Material code recognition system in iron and steel industry

Abstract

The invention provides a material code recognition system in the iron and steel industry, and the system comprises a data obtaining module which is configured to obtain a material code original data set on the surface of a material, and carries out the preprocessing of the material code original data set, and obtains a material code target data set; the preprocessing comprises image rotation expansion processing; the data positioning module is configured to perform effective positioning processing according to the material coding target data set to obtain a coding character region image; the character correction module is configured to perform correction processing on the coded character region image to obtain a corrected character region image; and the character recognition module is configured to perform character recognition according to the corrected character region image to obtain at least one coded character. According to the scheme, the problem that an existing steel industry material recognition technology cannot efficiently recognize coded characters at any angle is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of material coding in the steel industry, and in particular to a material coding identification system in the steel industry. Background Art

[0002] As a key link in the deep integration of automation and informatization in steel enterprises, material tracking has immeasurable value in comprehensively improving the production management efficiency of enterprises and promoting production management to move towards online, integrated and rational directions. In the complex and huge system of steel production, accurate and efficient tracking of various materials such as iron ladles, steel ladles, steel billets, slabs, wire rods, coils, plates, etc. is the basis for optimizing production processes, rationally allocating resources and tracing product quality. However, limited by historical technical accumulation and current technical bottlenecks, most steel enterprises still rely on traditional manual operation mode in material management. This mode not only consumes a lot of human resources and is inefficient, but also has problems such as delayed response and huge work intensity. More importantly, the untimely data update and error-prone characteristics caused by manual management have seriously restricted the pace of steel enterprises' transformation to intelligent and refined production, and cannot meet the urgent needs of modern steel industry for efficient, accurate and real-time production information management.

[0003] The character recognition technology currently widely used in the market is designed with a focus on the discrete manufacturing field with a relatively controllable environment, and has relatively strict requirements for the standardization, clarity and recognition environment of characters. However, the steel manufacturing environment is completely different. The diversity and complexity of its material coding pose unprecedented challenges to character recognition technology. In the steel manufacturing process, material coding is widely used on the surfaces of metal objects such as steel billets and steel plates. These codes not only have different forms, including inkjet coding, labeling, steel stamping, QR codes, RFID tags, customized labels and even handwritten characters, but also the metal surface environment they are in is harsh and easily affected by various physical factors such as metal deformation, iron oxide shedding, stains, and light reflection. In addition, the material may be at any angle (0 to 360 degrees) on the production line, which undoubtedly further increases the difficulty of character recognition.

[0004] The production sites of steel enterprises are often accompanied by extreme conditions of high temperature and high dust. Such an environment poses a severe test to the stability and accuracy of imaging equipment, making it difficult to ensure consistent imaging quality, which directly affects the accuracy and reliability of character recognition. In view of the above multiple challenges, existing general character recognition products have encountered significant obstacles in the application of the steel industry, and direct application often has poor results. Therefore, in order to meet the specific material tracking needs of steel enterprises, in-depth model customization and development are usually required for the actual production environment. This process requires not only highly professional technical support, but also a lot of on-site debugging and manual adaptation work to adapt to the ever-changing production conditions and material characteristics. This not only increases the complexity of technology implementation, but also prolongs the cycle of technology application. It is a technical problem that needs to be solved urgently in the process of promoting material tracking automation and informatization of steel enterprises. Summary of the invention

[0005] The present application provides a material coding identification system for the steel industry to solve the problem that the existing material identification technology in the steel industry cannot efficiently identify coded characters at any angle.

[0006] The system comprises:

[0007] A data acquisition module, wherein the data acquisition module is configured to acquire a material coding original data set of a material surface, and preprocess the material coding original data set to obtain a material coding target data set; the preprocessing includes image rotation and expansion processing;

[0008] A data positioning module, wherein the data positioning module is configured to perform effective positioning processing according to the material coding target data set to obtain a coding character area image;

[0009] A character correction module, wherein the character correction module is configured to perform correction processing on the coded character area image to obtain a corrected character area image;

[0010] A character recognition module is configured to perform character recognition according to the corrected character area image to obtain at least one coded character.

[0011] Preferably, the data acquisition module includes:

[0012] A photoelectric sensor unit, wherein the photoelectric sensor unit is configured to generate a material arrival signal when the material moves to a specified position;

[0013] An image acquisition unit, wherein the image acquisition unit is configured to acquire a material coded original data set of a material surface according to the material in place signal;

[0014] An image preprocessing unit is configured to preprocess the material coding original data set to obtain the material coding target image data set.

