Digitized intelligent management method and system for bidder qualification certificates

Abstract

The invention belongs to the technical field of digital intelligent management, and provides a digital intelligent management method and system for bidder qualification certificates, and the method comprises the steps: constructing an auxiliary collection unit, and completing quality evaluation through edge detection and TTS voice guide standard collection in combination with definition, brightness uniformity and unnecessary shielding ratio. According to the method, defects such as yellowing and wrinkles of standard images are detected, technologies such as LAB space correction, Pix2Pix character complementation, HRNet and transfer learning are adopted for repairing, key fields are positioned based on a certificate type-format feature library, after CRNN extraction, logic semantic verification is completed in combination with bidding qualification requirements, and results are output. Manual rechecking is reduced, and bidding and tendering auditing efficiency and fairness are guaranteed.

Description

Technical Field

[0001] This invention belongs to the field of digital intelligent management technology, specifically a digital intelligent management method and system for the qualification certificates of bidding personnel. Background Technology

[0002] In the digital management system for the qualification certificates of bidding personnel, optical character recognition (OCR) technology is the core means to realize the digital transformation of paper certificates, lays the data foundation for subsequent intelligent verification and file management, significantly reduces the cost of manual data entry, and is a key support for promoting the intelligent upgrade of bidding qualification review.

[0003] However, in actual business scenarios, bidding qualification certificates generally suffer from physical damage and information obstruction: some certificates have yellowed, faded, wrinkled, or worn and blurred key areas due to long-term storage and multiple transfers; the core information column is often covered by stamp ink or the characters are missing due to binding holes, which poses a serious challenge to digitization and OCR recognition.

[0004] The existing technical solutions have obvious shortcomings: the imaging accuracy of conventional acquisition equipment such as mobile phone photography and ordinary scanners relied upon by bidders is limited, making it difficult to restore the details of damaged documents; the basic image enhancement algorithm on the system can only complete simple brightness adjustment and noise removal, but cannot eliminate wrinkles, shadows, ink obscuration, or repair missing information, which directly leads to substandard digital image quality and fails to meet the accuracy requirements of OCR recognition.

[0005] This problem has triggered a chain reaction: the error rate of OCR character extraction has increased significantly, causing information matching failures and deviations in the review conclusions during the intelligent verification process. This not only requires a large amount of manpower to review, offsetting the efficiency advantages of digitalization, but may also misjudge the qualifications of compliant bidders, undermining the fairness of bidding activities.

[0006] In summary, existing acquisition and image enhancement technologies are insufficient in processing damaged or obscured qualification certificates, which has become a key technical bottleneck restricting the stability and reliability of the digital intelligent management system for bidder qualification certificates.

[0007] Therefore, the present invention provides a digital intelligent management method and system for the qualification certificates of bidding personnel. Summary of the Invention

[0008] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0009] The technical solution adopted by this invention to solve its technical problem is:

[0010] One objective of this invention is to provide a digital intelligent management method for the qualification certificates of bidding personnel, comprising:

[0011] Step S10: Construct an auxiliary data collection unit for bidder qualification certificates, collect bidder qualification certificates, and evaluate the quality of the collected bidder qualification certificate images;

[0012] Step S20: Analyze the necessity of repairing the qualification certificate images of qualified bidders, repair the qualification certificate images of bidders that need repair, and improve the OCR recognition accuracy;

[0013] Step S30: Extract key fields from the repaired image of the bidder's qualification certificate, perform logical semantic verification on the extracted key fields, and output the recognition result.

[0014] As a further improvement of the present invention, the specific process of constructing the auxiliary data collection unit for bidder qualification certificates and collecting bidder qualification certificates is as follows:

[0015] Based on the edge detection algorithm, the device captures the physical outline of the bidder's qualification certificate and dynamically generates adaptive auxiliary lines on the acquisition interface. The auxiliary lines are set to high-contrast red, and the interface displays the text prompt to align the certificate with the red frame.

[0016] When the tilt angle of the bidder's qualification certificate is greater than 5°, the auxiliary line will flash and indicate the tilt of the certificate. Please adjust the angle until the tilt angle of the bidder's qualification certificate is less than or equal to 3°.

