Large model-based road place name sign standardized detection method

Abstract

The invention discloses a road place name sign standardization detection method based on a large model, and the method is characterized in that the method comprises the following steps: S1, building a standardization place name information database, and issuing a place name standardization information online query service interface; s2, performing automatic identification and extraction, structured coding and cloud storage on Chinese names and Roman alphabet spelling in the place name marks; and S3, comparing the place name sign information stored in the cloud with the standardized place name information, identifying a non-standard place name sign, and outputting a non-standard place name sign ledger. According to the method, large model technologies such as image recognition, character extraction and standardized comparison are utilized, rapid recognition and detection of non-standard road type place name signs are achieved, corresponding ledgers are generated, standardization of the place name signs is promoted, the modern digital management level of place names is greatly improved, and the service efficiency of the place names is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of standardized detection, and specifically is a large-model-based standardized detection method for road place name signs. Background Art

[0002] According to the relevant provisions of the "Regulations on the Management of Place Names", the setting of place name signs must not only be consistent with the approved standard names of place names, but also the Romanization of place names should be standardized in accordance with the "Chinese Pinyin Scheme". However, some place name signs currently have problems such as inconsistent styles, incomplete content, and non-standard Romanization, which are mainly reflected in road place name signs. For example, when setting place name signs, the road "LU" is irregularly marked as "Rd" and the street "JIE" is irregularly marked as "St". At the same time, the place name management department lacks effective technical means to detect and identify non-standard road place name signs, which not only affects communication and the optimization of the business environment, but also affects the efficiency and service level of place name management. Summary of the invention

[0003] In view of the above problems, the present invention is proposed to provide a large model-based road place name sign normalization detection method that overcomes the above problems or at least partially solves the above problems.

[0004] To achieve the above object, the present invention adopts the following technical solutions: A method for detecting road place name signs based on a large model, the method comprising the following steps: S1. Establish a standardized place name information database and publish an online query service interface for place name standardization information; S2, automatic recognition and extraction of Chinese names and Roman spellings in place names, structured coding and cloud storage; S3. Compare the place name sign information stored in the cloud with the standardized place name information, identify non-standard place name signs, and output the non-standard place name sign ledger.

[0005] Optionally, step S2 includes: S201, capturing and recording image information and location information of place name signs along the road through a mobile acquisition device; S202, establishing a lightweight deep learning image recognition model and deploying it in a mobile acquisition device to detect and identify in real time whether the image captured in step S201 is a place name sign; S203, establishing a lightweight deep learning text recognition model and deploying it in a mobile acquisition device to extract relevant characters of place name signs in real time; S204, deploying the big model in the cloud and publishing a service interface. The acquisition device calls the big model in real time through the interface request, and performs semantic understanding on the recognition result of step S203 to achieve accurate correction of the image recognition result; S205, based on the recognition result of step S203 and the large model understanding result of step S204, the pictures and the results are clustered based on the spatial coordinates to form clusters indexed by spatial features, and these clusters are encoded for subsequent data management and analysis; S206, uploading the encoded result to the cloud database, and establishing a spatial feature index after decoding for subsequent query and analysis.

[0006] Optionally, step S201 includes using a high-resolution camera and a high-precision global positioning system acquisition device to capture real-time images of place name signs at a designated location or along a preset route.

[0007] Optionally, step S202 includes: Develop and train a deep learning image recognition model based on convolutional neural networks to identify and judge place name signs; Use pruning, quantization, and lightweight technologies to optimize the convolutional neural network model, reduce the model's computing resource requirements, and increase the running speed to adapt to the processing capabilities of mobile devices; Deploy the optimized convolutional neural network model to a mobile device with image acquisition capabilities; The built-in software of the mobile device contains a model inference engine. After the captured image is input into the software, the software calls the deployed deep learning model to analyze the image and determine whether it contains place name signs.

[0008] Optionally, step S203 includes: Build a deep learning model for optical character classification and recognition to identify and classify text content on place name signs; A dataset of place name images was established, and random cropping, perspective transformation, random scaling and translation techniques were used to enhance sample diversity. At the same time, illumination and contrast were simulated to increase image Gaussian noise and salt and pepper noise to enhance data robustness. The place name sign image is processed by an image encoder, and the pre-processed image features are obtained; Using convolutional networks, we optimized the design of repeated gradient information and extracted the basic coding features of the feature images of place name signs to complete the segmentation of the Chinese name and Roman alphabet spelling areas of the image. The image features and task instructions are input into the structured center point sequence decoder to generate a structured information sequence; Using structured center point sequences and image features, input text content decoder for text recognition and parsing; Based on the decoded text content and structured center point sequence, the final image parsing result is output.

[0009] Optionally, step S204 includes: Collect text images of different scenes, fonts, sizes, and blur levels as model training sets, and verify and modify the annotation information to ensure the quality of the training set; When text is misrecognized due to blurriness or damage, the semantic understanding capability of the large model is used to correct the recognition result.

[0010] Optionally, step S205 includes dividing the area into several grids according to spatial positions, corresponding the coordinate points of the image to each grid, and integrating the corresponding place name classification and recognition results into the grid to form a cluster indexed by spatial features and the image and its place name classification result value.

