Steel coil number recognition method and system based on DBNet and CRNN
By combining DBNet and CRNN models, steel coil numbers are automatically extracted and identified, solving the problems of time-consuming and error-prone identification in traditional methods, and achieving efficient and accurate steel coil number identification and management.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- 北京瓦特曼智能科技有限公司
- Filing Date
- 2023-07-25
- Publication Date
- 2026-07-21
AI Technical Summary
Traditional steel coil number identification methods are time-consuming and error-prone. Ambient lighting and background interference affect accuracy. Existing deep learning solutions have difficulty distinguishing steel coil number information, resulting in low identification reliability.
The DBNet model is used to extract the target bounding box region in the steel coil image, and the CRNN network model is used for character recognition. Combined with image preprocessing and secondary cropping, automated steel coil number recognition is achieved.
It improves the accuracy and robustness of steel coil number identification, reduces labor costs, enables rapid batch processing, and is suitable for automated management of steel coil-related processes.
Smart Images
Figure CN117011850B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image recognition and detection of steel ladles, specifically to a method and system for recognizing steel coil numbers based on DBNet and CRNN. Background Technology
[0002] The coil number is typically a key indicator for tracking the logistics of steel coil products. In traditional steel coil production processes, identifying and entering the coil number requires operators to manually watch on-site production line videos. This method is time-consuming and prone to errors, and the results also need to be manually entered into the electronic system, making the process cumbersome and inefficient.
[0003] To address this issue, image recognition technology has emerged as a viable solution. However, the complex factory environment and ambient lighting can affect the accuracy of roll number recognition, easily leading to identification anomalies. Furthermore, the roll numbers are distributed in an arc, making direct character extraction prone to errors in character order. Additionally, the paint used to print roll numbers on steel coils may be fragmented or discontinuous. Conventional deep learning-based roll number detection schemes, which directly recognize characters in images, are easily affected by background objects, making it difficult to distinguish roll number information from different steel coils, resulting in low reliability.
[0004] Based on the above, this application is hereby submitted. Summary of the Invention
[0005] In order to solve the technical problems existing in the prior art, this application provides a steel coil number recognition method and system based on DBNet and CRNN that has high versatility and stability.
[0006] According to a first aspect of this application, a method for recognizing coil numbers based on DBNet and CRNN is provided, comprising: acquiring a coil image; using a DBNet model to extract the target bounding box region of the coil number in the coil image and returning the coordinate information of the corresponding target bounding box region; cropping the coil image according to the coordinate information to obtain a region image for each coil number; using each cropped region image as input, and using a CRNN network model to recognize the coil number text to obtain a character sequence.
[0007] In a further embodiment of this application, the method of using the DBNet model to extract the target bounding box region of the steel coil number in the steel coil image and returning the coordinate information of the corresponding target bounding box region includes: preprocessing the steel coil image once; passing the preprocessed steel coil image through the forward propagation process of the DBNet model, in which the DBNet model obtains a steel coil confidence map based on feature extraction and contextual understanding of the steel coil image; filtering out target pixels in the steel coil confidence map that are higher than a preset first threshold; connecting the target pixels through a connected component analysis algorithm to obtain the target bounding box region, and feeding back the coordinate information of the target bounding box region, wherein the coordinate information includes the coordinates of the corner points and the coordinates of the center point of the target bounding box.
[0008] In a further embodiment of this application, cropping the steel coil image based on coordinate information to obtain the region image for each steel coil number includes: determining a rectangular region based on coordinate information, cropping the region from the steel coil image to obtain a rectangular region image; converting the rectangular region image into a grayscale image; and performing secondary preprocessing on the grayscale image to obtain the region image for each steel coil number.
[0009] In a further embodiment of this application, the method of using each cropped region image as input to obtain a character sequence by recognizing the serial number text through a CRNN network model includes: inputting the region image into the CRNN network model, extracting the feature sequence of the region image through the convolutional and pooling layers in the CRNN network model; inputting the extracted feature sequence into the recurrent neural network in the CRNN network model to obtain a text classification dot matrix; calculating the character classification result with the highest probability at each position in the text classification dot matrix and obtaining the corresponding confidence score; and determining the character sequence based on the character classification result and the confidence score.
