNR) is the ratio of the peak energy information to the noise energy signal. The calculation formula of the peak signal-to-noise ratio is:

[0084]

[0085] Where, MAE is the noise energy signal value, It is the square of the maximum value of the pixel color.

[0086] like Figure 1 to Figure 3 As shown, this embodiment proposes a horizontal continuous casting cutting and screening device, comprising:

[0087] The image acquisition module, when detecting the bar material 6, captures the image of the bar material 6 at each time interval according to a preset time interval;

[0088] The image stitching module uses the RANSAC algorithm to fit and stitch images sorted by time intervals, and performs shadow preprocessing on the stitched images;

[0089] The adversarial network fitting module obtains the preprocessed image to construct a data set, generates an adversarial network fitting based on the image data in the data set, and re-fits the image data that fails to fit the adversarial network until the adversarial network fitting succeeds;

[0090] The edge detection module obtains the image that the adversarial network successfully fits, extracts and measures the edge of the image, compares the edge data of the image with the system database, and determines whether it is a bar section that does not meet the standards;

[0091] The cutting and separation module cuts and removes the 6 sections of bar materials that do not meet the standards, and cuts and collects the 6 sections of bar materials that meet the standards into specified lengths.

[0092] In this embodiment, the image acquisition module includes a laser sensor 7 and a high-definition camera 5. The laser sensor 7 is used to detect the position and length of the bar 6, and the high-definition camera 5 is used to capture the image of the bar 6. The image acquisition module is also provided with a sensor controller for controlling the laser sensor 7.

[0093] The cutting and separation module includes a robotic arm 1, a cutting box 3, a front clamping device 2, and a rear clamping device 4. The robotic arm 1 is used to pick up the rod material 6 that meets the standard or does not meet the standard and place it in the corresponding position. The cutting box 3 includes a tool moving device 8 and a cutting tool 9. The tool moving device 8 is used to drive the cutting tool 9 to move. The cutting tool 9 is used to cut the rod material 6. The front clamping device 2 and the rear clamping device 4 are used to clamp the rod material 6 so that the cutting tool 9 can perform stable cutting.

[0094] The horizontal continuous casting cutting and screening device is also provided with a master control for coordinating and controlling the entire horizontal continuous casting cutting and screening process. A feeding device 10 is provided at the lower side of the cutting and separation module for conveying the bar material 6. The feeding device 10 is provided with multiple material ports, and the multiple material ports simultaneously convey multiple materials for cutting and screening. A debris collection box is provided at the lower side of the feeding device 10 for collecting materials during the sawing process.

[0095] Specifically, the working process is as follows: the feeding device 10 is started, each feeding port 11 feeds at the same time (each feeding port 11 has an independent power), the laser sensor 7 is started to start measuring, the high-definition camera 5 takes an indiscriminate snapshot every interval △t, the image is fitted using the RANSAC algorithm to complete the image stitching, the image is pre-processed by the pre-processing system to remove the shadow, the processed image constructs a data set, generates an adversarial grid fitting, the image that fails to fit is reprocessed and the above process is repeated, the edge extraction measurement of the successfully fitted image is compared with the system data by two methods: structural similarity (SSIM) and peak signal-to-noise ratio (PSNR). The library is compared and identified. When it is identified as an error part, the main control sends instructions to the sensor controller and the corresponding feed port 11 power system. The sensor controller controls the laser sensor 7 to stop distance measurement, the corresponding feed port 11 to stop feeding, the front clamping system and the rear clamping mechanism to clamp, the tool moving device 8 to start, and the cutting tool 9 to move. When the laser sensor 7 detects that the cutting tool 9 reaches the specified error bar 6 section, the cutting tool 9 starts cutting. After the cutting is completed, the cutting tool 9 is reset, and the robot arm 1 starts to grab and move the error part to the specified waste bin; if there is no bar 6 at the identified material port, the feed port 11 is started, and the laser sensor 7 starts to measure the distance. If no error is found during the work process, the laser sensor 7 detects that the length of the bar 6 meets the standard, then the main control controls the feed port 11 to stop feeding, the front clamping system and the rear clamping mechanism to clamp, the tool moving device 8 to start, the cutting tool 9 to move to the corresponding position, the cutting tool 9 to start cutting, the cutting tool 9 to reset after the cutting is completed, and the robot arm 1 starts to grab and move the correct part to the collection box.

[0096] This embodiment proposes a horizontal continuous casting cutting and screening system, the system comprising:

[0097] one or more memories for storing instructions; and

[0098] One or more processors are used to call and run instructions from the memory to execute the horizontal continuous casting cutting and screening method.

[0099] This embodiment provides a computer-readable storage medium, the computer-readable storage medium comprising:

[0100] The program, when the program is run by the processor, the horizontal continuous casting cutting screening method is executed.

