Vehicle-mounted information processing method and device of environment-friendly vehicle, medium and equipment

Abstract

The invention provides a vehicle-mounted information processing method and device of an environment-friendly vehicle, a medium and equipment, and belongs to the technical field of data processing. The method comprises the following steps: acquiring a to-be-identified vehicle-mounted list picture; identifying a corresponding first vehicle identifier from an identification code in the vehicle-mounted list picture; performing multi-modal resolution enhancement processing on the vehicle-mounted list picture to generate at least two enhanced images, and calling at least one OCR model to extract character information in the vehicle-mounted list picture from the at least two enhanced images; identifying corresponding first environmental protection key information according to the extracted character information; and comparing the first vehicle identifier and the first environmental protection key information with known vehicle information, and judging whether the corresponding environmental protection vehicle passes verification or not. The accuracy of environmental protection verification can be improved.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and particularly to a method, device, medium, and equipment for processing vehicle information of environmental protection vehicles. Background Art

[0002] With the continuous improvement of the national environmental protection policy system and the continuous strengthening of supervision, the control of environmental protection information of vehicles entering and leaving factories and parks has become an important part of environmental governance. According to the current regulations, vehicles with unqualified environmental protection information will be prohibited from entering the factory to ensure the green compliance of production activities. In the currently commonly used control process, the applicant vehicle needs to fill in environmental protection information through a dedicated information system and upload pictures of the vehicle list (covering images of paper lists or electronic documents), and can obtain factory entry authorization only after manual review. However, this mode has significant defects: due to the influence of year-on-year changes and regional policy differences, the styles of vehicle lists show diverse characteristics, and manual review is prone to problems such as information omission and misjudgment, resulting in difficulties in ensuring the review efficiency and accuracy.

[0003] To solve the above problems, the industry has introduced optical character recognition (OCR) technology. By automatically identifying the content of the vehicle list pictures uploaded by the mobile terminal, it assists manual review to improve work efficiency. However, this technology still faces many challenges in actual application: on the one hand, physical damages such as wrinkles and damages are easily generated during the circulation of paper documents of vehicle lists; on the other hand, due to the constraints of device performance, operation specifications, and environmental conditions during user shooting, problems such as insufficient picture resolution and blurred content often occur. These factors have greatly reduced the recognition accuracy of OCR technology for environmental protection information, and then led to incorrect review results, affecting the reliability and effectiveness of the control process. Summary of the Invention

[0004] The purpose of this application is to provide a method, device, medium, and equipment for processing vehicle information of environmental protection vehicles to solve at least one of the above technical problems.

[0005] In the first aspect of this application, a method for processing vehicle information of environmental protection vehicles is provided. The method includes: Obtain the vehicle list picture to be recognized; Identify the corresponding first vehicle identifier from the identification codes in the vehicle list picture; Perform multi-modal resolution enhancement processing on the vehicle list picture to generate at least two enhanced images, and call at least one OCR model to extract the text information in the vehicle list picture from at least two enhanced images; Identify the corresponding first environmental protection key information according to the extracted text information; Compare the first vehicle identifier, the first environmental protection key information, and the known vehicle information to determine whether the corresponding environmental protection vehicle passes the verification.

[0006] Optionally, the comparing the first vehicle identifier, the first environmental protection key information, and the known vehicle information to determine whether the corresponding environmental protection vehicle passes the verification includes: Compare whether the first vehicle identifier is exactly the same as the second vehicle identifier in the known vehicle information; Verify the authenticity of the corresponding vehicle inventory list based on the first environmental protection key information; When the vehicle identifiers are exactly the same and the vehicle inventory list is a genuine list, compare whether the first environmental protection key information matches the second environmental protection key information in the known vehicle information. If they match, determine that the verification passes; If the vehicle identifiers are not exactly the same, or the vehicle inventory list is not a genuine list, or the environmental protection key information does not match, determine that the verification fails.

[0007] Optionally, the verifying the authenticity of the corresponding vehicle inventory list based on the first environmental protection key information includes: Extract the corresponding first keyword field and the first field content corresponding to the first keyword field from the first environmental protection key information; Identify the coincidence rate between the extracted keyword field and the standard keyword fields in the set of standard keyword fields required for the vehicle inventory list; Identify whether each first field content conforms to the field content specification of the corresponding coincident standard keyword field, and calculate the corresponding content specification degree according to the specification result; When the coincidence rate exceeds the corresponding coincidence rate threshold and the content specification degree exceeds the corresponding specification degree threshold, determine that the vehicle inventory list is a genuine list; When the coincidence rate does not exceed the corresponding coincidence rate threshold, or the content specification degree does not exceed the corresponding specification degree threshold, determine that the vehicle inventory list is not a genuine list.

[0008] Optionally, for the first type of enhanced image, the vehicle inventory list image is upsampled to the target resolution using nearest neighbor interpolation and then Gaussian filtered. For the second type of enhanced image, the image is upsampled to the target resolution using bicubic interpolation. For the third type of enhanced image, the image is upsampled to the target resolution using bilinear interpolation; Invoking at least one OCR model to extract the text information in the vehicle inventory picture from at least two enhanced images includes: invoking at least one preset OCR model to perform text recognition on each enhanced image, and each OCR model outputs a character recognition result for each enhanced image, and the OCR model includes the Tesseract-OCR model; determining the text information in the vehicle inventory picture according to each character recognition result.

[0009] Optionally, each character recognition result includes the recognized character and the original confidence of the corresponding character; determining the text information in the vehicle inventory picture according to each character recognition result includes: Performing cross-model calibration on the original confidence output by each OCR model to obtain the standard confidence; Selecting characters according to the recognized characters and the corresponding standard confidence in each character recognition result; Forming the text information in the vehicle inventory picture based on the selected characters.

[0010] Optionally, selecting characters according to the recognized characters and the corresponding standard confidence in each character recognition result includes: for each character recognition result, when the characters recognized under the same position are inconsistent, performing weighted voting calculation on each character under the same position based on the standard confidence, and selecting candidate characters from multiple characters based on the weighted voting calculation result; Forming the text information in the vehicle inventory picture based on the selected characters includes: Forming initial text information based on the selected candidate characters; Performing vehicle information standardization verification on the initial text information, and identifying the candidate characters that do not conform to the vehicle information standardization in the initial text information as the characters to be corrected; Correcting the characters to be corrected based on each character recognition result to obtain the corrected characters corresponding to the characters to be corrected; Forming the text information in the vehicle inventory picture according to the initial text information and the corrected characters.