[0010] In a further embodiment of this application, the steel coil number recognition method further includes: performing text arrangement transformation on the character sequence structure to make it linearly distributed, thereby obtaining the final steel coil number text.
[0011] In a further embodiment of this application, the text arrangement transformation of the character sequence structure to make it linearly distributed includes: estimating the position information of each character in the image through geometric transformation; performing text arrangement transformation on the character sequence structure to make it linearly distributed; and reconstructing the steel coil number text based on the sorted character sequence to form the final steel coil number text.
[0012] In a further embodiment of this application, calculating the character classification result with the highest probability at each position in the text classification dot matrix and obtaining the corresponding confidence score includes: using the argmax function to return the index of the character with the highest probability in the text classification dot matrix; and obtaining the corresponding character label and the confidence score of the corresponding character label through the index.
[0013] In a further embodiment of this application, the coil number identification method further includes associating the coil number text with the corresponding coil sequence in a linear layout.
[0014] A second aspect of this application provides a steel coil number recognition device based on DBNet and CRNN. The device includes: an acquisition module for acquiring images of steel coils; an extraction module for extracting target bounding boxes of steel coil numbers from the images using a DBNet model and returning the coordinate information of the corresponding target bounding boxes; a cropping module for cropping the images of steel coils according to the coordinate information to obtain region images for each steel coil number; and a recognition module for using each cropped region image as input and recognizing the coil number text through a CRNN network model to obtain a character sequence.
[0015] In a further embodiment of this application, the coil number identification device further includes an association module, used to associate the coil number text with the corresponding coil sequence in a linear layout.
[0016] The steel coil number recognition method based on DBNet and CRNN provided in this invention uses a DBNet model to obtain the target bounding box region, effectively detecting the target bounding box region of the steel coil number in the image, exhibiting high accuracy and robustness. Based on the target bounding box coordinate information returned by the DBNet model, the region of each steel coil number can be accurately cropped from the image, avoiding errors that may be introduced by manual or rule-based cropping operations. The CRNN network model can perform text recognition on the cropped steel coil number region image, handling different fonts, sizes, and complex backgrounds well, achieving high recognition accuracy. The entire process is automated, enabling rapid and batch processing of large numbers of steel coil images and outputting corresponding character sequences, improving work efficiency and reducing labor costs. In summary, the steel coil number recognition method provided in this invention can achieve automated, efficient, and accurate steel coil number extraction and recognition, suitable for automated processing and management scenarios related to steel coils.
[0017] The features and advantages of the present invention will be described in the following detailed description section. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, the collected drawings can be obtained from these drawings without creative effort.
[0019] Figure 1 The flowchart below shows the overall process of the steel coil number recognition method based on DBNet and CRNN provided in this embodiment of the invention.
[0020] Figure 2 This is a flowchart of step S200 in the steel coil number recognition method based on DBNet and CRNN provided in the embodiments of the present invention;
[0021] Figure 3 This is a flowchart of step S300 in the steel coil number recognition method based on DBNet and CRNN provided in the embodiments of the present invention;
[0022] Figure 4 This is a flowchart of step S400 in the steel coil number recognition method based on DBNet and CRNN provided in the embodiments of the present invention;
[0023] Figure 5 The flowchart is a step S403 in the steel coil number recognition method based on DBNet and CRNN provided in the embodiment of the present invention.
[0024] Figure 6 A flowchart of step S500 in the steel coil number recognition method based on DBNet and CRNN provided in this embodiment of the invention; and
[0025] Figure 7 This is a schematic diagram of the module of the steel coil number recognition device based on DBNet and CRNN provided in an embodiment of the present invention.
[0026] Marker description
[0027] 100. Steel coil number identification device;
[0028] 10. Acquisition module; 20. Extraction module;
[0029] 30. Cropping module; 40. Recognition module;
[0030] 50. Related modules. Detailed Implementation
[0031] To make the above-described features and advantages of this application clearer, the application is further described below with reference to the accompanying drawings. It should be understood that the specific embodiments given herein are for the purpose of explanation to those skilled in the art and are exemplary only, not restrictive.