[0015] Preferably, the image preprocessing module is further configured to:

[0016] The material code original data set and each corresponding character area are intercepted to obtain a code character background data set and a single character mask data set;

[0017] Respectively labeling the coded character background data set and the single character mask data set to obtain the material code labeling data;

[0018] The material code original data set is subjected to 360° random rotation processing, random scaling processing and random filling processing according to the material code annotation data to obtain the material code target data set.

[0019] The image preprocessing module is further configured to:

[0020] Create a character background image, character length, character content and font color index list;

[0021] Randomly extract the character background image and the character length, and combine them to obtain a character string of fixed length;

[0022] scaling the character background image according to the character length, selecting a suitable font type and size, and drawing a character mask image in combination with a fixed-length character string;

[0023] The character mask image is subjected to salt and pepper noise processing and mean filtering to obtain the material code target data set.

[0024] Preferably, the data positioning module includes:

[0025] An effective area positioning unit, the effective area positioning unit is configured to perform effective area detection of coded characters according to the material coding target data set to obtain an effective area image;

[0026] A coded character locating unit is configured to perform coded character detection according to the effective area image to obtain a coded character area image.

[0027] Preferably, the character correction module is further configured to:

[0028] Calculating the tilt angle of the coded character in the coded character area image using the invariant feature information inherent to the character;

[0029] The coded character area image is corrected using the tilt angle to obtain the corrected character area image.

[0030] Preferably, the character correction module is further configured to:

[0031] Calculating the angle between the longest side of the smallest outer rectangle of each line of the coded character area in the coded character area image and the inherent invariant feature information of the character to obtain a plurality of rectangular angles;

[0032] The rectangular angle corresponding to the longest minimum outer rectangle, the median of all rectangular angles, or the average of all rectangular angles is used as the tilt angle.

[0033] Preferably, the character correction module is further configured to:

[0034] Performing angle classification on each row of coded characters in the coded character region image to obtain a data set that does not require correction and a data set that requires correction;

[0035] Calculating the tilt angle of each row of coded character area in the data set that needs to be corrected;

[0036] The data set to be corrected is corrected using the tilt angle to obtain the corrected character area image.

[0037] Preferably, the character recognition module is further configured as follows:

[0038] Performing training recognition on the corrected character region image to obtain a coded character recognition model;

[0039] The corrected character region image is imported into the coded character recognition model for real recognition to obtain a number of rows of coded characters corresponding to the corrected character region image.

[0040] Preferably, the coded character categories in the corrected character area image include one or more of stamped characters, inkjet characters, label characters and handwritten characters, and the coded character recognition model includes a dedicated recognition model and a dual-purpose recognition model;

[0041] The character recognition module is also configured to:

[0042] Performing category recognition on the corrected character region image, marking the corrected character region image with a single coded character category as a first-category image, and marking the corrected character region image with more than one coded character category as a second-category image;

[0043] The dedicated recognition model is used to truly recognize the first type of image, and the dual-use recognition model is used to truly recognize the second type of image.

[0044] It can be seen from the above content that the present application provides a material coding identification system for the steel industry, the system comprising a data acquisition module, the data acquisition module is configured to acquire a material coding original data set of the material surface, and preprocess the material coding original data set to obtain a material coding target data set; the preprocessing includes image rotation and expansion processing; a data positioning module, the data positioning module is configured to perform effective positioning processing according to the material coding target data set to obtain a coded character area image; a character correction module, the character correction module is configured to perform correction processing on the coded character area image to obtain a corrected character area image; a character recognition module, the character recognition module is configured to perform character recognition according to the corrected character area image to obtain at least one coded character. The present application solves the problem that the existing material recognition technology in the steel industry cannot efficiently recognize coded characters at any angle through the above scheme. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0046] Figure 1 A schematic diagram of a material coding identification system for the steel industry is provided for this application;

[0047] Figure 2 This is an example diagram of common material coding;

[0048] Figure 3 This is an example image of label characters;

[0049] Figure 4 This is an example of a background and single character area.