[0017] Integrated TTS voice synthesis technology: when the camera detects light intensity <80 lux, it plays instructions to avoid shooting in backlight.

[0018] When the distance between the bidder's qualification certificate and the camera is greater than 15cm, the playback instruction should be to move closer to the certificate and maintain a distance of 5-10cm; when obstructions such as hands or other objects are detected in the image of the bidder's qualification certificate, the playback instruction should be to remove the obstructions.

[0019] As a further improvement of the present invention, the specific process of quality assessment of the collected images of the bidders' qualification certificates is as follows:

[0020] The variance gradient method is used to calculate the sharpness value of the bidders' qualification certificate images. The preset sharpness threshold is ≥0.8. The higher the sharpness value, the clearer the image details.

[0021] The standard deviation of image grayscale values ​​is calculated and converted into brightness uniformity, with a preset brightness uniformity threshold of ≥0.7.

[0022] Based on semantic segmentation technology, necessary occlusion and unnecessary occlusion are distinguished, and the proportion of unnecessary occlusion area is calculated. The threshold for the proportion of unnecessary occlusion area is preset to be ≤30%.

[0023] If the image of the bidder's qualification certificate does not meet any of the thresholds corresponding to the clarity threshold, brightness uniformity threshold, or unnecessary occlusion area ratio threshold, the image quality of the bidder's qualification certificate is deemed substandard; otherwise, it is considered compliant.

[0024] As a further improvement of the present invention, the specific process of analyzing the necessity of restoring the qualification certificate images of qualified bidders is as follows:

[0025] Obtain pre-processed images of the bidders' qualification certificates that meet the quality standards. For any qualified bidders' qualification certificate image, if the color offset value is ≤15, the image is considered to have a yellowish tint; if the wrinkle width is ≤2px, the image is considered to have wrinkles; if the red pixel ratio is ≥3% and a circle / ellipse with a diameter of 2-5mm is detected and the gray value of the dark area of ​​the circle / ellipse is ≤50, the image is considered to have necessary obstruction by a seal / binding hole.

[0026] If any qualified bidder's qualification certificate image has any of the following issues: yellowing, wrinkles, or necessary obstruction of the seal / binding hole, then the bidder's qualification certificate image needs to be repaired.

[0027] As a further improvement to the present invention, the necessity analysis for restoring the qualification certificate images of qualified bidders specifically includes:

[0028] The images of the qualification certificates of bidders that meet the quality standards are preprocessed. Based on Hough transform, the images of the qualification certificates of bidders that do not meet the quality standards are tilted and corrected. The multi-scale CLAHE algorithm is used to perform brightness equalization processing on the images of the qualification certificates of bidders.

[0029] As a further improvement of the present invention, the specific process of repairing the image of the bidder's qualification certificate that needs to be repaired is as follows:

[0030] The images of the bidders' qualification certificates are converted from RGB color space to LAB color space. Histogram equalization is performed on the A red-green difference and B blue-yellow difference channels to correct color shift. Brightness is improved by gamma correction. Canny edge detection is used to locate wrinkled areas. Gaussian mixture model (GMM) is used to separate shadow pixels from normal pixels and perform brightness compensation. The Pix2Pix algorithm takes a 5×5 pixel normal image block around the wrinkled area as input to complete the blurred characters in the wrinkled area.

[0031] As a further improvement of the present invention, the specific process of repairing the image of the bidder's qualification certificate that needs to be repaired also includes:

[0032] The location coordinates of the binding hole are located using the HRNet keypoint detection algorithm, and the binding hole area is marked.

[0033] Call the feature library of complete document layout of the same type, match the field type corresponding to the binding hole area, and transfer the field features corresponding to the complete document layout features to the binding hole area based on transfer learning technology. Then, use the neighbor pixel interpolation method to complete the missing characters or borders.

[0034] As a further improvement of the present invention, the specific process of extracting key fields from the repaired image of the bidder's qualification certificate is as follows:

[0035] A document type-format feature database is pre-established to store the key field locations, font types, and numbering rules of documents of different types and regions;

[0036] The repaired images of the bidders' qualification certificates are compared with the features in the certificate type-format feature database to accurately match the certificate type and specific format.

[0037] Based on the features in the document type-format feature database, the key fields are located and extracted using a CRNN model.

