Machine scene-oriented character recognition method and system

Abstract

The invention discloses a character recognition method and system for a machine tool scene, and relates to the technical field of computer vision. The method comprises the steps of obtaining an image of a machine tool numbering area through mobile terminal imaging equipment; substituting the image into a double-branch decoupling module to obtain an optimized image; substituting the optimized image into the character recognition intelligent model to obtain a character recognition result; performing semantic recognition on the character recognition result to judge whether the text is inverted or not; if the text is inverted, performing normalized direction correction on a character recognition result; and if the text is not inverted, outputting a character recognition result to the terminal. According to the invention, the definition of the image is improved, and the machine character recognition capability is improved; the recognition precision is improved through the character recognition intelligent model, and character inversion is recognized and judged, so that the problem that the tool character inversion cannot be recognized is solved, and the tool character recognition efficiency is improved.

Description

Technical Field

[0001] This invention belongs to the field of computer vision technology, specifically relating to a character recognition method and system for machine-based scenarios. Background Technology

[0002] With the development of digitalization, the current task is to extract and recognize character information in machinery scenarios, ensuring that this character data can be accurately and efficiently converted into readable digital information. This method and system utilize a series of technical means, such as image preprocessing, target character localization, feature extraction, and classification recognition, to separate and parse characters (such as equipment identifiers and parameter numbers) in complex environments within machinery scenarios from images. This constructs a recognition process adapted to machinery scenarios, providing accurate and reliable character information support for intelligent management and status monitoring of machinery.

[0003] Existing technology (publication number: CN111833513A) discloses a multi-channel serial number recognition method and apparatus. Step 1 involves acquiring images of the banknote to be identified under different spectra, and selecting two or more spectral images based on the contrast of the serial number under different spectra. Step 2 involves recognizing the serial number in the selected spectral images, identifying the identifiable serial number characters in each spectral image. Step 3 involves concatenating the identified serial number characters from each spectral image to form a complete serial number. By flexibly selecting spectral images based on the contrast of the serial number under different spectra, identifying the identifiable serial number characters in each spectral image, and then recognizing the complete serial number through coordinate mapping and serial number concatenation, this method can be applied to all financial equipment products that require serial number recognition, significantly improving serial number recognition efficiency.

[0004] The aforementioned patent improves recognition efficiency by recognizing products with serial numbers. However, during the recognition of machine (product) characters, the images captured by the camera device of the machine in operation are blurry, resulting in low image clarity and thus reducing the ability to recognize machine characters. Furthermore, it cannot accurately recognize machine characters when they are inverted, further reducing the recognition efficiency of machine characters. Summary of the Invention

[0005] The purpose of this invention is to solve the problem that images captured by camera devices of operating machinery are blurry, resulting in low image clarity and thus reducing the ability to recognize machine characters. Furthermore, the invention cannot accurately recognize inverted machine characters, thereby reducing the efficiency of machine character recognition. Therefore, this invention proposes a character recognition method and system for machine scenarios.

[0006] In a first aspect of this invention, a character recognition method for machine-based scenarios is proposed, the method comprising: Image of the equipment number area is acquired using a mobile imaging device; The optimized image is obtained by substituting the image into the dual-branch decoupling module; The optimized image is then substituted into the character recognition intelligent model to obtain the character recognition result; The character recognition results are used to perform semantic recognition to determine whether the text is reversed; If the text is reversed, the character recognition result will be normalized and the direction corrected. If the text is not reversed, the character recognition result will be output to the terminal.

[0007] Optionally, substituting the image into the dual-branch decoupling module to obtain the optimized image includes: The image is subjected to local blur feature extraction and global blur feature extraction respectively to obtain local branch features and global branch features; The image, the local branch features, and the global branch features are subjected to branch attention to obtain attention features; The attention features are convolved to obtain the first feature; The second feature is obtained by multiplying the local feature and the first feature element by element. The third feature is obtained by adding the global feature and the second feature element by element. The target image is obtained by convolving the third feature.