[0011] Optionally, performing cross-model calibration on the original confidence output by each OCR model to obtain the standard confidence includes: mapping the original confidence c0 to the corresponding standard confidence c1 according to the following piecewise linear function: , where k1, k2, b1, b2, c2 are preset confidence calibration parameters.

[0012] In the second aspect of the present application, a vehicle-mounted information processing device for an environmentally friendly vehicle is provided. The device includes: A picture acquisition module for acquiring a vehicle-mounted list picture to be recognized; An identification extraction module for identifying a corresponding first vehicle identification from the identification code in the vehicle-mounted list picture; An OCR recognition module for performing multi-modal resolution enhancement processing on the vehicle-mounted list picture to generate at least two enhanced images, invoking at least one OCR model to extract the text information in the vehicle-mounted list picture from at least two enhanced images; and identifying corresponding first environmental protection key information according to the extracted text information; A verification module for comparing based on the first vehicle identification, the first environmental protection key information, and known vehicle information to determine whether the corresponding environmentally friendly vehicle passes the verification.

[0013] In the third aspect of the present application, a computer-readable storage medium is provided. An executable instruction is stored on the computer-readable storage medium. When the executable instruction is executed by a processor, the processor executes the method described in any embodiment of the present application.

[0014] In the fourth aspect of the present application, an electronic device is provided, including: one or more processors; a memory for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors execute the method described in any embodiment of the present application.

[0015] In the vehicle-mounted information processing method, device, medium, and equipment for the environmentally friendly vehicle in the present application, by identifying the first vehicle identification from the vehicle-mounted list picture, and performing multi-modal resolution enhancement processing on the vehicle-mounted list picture to generate at least two enhanced images, invoking at least one OCR model to extract the text information in the vehicle-mounted list picture from at least two enhanced images, extracting the first environmental protection key information from the text information, and comparing based on the first vehicle identification and the first environmental protection key information with the known vehicle information for environmental protection verification, the efficiency of environmental protection verification can be improved. In addition, by using at least one OCR model to extract the text information therein, the accuracy of the extracted text information can also be improved under a vehicle-mounted list picture with low resolution, thereby improving the accuracy of the subsequent extraction of the first environmental protection key information, and ultimately improving the accuracy of environmental protection verification. Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required to be used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present application and should not be regarded as limiting the scope of the present application.

[0017] Figure 1 is a schematic flow chart of a method for processing vehicle - associated information of an environment - friendly vehicle in an embodiment; Figure 2 is a schematic flow chart for comparing the first vehicle identifier, the first environmental protection key information with known vehicle information to determine whether the corresponding environment - friendly vehicle passes the verification in an embodiment; Figure 3 is a schematic flow chart for verifying the authenticity of the corresponding vehicle - associated list based on the first environmental protection key information in an embodiment; Figure 4 is a schematic flow chart for calling at least one OCR model to extract the text information in the vehicle - associated list picture from at least two enhanced images in an embodiment; Figure 5 is a schematic flow chart for determining the text information in the vehicle - associated list picture according to each character recognition result in an embodiment; Figure 6 is a schematic flow chart for forming the text information in the vehicle - associated list picture based on the selected characters in an embodiment; Figure 7 is a schematic structural diagram of a vehicle - associated information processing device for an environment - friendly vehicle in an embodiment; Figure 8 is a schematic structural diagram of an electronic device in an embodiment. Detailed implementation manners

[0018] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0019] All terms used in the present application (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0020] For example, terms such as "first", "second", etc. used in the present application may be used herein to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from another element.

[0021] For another example, terms such as "including", "comprising", etc. used in the present application indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0022] The present application provides a method for processing vehicle - associated information of an environment - friendly vehicle, asFigure 1 As shown in Figure 1 , the method includes: Step 110: Obtain the vehicle inventory list picture to be recognized.

[0023] In this embodiment, the vehicle inventory list picture is a picture formed by photographing or scanning the vehicle environmental protection information inventory list. The vehicle environmental protection information inventory list is a document that details the environmental protection information of a vehicle, covering environmental protection data of various types of motor vehicles such as gasoline vehicles, diesel vehicles, and electric vehicles. It is an important procedure during the vehicle registration process, including but not limited to relevant fields such as vehicle basic information, emission standards, fuel efficiency, and key environmental protection configurations, as well as the field contents in the corresponding fields.

[0024] Among them, the vehicle inventory list picture can be an image obtained by the client photographing the paper vehicle environmental protection information inventory list, or an image converted from a scanned copy of the paper vehicle environmental protection information inventory list or an electronic version list in a format such as PDF. For example, the user forms a corresponding image by photographing the paper vehicle environmental protection information inventory list with the client, uploads the image to the electronic device, and the electronic device takes the received image as the vehicle inventory list picture.

[0025] Step 120: Identify the corresponding first vehicle identifier from the identification code in the vehicle inventory list picture.

[0026] In this embodiment, the vehicle inventory list picture includes the environmental protection information of the above-mentioned various fields and field contents, and also includes the identification code of the corresponding environmental protection vehicle. The vehicle identifier is loaded in the identification code. The identification code can be a barcode or a QR code.

[0027] For the received vehicle inventory list picture, the electronic device calls the corresponding identification code recognition module to scan and recognize the identification code, and reads the content carried therein as the first vehicle identifier. The vehicle identifier is used to uniquely identify a vehicle and can be composed of a preset number of digits, letters, or other special characters. Specifically, the vehicle identifier can be the VIN code of the environmental protection vehicle.

[0028] Step 130: Perform multi-modal resolution enhancement processing on the vehicle inventory list picture to generate at least two enhanced images, and call at least one OCR model to extract the text information in the vehicle inventory list picture from the at least two enhanced images.

[0029] In this embodiment, the electronic device presets one or more OCR models, which may include, but are not limited to, one or more of the following: Tesseract-OCR model, Transformer-based OCR model, end-to-end OCR model, etc. Each OCR model includes an image preprocessing module, an image enhancement module, and an OCR recognition framework. The image preprocessing module performs operations such as grayscale conversion, binarization, denoising, and skew correction on the input image; the image enhancement module is used to perform multi-modal resolution enhancement processing on the preprocessed image to obtain at least one resolution-enhanced image, so as to improve the recognition accuracy of the text information output by the subsequent OCR recognition framework. For example, various image enhancement processing methods can be used to obtain corresponding enhanced images (such as using nearest neighbor interpolation, bilinear interpolation, bicubic interpolation, etc. for image enhancement processing), and text information recognition is independently performed on each enhanced image to obtain corresponding recognition results of multiple text information.