[0050] Figure 5 This is a binary example diagram of a single character;

[0051] Figure 6 This is an example diagram of the effect of character recognition data expansion;

[0052] Figure 7 An example diagram for locating the effective area of ​​the encoding;

[0053] Figure 8 This is an example of the character positioning effect;

[0054] Fig. 9 This is an example of the character correction effect;

[0055] Fig.10 This is an example of the encoding character recognition effect. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in 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] Figure 1 This is a schematic diagram of a material coding and identification system for the steel industry in this application.

[0058] See also Figure 1 It can be seen that this embodiment provides a material coding identification system for the steel industry, and the system includes:

[0059] The data acquisition module 100 is configured to acquire a material coding original data set of a material surface, and preprocess the material coding original data set to obtain a material coding target data set; the preprocessing includes image rotation and expansion processing. Specifically, in this embodiment, the data acquisition module 100 is used to acquire images of the material surface of the steel industry to form the material coding original data set, and perform relevant preprocessing on the material coding original data set to obtain the material coding target data set.

[0060] Wherein, the data acquisition module 100 includes:

[0061] A photoelectric sensor unit 110, wherein the photoelectric sensor unit 110 is configured to generate a material arrival signal when the material moves to a specified position;

[0062] An image acquisition unit 120, wherein the image acquisition unit 120 is configured to acquire a material code original data set of a material surface according to the material arrival signal;

[0063] The image preprocessing unit 130 is configured to preprocess the material coding original data set to obtain the material coding target image data set.

[0064] Specifically, in this embodiment, the photoelectric sensing unit 110 is used to obtain the material arrival signal, and the image acquisition unit 120 is used to dynamically acquire the original coded image of the material surface to establish a material coding original data set. The coded characters include stamped characters, inkjet characters, label characters, and regular handwritten characters, among which regular handwritten characters refer to handwritten characters written in lines.

[0065] For example, a photoelectric sensor is placed at the inspection station of the billet conveyor line, and the photoelectric sensor is directly connected to the camera through an IO signal line or the photoelectric signal is obtained through a PLC. When the billet arrives at the inspection station, the camera is immediately triggered to take a photo and capture the coded image of the end face of the billet. Figure 2 As shown in the figure, it is a common material coding image collected. Figure 3 As shown, there are four types of characters that the present invention focuses on, and a material coding original data set consisting of four types of character images is established.

[0066] The image preprocessing unit 130 combines the material code original data set with image augmentation and fusion strategies to expand the image data set and establish a material code annotation data set, a code character background data set and a single character mask data set.

[0067] The specific steps of preprocessing by the image preprocessing unit 130 include:

[0068] The coded character background data set is composed of intercepted background area images according to the pixel size of the coded character row, and the single-character mask data set is composed of intercepted single-character area images.

[0069] Take label character images as an example, Figure 3 As shown, in a specific embodiment, characters such as numbers, uppercase and lowercase characters, common logos, etc. are set as category labels, and a background class is added. A labeling tool is used to label the label image with a rectangular frame, and the background area and the character area are intercepted and saved as an image through the labeling file, such as Figure 4 As shown, the background area image and the single character area image are cut out from the label image.

[0070] The traditional threshold segmentation method can be used to process the single-character area image. At the same time, by building a coded character image segmentation model, the binary image of the single character can be segmented from the original image to establish a single-character mask dataset.

[0071] In a specific embodiment, the single character region image is thresholded using the maximum inter-class variance method, and morphological processing is performed to fill holes, thereby obtaining a binary image of the single character, such as Figure 5 shown.

[0072] The material coding and labeling dataset is composed of a coding area positioning dataset, a coding character positioning dataset, and a coding character recognition dataset. In a specific embodiment, a coding character positioning dataset and a coding character recognition dataset are established for each type of characters such as steel stamps, inkjet codes, labels, and regular handwriting. The coding character positioning dataset uses a 360° random rotation method to expand the character positioning dataset. In a specific embodiment, the 500-image label character positioning dataset is expanded to 2,000 at a ratio of 1:3 by using a random rotation image method. The coding character recognition dataset uses random scaling and random filling, while integrating color indexing, noise generation, and image filtering methods to directionally generate and expand the character recognition dataset. For label and inkjet characters, a random sampling and combination method is used. The corresponding font library is selected according to the standard font type for printing. Combined with the coded character background data set, characters of different lengths are randomly printed on the background image to obtain a character mask image; for stamped characters, inkjet characters, and regular handwritten characters, a random sampling and combination method is used. The coded character background data set and the single character mask data set are integrated, and the background image is used as the mask. The single character image is pasted on the background image to obtain the character mask image; or the single character image is directly subjected to random extraction and combination methods to obtain the character mask image; and the character mask image is further subjected to noise generation and filtering processing to expand the character recognition data set.