[0008] Optionally, the specific processes for local fuzzy feature extraction and global fuzzy feature extraction include: The local fuzzy feature extraction specifically includes: Substituting the image into a 5×5 convolutional layer yields the first convolutional feature; Substituting the first convolutional feature into a 3×3 convolutional layer yields the second convolutional feature. Substituting the second convolutional feature into a 1×1 convolutional layer yields the third convolutional feature; Substituting the third convolutional feature into a 3×3 convolutional layer yields the fourth convolutional feature; Substituting the fourth convolutional feature into a 5×5 convolutional layer yields local branch features; The global fuzzy feature extraction specifically includes: Substituting the image into a 7×7 convolutional layer yields the fifth convolutional feature; Substituting the fifth convolutional feature into a 5×5 convolutional layer yields the sixth convolutional feature; Substituting the sixth convolutional feature into a 3×3 convolutional layer yields the seventh convolutional feature; Substituting the seventh convolutional feature into a 5×5 convolutional layer yields the eighth convolutional feature; Substituting the eighth convolutional feature into a 7×7 convolutional layer yields the global branch feature.

[0009] Optionally, the character recognition intelligent model is an improvement based on the original YOLOv8 model, specifically including: In the neck structure, replace all C2f modules with CCR_C2f modules; The working principle of the CCR_C2f module specifically includes: The feature map input to the CCR_C2f module is determined as the input feature map; The input feature map is convolved into a 3×3 convolutional layer to obtain the first convolutional feature map; The first convolutional feature map is normalized to obtain the second convolutional feature map; The second convolutional feature map is activated using the ReLU function to obtain the activated feature map; The activated feature map is segmented to obtain three feature maps with the same number of channels, namely the first segmentation feature map, the second segmentation feature map, and the third segmentation feature map, and the number of channels in each map is C; The first segmentation feature map is subjected to a 3×3 deep convolutional layer to obtain a second convolutional feature map with 2C channels; the second segmentation feature map is subjected to a 1×1 deep convolutional layer to obtain a third convolutional feature map with C channels; the third segmentation feature map is subjected to a 1×1 deep convolutional layer to obtain a fourth convolutional feature map with C channels. The second, third, and fourth convolutional feature maps are fused to obtain a fused feature map; the fused feature map is then subjected to a 1×1 convolutional layer to obtain a fifth convolutional feature map with 2C channels; the fifth convolutional feature map and the input feature map are added element by element to obtain the target feature map.

[0010] Optionally, the semantic recognition process includes: Retrieve the set of character text from the preset equipment database; The target recognized character text is obtained by verifying and comparing the character recognition results based on the character text set and then outputting the target recognized character text. The target character text is filtered to obtain the first character text; The target character text is obtained by semantically sorting the first character text.

[0011] In a second aspect of this invention, a character recognition system for machine-based scenarios is proposed, the system comprising: Image acquisition module: Acquires images of the equipment number area via a mobile imaging device; Image optimization module: Substitute the image into the dual-branch decoupling module to obtain the optimized image; Character recognition intelligent module: Substitutes the optimized image into the character recognition intelligent model to obtain the character recognition result; Machine character judgment module: performs semantic recognition on the character recognition result to determine whether the text is reversed; if the text is reversed, the character recognition result is normalized and corrected; if the text is not reversed, the character recognition result is output to the terminal.

[0012] Optionally, the image optimization module includes: a blur extraction module, a branch attention module, a feature convolution module, a multiplication module, and an addition module. The fuzz extraction module is used to extract local fuzz features and global fuzz features from the image to obtain local branch features and global branch features, respectively. The branch attention module is used to perform branch attention on the image, the local branch features, and the global branch features to obtain attention features; The feature convolution module is used to convolve the attention features to obtain the first feature; The multiplication module is used to multiply the local feature and the first feature element by element to obtain the second feature; The addition module is used to add the global feature and the second feature element by element to obtain the third feature; The target image is obtained by convolving the third feature.

