Environment evaluation form intelligent identification method and system

Abstract

The invention relates to the technical field of computer vision, and provides an environment evaluation form intelligent identification method and system. The method comprises the following steps: acquiring image information of an environment evaluation form according to a preset definition requirement by using image acquisition equipment, and constructing an environment evaluation form image library; preprocessing the image, including de-noising, enhancing and the like; segmenting the image according to a preset requirement by using an image segmentation technology, and labeling attributes; identifying the segmentation result by using an optical character identification technology to obtain an identification result; performing evaluation and correction according to the attribute label to obtain a corrected identification result; key information is extracted by using a natural language processing technology and is stored in a structured form. According to the method, the image segmentation result is identified through an optical character identification technology, and the identification result is evaluated and corrected based on attribute labeling, so that the technical problem of low identification efficiency and accuracy caused by form content diversity is solved, and efficient and accurate environment evaluation work is ensured.

Description

Technical Field

[0001] The present application relates to the field of computer vision technology, specifically to the field of image recognition and optical character recognition technology, and in particular to a method and system for intelligent recognition of environmental assessment forms. Background Art

[0002] In environmental protection work, intelligent processing of environmental assessment forms is crucial for data analysis and decision-making. The traditional environmental assessment form processing method mainly relies on the computer's own algorithms and preset programs. Although it avoids a lot of manual operations, in actual applications, there are still some problems that affect the efficiency and accuracy of form recognition when the computer processes environmental assessment forms, such as the limitations of algorithms and models, and insufficient accuracy in image data recognition and segmentation.

[0003] In this context, an intelligent recognition method and system for environmental assessment forms came into being. The system aims to identify the image segmentation results through optical character recognition technology, and evaluate and correct the recognition results based on attribute annotation, so as to solve the technical problems of low recognition efficiency and accuracy caused by the diversity of form content. Summary of the invention

[0004] The present application provides an intelligent recognition method and system for environmental assessment forms, aiming to identify image segmentation results through optical character recognition technology, and evaluate and correct the recognition results based on attribute annotation, so as to solve the technical problem of low recognition efficiency and accuracy caused by the diversity of form content.

[0005] In view of the above problems, the present application provides an environmental evaluation form intelligent recognition method and system.

[0006] The first aspect disclosed in the present application provides an intelligent recognition method for an environmental assessment form, the method comprising: using an image acquisition device to acquire image information of an environmental assessment form according to a preset clarity requirement, and constructing an environmental assessment form image library; preprocessing the images in the environmental assessment form image library, including denoising and edge detection; using image segmentation technology to segment the preprocessed environmental assessment form image according to preset segmentation requirements, and annotating according to the attributes corresponding to the segmentation results; using optical character recognition technology to sequentially identify the segmentation results of the environmental assessment form to obtain recognition results; based on the attribute annotation, evaluating and correcting the recognition results to obtain a corrected recognition result; according to the structural content of the environmental assessment form, extracting key information in the corrected recognition result through natural language processing technology, and storing it in a structured form based on the structural content.

[0007] Furthermore, the image in the environmental assessment form image library is preprocessed, including the method of denoising and edge detection, and further includes: Performing denoising processing on the images in the environmental assessment form image library; Use the denoised images to perform edge detection, extract images that do not meet the edge detection requirements, and build an abnormal image library; Confirming the number of the acquired image based on the abnormal image library generates a re-acquisition instruction, sends it to the image acquisition device, and establishes a mapping relationship between the re-acquisition image and the abnormal image; The edge detection of the re-collected images is performed based on the abnormal image library until the preset requirements are met, and all images that meet the preset requirements are stored in the environmental assessment form image library.

