Steel industry material steel seal character recognition method and device and storage medium

Abstract

The invention discloses a steel industry material steel seal character recognition method and device and a storage medium, and belongs to the field of motor control. According to the method, an original steel seal character image is subjected to low-light image enhancement by adopting a low-light enhancement model, so that the brightness of a steel seal character area in a picture is improved, meanwhile, the picture is prevented from overexposure, and after a low-light enhanced image is obtained, character positioning and tilt correction are performed, and finally, each row of steel seal characters are identified. The problem that the material character recognition accuracy is low due to the fact that the low-illumination image is collected under the condition that the field environment and the imaging condition are poor is solved, and the steel seal character recognition efficiency is effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of motor control, and in particular to a method for recognizing steel-printed characters on materials in the steel industry. Background Art

[0002] In recent years, with the advancement of automation and information management in the steel industry, character recognition technology for steel industry materials has become a key tool for improving quality management throughout the entire process, from billet production to delivery, and is crucial for driving enterprises' transformation towards intelligent manufacturing. However, due to technical limitations, most steel companies still rely on manual management to identify and track materials such as ladles, steel ladles, billets, slabs, wire rod, coils, and plates throughout the production process. This approach not only consumes a significant amount of human resources, is inefficient, slow to respond, and labor-intensive, but also suffers from slow data updates and error-proneness, making it difficult to meet the requirements of modern steel companies for efficient and accurate production management.

[0003] Most character recognition technologies currently on the market are suitable for applications with stable imaging environments and good lighting, such as medicine boxes and product nameplates. These technologies place high demands on character standardization and the specific operating environment. In contrast, character recognition applications in the steel manufacturing industry are more complex. Stamped character recognition is primarily used on the surfaces of items such as steel billets and plates. These characters can become difficult to identify due to factors such as flaking oxide layers and surface textures generated during the manufacturing process. Furthermore, the complex working environment and lighting conditions in steel mills place strict requirements on the location and method of camera light source installation, increasing the difficulty of obtaining stable, high-quality images.

[0004] These factors pose significant challenges to obtaining stable, high-quality images. Existing character recognition technologies are difficult to directly apply to steel industry scenarios, often requiring customized models tailored to specific scenarios. To address these challenges, there is an urgent need to develop a method that can overcome image quality issues caused by factors such as harsh imaging environments and rough surfaces, thereby reducing character recognition errors.

[0005] Existing character recognition products perform poorly in steel industry environments. This is primarily due to the fact that their recognition process directly feeds raw images into a character localization and detection network to achieve character recognition. This approach works well in well-defined imaging environments, but in complex environments, particularly low-light conditions, character clarity is significantly reduced, affecting the accuracy of character localization and detection. Consequently, current character recognition methods have significant limitations when working with low-light images.

[0006] Due to factors such as insufficient lighting in the imaging environment and large interference on the character surface, it is difficult to use directly and requires more manpower to adapt to the new production environment. Summary of the Invention

[0007] In order to better adapt to the steel stamp character recognition in the steel industry scenario, this application provides a steel industry material steel stamp character recognition method based on low-light enhancement technology, which can compensate for the situation where insufficient on-site lighting causes a decrease in character recognition accuracy from an algorithm perspective, thereby greatly improving the initial recognition rate of steel industry character coding, while avoiding too much time spent on purchasing and installing light sources, thereby improving the efficiency of character recognition.

[0008] The present application embodiment first provides a method for recognizing characters on steel industry material stamps, which is characterized by comprising the following steps:

[0009] S1, image acquisition;

[0010] Use photoelectric sensors to obtain material arrival signals, trigger industrial cameras to take photos and collect images of the area where the steel-stamped characters are located on the material surface, and establish an original steel-stamped character image dataset;

[0011] S2, image preprocessing;

[0012] The original stamped character image dataset was expanded using three random rotation methods of 90°, 180° and 270°;

[0013] S3, low-light image enhancement;

[0014] Perform low-light image enhancement on the original stamped character image to increase the brightness of the stamped character area in the image while ensuring that the image is not overexposed, thus obtaining a low-light enhanced image.

[0015] S4, character positioning;

[0016] Perform character positioning on the low-light enhanced image, detect the position and number of the stamped characters, obtain the regional position information of each line of stamped characters, and obtain the effective regional image of the stamped characters;

[0017] S5, character tilt correction;

[0018] Correct the tilt and direction of the stamped characters to ensure that all stamped characters are in a horizontal state;

[0019] S6, character recognition;

[0020] The character recognition model is used to identify each line of stamped characters to obtain the final stamped character recognition result.