[0038] As a further improvement of the present invention, the specific process of outputting the recognition result is as follows:

[0039] Obtain the qualification requirements of the current bidding project and compare them with the OCR recognition results. If the social security payment period does not meet the bidding requirements, the qualification level is not up to standard, or the certificate validity period does not cover the bidding deadline, any one of these conditions is not met, and the OCR recognition results are judged to have logical deviations.

[0040] The second objective of this invention is to provide a digital intelligent management system for the qualification certificates of bidding personnel, comprising:

[0041] Bidder Qualification Document Collection and Quality Assessment Module: Constructs an auxiliary collection unit for bidder qualification documents, collects bidder qualification documents, and performs quality assessment on the collected bidder qualification document images;

[0042] Image restoration analysis and optimization module for qualification certificates: Analyzes the necessity of restoring qualification certificate images of bidders that meet the quality standards, restores qualification certificate images of bidders that need restoration, and improves OCR recognition accuracy;

[0043] The module for extracting and verifying key fields of qualification certificates extracts key fields from the repaired images of the bidders' qualification certificates, performs logical semantic verification on the extracted key fields, and outputs the recognition results.

[0044] The beneficial effects of this invention are as follows:

[0045] 1. By using adaptive guide lines, voice guidance, and triple quality threshold assessment, unqualified images such as tilted, backlit, and occluded images are filtered out at the source, which greatly improves the standardization of document collection and the basic image quality, and reduces post-processing costs.

[0046] 2. Specifically addresses issues such as yellowing, wrinkles, and obstruction of seals / binding holes on documents. By combining multiple algorithms to repair incomplete / blurred information, it significantly improves OCR recognition accuracy and overcomes the bottleneck of insufficient processing capabilities for damaged documents in existing technologies.

[0047] 3. By relying on the document type-format feature database to locate key fields, and combining semantic logic with bidding qualification requirements for dual verification, we can reduce problems such as character confusion and format errors, reduce recognition bias, and avoid misjudging the qualifications of bidders.

[0048] 3. The entire process of data collection, repair, identification, and verification is automated, which greatly reduces the workload of manual data entry and review, improves the efficiency of bidding qualification review, and ensures the fairness and reliability of the review results. Attached Figure Description

[0049] The invention will now be further described with reference to the accompanying drawings.

[0050] Figure 1 This is a flowchart illustrating the steps of a digital intelligent management method for the qualification certificates of bidding personnel according to the present invention;

[0051] Figure 2 This is a system module diagram of a digital intelligent management system for the qualification certificates of bidding personnel according to the present invention. Detailed Implementation

[0052] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0053] Example 1

[0054] like Figure 1 As shown in the embodiment of the present invention, a method for annotating audio and video recorder recording files includes:

[0055] Step S10: Construct an auxiliary data collection unit for bidder qualification certificates, collect bidder qualification certificates, and evaluate the quality of the collected bidder qualification certificate images;

[0056] In step S10, an auxiliary data collection unit for bidder qualification certificates is constructed. The specific process for collecting bidder qualification certificates is as follows:

[0057] Based on the edge detection algorithm (Canny operator), the physical outline of the bidder's qualification certificate is captured in real time by the terminal device's camera. Adaptive auxiliary lines are dynamically generated on the acquisition interface (the outline ratio is automatically adjusted according to the size of the certificate, supporting different specifications of certificates such as ID card and A4 paper professional certificate). The auxiliary line color is set to high-contrast red, and the interface displays the text prompt to align the certificate with the red outline.

[0058] When the tilt angle of the bidder's qualification certificate is greater than 5°, the auxiliary line will flash and indicate that the certificate is tilted. Please adjust the angle and guide the user to rotate the terminal device until the tilt angle of the bidder's qualification certificate is less than or equal to 3° to ensure the correct position of the captured image.

[0059] It integrates TTS (Text-to-Speech) speech synthesis technology and presets 3 core guidance commands, which are dynamically triggered according to the shooting scene: when the camera detects light intensity <80 lux (backlight scene), it automatically plays the command to avoid backlight shooting and suggests adjusting the shooting direction;

[0060] When the distance between the bidder's qualification certificate and the camera is greater than 15cm, the playback instruction should be to move closer to the certificate and maintain a distance of 5-10cm; when obstructions such as hands or other objects are detected in the image of the bidder's qualification certificate (obstruction area > 5%), the playback instruction should be to remove the obstructions to ensure the integrity of the certificate information; the voice prompt volume is consistent with the terminal device system volume by default, but can be manually adjusted by the user.