[0073] Among them, the character recognition data expansion method includes:

[0074] Method 1: Establish a list of coded character background images, a character length index list, a character content index list, and a font color index list; randomly extract a background image from the coded character background image list; randomly select a character length from the character length index list; according to the selected character length, use a random combination method from the character content index list to obtain a fixed-length string; according to the selected character length and the size of the coded character, scale the extracted background image to an appropriate scale to obtain a scaled background image; select a font type that is similar to the coded character morphology from the standard font library and set an appropriate font size; use the scaled background image as a mask, select a color from the font color index list, draw a fixed-length string, and obtain a character mask image; use the salt and pepper noise generation method to process the character mask image to obtain a character mask noise image; set several groups of convolution kernels of different sizes and randomly select a group of convolution kernels. Use the mean filter method to process the character mask noise image to obtain an expanded character recognition image.

[0075] Method 2: Establish a list of coded character background images, a list of single character mask images, a list of character length indexes, a list of character content indexes, and a list of font color indexes; randomly extract a background image from the list of coded character background images; randomly select a character length from the list of character length indexes; according to the selected character length, use a random combination method from the list of character content indexes to obtain a fixed-length string; scale the extracted background image to a suitable scale according to the selected character length and the size of the coded character to obtain a scaled background image; according to the content of the fixed-length string, use a random combination method from the list of single character mask images to obtain a set of single character mask images; according to the selected character length and the size of the coded character, scale this set of character mask images to a suitable scale to obtain a scaled single character mask image; use the scaled background image as a mask, randomly select a color from the list of font color indexes as the character color, and sequentially paste the scaled single character mask images into the subsequent background image in order to obtain a character mask image; use the salt and pepper noise generation method to process the character mask image. Obtain a character mask noise image; set several groups of convolution kernels of different sizes and randomly select a group of convolution kernels. The character mask noise image is processed by the mean filtering method to obtain an expanded character recognition image.

[0076] Through the above method, taking the label image as an example, the generated character content is set to include 10 numbers and 52 uppercase and lowercase characters. The generated effect of character recognition data is as follows Figure 6 shown.

[0077] The system further comprises:

[0078] The data positioning module 200 is configured to perform effective positioning processing according to the material coding target data set to obtain a coded character area image. Specifically, in this embodiment, the data positioning module 200 is used to perform relevant image positioning on each image in the material coding target data set.

[0079] The data positioning module 200 includes:

[0080] An effective area positioning unit 210, wherein the effective area positioning unit 210 is configured to perform effective area detection of coded characters according to the material coding target data set to obtain an effective area image;

[0081] The coded character locating unit 220 is configured to perform coded character detection according to the effective area image to obtain a coded character area image.

[0082] Specifically, in this embodiment, the original coded image of the material surface is sent to the constructed effective area positioning model through the effective area positioning unit 210 to detect the location, quantity and type of the code and obtain the effective area image of the coded character.

[0083] Taking the coding area positioning dataset as the training set, by constructing a coding effective area positioning target detection model, four types of labels, namely steel stamp, inkjet code, label, and handwritten character, are set to detect and classify the coding character area, and determine the coding character type and the location of the effective area.

[0084] In a specific embodiment, considering the advantages of yolov8 such as high efficiency, ease of use and robustness, the target detection model of the yolov8 series is selected as the coding effective area positioning model; Figure 7 As shown, for the three types of characters, namely, stamped characters, inkjet characters, and handwritten characters, the entire area where the characters are located (there is only one circumscribed rectangle) is used as the valid area; for label characters, the area where each label is located (determined by the number of labels) is used as the valid area.

[0085] The effective area image is sent to the constructed character positioning model through the coded character positioning unit 220 to detect the area position of each line of coded characters, and the area image of each line of coded characters is intercepted in turn.

[0086] Using the coded character positioning dataset as the training set, separate coded character positioning semantic segmentation models are constructed for stamped characters, inkjet characters, label characters, and regular handwritten characters. Two types of labels are set for the character area and the background area. The character area is treated as a target entity according to the character row, and the position of the coded characters in each row is located.