[0021] An embodiment of the present application also provides an electronic device, including a processor and a memory, wherein the memory stores a program that can be run on the processor, and is characterized in that when the program is executed by the processor, the steps of the above-mentioned steel industry material steel stamp character recognition method are implemented.

[0022] Also, a computer-readable storage medium is provided, storing at least one program, characterized in that the at least one program can be executed by at least one processor to implement the steps of the above-mentioned steel industry material steel stamp character recognition method.

[0023] The present invention has the following beneficial effects:

[0024] This application is based on character recognition in the steel industry scenario, and provides a method for steel industry material steel stamp character recognition. A low-light enhancement model is used to perform low-light image enhancement on the original steel stamp character image to improve the brightness of the steel stamp character area in the picture while ensuring that the picture is not overexposed. After obtaining the low-light enhanced image, the character positioning and tilt correction are performed, and finally each line of steel stamp characters are recognized. This solves the problem of low material character recognition accuracy caused by low-light images collected under poor field environment and imaging conditions, and effectively improves the efficiency of steel stamp character recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a flow chart of the method of the present invention;

[0026] Figure 2 It is the low-light character image on the surface of the blank of the present invention;

[0027] Figure 3 is a schematic diagram of the low-light image enhancement module of the present invention;

[0028] Figure 4 The low-light image of the present invention is an image after the low-light image enhancement module is used;

[0029] Figure 5 This is the result graph after the image of the present invention is processed by the semantic segmentation model;

[0030] Figure 6 is a flowchart of character tilt correction of the present invention;

[0031] Figure 7 It is the final character detection result of the present invention. DETAILED DESCRIPTION

[0032] To make the technical solution of the present invention easier to understand, the technical solution of the present invention is further described in detail below based on specific embodiments and in conjunction with the accompanying drawings. Obviously, the following embodiments are only some of the embodiments of the present invention and do not constitute a limitation of the technical solution of the present invention. For those of ordinary skill in the art, other embodiments can be obtained based on the embodiments of the present application without expending creative work.

[0033] It should be noted that the order of steps described in the method embodiments in the following embodiments does not constitute a limitation on the technical solution of the present invention. For ordinary technicians in this field, without paying any creative labor, they can also reasonably adjust the order of certain steps according to the embodiments of this application.

[0034] like Figure 1 As shown, the method for recognizing characters on steel industry material stamps described in this embodiment includes the following steps:

[0035] S1. Image acquisition

[0036] The photoelectric sensor is used to obtain the material arrival signal, trigger the industrial camera to take pictures, dynamically collect the image of the area where the steel stamp characters are located on the surface of the material (blank), and establish the original steel stamp character image data set. Figure 2 As shown, the low-light character image collected on the blank surface.

[0037] S2. Image preprocessing

[0038] By observing the on-site production situation, it is found that the square blank may be flipped 90°, 180° and 270°. In order to improve the accuracy of character positioning, the image augmentation strategy is used to expand the original steel stamp character image dataset obtained in step S1.

[0039] Specifically, the original steel stamp character image dataset is expanded by using three random angle rotation methods of 90°, 180° and 270°, which can expand the original steel stamp character image dataset to 4 times the original size.

[0040] S3, low-light image enhancement

[0041] The low-light enhancement module is used to perform low-light image enhancement on the original steel-stamped character image to increase the brightness of the steel-stamped character area in the picture while ensuring that the image is not overexposed, thereby obtaining a low-light enhanced image.

[0042] like Figure 3 As shown in the figure, the low-light enhancement module mainly includes a low-light enhancement network, which combines the exposure control loss function, texture consistency loss function and spatial consistency loss function, and uses the original steel-printed character image as the training set to train the model.

[0043] The low-light enhancement network includes multiple convolutional layers with the same parameters, which extract low-level contour features and high-level semantic features of the image layer by layer. Then, according to the Retinex theory, the image features are decomposed into illumination and reflection components through the last two convolutional layers. As shown in the figure, the multiple convolutional layers with the same parameters can be divided into four modules, including a low-level contour feature extraction module, a high-level semantic feature extraction module, an illumination component feature extraction module, and a reflection component feature extraction module. After the illumination and reflection components of the input image are extracted, they are dynamically optimized using exposure control loss and texture consistency loss. These two components are then multiplied to obtain the enhanced image. Finally, the network parameters are optimized between the enhanced image and the original image using spatial consistency loss to obtain the final low-light enhanced image.