[0061] It should be noted that the construction of the bidder qualification certificate auxiliary collection system filters the quality of bidder qualification certificate images from the source, reduces the later preprocessing process of bidder qualification certificate images, and improves the management efficiency of bidder qualification certificates.

[0062] In step S10, the specific process of quality assessment of the collected images of the bidders' qualification certificates is as follows:

[0063] The variance gradient method is used to calculate the sharpness value (range 0-1) of the bidder's qualification certificate image. The preset sharpness threshold is ≥0.8. The higher the sharpness value, the clearer the image details.

[0064] The standard deviation of the image grayscale values ​​is calculated (the smaller the standard deviation, the more uniform the brightness), and then converted into brightness uniformity (the value range is 0-1). The preset brightness uniformity threshold is ≥0.7.

[0065] Based on semantic segmentation techniques (such as the lightweight SegNet model), necessary occlusions (such as seals and binding holes) and unnecessary occlusions (such as hands and clutter) are distinguished. The proportion of unnecessary occlusion area is calculated as (number of pixels in the unnecessary occlusion area / total number of pixels in the bidder's qualification certificate image × 100%), with a preset threshold of ≤30% for the proportion of unnecessary occlusion area.

[0066] If the image of the bidder's qualification certificate does not meet any of the corresponding thresholds (e.g., clarity 0.6 < 0.8, non-necessary occlusion area ratio 40% > 30%), the image quality of the bidder's qualification certificate is deemed substandard, and the system immediately returns to the bidder's qualification certificate auxiliary acquisition unit to re-acquire the image.

[0067] If all the images of the bidders' qualification certificates meet the corresponding thresholds, then the quality of the bidders' qualification certificate images is deemed to be up to standard.

[0068] In some embodiments, necessary occlusions (stamps, binding holes) and unnecessary occlusions (hands, debris) in bidding qualification certificate images are accurately distinguished to provide data support for image quality assessment. A lightweight SegNet model is used to accurately distinguish between necessary and unnecessary occlusions. The basic architecture of the lightweight SegNet model consists of an encoder-decoder and skip connections (preserving edge localization accuracy). MBConv blocks are used to replace standard convolutions, and low-frequency hybrid wavelet pooling downsampling is used. The number of parameters is ≤0.5MB, and the inference time is ≤100ms / image (terminal adaptation). Necessary occlusion: red pixel ratio ≥3% + 2-5mm circular / elliptical dark area grayscale ≤50; unnecessary occlusion: irregular contours and standard deviation >20. The input is a certificate image with adequate lighting; the output is the proportion of unnecessary occlusion area.

[0069] Step S20: Analyze the necessity of repairing the qualification certificate images of qualified bidders, repair the qualification certificate images of bidders that need repair, and improve the OCR recognition accuracy;

[0070] In step S20, the specific process of analyzing the necessity of restoring the qualification certificate images of qualified bidders is as follows:

[0071] First, the qualification certificate images of qualified bidders are preprocessed. Based on Hough transform, the qualification certificate images of unqualified bidders are tilted (high-precision horizontal alignment) to correct the tilt distortion of the qualification certificate images captured by the terminal device camera. The multi-scale CLAHE (Contrast Limited Adaptive Histogram Equalization) algorithm is used to perform brightness equalization processing on the qualification certificate images of bidders to avoid local overexposure or underexposure caused by global brightness adjustment and ensure the brightness uniformity of the qualification certificate images of bidders.

[0072] Obtain images of the qualification certificates of the bidders after preprocessing to meet the quality standards. Perform defect detection on the images of the qualification certificates of the bidders that meet the quality standards. Defect detection includes: whether the images of the qualification certificates of the bidders are yellowed, whether there are wrinkles, and whether there is necessary obstruction of the seal / binding hole.