[0030] Specifically, for each OCR model, the electronic device prepares corresponding training datasets and test datasets in advance, iteratively trains each OCR model based on the training datasets. After completing the iterative training, the test datasets are called to test and verify each iteratively trained OCR model. After passing the verification, a trained OCR model is obtained; if the verification fails, the OCR model that fails the verification continues to be trained and verified until the verification passes. The vehicle checklist pictures are recognized based on each trained OCR model, and the corresponding text information is recognized from the vehicle checklist pictures.

[0031] Among them, the training datasets and test datasets may include a large number of historical vehicle checklist pictures and corresponding text recognition results. The resolutions of the pictures in the training datasets and test datasets are different, and a small amount of Gaussian noise is randomly added to some pictures. At the same time, some pictures are randomly rotated (±3°), skewed (±2°), and perspective distorted (0.05 - 0.1), etc., to simulate the slight tilt effect generated during picture scanning or shooting.

[0032] Taking the Tesseract-OCR model as an example, Tesseract-OCR uses a long short-term memory network LSTM to train the tessdata_fast network in the Tesseract-OCR model. This model can automatically identify and extract key information on the vehicle checklist. If it is found that the recognition effect does not reach the ideal effect during use, the model can be further optimized. Use the tessdata_best network (this network can be obtained from the Tesseract repository), which is a model for continued training and optimization, and can improve recognition accuracy and reliability.

[0033] Specifically, the Tesseract-OCR model can specifically be based on the open-source Tesseract 5.3.3 engine, and on this basis, the original LSTM network is replaced with a bidirectional LSTM network to add an attention mechanism layer. The OCR model based on Transformer can specifically be a CNN+Transformer attention integration model. The OCR model based on end-to-end can specifically be an LSTM-CTC end-to-end model, and its feature extraction layer can adopt a convolutional layer with 7 convolutional blocks + 3 2×2 max pooling layers, and the sequence processing layer includes a bidirectional LSTM network and a Bahdanau attention mechanism.

[0034] The training mode of each OCR model can include fine-tuning training and training from scratch. For the OCR model for fine-tuning training, it can start from an existing trained OCR model, use the above training dataset as specific additional data for iterative training to achieve fine-tuning of the model, so as to obtain an OCR model that has completed training and passed verification. By introducing the knowledge graph in the environmental protection field, during the fine-tuning process, a greater weight (such as applying an attention weight offset) is applied to the samples containing terms related to the environmental protection field identified in the picture, and an environmental protection term matching reward item is added to the loss function, thereby improving the accuracy of the recognition of the vehicle checklist picture.

[0035] If the OCR model obtained by fine-tuning training fails to pass verification, then from the corresponding OCR model, the top layer (or any number of layers) can be cut off from the network and a new top layer can be retrained using a new dataset until a trained OCR model is obtained.

[0036] If it still cannot be used, then it can be trained from scratch. By increasing the number of iterations (such as using 800 epochs of training round iterations) and the number of training samples, various trained OCR models can be finally obtained.

[0037] By integrating the trained OCR model into the information system, the function of automatically recognizing the vehicle checklist is realized. The user only needs to upload a picture of the vehicle checklist, and the corresponding OCR model can output the text information in the vehicle checklist picture. Among them, each OCR model can output one or more corresponding text recognition results of the vehicle checklist picture, and the electronic device finally forms the corresponding text information by combining multiple text recognition results. For example, the characters contained in each text recognition result can be referred to, and the most accurately recognized characters can be selected through a voting mechanism, and the selected characters are combined to form the corresponding text information.

[0038] In this embodiment, for the photographed vehicle inventory picture, the resolution can be identified. When the identified resolution is lower than the preset resolution threshold, the above-mentioned multiple resolution enhancement processes are adopted. Each resolution enhancement process will output an enhanced image, so as to convert a vehicle inventory picture into multiple enhanced images with the target resolution. For each enhanced image, one or more of the above-mentioned OCR models are called for OCR recognition. Each OCR model outputs a corresponding character recognition result for each enhanced image, and the text information is determined from the multiple character recognition results to improve the accuracy of OCR recognition of the vehicle inventory picture at low resolution.

[0039] Among the multiple OCR models, one of the OCR models can be used as the default OCR model. For example, the Tesseract-OCR model can be used as the default OCR model; among the multiple resolution enhancement processing modes, one of the resolution enhancement processing modes is used as the default processing mode. For example, the bicubic interpolation image enhancement processing can be used as the default processing mode.

[0040] When the identified resolution exceeds the preset resolution threshold, only one of the resolution enhancement processes (such as adopting the default processing mode) can be used to output an enhanced image, and one of the OCR models (such as adopting the default OCR model) is called to perform OCR recognition on the enhanced image, and the text information is determined based on the identified character recognition result to improve the OCR recognition efficiency. The resolution threshold can be a suitable threshold preset according to the actual situation, such as any suitable value such as 75 dpi, 100 dpi, 120 dpi, 150 dpi, 200 dpi, etc.

[0041] Step 140, identify the corresponding first environmental protection key information according to the extracted text information.

[0042] In this embodiment, the first environmental protection key information refers to the information that needs to be environmentally verified, which can include the field contents in each environmental protection field that needs to be audited and verified. For example, it can include the specific contents corresponding to the vehicle basic information, emission standards, environmental protection key configurations, expiration date, fuel efficiency and other fields (that is, the first field contents in the following text), and these contents constitute the corresponding first environmental protection key information.

[0043] Specifically, for the extracted text information, semantic alignment and data verification can be performed. During the semantic alignment process, based on the keyword library in the field of motor vehicle environmental protection information, the fields recognized by OCR (such as "emission standard") are mapped to standardized fields, so that it is possible to know which fields the respective text contents belong to and what the contents are under the specific fields. During the data verification process, the compliance of the specific contents of the fields can be verified in real time to intercept invalid data with incorrect formats or missing fields, thereby improving the accuracy of the first environmental protection key information extracted. It can be understood that there are certain data specification requirements for the field contents (specific contents) corresponding to different fields. Based on this data specification requirement, the verification of the specific contents can be carried out.