[0013] Optionally, the specific processes for local fuzzy feature extraction and global fuzzy feature extraction include: The local fuzzy feature extraction specifically includes: Substituting the image into a 5×5 convolutional layer yields the first convolutional feature; Substituting the first convolutional feature into a 3×3 convolutional layer yields the second convolutional feature. Substituting the second convolutional feature into a 1×1 convolutional layer yields the third convolutional feature; Substituting the third convolutional feature into a 3×3 convolutional layer yields the fourth convolutional feature; Substituting the fourth convolutional feature into a 5×5 convolutional layer yields local branch features; The global fuzzy feature extraction specifically includes: Substituting the image into a 7×7 convolutional layer yields the fifth convolutional feature; Substituting the fifth convolutional feature into a 5×5 convolutional layer yields the sixth convolutional feature; Substituting the sixth convolutional feature into a 3×3 convolutional layer yields the seventh convolutional feature; Substituting the seventh convolutional feature into a 5×5 convolutional layer yields the eighth convolutional feature; Substituting the eighth convolutional feature into a 7×7 convolutional layer yields the global branch feature.

[0014] Optionally, the character recognition intelligent module is further configured to use an intelligent character recognition model that is an improvement upon the original YOLOv8 model, specifically including: In the neck structure, replace all C2f modules with CCR_C2f modules; The working principle of the CCR_C2f module specifically includes: The feature map input to the CCR_C2f module is determined as the input feature map; The input feature map is convolved into a 3×3 convolutional layer to obtain the first convolutional feature map; The first convolutional feature map is normalized to obtain the second convolutional feature map; The second convolutional feature map is activated using the ReLU function to obtain the activated feature map; The activated feature map is segmented to obtain three feature maps with the same number of channels, namely the first segmentation feature map, the second segmentation feature map, and the third segmentation feature map, and the number of channels in each map is C; The first segmentation feature map is subjected to a 3×3 deep convolutional layer to obtain a second convolutional feature map with 2C channels; the second segmentation feature map is subjected to a 1×1 deep convolutional layer to obtain a third convolutional feature map with C channels; the third segmentation feature map is subjected to a 1×1 deep convolutional layer to obtain a fourth convolutional feature map with C channels. The second, third, and fourth convolutional feature maps are fused to obtain a fused feature map; the fused feature map is then subjected to a 1×1 convolutional layer to obtain a fifth convolutional feature map with 2C channels; the fifth convolutional feature map and the input feature map are added element by element to obtain the target feature map.

[0015] Optionally, the machine character judgment module includes: a character text module, a verification module, a filtering module, and a semantic sorting module. The character text module is used to obtain a set of character texts from a preset equipment database; The verification module is used to verify and compare the character recognition results based on the character text set and output the target recognized character text. The filtering module is used to filter the target recognized character text to obtain the first character text; The semantic sorting module is used to perform semantic sorting on the first character text to obtain the target character text.

[0016] The beneficial effects of this invention are: This invention proposes a character recognition method and system for machinery scenarios. Firstly, addressing the common image blurring problem in machinery scenarios, a dual-branch decoupling module is used to specifically optimize blurred images, effectively improving texture details and clarity. This provides a high-quality image input foundation for subsequent character recognition, significantly enhancing the basic character recognition capability for machinery. Secondly, an intelligent character recognition model is introduced, which, while greatly improving character recognition accuracy, also possesses the ability to specifically recognize and judge scenarios where characters are reversed. This method successfully overcomes the technical bottleneck of traditional recognition technologies, which cannot effectively recognize reversed characters on machinery. It not only ensures the accuracy of character recognition in complex scenarios but also further improves the overall efficiency of character recognition on machinery. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 A flowchart illustrating a character recognition method for machine-based scenarios provided in an embodiment of the present invention; Figure 2 A network structure diagram of a dual-branch decoupling module provided in an embodiment of the present invention; Figure 3 This is a network structure diagram of a character recognition intelligent model provided in an embodiment of the present invention; Figure 4 A network structure diagram of the original YOLOv8 model provided in an embodiment of the present invention. Figure 5 This is a framework diagram of a character recognition system for machine-based scenarios provided in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and B can represent: A alone, A and B simultaneously, and B alone. Furthermore, descriptions involving "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" can explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0020] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] This invention provides a character recognition method for machine-based scenarios. See also... Figure 1 , Figure 1 A flowchart illustrating a character recognition method for machine-based scenarios provided by an embodiment of the present invention. The method includes the following steps: Image of the equipment number area is acquired using a mobile imaging device; The optimized image is obtained by substituting the image into the dual-branch decoupling module; The optimized image is then fed into the character recognition intelligent model to obtain the character recognition result; Perform semantic recognition on the character recognition results to determine whether the text is reversed; If the text is reversed, the character recognition results will be normalized and the direction corrected. If the text is not reversed, the character recognition result will be output to the terminal.