[0073] For any qualified bidder's qualification certificate image, if the color offset value of the bidder's qualification certificate image is ≤15, the bidder's qualification certificate image is judged to have yellowing; if the wrinkle width of the bidder's qualification certificate image is ≤2px, the bidder's qualification certificate image is judged to have wrinkles; if the proportion of red pixels in the bidder's qualification certificate image is ≥3% and a circle / ellipse with a diameter of 2-5mm is detected and the gray value of the circle / ellipse dark area is ≤50, the bidder's qualification certificate image is judged to have necessary obstruction of the seal / binding hole.

[0074] If any qualified bidder's qualification certificate image has any of the following issues: yellowing, wrinkles, or necessary obstruction of the seal / binding hole, then the bidder's qualification certificate image needs to be repaired.

[0075] In step S20, the specific process of repairing the images of the bidders' qualification certificates that need to be repaired and improving the OCR recognition accuracy is as follows:

[0076] For the images of the bidders' qualification certificates that need to be repaired, the images are converted from RGB color space to LAB color space. Histogram equalization is performed on the A (red-green difference) and B (blue-yellow difference) channels to correct color shift. Gamma correction (gamma value = 1.2) is used to improve brightness and restore the original colors of the bidders' qualification certificates.

[0077] Canny edge detection locates wrinkled regions (continuous regions with edge gradient values ​​> 80), Gaussian mixture model (GMM) separates shadow pixels from normal pixels and performs brightness compensation, and Pix2Pix algorithm takes a 5×5 pixel normal image block around the wrinkled region (edge ​​of the wrinkled region) as input to complete the blurred characters at the wrinkled region.

[0078] The location coordinates of the binding hole are located using the HRNet keypoint detection algorithm, and the binding hole area is marked.

[0079] Call the complete document format feature library of the same type (stores information such as field position, font size, character spacing, etc.) to match the field type (such as document number, validity period) corresponding to the binding hole area range.

[0080] Based on transfer learning technology, the field features corresponding to the complete document layout features are transferred to the binding hole area. The missing characters or borders are filled in by neighbor pixel interpolation to ensure that no key information is lost.

[0081] By performing restoration processing on the qualification certificate images of bidders that need to be repaired, the output qualification certificate images of bidders can significantly improve the OCR recognition accuracy;

[0082] Step S30: Extract key fields from the restored qualification certificate images of bidders, perform logical semantic verification on the extracted key fields, and output the recognition result;

[0083] In step S30, the specific process of extracting key fields from the restored qualification certificate images of bidders is as follows:

[0084] Pre-establish a certificate type-layout feature database to store the positions of key fields (such as the name of the ID card is in the upper left area of the image, the validity period of the professional title certificate is in the lower right corner, and the payment duration of the social security certificate is in the middle area), font types (Song typeface, Black typeface, Regular script), and numbering rules (such as the professional qualification certificate number is the regional code (2 digits) + year (4 digits) + serial number (6 digits)) of certificates of different types and regions;

[0085] Compare the restored qualification certificate images of bidders with the features in the certificate type-layout feature database to accurately match the certificate type and specific layout;

[0086] Locate the positions of key fields (name, certificate number, validity period, issuing authority, social security payment duration, qualification level, etc.) according to the features in the certificate type-layout feature database, and extract the key fields through the CRNN (Convolutional Recurrent Neural Network) model;

[0087] For the still slightly blurred areas (character gray contrast 30-40) after restoration, crop the blurred area and enlarge it by 2 times, then perform OCR recognition, and at the same time correct the recognition result in combination with the semantic logic of the front and back characters (such as there is no letter O in the certificate number, only contains the number 0; there are no special symbols in the name) to reduce the error rate;

[0088] In step S30, the specific process of performing logical semantic verification on the extracted key fields and outputting the recognition result is as follows:

[0089] Verify the length of the ID card number, and automatically correct obvious confusion errors such as recognizing the number 0 as the letter O, the number 1 as the number 7, and the number 6 as the number 9;

[0090] Verify the validity period, and the verification format is YYYY-MM-DD or YYYY year MM month DD day, and automatically correct non-standard formats such as 2026 / 05 / 01, 2026-5-1, 2026 year 5 month 1 day to the unified standard format;

[0091] Verify the certificate number according to the numbering rules of different certificates (e.g., professional title certificate number starts with ZC, social security payment certificate number is 10 digits, and professional qualification certificate number contains 12 characters), and mark the number that does not conform to the rules (e.g., ZC202312345 is missing 1 serial number and is marked as suspected error).