[0044] Step 150: Compare the first vehicle identifier, the first environmental protection key information with the known vehicle information to determine whether the corresponding environmental protection vehicle passes the verification.

[0045] In this embodiment, for the identified first vehicle identifier and the first environmental protection key information, they can be compared with the known vehicle information. Specifically, the vehicle can be searched in the vehicle information database based on the first vehicle identifier to query the target vehicle that needs to be subjected to environmental protection verification, and then the vehicle information of the target vehicle stored in the vehicle information database and the first environmental protection key information are compared for data to query whether the corresponding environmental protection data is consistent. Based on the query result, it is determined whether the corresponding environmental protection vehicle can pass the environmental protection verification.

[0046] For example, if the identified first vehicle identifier (VIN code) is "LSVNV133X12345678", then the vehicle with the VIN code "LSVNV133X12345678" is queried from the vehicle information database as the target vehicle. For this target vehicle, the environmental protection information or environmental protection information requirements (such as the index requirements of some environmental protection parameters) of the target vehicle that have been collected are compared with the first environmental protection key information uploaded by the user to check whether the two are consistent, or to detect whether the first environmental protection key information meets the environmental protection information requirements of the corresponding vehicle. If the consistency is met or the environmental protection information requirements are satisfied, it is determined that the environmental protection vehicle passes the verification.

[0047] The method for processing vehicle - related information of the environmental - protection vehicle in this application can improve the efficiency of environmental - protection verification by identifying the first vehicle identifier from the vehicle inventory picture and calling at least one OCR model to identify the text information in the picture, extracting the first environmental - protection key information from the text information, and comparing the first vehicle identifier and the first environmental - protection key information with the known vehicle information. In addition, by using at least one OCR model to extract the text information, the accuracy of the text information extracted from the vehicle inventory picture with low resolution can be improved, thereby improving the accuracy of the subsequent extraction of the first environmental - protection key information, and ultimately improving the accuracy of environmental - protection verification.

[0048] In one embodiment, as Figure 2 shown, step 150 includes: Step 210, comparing whether the first vehicle identifier is exactly the same as the second vehicle identifier in the known vehicle information.

[0049] In this embodiment, the first vehicle identifier has specific format requirements, such as having a specific number of digits, and the character content in some digits must be specific numbers or letters, etc. For the identified first vehicle identifier, it can be checked whether it meets the preset format requirements. When it does not meet the format requirements, it can be directly determined that the vehicle identifiers are inconsistent. When it meets the format requirements, then search for this vehicle identifier in the vehicle information database. If an exactly the same vehicle identifier (i.e., the second vehicle identifier) is found, it is determined that the vehicle identifiers are consistent. If no corresponding consistent vehicle identifier can be found, it is determined that the vehicle identifiers are inconsistent.

[0050] Step 220, verifying the authenticity of the corresponding vehicle inventory based on the first environmental - protection key information.

[0051] The environmental - protection key information refers to the core data directly related to the environmental - protection performance of the vehicle, and is the basis for determining whether the vehicle meets the environmental - protection standards and can legally drive on the road or enter a specific area. For the identified first environmental - protection key information, the vehicle identifier included in the first environmental - protection key information can be verified with the first vehicle identifier identified from the above - mentioned identification code to check whether they are exactly the same. After they are exactly the same, further identify the first key field and the corresponding field content in the environmental - protection key information, check whether the first key field is a field related to environmental - protection verification, and whether the corresponding field content meets the verification requirements, so as to determine whether the vehicle inventory is a real inventory. For example, for the validity - period field, it can be identified whether the validity - period content is within the valid period required by environmental protection. If it is not within the valid period, it means that the vehicle inventory is an expired vehicle inventory and does not meet the authenticity requirements.

[0052] In one embodiment, as Figure 3As shown, step 220 includes: Step 310, extracting corresponding first key fields and the first field content corresponding to the first key fields from the first environmental protection key information.

[0053] Among them, the first key fields represent the names of data fields related to vehicle environmental protection verification identified from the vehicle inventory picture, such as fields like emission standards and environmental protection numbers, and the first field content represents the specific data values corresponding to the first key fields. For example, some of the first key fields and their first field content extracted from the first environmental protection key information are as follows: Emission standards: National VI; Environmental protection number: Beijing Environmental Label 20230616; Expiry date: 2025-12-31; Emission limit: 0.5mg / m³; Testing agency: XX City Environmental Monitoring Center. Step 320, identifying the coincidence rate between the extracted key fields and the standard key fields in the set of standard key fields required for the vehicle inventory.

[0054] In this embodiment, the electronic device has predefined the standard key fields that the vehicle inventory needs to contain. For the first key fields identified from the picture, it can detect whether they conform to the standard key fields. It can be understood that due to possible wrinkles or wear on the user's vehicle inventory, the identified first key fields may be incomplete, or there may be some misidentified key fields. For example, "emission standards" may be misidentified as "emission standards". Based on this, the electronic device can perform semantic recognition for each identified key field and correct the misidentified fields, thereby improving the accuracy of key field recognition.

[0055] The standard key fields can be extracted from a predefined standard template. By comparing the first key fields with the standard key fields, the coincidence rate between the two can be obtained. Due to possible incomplete or missing fields in the vehicle inventory, an appropriate coincidence rate threshold can be set. When the calculated coincidence rate exceeds the corresponding coincidence rate threshold, step 330 can be executed. If the coincidence rate is less than the corresponding coincidence rate threshold, it is directly determined that the vehicle inventory is not authentic, and the user needs to provide a new vehicle inventory. For example, the user can take a new picture of the vehicle inventory with higher quality to re-perform the environmental protection verification.

[0056] It can be understood that the coincidence rate threshold can be adaptively adjusted according to the quality of the uploaded vehicle inventory picture, or based on business requirements and actual recognition accuracy, so as to improve the applicability of vehicle environmental protection verification. For example, the coincidence rate threshold can be set to 80%.

[0057] For example, the identified keyword fields include: {"Emission standard", "Environmental protection number", "Expiry date", "Emission limit", "Testing agency"}, a total of 5. In addition to these 5, the required standard keyword fields also include the engine model. Based on this, the corresponding coincidence rate can be calculated as 5 / 6 ≈ 83.3%.