[0022] Based on the character recognition method for machine scenarios provided by the embodiments of the present invention, the method optimizes blurred images through a dual-branch decoupling module, thereby improving image clarity and the ability to recognize machine characters; it improves recognition accuracy through a character recognition intelligent model, and solves the problem of unrecognizable machine characters when they are reversed by recognizing them, thus improving the recognition efficiency of machine characters.

[0023] Specifically, the normalization direction correction includes: The character recognition results (machine text character images) are divided into two categories at 0° or 180°. In the rectangular viewfinder recognition frame of the mobile device, the machine characters actually appear at the classification angle (e.g., 0° or 180°). Therefore, in actual shooting, the characters are easily inverted.

[0024] In one implementation, Figure 2 This invention provides a network structure diagram of a dual-branch decoupling module. Substituting the image into the dual-branch decoupling module to obtain an optimized image includes: Local and global fuzzy feature extractions are performed on the image to obtain local and global branch features, respectively. Attention features are obtained by applying branch attention to the image, local branch features, and global branch features; The first feature is obtained by convolving the attention features; The second feature is obtained by element-wise multiplying the local feature and the first feature; The third feature is obtained by adding the global feature and the second feature element by element. The target image is obtained by convolving the third feature.

[0025] In one implementation, global and local branch features of the input image are extracted through two-branch feature extraction; then, attention features are obtained through dynamic fusion of branch attention, and multiplied by local branch features at the element level, so that the local branch module focuses on local features; finally, the image clarity of the machine number area is improved through the combined effect of local branch features and branch features.

[0026] In one implementation, the specific processes for local fuzzy feature extraction and global fuzzy feature extraction include: Local fuzzy feature extraction specifically includes: Substituting the image into a 5×5 convolutional layer yields the first convolutional feature; Substituting the first convolutional feature into a 3×3 convolutional layer yields the second convolutional feature; Substituting the second convolutional feature into a convolutional layer with a 1×1 kernel yields the third convolutional feature; Substituting the third convolutional feature into a 3×3 convolutional layer yields the fourth convolutional feature; Substituting the fourth convolutional feature into a 5×5 convolutional layer yields local branch features; Global fuzzy feature extraction specifically includes: Substituting the image into a 7×7 convolutional layer yields the fifth convolutional feature; Substituting the fifth convolutional feature into a 5×5 convolutional layer yields the sixth convolutional feature; Substituting the sixth convolutional feature into a 3×3 convolutional layer yields the seventh convolutional feature; Substituting the seventh convolutional feature into a 5×5 convolutional layer yields the eighth convolutional feature; The eighth convolutional feature is fed into a 7×7 convolutional layer to obtain the global branch feature.

[0027] In one implementation, the local branch uses a sequence of convolutional kernels with kernel sizes of 5×5, 3×3, 1×1, 3×3, and 5×5. The 1×1 convolutional kernel achieves feature fusion and dimensionality reduction in the channel dimension, reducing computational redundancy. At the same time, it focuses on local details and edge information, and strengthens the detail representation capability through multiple rounds of convolution iterations, accurately extracting fine-grained features such as target texture and contour, providing accurate local support for subsequent recognition, segmentation, and other tasks. The global branch mainly uses large-size convolutional kernels of 7×7, 5×5, 3×3, 5×5, and 7×7. The large convolutional kernels expand the receptive field, enabling the capture of the overall shape, spatial layout, and contextual information of the target across a larger pixel range, avoiding the one-sidedness of local features. It is especially suitable for global association modeling of targets and backgrounds in complex scenes, improving the global consistency of features. By complementing the convolutional kernel sizes of local and global branches, local detail features and global contextual features work together to solve the problems of insufficient receptive field and lack of global information in a single small convolutional kernel, while also making up for the weakness of a single large convolutional kernel in capturing details. This results in the final fused features having both accuracy and comprehensiveness, and enhances the adaptability of acquiring images of machine numbered regions in complex scenes (such as when the camera is shaking or the machine is in motion), thereby improving the task's generalization performance.