[0032] In the following description, numerous specific details are set forth to provide a thorough understanding of this application. However, it will be apparent to those skilled in the art that these specific details are not required to practice this application. Where known steps or operations have been omitted from the description, they are not detailed to avoid obscuring this application.
[0033] Please refer to the following first. Figure 1 , Figure 1 The flowchart below shows the overall process of the steel coil number recognition method based on DBNet and CRNN provided in this embodiment of the invention.
[0034] The inventive concept of this invention provides a steel coil number recognition method based on DBNet and CRNN, comprising the following steps:
[0035] Step S100: Obtain an image of the steel coil;
[0036] Step S200: Use the DBNet model to extract the target bounding box region of the steel coil number in the steel coil image and return the coordinate information of the corresponding target bounding box region;
[0037] Step S300: Crop the steel coil image according to the coordinate information to obtain the region image for each steel coil number;
[0038] Step S400: Take each cropped region image as input and use the CRNN network model to recognize the volume number text to obtain the character sequence.
[0039] Understandably, in step S100, an image acquisition device is used to acquire an image of the steel coil, which can be an industrial camera.
[0040] In step S200, the training dataset is prepared: a dataset containing images of steel coils and corresponding bounding box annotations is collected. To ensure dataset quality, the training dataset is first preprocessed, such as image resizing and normalization. The prepared dataset is then used to train the DBNet model. The trained DBNet model is then used to detect bounding boxes in the steel coil images and returns the coordinate information of the bounding boxes.
[0041] It is understood that the DBNet model is a deep learning-based text detection model, which is used in this invention to detect text target boxes in images.
[0042] In step S300, the steel coil image is cropped according to the coordinate information to obtain the region image of each steel coil number;
[0043] Specifically, the bounding box coordinates of each steel coil number are determined based on the target bounding box coordinates output by the DBNet model. Using the bounding box coordinates, the steel coil image is cropped to obtain the region image for each steel coil number.
[0044] Then, step S400 is executed: A CRNN network model is used to perform steel coil number text recognition on each region image to obtain the character sequence.
[0045] First, prepare the training dataset, which involves collecting images containing coil numbers and corresponding character sequence annotations. Preprocess the region images, including resizing and normalization. Train the CRNN model using the prepared dataset. For each cropped region image, use the trained CRNN model to recognize the coil number text and output the corresponding character sequence.
[0046] As you can understand, the CRNN model is a deep learning-based text recognition model used to identify character sequences from images.
[0047] The steel coil number recognition method based on DBNet and CRNN provided in this invention uses a DBNet model to obtain the target bounding box region, effectively detecting the target bounding box region of the steel coil number in the image, exhibiting high accuracy and robustness. Based on the target bounding box coordinate information returned by the DBNet model, the region of each steel coil number can be accurately cropped from the image, avoiding errors that may be introduced by manual or rule-based cropping operations. The CRNN network model can perform text recognition on the cropped steel coil number region image, handling different fonts, sizes, and complex backgrounds well, achieving high recognition accuracy. The entire process is automated, enabling rapid and batch processing of large numbers of steel coil images and outputting corresponding character sequences, improving work efficiency and reducing labor costs. In summary, the steel coil number recognition method provided in this invention can achieve automated, efficient, and accurate steel coil number extraction and recognition, suitable for automated processing and management scenarios related to steel coils.
[0048] By following the steps above, the coil number can be extracted and identified from the coil image.
[0049] Please refer to the following first. Figure 2 , Figure 2 This is a flowchart of step S200 in the steel coil number recognition method based on DBNet and CRNN provided in the embodiments of the present invention.
[0050] In step S200, the DBNet model is used to extract the target bounding box region of the steel coil number in the steel coil image, and the coordinate information of the corresponding target bounding box region is returned, including:
[0051] Step S201: Perform one preprocessing on the steel coil image;
[0052] Step S202: The preprocessed steel coil image is passed through the forward propagation process of the DBNet model. Based on the feature extraction and contextual understanding of the steel coil image, the DBNet model obtains the steel coil confidence dot matrix.