[0044] The exposure control loss function formula is as follows: L exp =|I l -E|, where, I l represents the reflection component image, and E represents the average grayscale value of multiple high-brightness images collected under good lighting conditions.

[0045] The texture consistency loss function formula is as follows: L tc =cos(φ vgg (I r )-φ vgg (I)), where cos() represents the cosine similarity function, I r represents the reflection component image, I represents the input image, φ vgg (·) denotes the VGG network.

[0046] The spatial consistency loss function is as follows: Where K represents the number of image blocks,

[0047] Y and I are the average pixel values of the local region in the low-light enhanced image and the input image, respectively, and Ω(i) represents the four adjacent regions (upper, lower, left, and right) centered on region i. Figure 4 for Figure 2 The low-light enhanced image is obtained after low-light enhancement by the low-light enhancement module.

[0048] S4. Character positioning

[0049] The low-light enhanced image obtained after the low-light image enhancement in step S3 is used as a training set, and a semantic segmentation model for character positioning is trained for the steel-stamped characters on the blank. In this embodiment, the semantic segmentation model uses yolov8, which detects two lines of characters as individuals. Figure 5 Shown is the result image after semantic segmentation model.

[0050] S5, Character tilt correction

[0051] Figure 6 This is a flowchart for character tilt correction. Utilizing the feature that each line of steel-stamped characters on the blank is always continuous and on the same straight line, the mask image of each line of characters is extracted through the semantic segmentation results of yolov8. The mask images are then binarized and the long side and angle of the minimum circumscribed rectangle are solved, which means that the long side and deflection angle of each line of characters are calculated. The deflection angles of the two lines of characters are averaged as the deflection angle of the entire character, and the characters can be corrected to a horizontal or flipped state. The corrected character image is then voted on for character direction classification using the character direction classification model, and the category with the most votes is selected as the direction classification result of the character image. If the number of votes for the horizontal direction (0°) is high, the character image does not need to be flipped, and the final corrected character image is obtained. Otherwise, the character image needs to be flipped 180° as the final corrected character image. In this implementation case, the character direction classification model uses the direction classification model in the PaddleOCRv2.6 character recognition algorithm library.

[0052] S6. Character Recognition

[0053] The horizontal character area image obtained by S5 correction is input into the character recognition model to recognize each row of coded characters in turn.

[0054] The character recognition model of this embodiment adopts the ppocrv3 algorithm in the PaddleOCRv2.6 character recognition library. The low-light enhanced image in step S3 is used as the training set to train the character recognition model. Then, the trained character recognition model is used to recognize each line of coded characters in turn. The final recognition result is as follows: Figure 7 shown.

[0055] This paper is based on the demand for material character recognition in the application scenario of the steel industry, and provides a steel industry material stamp character recognition method based on low-light enhancement. It focuses on solving the problem of low material character recognition accuracy caused by low-light images collected under poor field environment and imaging conditions. The present invention first preprocesses the collected original image and expands the original image data set by randomly rotating 90°, 180° and 270°; then decomposes the image according to the Retinex theory, and enhances the original low-light image from the aspects of lighting and texture through three loss functions; then, by constructing a character positioning semantic segmentation model, the location of each line of characters is detected; the final character's tilt angle is obtained by solving the angle corresponding to the long side of the minimum circumscribed rectangle of each located character area and taking the average, and then performing direction classification to convert all characters to the horizontal direction; finally, the ppocrv3 algorithm in the PaddleOCRv2.6 character recognition library is used to recognize each line of characters. The method of the present invention compensates for the decrease in character recognition accuracy caused by insufficient on-site lighting from an algorithmic perspective, thereby greatly improving the initial recognition rate of steel industry character codes. At the same time, it avoids wasting too much time on purchasing and installing light sources, effectively improving the efficiency of steel stamp character recognition.

[0056] Example 2

[0057] This embodiment provides an electronic device, including a processor and a memory communicatively connected to the processor, wherein the memory stores a program that can be run on the processor, and when the program is executed by the processor, the steps of the steel industry material steel stamp character recognition method provided in the above embodiment are implemented.

[0058] The program may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, and C++, as well as conventional procedural programming languages such as "C" or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server.

[0059] Example 3

[0060] This embodiment provides a computer-readable storage medium, which stores at least one program. The at least one program can be executed by at least one processor to implement the steps of the steel industry material steel stamp character recognition method provided in the above embodiment.