[0028] In one implementation, Figure 3 This is a network structure diagram of a character recognition intelligent model provided in an embodiment of the present invention. Figure 4 This invention provides a network structure diagram of the original YOLOv8 model. The character recognition intelligent model is an improvement on the original YOLOv8 model, specifically including: In the neck structure, replace all C2f modules with CCR_C2f modules; The working principle of the CCR_C2f module specifically includes: The feature map input to the CCR_C2f module is determined as the input feature map; The first convolutional feature map is obtained by performing a 3×3 convolutional layer on the input feature map. The second convolutional feature map is obtained by normalizing the first convolutional feature map. The activated feature map is obtained by applying the ReLU function to the second convolutional feature map; The activation feature map is segmented to obtain three feature maps with the same number of channels: the first segmentation feature map, the second segmentation feature map, and the third segmentation feature map, and the number of channels in each map is C. The first segmentation feature map is convolved with a 3×3 deep convolutional layer to obtain a second convolutional feature map with 2C channels; the second segmentation feature map is convolved with a 1×1 deep convolutional layer to obtain a third convolutional feature map with C channels; the third segmentation feature map is convolved with a 1×1 deep convolutional layer to obtain a fourth convolutional feature map with C channels. The second, third, and fourth convolutional feature maps are fused to obtain a fused feature map; the fused feature map is then subjected to a 1×1 convolutional layer to obtain a fifth convolutional feature map with 2C channels; the fifth convolutional feature map and the input feature map are added element-wise to obtain the target feature map.

[0029] In one implementation, the head component of YOLOv8 is enhanced by replacing the traditional C2f module with a scaled-down directional residual (CCR_C2f) module. For machine character image data, which typically contains targets of varying sizes—large, medium, and small—the standard YOLOv8n model may face challenges such as information loss, high computational complexity, and insufficient generalization ability when processing multi-scale contextual information. These problems can lead to insufficient object recognition and detection at different scales, potentially resulting in omissions or misjudgments. Therefore, this approach enhances the extraction of multi-scale contextual information based on the DWR module: region residualization and semantic residualization. This method not only improves the model's ability to detect objects at multiple scales but also enhances the overall detection performance of machine characters, effectively addressing the inherent limitations of traditional single-scale detection methods.

[0030] In one implementation, the semantic recognition process specifically includes: Retrieve the set of character text from the preset equipment database; The target recognized character text is obtained by verifying and comparing the character recognition results based on the character text set. The target character text is filtered to obtain the first character text; The target character text is obtained by semantically sorting the first character text.

[0031] In one implementation, specifically, a character text set is preset in the equipment database (e.g., only A–Z, 0–9, and some symbols such as "-", "*", etc.), and the target character text is output through verification and comparison (to avoid identifying irrelevant Chinese characters, punctuation marks, or other interfering characters from the source). The specific filtering method is to remove individual characters (such as "1") misidentified by the background using regular expressions (^[A-Z0-9-*.]+$) and length rules (len>1). The semantic sorting is specifically to ensure that "prefixes" starting with letters (such as "QZ") are always placed before numeric "suffixes" (such as "-02287"), thereby solving the problem of reversed order.

[0032] Based on the same inventive concept, this invention also provides a character recognition system for machine-based scenarios. See also Figure 5 , Figure 5 A framework diagram of a character recognition system for machine-based scenarios provided in this embodiment of the invention includes: Image acquisition module: Acquires images of the equipment number area via a mobile imaging device; Image optimization module: Substitute the image into the dual-branch decoupling module to obtain the optimized image; Character recognition intelligent module: This module inputs the optimized image into the character recognition intelligent model to obtain the character recognition result. Machine character judgment module: Performs semantic recognition on the character recognition results to determine whether the text is reversed; if the text is reversed, the character recognition results are normalized and the direction is corrected; if the text is not reversed, the character recognition results are output to the terminal.

[0033] Based on the character recognition system for machine scenarios provided by the embodiments of the present invention, the system optimizes blurred images through a dual-branch decoupling module, thereby improving image clarity and the ability to recognize machine characters; it improves recognition accuracy through a character recognition intelligent model, and solves the problem of unrecognizable machine characters when they are reversed by recognizing them, thus improving the recognition efficiency of machine characters.