[0053] Step S203: Filter out target pixels in the steel coil confidence dot matrix that are higher than the preset first threshold;
[0054] Step S204: Connect the target pixels using a connected component analysis algorithm to obtain the target bounding box region, and feed back the coordinate information of the target bounding box region, including the coordinates of the corner points and the coordinates of the center point of the target bounding box.
[0055] In step 201, some image processing techniques can be used to preprocess the steel coil image, such as image resizing, brightness and contrast adjustment, or image smoothing using filters, thereby improving image quality, highlighting target features, and providing better input conditions so that the subsequent processing and analysis of the DBNet model can be accurate.
[0056] In step S202, the preprocessed steel coil image is then input into the DBNet model for forward propagation. Based on feature extraction and contextual understanding of the steel coil image, the DBNet model generates a confidence dot matrix in its output.
[0057] In step S203, based on a set first threshold, target pixels with confidence levels higher than the first threshold are filtered from the confidence matrix generated by the DBNet model. These target pixels are considered regions that may contain the coil number, and regions with low confidence levels can be filtered out, retaining only the important target regions. This helps reduce the false detection rate and improves the accuracy and efficiency of subsequent processing.
[0058] Finally, a connected component analysis algorithm is used to connect target pixels above a threshold to form target bounding boxes. Then, the corner coordinates and center coordinates of the target bounding boxes are extracted from the connected target bounding boxes as the coordinate information of the target bounding boxes.
[0059] Please refer to the following first. Figure 3 , Figure 3 This is a flowchart of step S300 in the steel coil number recognition method based on DBNet and CRNN provided in an embodiment of the present invention.
[0060] In step S300, the steel coil image is cropped according to the coordinate information to obtain the region image for each steel coil number, including:
[0061] Step S301: Determine a rectangular region based on the coordinate information, and crop the region from the steel coil image to obtain a rectangular region image;
[0062] Step S302: Convert the rectangular region image into a grayscale image;
[0063] Step S303: Perform secondary preprocessing on the grayscale image to obtain the region image for each steel coil number.
[0064] It is understandable that step S301 determines a rectangular region based on coordinate information and crops this region from the coil image. This helps extract the target region of interest, reduces the processing range, and improves the efficiency and accuracy of subsequent processing. In step S302, the rectangular region image is converted into a grayscale image. Grayscale images only contain brightness information, removing the interference of color information, which is beneficial for subsequent secondary preprocessing and regional image analysis of the coil number. In step S303, secondary preprocessing is performed on the grayscale image to obtain the regional image of each coil number. Secondary preprocessing may include image enhancement, denoising, edge detection, and other operations to further optimize image quality and highlight the features of the coil number, providing better input conditions for subsequent recognition and analysis. Through the above steps, the regional image of each coil number can be located and extracted. At the same time, through grayscale conversion and secondary preprocessing operations, noise and interference are reduced, the features of the coil number are highlighted, providing a more reliable data foundation for subsequent recognition and analysis, and improving the accuracy and reliability of recognition.
[0065] Please refer to the following first. Figure 4 , Figure 4 This is a flowchart of step S400 in the steel coil number recognition method based on DBNet and CRNN provided in the embodiments of the present invention.
[0066] In step S400, each cropped region image is used as input, and the character sequence is obtained by recognizing the volume number text through a CRNN network model, including:
[0067] Step S401: Input the region image into the CRNN network model, and extract the feature sequence of the region image through the convolutional layer and pooling layer in the CRNN network model;
[0068] Step S402: Input the extracted feature sequence into the recurrent neural network in the CRNN network model to obtain the text classification dot matrix;
[0069] Step S403: Calculate the character classification result with the highest probability at each position in the text classification dot matrix and obtain the corresponding confidence score;
[0070] Step S404: Determine the character sequence based on the character classification results and confidence level.