[0061] The aforementioned storage media include any medium capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory, a random access memory, a magnetic disk, or an optical disk, and any combination thereof. With the advancement of science and technology, the meaning of storage media may become increasingly broad and not limited to tangible media.

[0062] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing the present invention. In addition to the above embodiments, the present invention may also have other implementation methods; any technical solutions formed by equivalent replacement or equivalent transformation fall within the scope of protection required by the present invention.

Claims

1. A method for recognizing characters on steel industry material stamps, characterized in that: The steps include: S1, image acquisition; Use photoelectric sensors to obtain material arrival signals, trigger industrial cameras to take photos and collect images of the area where the steel-stamped characters are located on the material surface, and establish an original steel-stamped character image dataset; S2, image preprocessing; The original stamped character image dataset was expanded using three random rotation methods of 90°, 180° and 270°; S3, low-light image enhancement; Perform low-light image enhancement on the original stamped character image to increase the brightness of the stamped character area in the image while ensuring that the image is not overexposed, thus obtaining a low-light enhanced image. S4, character positioning; Perform character positioning on the low-light enhanced image, detect the position and number of the stamped characters, obtain the regional position information of each line of stamped characters, and obtain the effective regional image of the stamped characters; S5, character tilt correction; Correct the tilt and direction of the stamped characters to ensure that all stamped characters are in a horizontal state; S6, character recognition; The character recognition model is used to identify each line of stamped characters to obtain the final stamped character recognition result.

2. The steel industry material steel stamp character recognition method according to claim 1, characterized in that: The step S3 comprises: The original steel-stamped character image is used as the training set. According to the Retinex theory, the input original steel-stamped character image is decomposed into reflection component and illumination component. A low-light image enhancement model is constructed and trained to enhance the low-light image of the original steel-stamped character image and the texture detail information of the illumination component to obtain a low-light enhanced image.

3. The steel industry material steel stamp character recognition method according to claim 2, characterized in that: The low-light enhancement network is mainly composed of multiple convolutional layers with the same parameters. It extracts low-level contour features and high-level semantic features of the image layer by layer. Then, according to the Retinex theory, the image features are decomposed into illumination and reflection components through the last two convolutional layers. The illumination and reflection components are dynamically optimized through exposure control loss and texture consistency loss. The two components are then multiplied to obtain an enhanced image. Finally, the network parameters are optimized between the enhanced image and the original image through spatial consistency loss to obtain the final low-light enhanced image.

4. The steel industry material steel stamp character recognition method according to claim 3, characterized in that: The exposure control loss function formula is as follows: L exp =|I l -E|, where, I l represents the reflection component image, and E represents the average grayscale value of multiple high-brightness images collected under good lighting conditions.

5. The steel industry material steel stamp character recognition method according to claim 3, characterized in that: The texture consistency loss function formula is as follows: L tc =cos(φ vgg (I r )-φ vgg (I)), where cos() represents the cosine similarity function, I r represents the reflection component image, I represents the input image, φ vgg (·) denotes the VGG network.

6. The steel industry material steel stamp character recognition method according to claim 3, characterized in that: The spatial consistency loss function is as follows: Where K represents the number of image blocks, Y and I represent the average pixel values of the local area in the low-light enhanced image and the input image, respectively, and Ω(i) represents the four adjacent areas above, below, left, and right of area i.

7. The steel industry material steel stamp character recognition method according to claim 1, characterized in that: The step S4 comprises: The low-light enhanced images are used as training sets, and a character positioning semantic segmentation model is trained for the stamped characters. A class 1 label is set for the stamped character area, and each line of stamped characters is positioned as a target as a whole.

8. The steel industry material steel stamp character recognition method according to claim 1, characterized in that: The step S5 includes: performing tilt correction on the steel-printed characters so that all the steel-printed characters are in a horizontal or flipped state; The original stamped character image with an angle of 0° and the stamped character image with an angle of 180° generated in step S2 are used as training sets to train the character direction classification model, and the angles of all stamped characters are corrected to a horizontal state.

9. An electronic device comprising a processor and a memory, wherein the memory stores a program that can be run on the processor, characterized in that: When the program is executed by the processor, the steps of the steel industry material stamp character recognition method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium storing at least one program, characterized in that: The at least one program can be executed by at least one processor to implement the steps of the steel industry material steel stamp character recognition method described in any one of claims 1 to 8.

Images