[0011] Optionally, step S206 includes decoding the uploaded data using a cloud interface, creating an index based on spatial features, and storing the index in a structured manner in a cloud database.

[0012] Optionally, step S3 includes reading the name and Roman letter spelling information of the structured place name signs in the cloud, and comparing them with the standardized place name information, marking the place names that are inconsistent between the two, using the spatial index to extract the generated non-standard place name ledger information from the cloud database, and providing feedback.

[0013] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: The present invention utilizes large-scale model technologies such as image recognition, text extraction, and standardized comparison to realize rapid recognition and detection of non-standard road place name signs, and generates corresponding ledgers, thereby promoting the standardization of place name signs, greatly improving the level of modern digital management of place names, and improving the efficiency of place name services. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A schematic flow chart of a large-model-based method for normalizing detection of road place name signs provided in an embodiment of the present application. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0016] See also Figure 1 This embodiment provides a method for detecting road place name signs based on a large model, and the method comprises the following steps: S1. Establish a standardized place name information database and publish an online query service interface for place name standardization information.

[0017] Extract road place name information approved by the Beijing place name administrative department, including standard name, Roman alphabet spelling, district, longitude and latitude, and release time. Establish a standard place name information database and publish an online query service interface for place name standardization information.

[0018] S2. Automatically identify and extract the Chinese names and Roman spellings in place names, and perform structured coding and cloud storage.

[0019] Step S2 includes: S201. Capture and record the image information and location information of place-name signs along the road through a mobile acquisition device.

[0020] S202. Establish a lightweight deep learning image recognition model and deploy it in a mobile acquisition device to detect and identify in real time whether the image captured in step S201 is a place name sign.

[0021] S203. Establish a lightweight deep learning text recognition model and deploy it in a mobile acquisition device to extract relevant characters of place name signs in real time.

[0022] S204: deploy the big model in the cloud and publish a service interface. The acquisition device calls the big model in real time through the interface request, and performs semantic understanding on the recognition result of step S203 to achieve accurate correction of the image recognition result.

[0023] S205: Based on the recognition result of step S203 and the large model understanding result of step S204, the images and the results are clustered based on the spatial coordinates to form clusters indexed by spatial features, and these clusters are encoded for subsequent data management and analysis.

[0024] S206, uploading the encoded result to the cloud database, and establishing a spatial feature index after decoding for subsequent query and analysis.

[0025] Step S201 includes using a high-resolution camera and a high-precision global positioning (Beidou) system acquisition device to capture real-time images of place name signs at a designated location or along a preset route.

[0026] Step S202 includes: Develop and train a deep learning image recognition model based on convolutional neural networks to identify and judge place name signs.

[0027] The convolutional neural network model is optimized using pruning, quantization, and lightweight techniques to reduce the model's computing resource requirements and increase its running speed to adapt to the processing capabilities of mobile devices.

[0028] Deploy the optimized convolutional neural network model to a mobile device with image acquisition capabilities.

[0029] The built-in software of the mobile device contains a model inference engine. After the captured image is input into the software, the software calls the deployed deep learning model to analyze the image and determine whether it contains place name signs.

[0030] Step S203 includes: Build a deep learning model for optical character classification and recognition to recognize and classify text content on place name signs.

[0031] A place name sign image dataset is established, and random cropping, perspective transformation, random scaling and translation techniques are used to enhance sample diversity. At the same time, lighting and contrast are simulated to increase image Gaussian noise and salt and pepper noise to enhance the robustness of the data.

[0032] The place name sign image is processed by an image encoder and the preprocessed image features are obtained.

[0033] The convolutional network is used to optimize the repeated gradient information design and extract the basic coding features of the place name sign feature image to complete the segmentation of the Chinese name and Roman letter spelling area of ​​the image.

[0034] The image features and task instructions are input into the structured center point sequence decoder together to generate a structured information sequence.

[0035] Using structured center point sequences and image features, the text content decoder is input for text recognition and parsing.

[0036] Based on the decoded text content and structured center point sequence, the final image parsing result is output.

[0037] Step S204 includes: Text images with different scenes, fonts, sizes and blur levels are collected as model training sets, and the annotation information is verified and modified by place name experts to ensure the quality of the training set.

[0038] When the text is misrecognized due to blur or damage, the semantic understanding ability of the large model is used to correct the recognition result. For example, "Xibeiwang East Road" is misrecognized as "XIBEIWAMG Eesl Rd". At this time, the semantic understanding ability of the large model can play a role and correct the recognition result to the correct "XIBEIWANG East Rd".

[0039] Step S205 includes dividing the area into several grids according to the spatial position, mapping the coordinate points of the image to each grid, and integrating the corresponding place name classification and recognition results into the grid to form a cluster indexed by the spatial feature and the image and its place name classification result value.

[0040] Step S206 includes decoding the uploaded data using the cloud interface, creating an index based on the spatial features, and storing the index in a structured manner in a cloud database.

[0041] S3. Compare the place name sign information stored in the cloud with the standardized place name information, identify non-standard place name signs, and output the non-standard place name sign ledger.