[0071] The process involves inputting a region image into a CRNN (Convolutional Recurrent Neural Network) model. The CRNN model combines convolutional layers and recurrent neural networks, simultaneously extracting both spatial and sequential features from an image. In this step, the input region image is processed through convolutional and pooling layers to extract a feature sequence. This extracted feature sequence is then fed into the CRNN model's recurrent neural network. Recurrent neural networks (RNNs) can capture dependencies between sequential data. In this step, the CRNN model further processes the feature sequence, learning higher-level semantic information. The CRNN model calculates the probability distribution of each character category at each position based on the input feature sequence. In this step, the character category with the highest probability is prioritized as the classification result for that position, and its corresponding confidence value is obtained. The character sequence is then determined based on the character classification result and the confidence value. Based on the character classification result at each position and the corresponding confidence value, the entire character sequence can be determined. A threshold can be set based on the confidence value, selecting only character results with a confidence value higher than the threshold to improve the reliability of recognition.
[0072] Please refer to the following first. Figure 5 , Figure 5 This is a flowchart of step S403 in the steel coil number recognition method based on DBNet and CRNN provided in the embodiments of the present invention.
[0073] In step S403, the character classification result with the highest probability at each position in the text classification dot matrix is calculated, and the corresponding confidence level is obtained, including:
[0074] Step S4031: Use the argmax function to return the index of the character with the highest probability in the text classification dot matrix;
[0075] Step S4032: The corresponding character label and the confidence level of the corresponding character label can be obtained through the index.
[0076] The `argmax` function is used to return the index of the character with the highest probability in the text classification dot matrix. The CRNN network model calculates the probability distribution of each character category at each position, and then uses the `argmax` function to find the index corresponding to the character with the highest probability. The index provides the corresponding character label and its confidence score. By mapping the character index to a predefined list of character labels, the character label at each position in the recognition result can be obtained. Furthermore, the CRNN network model can output the confidence score for each character category, which can be obtained from the index position.
[0077] In this embodiment of the invention, the steel coil number identification method further includes:
[0078] Step S500: Perform text arrangement transformation on the character sequence structure to make it linearly distributed, and obtain the final steel coil number text.
[0079] Since the coil number is arranged in an arc, the following steps can be taken to convert the coil number text from an arc distribution to a linear distribution:
[0080] Based on the character sequence obtained in the previous steps, all characters are extracted without considering their order. The extracted characters are then sorted, ideally in ascending order by their coordinates, arranging them in a straight line from left to right. The sorted characters are then connected sequentially to form the final coil number text. Spaces or other delimiters can be used to separate each character for easier recognition and reading. These steps transform the original circularly distributed character sequence into a linear distribution, resulting in the final coil number text. This arrangement facilitates subsequent recognition and analysis.
[0081] Please refer to the following first. Figure 6 , Figure 6 This is a flowchart of step S500 in the steel coil number recognition method based on DBNet and CRNN provided in the embodiments of the present invention.
[0082] Transforming a character sequence structure into a linear arrangement includes:
[0083] Step S501: Estimate the position information of each character in the character sequence in the image through geometric transformation;
[0084] Step S502: Perform text arrangement transformation on the character sequence structure to make it linearly distributed;
[0085] Step S503: Reconstruct the coil number text based on the sorted character sequence to form the final coil number text.
[0086] Understandably, the positional information of each character in the character sequence within the image is estimated through geometric transformation: the purpose of this step is to locate the characters and determine their position in the original image. Image processing techniques, such as contour detection, edge detection, and morphological operations, can be used to find the bounding rectangle or bounding box of the character. Using the position and size information of these bounding boxes, the specific position of the character in the image can be calculated. Subsequently, the character sequence structure undergoes text arrangement transformation to make it linearly distributed. In the previous steps, the positional information of the characters has already been obtained. By processing this positional information, the arrangement transformation can be performed according to the relative positional relationships between the characters, making the character sequence linearly distributed. For example, the characters can be sorted in ascending order according to their coordinates, thus arranging the characters in a straight line from left to right. Based on the sorted character sequence, the coil number text is reconstructed to form the final coil number text. In step S502, the linear arrangement of the character sequence has been completed. Based on this sorted character sequence, the coil number text can be reconstructed. The characters can be connected sequentially to form the final coil number text.
[0087] In this embodiment of the application, the steel coil number identification method further includes:
[0088] Step S600: Associate the coil number text with the corresponding coil sequence in a linear layout.