[0058] Step 330: Identify whether each first field content conforms to the field content specification of the corresponding coincident standard keyword field, and calculate the corresponding content compliance degree according to the specification result.

[0059] In this embodiment, different fields have corresponding content specification requirements. For example, for the environmental protection number, its content specification requirement is: provincial abbreviation + environmental label + 8-digit number; the content specification requirement corresponding to the expiry date field is: YYYY-MM-DD, and it has not expired; the content specification requirement corresponding to the emission limit field is 0.1 - 1.0 mg / m 3 . For each first field content, identify whether it conforms to the corresponding field content specification, and calculate the corresponding content compliance degree according to the identification result.

[0060] Step 340: When the coincidence rate exceeds the corresponding coincidence rate threshold and the content compliance degree exceeds the corresponding compliance degree threshold, determine that the vehicle inventory list is a genuine list.

[0061] Step 350: When the coincidence rate does not exceed the corresponding coincidence rate threshold, or the content compliance degree does not exceed the corresponding compliance degree threshold, determine that the vehicle inventory list is not a genuine list.

[0062] In this embodiment, by performing coincidence rate and compliance degree verification, the accuracy of list authenticity determination can be improved.

[0063] Step 230: When the vehicle identification is exactly the same and the vehicle inventory list is a genuine list, compare whether the first environmental protection key information matches the second environmental protection key information in the known vehicle information. If it matches, determine that the verification is passed.

[0064] If the vehicle identification is exactly the same and after initially identifying that the vehicle inventory list is a genuine list, the first environmental protection key information can be compared with the second environmental protection key information corresponding to the target vehicle. If it matches, determine that the verification is passed.

[0065] Specifically, for different environmental protection key information, the matching requirements are not necessarily the same. For example, some environmental protection key information requires exact matching, while some only need to meet fuzzy matching. For different environmental protection key information, make judgments according to the corresponding matching requirements. When each environmental protection key information meets the corresponding matching requirements, it can be determined that the verification is passed. Or when more than the preset number of first environmental protection key information meet the matching requirements, it is determined that the verification is passed.

[0066] By performing environmental protection key information verification and vehicle identification verification, the accuracy of environmental protection vehicle verification can be improved.

[0067] Step 240, if the vehicle identifications are not completely consistent, or the vehicle inventory list is not a genuine list, or the environmental protection key information does not match, it is determined that the verification fails.

[0068] In this embodiment, when any one of the situations such as the vehicle identifications not being completely consistent, the vehicle inventory list not being a genuine list, and the environmental protection key information not matching occurs, it is determined that the verification fails, and the reason for the failed verification is prompted, so as to facilitate the user to submit genuine environmental protection information or re-perform vehicle environmental protection verification to ensure the safety of the vehicle.

[0069] In one embodiment, the first enhanced image performs Gaussian filtering after upsampling the vehicle inventory list image to the target resolution using nearest neighbor interpolation, the second enhanced image upsamples the image to the target resolution using bicubic interpolation; the third enhanced image upsamples the image to the target resolution using bilinear interpolation; as Figure 4 shown, at least one OCR model is called to extract the text information in the vehicle inventory list picture from at least two enhanced images, including: Step 410, call at least one preset OCR model to perform text recognition on each enhanced image, and each OCR model outputs a character recognition result for each enhanced image.

[0070] Step 420, determine the text information in the vehicle inventory list picture according to each character recognition result.

[0071] The multi-modal resolution enhancement processing may include at least two or three enhancement processing methods such as nearest neighbor interpolation, bilinear interpolation, and bicubic interpolation. Among them, the first enhanced image upsamples the vehicle inventory list image to the target resolution using nearest neighbor interpolation and then performs Gaussian filtering, the second enhanced image upsamples the image to the target resolution using bicubic interpolation; the third enhanced image upsamples the image to the target resolution using bilinear interpolation and then performs histogram equalization.

[0072] For the first enhanced image, during the upsampling process, each pixel of the preprocessed vehicle inventory list picture is mapped to the nearest position of the target resolution (such as 300 dpi) without generating new pixel values. In this upsampling process, the sharp edges of the original image (i.e., the preprocessed vehicle inventory list picture) are retained but the magnification jaggedness is present; based on this, Gaussian filtering is used to perform edge processing on it, such as Gaussian low-pass filtering (such as the standard deviation of the Gaussian filtering parameter value is any suitable value such as 0, 0.5, 1.0, 1.5, or 2, and the convolution kernel size is 3×3 or any other suitable value), to obtain the first enhanced image.

[0073] For the second enhanced image, during the upsampling process, the corresponding new pixel values in the enhanced image are calculated based on the weighted average of the nearest neighbor pixels to obtain the second enhanced image. The weights can be determined by a cubic function. Among them, the number of nearest neighbor pixels selected can be determined according to the resolution of the original image, so that the resolution of the enhanced image obtained by upsampling can reach the target resolution. For example, it is determined to perform weighted average based on 9 nearest neighbor pixels (3×3 neighborhood).

[0074] For the third enhanced image, during the upsampling process, the corresponding new pixel values in the enhanced image are also calculated based on the weighted average of the nearest neighbor pixels, but its weight is inversely proportional to the distance between the nearest neighbor pixel and the target pixel. For the formed new pixel values, histogram equalization processing is adopted to obtain the third enhanced image. Similarly, the number of nearest neighbor pixels selected can be determined according to the resolution of the original image.

[0075] Each OCR model performs OCR recognition on one of the formed enhanced images to obtain the character recognition result corresponding to the enhanced image. For example, if M OCR models and N enhanced images are set, then M×N character recognition results can be obtained correspondingly. For example, if M = 2 (Tesseract-OCR model, Transformer-based OCR model) and N = 3, then 6 character recognition results can be output.

[0076] Taking the Tesseract-OCR model as an example of the OCR model, for the output enhanced image, the Tesseract-OCR model uses linear interpolation and unsharp masking to segment the text line to form a single-line image, then calls the trained LSTM network for feature extraction, and calls CTC decoding for the obtained feature vector to obtain the corresponding recognized character and the confidence of the character, and outputs the character recognition result.

[0077] For the Transformer-based OCR model, after image segmentation, visual features are extracted through its encoder, and the decoder decodes the extracted feature vector to obtain the corresponding recognized character and the confidence of the character, and outputs the character recognition result.