[0092] Obtain the qualification requirements of the current bidding project (such as social security payment duration ≥ 6 months, qualification level of intermediate or above, and certificate validity period covering the bid deadline), and compare them with the OCR recognition results;

[0093] If the recognition result is: social security payment period of 3 months, qualification level of primary level, certificate validity period until December 31, 2024 (bid deadline is January 10, 2025), then the OCR recognition result is judged to have logical deviation.

[0094] Specifically, the criteria for judging logical deviation are as follows:

[0095] If the duration of social security contributions does not meet the bidding requirements, the qualification level is not up to standard, or the validity period of the certificate does not cover the bid deadline, any one of these conditions will be judged as a logical deviation.

[0096] Example 2

[0097] like Figure 2 As shown, based on the specific implementation process of Embodiment 1, the present invention provides a digital intelligent management system for the qualification certificates of bidding personnel, including:

[0098] Bidder Qualification Document Collection and Quality Assessment Module: Constructs an auxiliary collection unit for bidder qualification documents, collects bidder qualification documents, and performs quality assessment on the collected bidder qualification document images;

[0099] Image restoration analysis and optimization module for qualification certificates: Analyzes the necessity of restoring qualification certificate images of bidders that meet the quality standards, restores qualification certificate images of bidders that need restoration, and improves OCR recognition accuracy;

[0100] The module for extracting and verifying key fields of qualification certificates extracts key fields from the repaired images of the bidders' qualification certificates, performs logical semantic verification on the extracted key fields, and outputs the recognition results.

[0101] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A digital intelligent management method for the qualification certificates of bidding personnel, characterized in that: include: Step S10: Construct an auxiliary data collection unit for bidder qualification certificates, collect bidder qualification certificates, and evaluate the quality of the collected bidder qualification certificate images; Step S20: Analyze the necessity of repairing the qualification certificate images of qualified bidders, repair the qualification certificate images of bidders that need repair, and improve the OCR recognition accuracy; Step S30: Extract key fields from the repaired image of the bidder's qualification certificate, perform logical semantic verification on the extracted key fields, and output the recognition result.

2. The digital intelligent management method for bidder qualification certificates according to claim 1, characterized in that: The specific process of constructing the auxiliary data collection unit for bidder qualification certificates is as follows: Based on the edge detection algorithm, the device captures the physical outline of the bidder's qualification certificate and dynamically generates adaptive auxiliary lines on the acquisition interface. The auxiliary lines are set to a high-contrast red color, and the interface displays a text prompt to align the certificate with the red frame. When the tilt angle of the bidder's qualification certificate is greater than 5°, the auxiliary line will flash and indicate the tilt of the certificate. Please adjust the angle until the tilt angle of the bidder's qualification certificate is less than or equal to 3°. Integrated TTS voice synthesis technology: when the camera detects light intensity <80 lux, it plays instructions to avoid shooting in backlight. When the distance between the bidder's qualification certificate and the camera is greater than 15cm, the playback instruction should be to move closer to the certificate and maintain a distance of 5-10cm; when obstructions such as hands or other objects are detected in the image of the bidder's qualification certificate, the playback instruction should be to remove the obstructions.

3. The digital intelligent management method for bidder qualification certificates according to claim 1, characterized in that: The specific process for quality assessment of the collected images of bidders' qualification certificates is as follows: The variance gradient method is used to calculate the sharpness value of the bidders' qualification certificate images. The preset sharpness threshold is ≥0.

8. The higher the sharpness value, the clearer the image details. The standard deviation of image grayscale values ​​is calculated and converted into brightness uniformity, with a preset brightness uniformity threshold of ≥0.

7. Based on semantic segmentation technology, necessary occlusion and unnecessary occlusion are distinguished, and the proportion of unnecessary occlusion area is calculated. The threshold for the proportion of unnecessary occlusion area is preset to be ≤30%. If the image of the bidder's qualification certificate does not meet any of the thresholds corresponding to the clarity threshold, brightness uniformity threshold, or unnecessary occlusion area ratio threshold, the image quality of the bidder's qualification certificate is deemed substandard; otherwise, it is considered compliant.