[0034] The foregoing has described one embodiment of the present invention in detail, but this content is merely a preferred embodiment and should not be considered as limiting the scope of the present invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the scope of the claims of this invention.

Claims

1. A character recognition method for machine-based scenarios, characterized in that, The method includes: Image of the equipment number area is acquired using a mobile imaging device; The optimized image is obtained by substituting the image into the dual-branch decoupling module; The optimized image is then substituted into the character recognition intelligent model to obtain the character recognition result; The character recognition results are used to perform semantic recognition to determine whether the text is reversed; If the text is reversed, the character recognition result will be normalized and the direction corrected. If the text is not reversed, the character recognition result will be output to the terminal.

2. The character recognition method for machine-based scenarios according to claim 1, characterized in that, The step of substituting the image into the dual-branch decoupling module to obtain the optimized image includes: The image is subjected to local blur feature extraction and full blur feature extraction respectively to obtain local branch features and global branch features; The image, the local branch features, and the global branch features are subjected to branch attention to obtain attention features; The attention features are convolved to obtain the first feature; The second feature is obtained by multiplying the local feature and the first feature element by element. The third feature is obtained by adding the global feature and the second feature element by element. The target image is obtained by convolving the third feature.

3. The character recognition method for machine-based scenarios according to claim 2, characterized in that, The specific processes for local fuzzy feature extraction and global fuzzy feature extraction include: The local fuzzy feature extraction specifically includes: Substituting the image into a 5×5 convolutional layer yields the first convolutional feature; Substituting the first convolutional feature into a 3×3 convolutional layer yields the second convolutional feature. Substituting the second convolutional feature into a 1×1 convolutional layer yields the third convolutional feature; Substituting the third convolutional feature into a 3×3 convolutional layer yields the fourth convolutional feature; Substituting the fourth convolutional feature into a 5×5 convolutional layer yields local branch features; The global fuzzy feature extraction specifically includes: Substituting the image into a 7×7 convolutional layer yields the fifth convolutional feature; Substituting the fifth convolutional feature into a 5×5 convolutional layer yields the sixth convolutional feature; Substituting the sixth convolutional feature into a 3×3 convolutional layer yields the seventh convolutional feature; Substituting the seventh convolutional feature into a 5×5 convolutional layer yields the eighth convolutional feature; Substituting the eighth convolutional feature into a 7×7 convolutional layer yields the global branch feature.

4. The character recognition method for machine-based scenarios according to claim 1, characterized in that, The character recognition intelligent model is an improvement on the original YOLOv8 model, with specific improvements including: In the neck structure, replace all C2f modules with CCR_C2f modules; The working principle of the CCR_C2f module specifically includes: The feature map input to the CCR_C2f module is determined as the input feature map; The input feature map is convolved into a 3×3 convolutional layer to obtain the first convolutional feature map; The first convolutional feature map is normalized to obtain the second convolutional feature map; The second convolutional feature map is activated using the ReLU function to obtain the activated feature map; The activated feature map is segmented to obtain three feature maps with the same number of channels, namely the first segmentation feature map, the second segmentation feature map, and the third segmentation feature map, and the number of channels in each map is C; The first segmentation feature map is subjected to a 3×3 deep convolutional layer to obtain a second convolutional feature map with 2C channels; the second segmentation feature map is subjected to a 1×1 deep convolutional layer to obtain a third convolutional feature map with C channels; the third segmentation feature map is subjected to a 1×1 deep convolutional layer to obtain a fourth convolutional feature map with C channels. The second, third, and fourth convolutional feature maps are fused to obtain a fused feature map; the fused feature map is then subjected to a 1×1 convolutional layer to obtain a fifth convolutional feature map with 2C channels; the fifth convolutional feature map and the input feature map are added element by element to obtain the target feature map.

5. A character recognition method for machine-based scenarios according to claim 1, characterized in that, The specific process of semantic recognition includes: Retrieve the set of character text from the preset equipment database; The target recognized character text is obtained by verifying and comparing the character recognition results based on the character text set and then outputting the target recognized character text. The target character text is filtered to obtain the first character text; The target character text is obtained by semantically sorting the first character text.