[0078] In the character recognition result output by each OCR model, in addition to the character, it also includes the confidence of each recognized character.

[0079] It is understandable that the characters in the output of various character recognition results are not completely consistent. There must be some characters that are recognized incorrectly and some that are recognized correctly. Based on this, for the inconsistent characters in the recognition record, a weighted voting method can be used to select characters, and one of the characters is selected from the inconsistent characters to finally form the text information in the vehicle list picture.

[0080] For example, for the word "准" in the "Emission Standard" field in the vehicle list, the six character recognition results outputted are: 准,准,淮,桩,准,淮; the corresponding confidence levels are 83%, 81%, 82%, 78%, 79%, and 77% respectively. Through voting analysis, the corresponding character is finally determined to be "准".

[0081] In this embodiment, by setting at least one OCR model and performing multiple image enhancement processing on the vehicle list picture with low resolution, each OCR model outputs a character recognition result for each enhanced processed image, and the text information in the vehicle list picture is finally determined based on the multiple character recognition results, the accuracy of character recognition of low-resolution images can be improved, thereby ultimately improving the accuracy of environmentally friendly vehicle verification.

[0082] In one embodiment, each character recognition result includes the recognized character and the original confidence of the corresponding character. Figure 5 As shown, step 430 includes: Step 510 , cross-model calibration is performed on the original confidence output by each OCR model to obtain a standard confidence.

[0083] Step 520 , character selection is performed based on the characters recognized in each character recognition result and the corresponding standard confidence level.

[0084] Step 530, forming text information in the vehicle list image based on the selected characters.

[0085] In this embodiment, the original confidence represents the confidence calculated by the corresponding OCR model according to its own confidence evaluation logic. It is understandable that the confidence evaluation logic of different OCR models is not necessarily the same. For example, the confidence of a certain model is generally high, while the confidence given by a certain model is generally conservative, but the recognition capabilities of the two are basically the same. Based on this, the confidence of each character obtained by each model can be calibrated across models to obtain a normalized standard confidence, so that the confidence of the output characters between each model can be compared.

[0086] Specifically, a calibration dataset independent of the above training dataset and test dataset can be further set. This calibration dataset contains calibration images and their corresponding true character sequences. Each OCR model runs on this calibration dataset, records the original confidence c0 of each output character, and whether the character prediction is accurate (for example, outputting "1" means accurate and "0" means incorrect). By configuring the corresponding calibrator (such as using Platt Scaling), a logistic regression model is trained on the original confidence to map the original confidence to the actual accuracy rate, and the original confidence is converted into a calibrated confidence through this calibrator.

[0087] For each of the obtained calibrated confidences, normalization processing is performed on them, and finally a standard confidence that can be compared among various OCR models is formed.

[0088] Specifically, the original confidence c0 can be mapped to the corresponding standard confidence c1 according to the following piecewise linear function: , where k1, k2, b1, b2, c2 are preset confidence calibration parameters, c2 represents the segmentation threshold of the original confidence c0, k1, b1 are the first piecewise linear function parameters when the original confidence c0 is less than the segmentation threshold c2, and k2, b2 are the second piecewise linear function parameters when the original confidence c0 is greater than or equal to the segmentation threshold c2. The values of the above calibration parameters can be calculated according to the values calculated by the above calibrator during the calibration process, and the calibration parameters of different OCR models are not necessarily the same. For example, the piecewise linear function of the standard confidence c1 of the obtained Tesseract-OCR model is: .

[0089] For each of the calculated standard confidences, for each character, the character with the highest standard confidence can be selected as the character in the text information of the vehicle inventory picture, or character selection can be performed by weighted voting according to each standard confidence.

[0090] Furthermore, based on various character recognition results output by each OCR model, one can be determined from multiple character recognition results as the main recognition result based on the standard confidence of the corresponding characters. When the standard confidence of the character in the main recognition result exceeds the preset confidence threshold, the character is used as the character in the text information of the vehicle inventory picture. For characters that do not exceed the preset confidence threshold, the characters recognized in other character recognition results and their standard confidences are further referred to, and the final character is determined from them as the character in the text information of the vehicle inventory picture, thereby further improving the accuracy of character recognition.

[0091] Further, it is also possible to perform weighted voting calculation on the standard confidence levels of the characters to be selected in the various character recognition results output by each OCR model, calculate the weighted voting value of the standard confidence level of each character to be selected, and then select the character to be selected with the largest weighted voting value as the character in the text information of the vehicle checklist picture, so as to further improve the accuracy of character recognition.

[0092] In one embodiment, the weighted voting value can be calculated by the formula: ; .

[0093] Wherein, is the set of all characters to be selected for which character selection is required. is the set of all characters to be selected and is the sample value of one character to be selected in the set; represents the i-th character to be selected that is equal to the sample value of the character to be selected ; represents the index set of all characters to be selected that satisfy ; represents the number of characters to be selected that satisfy ; represents the standard confidence level of the i-th character to be selected that is equal to the sample value of the character to be selected ; w represents selecting the sample value of the character to be selected with the largest weighted voting value best as the character finally selected by the voting. ; as the character finally selected by the voting.

[0094] In this embodiment, the above-mentioned weighted voting calculation of the weighted voting value can avoid the over-preference selection of characters to be selected with a small sample size due to the relatively high average value of the standard confidence level; for the sample values of multiple characters to be selected with the same sample size, it is more inclined to the sample value of the character to be selected with a higher average value of the standard confidence level. In the case where the sample size of each character to be selected is different, the above-mentioned weighted voting mechanism makes a trade-off between the average value and the sample size, and can significantly improve the accuracy of the finally recognized characters.

[0095] For example, for a certain character to be selected, the 3 kinds of character recognition results participating in the selection and their corresponding standardized confidence levels are shown in Table 1 below.

[0096] Table 1

[0097] For the characters to be selected among the various character recognition results output by each OCR model in Table 1, two sample values were recognized, namely "National VI Emission" and "National V Emission". Through the weighted voting calculation described above, the weighted voting values of "National VI Emission" and "National V Emission" were obtained as 116 and 90 respectively, and the character w finally selected by voting was best "National VI Emission".