[0087] In a specific embodiment, taking the label image as an example, the yolov8 series segmentation model is selected as the character positioning model. The character positioning effect is as follows: Figure 8 shown.

[0088] The system further comprises:

[0089] The character correction module 300 is configured to perform correction processing on the coded character area image to obtain a corrected character area image. Specifically, in the present embodiment, the character correction module 300 calculates the inclination angle of each line of the coded character area image by utilizing the inherent invariant feature information of the character, and corrects the coded image according to the calculated inclination angle to obtain a corrected character area image.

[0090] By solving the angle corresponding to the long side of the minimum outer rectangle of each character area, as the inclination angle of the row of characters, the inclination angle of the longest character is selected or the median or average value is calculated to determine the inclination angle, which is used as the inclination angle of the coded character image to obtain the corrected character image of each row; at the same time, a coded character direction classification model is constructed, and two types of labels, 0° and 180°, are set. Each row of the corrected character image is classified into two categories, and the direction classification result is judged by voting. If the classification direction is 0°, no flipping is required. If the classification direction is 180°, the character needs to be flipped again to obtain the final corrected character area image.

[0091] In a specific embodiment, taking the label image as an example, the tilt angle of the longest character is used as the tilt angle of the coded character image, and at the same time, the direction classification model in the PaddleOCR library is selected as the direction classification model of the label character. Figure 8 The character image shown is corrected to obtain a corrected character image, such as Fig. 9 shown.

[0092] The system further comprises:

[0093] The character recognition module 400 is configured to perform character recognition based on the corrected character area image to obtain at least one coded character. Specifically, in the present embodiment, the coded character category in the corrected character area image includes one or more of stamped characters, inkjet characters, label characters and handwritten characters, and the coded character recognition model includes a dedicated recognition model and a dual-use recognition model; the character recognition module 400 is also configured to: perform category recognition on the corrected character area image, mark the corrected character area image with a single coded character category as a first-category image, and mark the corrected character area image with more than one coded character category as a second-category image; use the dedicated recognition model to perform true recognition on the first-category image, and use the dual-use recognition model to perform true recognition on the second-category image.

[0094] Exemplarily, the corrected character area image is sent to the constructed coded character recognition model to identify each line of coded characters in turn. Using the coded character recognition data set as the training set, the coded character recognition model is constructed in the following two ways: first: by fusing the steel stamp character, inkjet character, label character, and regular handwritten character image data, a universal coding recognition model is constructed, which is compatible with the recognition of multiple types of identification characters; second: special coding recognition models are constructed separately, including inkjet character recognition models, steel stamp character recognition models, label character recognition models, and handwritten character recognition models. According to the identification judgment result of the effective area positioning model, the corresponding special coding recognition model is selected to recognize different types of identification characters.

[0095] In a specific embodiment, considering that PaddleOCR has many advantages such as multi-language support, high performance, high accuracy, and ease of use, the present invention selects the recognition model in the PaddleOCR open source library as the code recognition model, and constructs dedicated code recognition models for different types of logos, and sends the corrected character images to the dedicated code recognition models for recognition in turn. Fig.10 As shown, the recognition effects of different types of logo characters are demonstrated.

[0096] In summary, the present invention is based on the material coding recognition needs in the application scenarios of the steel industry, focusing on commonly used identification type characters such as steel stamps, inkjet codes, labels, handwritten characters, etc., to establish a material coding character data set for the steel industry; in the preprocessing module, it mainly adopts random rotation, scaling, and filling methods, uses sampling and combination strategies, and integrates color indexing, noise generation, and image filtering methods to directionally expand the character data set; in the effective area positioning module, by constructing a coding effective area positioning model, the type of coding identification and the effective area position where it is located are detected; in the coding character positioning module, by constructing a coding character positioning semantic segmentation model, it is used to detect the location of each character; in the coding character correction module, the inherent invariant feature information of the character is used to solve the minimum circumscribed rectangle of the character row, and the strategy of obtaining the maximum value, median, and mean is selected. Based on the character direction classification model and combined with the clustering and voting method, the correction of the character angle and direction is realized; in the coding character recognition module, based on the strategy of multimodal data fusion and the strategy of multi-model fusion, a coding character recognition model is constructed, which is compatible with the recognition of multiple types of coding characters.