[0101] Specifically, the storage medium can be a general storage medium, such as a mobile disk, hard disk, etc. When the computer program on the storage medium is run, it can execute the above-mentioned horizontal continuous casting cutting and screening method, thereby solving the problem of "inability to automatically screen and detect bar materials", realizing fully automatic bar material detection and screening process, reducing errors and missed detection rates caused by human judgment, and improving production efficiency and finished product qualification rate.

[0102] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A horizontal continuous casting cutting and screening method, characterized in that: include: When the bar is detected, an image of the bar is captured at each time interval according to a preset time interval; The RANSAC algorithm is used to fit and splice the images sorted by time intervals, and shadow preprocessing is performed on the spliced ​​images. Obtain the preprocessed images to construct a data set, generate an adversarial network fitting based on the image data in the data set, and refit the image data where the adversarial network fitting fails until the adversarial network fitting succeeds; Obtain an image that has been successfully fitted by the adversarial network, extract and measure the edge of the image, compare the edge data of the image with the system database, and determine whether it is a bar segment that does not meet the standards; The bar segments that do not meet the standards are cut and removed, and the bar segments that meet the standards are cut into specified lengths and collected.

2. A horizontal continuous casting cutting and screening method according to claim 1, characterized in that: The step of using the RANSAC algorithm to fit and splice the images sorted according to the time intervals comprises: Randomly select feature points as internal points and determine the transformation matrix model through calculation; Repeat step 1 continuously, traverse all feature points, and establish multiple transformation matrices; The number of feature points in the transformation matrix is ​​counted, and the matrix with the largest number of feature points is the optimal transformation matrix.

3. A horizontal continuous casting cutting and screening method according to claim 2, characterized in that: The transformation matrix of the transformation matrix model determined by calculation is a homography matrix, and the matrix parameters are obtained through 4 sets of feature points and 8 linear equations. The specific expressions are as follows: Assume that the probability of a point selected from the image feature point set being an internal point is P, then the probability formula is: P=λ n , In the formula, N represents the total number of inliers, M represents the total number of feature points, and λ indicates that all randomly selected points are inliers. Probability, λ n It means iterating n times. The number of iterations n needs to take the logarithm of the formula. The derived formula is: The RANSAC algorithm is used to determine the internal points, the feature point set is traversed, the optimal transformation is calculated through the transformation matrix model, and finally the stitching of multiple images is completed.

4. A horizontal continuous casting cutting and screening method according to claim 1, characterized in that: The edge data of the image is compared with the system database to determine whether it is a bar segment that does not meet the standards based on the structural similarity between the compared image and the system database image and the peak signal-to-noise ratio between the compared image and the system database image.

5. A horizontal continuous casting cutting and screening method according to claim 4, characterized in that: The calculation formula of the structural similarity is: In the formula, μ is the value, which represents the brightness estimation; σ is the variance, which represents the contrast estimation; σ xy is the covariance, indicating the structural similarity estimation, C1=(k1L) 2 , C2=(k2L) 2 , k1 and k2 are 0.01 and 0.02 respectively, and L is the dynamic value range of the pixel.

6. A horizontal continuous casting cutting and screening method according to claim 4, characterized in that: The calculation formula of the peak signal-to-noise ratio is: Where, MAE is the noise energy signal value, It is the square of the maximum value of the pixel color.

7. A horizontal continuous casting cutting and screening device, characterized in that: include: The image acquisition module, when detecting the bar material, captures the image of the bar material at each time interval according to a preset time interval; The image stitching module uses the RANSAC algorithm to fit and stitch images sorted by time intervals, and performs shadow preprocessing on the stitched images; The adversarial network fitting module obtains the preprocessed image to construct a data set, generates an adversarial network fitting based on the image data in the data set, and re-fits the image data that fails to fit the adversarial network until the adversarial network fitting succeeds; The edge detection module obtains the image that has been successfully fitted by the adversarial network, extracts and measures the edge of the image, compares the edge data of the image with the system database, and determines whether it is a bar segment that does not meet the standards; The cutting and separation module cuts and removes the bar segments that do not meet the standards, and cuts and collects the bar segments that meet the standards into specified lengths.

8. A horizontal continuous casting cutting and screening device according to claim 7, characterized in that: The image acquisition module includes a laser sensor and a high-definition camera, wherein the laser sensor is used to detect the position and length of the bar material, and the high-definition camera is used to acquire the image of the bar material; The cutting and separation module includes a robotic arm, a cutting box, a front clamping device, and a rear clamping device. The robotic arm is used to pick up bars that meet the standards or do not meet the standards and place them in corresponding positions. The cutting box includes a tool moving device and a cutting tool. The tool moving device is used to drive the cutting tool to move. The cutting tool is used to cut the bars. The front clamping device and the rear clamping device are used to clamp the bars so that the cutting tool can perform stable cutting.

9. A horizontal continuous casting cutting and screening system, characterized in that: The system comprises: one or more memories for storing instructions; and One or more processors, configured to call and execute the instructions from the memory to perform the method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that: The computer readable storage medium comprises: The program, when the program is executed by a processor, the method according to any one of claims 1 to 6 is executed.

Images