[0098] In one embodiment, step 520 includes: for each character recognition result, when the characters recognized under the same position are inconsistent, a weighted voting calculation is performed on each character under the same position based on the standard confidence level, and a candidate character is selected from multiple characters based on the weighted voting calculation result.

[0099] In this embodiment, the weighted bidding calculation method can be calculated according to the above formula, and the character with the highest score is selected as the candidate character.

[0100] In one embodiment, as Figure 6 shown, step 530 includes: Step 610, forming initial text information based on the selected candidate characters.

[0101] Step 620, performing a vehicle information standardization check on the initial text information, and identifying the candidate characters that do not conform to the vehicle information standardization in the initial text information as characters to be corrected.

[0102] Step 630, correcting the characters to be corrected based on each character recognition result to obtain the corrected characters corresponding to the characters to be corrected.

[0103] Step 640, forming the text information in the vehicle list picture according to the initial text information and the corrected characters.

[0104] In this embodiment, for the formed initial text information, it can be detected whether it conforms to the vehicle information standardization check. For example, for the text "Emission Standard" in the initial text information, it can be recognized that it does not conform to the vehicle information standardization, and then the candidate character "Emission Standard" is used as the character to be corrected. For the character to be corrected, it can be corrected based on semantic recognition and in combination with the corresponding content in each character recognition result. For example, the candidate character "Emission Standard" can be finally corrected to "Emission Standard".

[0105] By further performing a standardization check on the formed initial text information and making corrections according to the standardization check results, the accuracy of the text information is further improved.

[0106] In one embodiment, as Figure 7As shown, a vehicle-mounted information processing device for an environmentally friendly vehicle is provided. The device includes: An image acquisition module 710 for acquiring an on-vehicle list image to be recognized; An identification extraction module 720 for identifying a corresponding first vehicle identification from the identification code in the on-vehicle list image; An OCR recognition module 730 for performing multi-modal resolution enhancement processing on the on-vehicle list image to generate at least two enhanced images, calling at least one OCR model to extract text information in the on-vehicle list image from at least two enhanced images; and identifying corresponding first environmentally friendly key information according to the extracted text information; A verification module 740 for comparing the first vehicle identification and the first environmentally friendly key information with known vehicle information to determine whether the corresponding environmentally friendly vehicle passes the verification.

[0107] In one embodiment, the verification module 740 is further configured to compare whether the first vehicle identification is exactly the same as the second vehicle identification in the known vehicle information; verify the authenticity of the corresponding on-vehicle list based on the first environmentally friendly key information; when the vehicle identifications are exactly the same and the on-vehicle list is a genuine list, compare whether the first environmentally friendly key information matches the second environmentally friendly key information in the known vehicle information. If they match, it is determined that the verification passes; if the vehicle identifications are not exactly the same, or the on-vehicle list is not a genuine list, or the environmentally friendly key information does not match, it is determined that the verification fails.

[0108] In one embodiment, the verification module 740 is further configured to extract corresponding first key fields and first field contents corresponding to the first key fields from the first environmentally friendly key information; identify the coincidence rate between the extracted key fields and the standard key fields in the standard key field set required by the on-vehicle list; identify whether each first field content conforms to the field content specification of the corresponding coincident standard key field, and calculate the corresponding content specification degree according to the specification result; when the coincidence rate exceeds the corresponding coincidence rate threshold and the content specification degree exceeds the corresponding specification degree threshold, it is determined that the on-vehicle list is a genuine list; when the coincidence rate does not exceed the corresponding coincidence rate threshold, or the content specification degree does not exceed the corresponding specification degree threshold, it is determined that the on-vehicle list is not a genuine list.

[0109] In one embodiment, the OCR recognition module 730 is further configured to perform multi-modal resolution enhancement processing on the vehicle inventory list picture to generate at least two enhanced images. Among them, the first enhanced image performs Gaussian filtering after upsampling the vehicle inventory list image to the target resolution using nearest neighbor interpolation. The second enhanced image upsamples the image to the target resolution using bicubic interpolation. The third enhanced image upsamples the image to the target resolution using bilinear interpolation. Call at least one preset OCR model to perform character recognition on each enhanced image. Each OCR model outputs a character recognition result for each enhanced image. The OCR model includes the Tesseract-OCR model. Determine the text information in the vehicle inventory list picture according to each character recognition result.

[0110] In one embodiment, each character recognition result includes the recognized character and the original confidence of the corresponding character. The OCR recognition module 730 is further configured to perform cross-model calibration on the original confidence output by each OCR model to obtain the standard confidence. Select characters according to the recognized characters and the corresponding standard confidence in each character recognition result. Form the text information in the vehicle inventory list picture based on the selected characters.

[0111] In one embodiment, the OCR recognition module 730 is further configured to, for each character recognition result, when the characters recognized under the same position are inconsistent, perform weighted voting calculation on each character under the same position based on the standard confidence, and select candidate characters from multiple characters based on the weighted voting calculation result.

[0112] In one embodiment, the OCR recognition module 730 is further configured to form initial text information based on the selected candidate characters. Perform vehicle information standardization verification on the initial text information, and identify the candidate characters that do not conform to the vehicle information standardization in the initial text information as the characters to be corrected. Correct the characters to be corrected based on each character recognition result to obtain the corrected characters corresponding to the characters to be corrected. Form the text information in the vehicle inventory list picture according to the initial text information and the corrected characters.

[0113] In one embodiment, the OCR recognition module 730 is further configured to map the original confidence c0 to the corresponding standard confidence c1 according to the following piecewise linear function: , where k1, k2, b1, b2, and c2 are preset confidence calibration parameters.

[0114] In one embodiment, a computer-readable storage medium is provided, on which executable instructions are stored. When the instructions are executed by a processor, the processor executes the steps in the above method embodiments.

[0115] In one embodiment, an electronic device is further provided, including one or more processors; a memory, where one or more programs are stored in the memory. When the one or more programs are executed by the one or more processors, the one or more processors are caused to execute the steps in the above method embodiments.

[0116] In one embodiment, as Figure 8 shown, it shows a schematic structural diagram of an electronic device for implementing the embodiments of the present application. The electronic device 800 includes a central processing unit (CPU) 801, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 802 or the program loaded from the storage section 808 into the random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the electronic device 800 are also stored. The CPU 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. The input / output (I / O) interface 805 is also connected to the bus 804.