6. A character recognition system for machinery scenarios, characterized in that, The system includes: Image acquisition module: Acquires images of the equipment number area via a mobile imaging device; Image optimization module: Substitute the image into the dual-branch decoupling module to obtain the optimized image; Character recognition intelligent module: Substitutes the optimized image into the character recognition intelligent model to obtain the character recognition result; Machine character judgment module: performs semantic recognition on the character recognition result to determine whether the text is reversed; if the text is reversed, the character recognition result is normalized and corrected; if the text is not reversed, the character recognition result is output to the terminal.

7. A character recognition system for machine-based scenarios according to claim 6, characterized in that, The image optimization module includes: a blur extraction module, a branch attention module, a feature convolution module, a multiplication module, and an addition module. The fuzz extraction module is used to extract local fuzz features and global fuzz features from the image to obtain local branch features and global branch features, respectively. The branch attention module is used to perform branch attention on the image, the local branch features, and the global branch features to obtain attention features; The feature convolution module is used to convolve the attention features to obtain the first feature; The multiplication module is used to multiply the local feature and the first feature element by element to obtain the second feature; The addition module is used to add the global feature and the second feature element by element to obtain the third feature; The target image is obtained by convolving the third feature.

8. A character recognition system for machine-based scenarios according to claim 7, characterized in that, The specific processes for local fuzzy feature extraction and global fuzzy feature extraction include: The local fuzzy feature extraction specifically includes: Substituting the image into a 5×5 convolutional layer yields the first convolutional feature; Substituting the first convolutional feature into a 3×3 convolutional layer yields the second convolutional feature. Substituting the second convolutional feature into a 1×1 convolutional layer yields the third convolutional feature; Substituting the third convolutional feature into a 3×3 convolutional layer yields the fourth convolutional feature; Substituting the fourth convolutional feature into a 5×5 convolutional layer yields local branch features; The global fuzzy feature extraction specifically includes: Substituting the image into a 7×7 convolutional layer yields the fifth convolutional feature; Substituting the fifth convolutional feature into a 5×5 convolutional layer yields the sixth convolutional feature; Substituting the sixth convolutional feature into a 3×3 convolutional layer yields the seventh convolutional feature; Substituting the seventh convolutional feature into a 5×5 convolutional layer yields the eighth convolutional feature; Substituting the eighth convolutional feature into a 7×7 convolutional layer yields the global branch feature.

9. A character recognition system for machine-based scenarios according to claim 6, characterized in that, The character recognition intelligent module is also used to improve the character recognition intelligent model based on the original YOLOv8 model. Specific improvements include: In the neck structure, replace all C2f modules with CCR_C2f modules; The working principle of the CCR_C2f module specifically includes: The feature map input to the CCR_C2f module is determined as the input feature map; The input feature map is convolved into a 3×3 convolutional layer to obtain the first convolutional feature map; The first convolutional feature map is normalized to obtain the second convolutional feature map; The second convolutional feature map is activated using the ReLU function to obtain the activated feature map; The activated feature map is segmented to obtain three feature maps with the same number of channels, namely the first segmentation feature map, the second segmentation feature map, and the third segmentation feature map, and the number of channels in each map is C; The first segmentation feature map is subjected to a 3×3 deep convolutional layer to obtain a second convolutional feature map with 2C channels; the second segmentation feature map is subjected to a 1×1 deep convolutional layer to obtain a third convolutional feature map with C channels; the third segmentation feature map is subjected to a 1×1 deep convolutional layer to obtain a fourth convolutional feature map with C channels. The second, third, and fourth convolutional feature maps are fused to obtain a fused feature map; the fused feature map is then subjected to a 1×1 convolutional layer to obtain a fifth convolutional feature map with 2C channels; the fifth convolutional feature map and the input feature map are added element by element to obtain the target feature map.

10. A character recognition system for machine-based scenarios according to claim 6, characterized in that, The machine character judgment module includes: a character text module, a verification module, a filtering module, and a semantic sorting module. The character text module is used to obtain a set of character texts from a preset equipment database; The verification module is used to verify and compare the character recognition results based on the character text set and output the target recognized character text. The filtering module is used to filter the target recognized character text to obtain the first character text; The semantic sorting module is used to perform semantic sorting on the first character text to obtain the target character text.

Images