[0042] Step S3 includes reading the name and Roman letter spelling information of the structured place name signs in the cloud, and comparing them with the place name standardization information, marking the place names that are inconsistent between the two, using the spatial index to extract the generated non-standard place name ledger information from the cloud database, and providing feedback.

[0043] Using the technical solution of this embodiment, Beijing has carried out the detection and identification of non-standard road place name signs, covering main roads, secondary roads, branch roads, secondary branches, and streets, and promptly fed back the verification results to the administrative department in charge of place names. By using modern digital management technology, standardized and intelligent management of place name signs is achieved, which is conducive to continuously optimizing the business environment, constantly improving the efficiency of place name management and services, and has achieved good results.

[0044] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A method for detecting road place name signs based on a large model, characterized in that: The method comprises the following steps: S1. Establish a standardized place name information database and publish an online query service interface for place name standardization information; S2, automatic recognition and extraction of Chinese names and Roman spellings in place names, structured coding and cloud storage; S3. Compare the place name sign information stored in the cloud with the standardized place name information, identify non-standard place name signs, and output the non-standard place name sign ledger.

2. A method for detecting road place name signs based on a large model as claimed in claim 1, characterized in that: Step S2 includes: S201, capturing and recording image information and location information of place name signs along the road through a mobile acquisition device; S202, establishing a lightweight deep learning image recognition model and deploying it in a mobile acquisition device to detect and identify in real time whether the image captured in step S201 is a place name sign; S203, establishing a lightweight deep learning text recognition model and deploying it in a mobile acquisition device to extract relevant characters of place name signs in real time; S204, deploying the big model in the cloud and publishing a service interface. The acquisition device calls the big model in real time through the interface request, and performs semantic understanding on the recognition result of step S203 to achieve accurate correction of the image recognition result; S205, based on the recognition result of step S203 and the large model understanding result of step S204, the pictures and the results are clustered based on the spatial coordinates to form clusters indexed by spatial features, and these clusters are encoded for subsequent data management and analysis; S206, uploading the encoded result to the cloud database, and establishing a spatial feature index after decoding for subsequent query and analysis.

3. A method for detecting road place name signs based on a large model as claimed in claim 2, characterized in that: Step S201 includes using a high-resolution camera and a high-precision global positioning system acquisition device to capture real-time images of place name signs at a designated location or along a preset route.

4. A method for detecting road place name signs based on a large model as claimed in claim 2, characterized in that: Step S202 includes: Develop and train a deep learning image recognition model based on convolutional neural networks to identify and judge place name signs; Use pruning, quantization, and lightweight technologies to optimize the convolutional neural network model, reduce the model's computing resource requirements, and increase the running speed to adapt to the processing capabilities of mobile devices; Deploy the optimized convolutional neural network model to a mobile device with image acquisition capabilities; The built-in software of the mobile device contains a model inference engine. After the captured image is input into the software, the software calls the deployed deep learning model to analyze the image and determine whether it contains place name signs.

5. A method for detecting road place name signs based on a large model as claimed in claim 2, characterized in that: Step S203 includes: Build a deep learning model for optical character classification and recognition to identify and classify text content on place name signs; A dataset of place name images was established, and random cropping, perspective transformation, random scaling and translation techniques were used to enhance sample diversity. At the same time, illumination and contrast were simulated to increase image Gaussian noise and salt and pepper noise to enhance data robustness. The place name sign image is processed by an image encoder, and the pre-processed image features are obtained; Using convolutional networks, we optimized the design of repeated gradient information and extracted the basic coding features of the feature images of place name signs to complete the segmentation of the Chinese name and Roman alphabet spelling areas of the image. The image features and task instructions are input into the structured center point sequence decoder to generate a structured information sequence; Using structured center point sequences and image features, input text content decoder for text recognition and parsing; Based on the decoded text content and structured center point sequence, the final image parsing result is output.

6. A method for detecting road place name signs based on a large model as claimed in claim 2, characterized in that: Step S204 includes: Collect text images of different scenes, fonts, sizes, and blur levels as model training sets, and verify and modify the annotation information to ensure the quality of the training set; When text is misrecognized due to blurriness or damage, the semantic understanding capability of the large model is used to correct the recognition result.

7. A method for detecting road place name signs based on a large model as claimed in claim 2, characterized in that: Step S205 includes dividing the area into several grids according to the spatial position, mapping the coordinate points of the image to each grid, and integrating the corresponding place name classification and recognition results into the grid to form a cluster indexed by the spatial feature and the image and its place name classification result value.

8. A method for detecting road place name signs based on a large model as claimed in claim 2, characterized in that: Step S206 includes decoding the uploaded data using the cloud interface, creating an index based on the spatial features, and storing the index in a structured manner in a cloud database.

9. A method for detecting road place name signs based on a large model as claimed in claim 1, characterized in that: Step S3 includes reading the name and Roman letter spelling information of the structured place name signs in the cloud, and comparing them with the place name standardization information, marking the place names that are inconsistent between the two, using the spatial index to extract the generated non-standard place name ledger information from the cloud database, and providing feedback.

Images