4. The digital intelligent management method for bidder qualification certificates according to claim 1, characterized in that: The specific process for analyzing the necessity of restoring the images of the qualification certificates of bidders that meet the quality standards is as follows: Obtain pre-processed images of the bidders' qualification certificates that meet the quality standards. For any qualified bidders' qualification certificate image, if the color offset value is ≤15, the image is considered to have a yellowish tint; if the wrinkle width is ≤2px, the image is considered to have wrinkles; if the red pixel ratio is ≥3% and a circle / ellipse with a diameter of 2-5mm is detected and the gray value of the dark area of ​​the circle / ellipse is ≤50, the image is considered to have necessary obstruction by a seal / binding hole. If any qualified bidder's qualification certificate image has any of the following issues: yellowing, wrinkles, or necessary obstruction of the seal / binding hole, then the bidder's qualification certificate image needs to be repaired.

5. The digital intelligent management method for bidder qualification certificates according to claim 4, characterized in that: The necessity analysis for restoring the qualification certificate images of qualified bidders also includes: The images of the qualification certificates of bidders that meet the quality standards are preprocessed. Based on Hough transform, the images of the qualification certificates of bidders that do not meet the quality standards are tilted and corrected. The multi-scale CLAHE algorithm is used to perform brightness equalization processing on the images of the qualification certificates of bidders.

6. The digital intelligent management method for bidder qualification certificates according to claim 1, characterized in that: The specific process for repairing the images of the bidders' qualification certificates that need to be restored is as follows: The images of the bidders' qualification certificates are converted from RGB color space to LAB color space. Histogram equalization is performed on the A red-green difference and B blue-yellow difference channels to correct color shift. Brightness is improved by gamma correction. Canny edge detection is used to locate wrinkled areas. Gaussian mixture model (GMM) is used to separate shadow pixels from normal pixels and perform brightness compensation. The Pix2Pix algorithm takes a 5×5 pixel normal image block around the wrinkled area as input to complete the blurred characters in the wrinkled area.

7. The digital intelligent management method for bidder qualification certificates according to claim 1, characterized in that: The specific process for repairing the images of the bidders' qualification certificates that need to be restored also includes: The location coordinates of the binding hole are located using the HRNet keypoint detection algorithm, and the binding hole area is marked. Call the feature library of complete document layout of the same type, match the field type corresponding to the binding hole area, and transfer the field features corresponding to the complete document layout features to the binding hole area based on transfer learning technology. Then, use the neighbor pixel interpolation method to complete the missing characters or borders.

8. The digital intelligent management method for bidder qualification certificates according to claim 1, characterized in that: The specific process for extracting key fields from the repaired images of the bidders' qualification certificates is as follows: A document type-format feature database is pre-established to store the key field locations, font types, and numbering rules of documents of different types and regions; The restored images of the bidders' qualification certificates are compared with features in the certificate type-format feature database to accurately match the certificate type and specific format. Based on the features in the document type-format feature database, the key fields are located and extracted using a CRNN model.

9. The digital intelligent management method for bidder qualification certificates according to claim 1, characterized in that: The specific process for outputting the recognition result is as follows: Obtain the qualification requirements of the current bidding project and compare them with the OCR recognition results. If the social security payment period does not meet the bidding requirements, the qualification level is not up to standard, or the certificate validity period does not cover the bidding deadline, any one of these conditions is not met, and the OCR recognition results are judged to have logical deviations.

10. A digital intelligent management system for bidder qualification certificates, used to execute the method described in any one of claims 1-9 above, comprising: Bidder Qualification Document Collection and Quality Assessment Module: Constructs an auxiliary collection unit for bidder qualification documents, collects bidder qualification documents, and performs quality assessment on the collected bidder qualification document images; Image restoration analysis and optimization module for qualification certificates: Analyzes the necessity of restoring qualification certificate images of qualified bidders, restores qualification certificate images of bidders that need restoration, and improves OCR recognition accuracy; The module for extracting and verifying key fields of qualification certificates extracts key fields from the repaired images of the bidders' qualification certificates, performs logical semantic verification on the extracted key fields, and outputs the recognition results.

Images