[0117] The following components are connected to the I / O interface 805: an input section 806 including a keyboard, a mouse, etc.; an output section 807 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc. and a speaker, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN card, a modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as required. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 810 as required, so that a computer program read from it can be installed into the storage section 808 as required.

[0118] Specifically, according to the embodiments of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments of the present application include a computer program product including a computer-readable medium carrying instructions. In such an embodiment, the instructions can be downloaded and installed from the network through the communication section 809, and / or installed from the removable medium 811. When the instructions are executed by the central processing unit (CPU) 801, the various method steps described in the present application are executed.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the various embodiments of the present application.

[0120] In addition, those skilled in the art can understand that although some embodiments herein include certain features included in other embodiments rather than other features, the combination of the features of different embodiments means that it is within the scope of the present application and forms different embodiments. For example, all the above embodiments can be used in any combination. The information disclosed in this background art section is only intended to deepen the understanding of the overall background art of the present application, and should not be regarded as an admission or any form of suggestion that this information constitutes prior art already known to those skilled in the art.

Claims

1. A method for processing vehicle information on an environmentally friendly vehicle, characterized in that, The method includes: Obtaining a vehicle inventory picture to be recognized; Recognizing a corresponding first vehicle identifier from the recognition codes in the vehicle inventory picture; Performing multi-modal resolution enhancement processing on the vehicle inventory picture to generate at least two enhanced images, and calling at least one OCR model to extract the text information in the vehicle inventory picture from at least two enhanced images; Recognizing corresponding first environmental protection key information according to the extracted text information; Based on the comparison between the first vehicle identifier, the first environmental protection key information, and the known vehicle information, determining whether the corresponding environmental protection vehicle passes the verification.

2. The method according to claim 1, wherein The determining whether the corresponding environmental protection vehicle passes the verification based on the comparison between the first vehicle identifier, the first environmental protection key information, and the known vehicle information includes: Comparing whether the first vehicle identifier is exactly the same as the second vehicle identifier in the known vehicle information; Verifying the authenticity of the corresponding vehicle inventory based on the first environmental protection key information; When the vehicle identifiers are exactly the same and the vehicle inventory is a genuine inventory, comparing whether the first environmental protection key information matches the second environmental protection key information in the known vehicle information. If they match, it is determined that the verification passes; If the vehicle identifiers are not exactly the same, or the vehicle inventory is not a genuine inventory, or the environmental protection key information does not match, it is determined that the verification fails.

3. The method according to claim 2, wherein The verifying the authenticity of the corresponding vehicle inventory based on the first environmental protection key information includes: Extracting a corresponding first key field and the first field content corresponding to the first key field from the first environmental protection key information; Identifying the coincidence rate between the extracted key field and the standard key fields in the set of standard key fields required for the vehicle inventory; Identifying whether each first field content conforms to the field content specification of the corresponding coincident standard key field, and calculating the corresponding content compliance degree according to the specification result; When the coincidence rate exceeds the corresponding coincidence rate threshold and the content compliance degree exceeds the corresponding compliance degree threshold, it is determined that the vehicle inventory is a genuine inventory; When the coincidence rate does not exceed the corresponding coincidence rate threshold, or the content compliance degree does not exceed the corresponding compliance degree threshold, it is determined that the vehicle inventory is not a genuine inventory.

4. The method according to any one of claims 1 to 3, characterized in that, The performing multi-modal resolution enhancement processing on the vehicle inventory picture to generate at least two enhanced images includes: The first enhanced image performs Gaussian filtering after upsampling the vehicle inventory image to the target resolution using nearest neighbor interpolation. The second enhanced image upsamples the image to the target resolution using bicubic interpolation; The third enhanced image upsamples the image to the target resolution using bilinear interpolation; The calling at least one OCR model to extract the text information in the vehicle inventory picture from at least two enhanced images includes: Calling at least one preset OCR model to perform text recognition on each enhanced image. Each OCR model outputs a character recognition result for each enhanced image. The OCR model includes the Tesseract-OCR model; Determining the text information in the vehicle inventory picture according to each character recognition result.

5. The method according to claim 4, wherein Each character recognition result includes the recognized character and the original confidence of the corresponding character; Determining the text information in the vehicle checklist picture according to each character recognition result includes: Performing cross-model calibration on the original confidence levels output by each OCR model to obtain standard confidence levels; Selecting characters according to the recognized characters and the corresponding standard confidence levels in each character recognition result; Forming the text information in the vehicle checklist picture based on the selected characters.

6. The method according to claim 5, wherein The selecting characters according to the recognized characters and the corresponding standard confidence levels in each character recognition result includes: for each character recognition result, when the recognized characters in the same position are inconsistent, performing weighted voting calculation on each character in the same position based on the standard confidence level, and selecting candidate characters from multiple characters based on the weighted voting calculation result; The forming the text information in the vehicle checklist picture based on the selected characters includes: Forming initial text information based on the selected candidate characters; Performing vehicle information standardization verification on the initial text information, and identifying candidate characters that do not conform to the vehicle information standardization in the initial text information as characters to be corrected; Correcting the characters to be corrected based on each character recognition result to obtain corrected characters corresponding to the characters to be corrected; Forming the text information in the vehicle checklist picture according to the initial text information and the corrected characters.

7. The method according to claim 5, characterized in that, The performing cross-model calibration on the original confidence levels output by each OCR model to obtain standard confidence levels includes: mapping the original confidence level c0 to the corresponding standard confidence level c1 according to the following piecewise linear function: , where k1, k2, b1, b2, and c2 are preset confidence calibration parameters.

8. An on-vehicle information processing device for an environmentally friendly vehicle, characterized in that, The device includes: A picture acquisition module for acquiring a vehicle checklist picture to be recognized; An identification extraction module for identifying a corresponding first vehicle identification from the identification code in the vehicle checklist picture; An OCR recognition module for performing multi-modal resolution enhancement processing on the vehicle checklist picture to generate at least two enhanced images, calling at least one OCR model to extract the text information in the vehicle checklist picture from at least two enhanced images; identifying corresponding first environmental protection key information according to the extracted text information; A verification module for comparing the first vehicle identification and the first environmental protection key information with known vehicle information to determine whether the corresponding environmental protection vehicle passes the verification.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores executable instructions, and when the executable instructions are executed by a processor, the processor executes the method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, Including: One or more processors; A memory for storing one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors execute the method according to any one of claims 1 to 7.

Images