Claims

1. A material coding identification system for the steel industry, characterized in that: The system comprises: A data acquisition module (100), the data acquisition module (100) being configured to acquire a material coding original data set of a material surface, and preprocess the material coding original data set to obtain a material coding target data set; the preprocessing includes image rotation and expansion processing; A data positioning module (200), the data positioning module (200) being configured to perform effective positioning processing according to the material coding target data set to obtain a coding character area image; A character correction module (300), the character correction module (300) being configured to perform correction processing on the coded character area image to obtain a corrected character area image; A character recognition module (400) is configured to perform character recognition based on the corrected character area image to obtain at least one coded character.

2. A material coding identification system for the steel industry according to claim 1, characterized in that: The data acquisition module (100) comprises: A photoelectric sensor unit (110), wherein the photoelectric sensor unit (110) is configured to generate a material arrival signal when the material moves to a specified position; An image acquisition unit (120), the image acquisition unit (120) being configured to acquire a material coded original data set of a material surface according to the material arrival signal; An image preprocessing unit (130), wherein the image preprocessing unit (130) is configured to preprocess the material coding original data set to obtain the material coding target image data set.

3. A material coding identification system for the steel industry according to claim 2, characterized in that: The image preprocessing module (130) is further configured to: The material code original data set and each corresponding character area are intercepted to obtain a code character background data set and a single character mask data set; Respectively labeling the coded character background data set and the single character mask data set to obtain material coding labeling data; The material code original data set is subjected to 360° random rotation processing, random scaling processing and random filling processing according to the material code annotation data to obtain the material code target data set.

4. A material coding identification system for the steel industry according to claim 3, characterized in that: The image preprocessing module (130) is further configured to: Create a character background image, character length, character content and font color index list; Randomly extract the character background image and the character length, and combine them to obtain a character string of fixed length; scaling the character background image according to the character length, selecting a suitable font type and size, and drawing a character mask image in combination with a fixed-length character string; The character mask image is subjected to salt and pepper noise processing and mean filtering to obtain the material code target data set.

5. A material coding identification system for the steel industry according to claim 3, characterized in that: The data positioning module (200) comprises: An effective area positioning unit (210), the effective area positioning unit (210) being configured to perform effective area detection of coded characters according to the material coding target data set to obtain an effective area image; A coded character locating unit (220) is configured to perform coded character detection according to the effective area image to obtain a coded character area image.

6. A material coding identification system for the steel industry according to claim 5, characterized in that: The character correction module (300) is further configured to: Calculating the tilt angle of the coded character in the coded character area image using the invariant feature information inherent to the character; The coded character area image is corrected using the tilt angle to obtain the corrected character area image.

7. A material coding identification system for the steel industry according to claim 6, characterized in that: The character correction module (300) is further configured to: Calculating the angle between the longest side of the smallest outer rectangle of each line of the coded character area in the coded character area image and the inherent invariant feature information of the character to obtain a plurality of rectangular angles; The rectangular angle corresponding to the longest minimum outer rectangle, the median of all rectangular angles, or the average of all rectangular angles is used as the tilt angle.

8. A material coding identification system for the steel industry according to claim 7, characterized in that: The character correction module (300) is further configured to: Performing angle classification on each row of coded characters in the coded character region image to obtain a data set that does not require correction and a data set that requires correction; Calculating the tilt angle of each row of coded character area in the data set that needs to be corrected; The data set to be corrected is corrected using the tilt angle to obtain the corrected character area image.

9. A material coding identification system for the steel industry according to claim 8, characterized in that: The character recognition module (400) is further configured to: Performing training recognition on the corrected character region image to obtain a coded character recognition model; The corrected character region image is imported into the coded character recognition model for real recognition to obtain a number of rows of coded characters corresponding to the corrected character region image.

10. A material coding identification system for the steel industry according to claim 9, characterized in that: The coded character categories in the corrected character area image include one or more of stamped characters, inkjet characters, label characters, and handwritten characters, and the coded character recognition model includes a dedicated recognition model and a dual-purpose recognition model; The character recognition module (400) is further configured to: Performing category recognition on the corrected character region image, marking the corrected character region image with a single coded character category as a first-category image, and marking the corrected character region image with more than one coded character category as a second-category image; The dedicated recognition model is used to truly recognize the first type of image, and the dual-use recognition model is used to truly recognize the second type of image.

Images