[0089] Associating coil number text with the corresponding coil sequence using a linear layout enables better management. This layout clearly presents coil number information and intuitively displays the order of coils within the sequence. The linear layout allows for quick searching and identification of specific coil numbers. By associating the coil number text with the corresponding coil sequence, the position of a specific coil within the sequence can be easily located for tracking, querying, or manipulation. The linear layout arranges coil numbers sequentially, making sorting and organization easier. You can sort, rearrange, or reorganize the coil sequence based on the position of characters in the linear layout, achieving more efficient management. Simultaneously, visually presenting coil number information allows for quick understanding of coil distribution, status, and related information, facilitating decision-making and planning.
[0090] Please refer to the following first. Figure 7 , Figure 7 This is a schematic diagram of the module of the steel coil number recognition device 100 based on DBNet and CRNN provided in an embodiment of the present invention.
[0091] A steel coil number identification device 100 based on DBNet and CRNN includes:
[0092] Acquisition module 10 acquires images of the steel coil;
[0093] Extraction module 20 uses the DBNet model to extract the target bounding box region of the steel coil number in the steel coil image and returns the coordinate information of the corresponding target bounding box region;
[0094] The cropping module 30 crops the steel coil image according to the coordinate information to obtain the region image for each steel coil number;
[0095] The recognition module 40 takes each cropped region image as input and uses a CRNN network model to recognize the volume number text and obtain the character sequence.
[0096] The steel coil number identification device 100 also includes:
[0097] The association module 50 is used to associate the coil number text with the corresponding coil sequence in a linear layout.
[0098] Understandably, the coil number recognition device 100 can acquire images of coils, extract the target area of the coil number, crop the image, and recognize the text of the coil number. It can also associate the coil number with the corresponding coil sequence through a linear layout, thereby achieving efficient coil number management.
[0099] Furthermore, those skilled in the art should understand that if all or part of the sub-modules involved in the steel coil number identification device 100 provided in the embodiments of the present invention are combined or replaced by means of fusion, simple changes, mutual transformation, etc., such as moving the position of each component; or setting the product they constitute as a whole; or having a detachable design; any combined components can form a device / apparatus / system with a specific function, and using such a device / apparatus / system to replace the corresponding components of the present invention also falls within the protection scope of the present invention.
[0100] It should be understood that the specific features, operations, and details described herein with respect to the methods of this application can also be similarly applied to the apparatus and system of this application, or vice versa. Furthermore, each step of the methods of this application described above can be performed by a corresponding component or unit of the apparatus or system of this application.
[0101] This application provides a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the above-described brightness and contrast adjustment method.
[0102] Those skilled in the art will understand that the method steps of this application can be executed by a computer program instructing related hardware, such as measuring devices or processors. The computer program implementing the above-described brightness and contrast adjustment method can be stored in a non-transitory computer-readable storage medium, and its execution causes the steps of this application to be performed. Depending on the context, any reference to memory, storage, or other media herein may include non-volatile or volatile memory. Examples of non-volatile memory include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, etc. Examples of volatile memory include random access memory (RAM), external cache memory, etc.
[0103] The technical features described above can be combined arbitrarily. Although not all possible combinations of these technical features are described, any combination of these technical features should be considered to be covered by this specification, provided that such combination does not contain contradictions.
[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.
Claims
1. A method for recognizing steel coil numbers based on DBNet and CRNN, characterized in that, include: Acquire images of steel coils; The DBNet model is used to extract the target bounding box region of the steel coil number in the steel coil image, and the coordinate information of the corresponding target bounding box region is returned; The steel coil image is cropped based on the coordinate information to obtain the region image for each steel coil number; Each cropped region image is used as input, and the character sequence is obtained by recognizing the volume number text through a CRNN network model. The step of using the DBNet model to extract the target bounding box region of the steel coil number in the steel coil image and returning the coordinate information of the corresponding target bounding box region includes: The image of the steel coil undergoes one preprocessing step; The preprocessed steel coil image is passed through the forward propagation process of the DBNet model. Based on feature extraction and contextual understanding of the steel coil image, the DBNet model obtains a steel coil confidence dot matrix. Filter out target pixels in the steel coil confidence matrix that are higher than a preset first threshold; The target pixels are connected by a connected component analysis algorithm to obtain the target bounding box region, and the coordinate information of the target bounding box region is fed back, wherein the coordinate information includes the coordinates of the corner points and the coordinates of the center point of the target bounding box. The step of using each cropped region image as input and then using a CRNN network model to recognize the volume number text to obtain the character sequence includes: The region image is input into the CRNN network model, and the feature sequence of the region image is extracted through the convolutional layer and pooling layer in the CRNN network model; The extracted feature sequences are input into the recurrent neural network in the CRNN network model to obtain a text classification dot matrix. Calculate the character classification result with the highest probability at each position in the text classification dot matrix and obtain the corresponding confidence score; determine the character sequence based on the character classification result and the confidence score.
2. The steel coil number identification method according to claim 1, characterized in that, The step of cropping the steel coil image based on the coordinate information to obtain the region image for each steel coil number includes: A rectangular region is determined based on the coordinate information, and this region is cropped from the image of the steel coil to obtain the rectangular region image. Convert the rectangular region image into a grayscale image; The grayscale image is preprocessed a second time to obtain the region image for each steel coil number.
3. The steel coil number identification method according to claim 1, characterized in that, The steel coil number identification method further includes: The character sequence structure is transformed into a linear distribution to obtain the final steel coil number text.
4. The steel coil number identification method according to claim 3, characterized in that, The step of performing text arrangement transformation on the character sequence structure to make it linearly distributed includes: The position information of each character in the character sequence in the image is estimated by geometric transformation; The character sequence structure is transformed into a linear distribution by performing a text arrangement conversion. Based on the sorted character sequence, the coil number text is reconstructed to form the final coil number text.
5. The steel coil number identification method according to claim 4, characterized in that, The calculation of the character classification result with the highest probability at each position in the text classification dot matrix, and the corresponding confidence score, includes: The argmax function is used to return the index of the character with the highest probability in the text classification dot matrix. The corresponding character label and the confidence level of the corresponding character label can be obtained through the index.
6. The steel coil number identification method according to claim 5, characterized in that, The steel coil number identification method further includes: The coil number text is associated with the corresponding coil sequence in a linear layout.
7. A steel coil number identification device based on DBNet and CRNN, characterized in that, include: The acquisition module obtains images of the steel coil; The extraction module uses the DBNet model to extract the target bounding box region of the steel coil number in the steel coil image and returns the coordinate information of the corresponding target bounding box region; The cropping module crops the steel coil image according to the coordinate information to obtain the region image for each steel coil number; The recognition module takes each cropped region image as input and uses a CRNN network model to recognize the volume number text and obtain the character sequence. The step of using the DBNet model to extract the target bounding box region of the steel coil number in the steel coil image and returning the coordinate information of the corresponding target bounding box region includes: The image of the steel coil undergoes one preprocessing step; The preprocessed steel coil image is passed through the forward propagation process of the DBNet model. Based on feature extraction and contextual understanding of the steel coil image, the DBNet model obtains a steel coil confidence dot matrix. Filter out target pixels in the steel coil confidence matrix that are higher than a preset first threshold; The target pixels are connected by a connected component analysis algorithm to obtain the target bounding box region, and the coordinate information of the target bounding box region is fed back, wherein the coordinate information includes the coordinates of the corner points and the coordinates of the center point of the target bounding box. The step of using each cropped region image as input and then using a CRNN network model to recognize the volume number text to obtain the character sequence includes: The region image is input into the CRNN network model, and the feature sequence of the region image is extracted through the convolutional layer and pooling layer in the CRNN network model; The extracted feature sequences are input into the recurrent neural network in the CRNN network model to obtain a text classification dot matrix. Calculate the character classification result with the highest probability at each position in the text classification dot matrix and obtain the corresponding confidence score; determine the character sequence based on the character classification result and the confidence score.
8. The steel coil number identification device according to claim 7, characterized in that, The coil number identification device further includes an association module, used to associate the coil number text with the corresponding coil sequence in a linear layout.