[0008] Preferably, the system terminal confirms the image number that needs to be recaptured based on the image information in the abnormal image library, and generates a corresponding recapture instruction. This instruction will be sent to the image acquisition device to instruct the device to reacquire these images. At the same time, in order to ensure the consistency and traceability of the data, a mapping relationship between the recaptured images and the abnormal images is established. In this way, in the subsequent processing and analysis, it can be clearly known which images are recaptured and their corresponding original abnormal images. After the recapture is completed, the system terminal will perform edge detection on these images again. This process will be repeated until all the recaptured images meet the preset edge detection requirements. Once the images meet the requirements, these images are stored in the environmental assessment form image library for subsequent use. Through this process, it can be ensured that the image quality in the environmental assessment form image library is effectively improved, providing accurate and reliable data support for subsequent environmental assessment work.

[0009] Use image segmentation technology to segment the preprocessed environmental assessment form image according to preset segmentation requirements, and mark it according to the attributes corresponding to the segmentation results; Furthermore, the method of segmenting the preprocessed environmental assessment form image according to preset segmentation requirements by using image segmentation technology and marking the attributes corresponding to the segmentation results also includes: According to the recognition task in the form recognition request, the preset segmentation requirements are set, including title, field, and data; Based on the preset segmentation requirement, generating the attribute annotation; The preset segmentation requirement is added to the image segmentation technology to generate a segmentation execution module, the pre-processed environment assessment form image is segmented, and the corresponding segmentation result is annotated based on the attribute annotation.

[0010] Preferably, the image segmentation technology is used to segment the pre-processed environmental assessment form image according to the preset segmentation requirements, and the segmentation result is marked according to the attributes corresponding to the segmentation result. The attribute marking refers to the marking of the category and meaning of the image area obtained after segmentation, including title marking, field marking, and data marking. Afterwards, the system terminal uses the integrated preset segmentation requirements as input parameters and embeds them into the preset image segmentation algorithm to form a segmentation execution module. Then, the pre-processed environmental assessment form image is input into the segmentation execution module. The segmentation execution module automatically segments the image according to the preset segmentation requirements and divides the form into multiple areas or objects. Then, according to the attribute marking generated before, a corresponding label is assigned to each area in the segmentation result to identify the functions and contents of different areas. Finally, the segmentation result with attribute marking is output in the form of a data structure. This output result not only contains the segmented image area, but also contains the attribute marking information of each area, which is convenient for subsequent information extraction and processing. Through this series of operations, the system terminal can realize the accurate segmentation and marking of the environmental assessment form image, laying a solid foundation for subsequent data extraction and analysis. This not only improves the efficiency and accuracy of form recognition, but also greatly improves overall work efficiency.

[0011] Furthermore, the method of sequentially recognizing the segmentation results of the environmental evaluation form using optical character recognition technology to obtain recognition results also includes: Converting the environmental assessment form into text information; Performing content recognition on the text information according to the segmentation result to obtain a recognition result; Based on the attribute annotation of the environmental evaluation form, the character structure of each segmentation result is determined, and the recognition result is converted according to the character structure to obtain the recognition result.

[0012] Furthermore, based on the attribute annotation, the recognition result is evaluated and corrected to obtain a corrected recognition result, and the method also includes: Based on the attribute annotation, determine the content format requirements and necessary parameter information of the segmentation result; According to the content format requirements and necessary parameter information, the recognition results are traversed and matched to determine abnormal matching results; When the abnormal matching result does not exist, a mapping relationship between the recognition result and the form is established and exported to the recognition database; When the abnormal matching result exists, the deviation information is obtained, and the deviation information is supplemented by using the attribute tolerance to obtain a corrected recognition result; for the deviation information exceeding the attribute tolerance, a correction recognition constraint condition is generated, and the constraint condition is added to the optical character recognition technology, and a re-recognition module is generated for recognition, and the recognition result that meets the requirements is stored in the recognition database.

[0013] Furthermore, the constraint condition is added to the optical character recognition technology to generate a re-recognition module for recognition, and the method further includes: Determine the image number of the environmental evaluation form based on the deviation information, and traverse from the abnormal image; Determine whether there is an abnormal graph according to the traversal result, and if so, locate the deviation information of the abnormal graph; Furthermore, the present application provides that before storing the recognition results that meet the requirements in the recognition database, the method also includes: When the re-identification result cannot meet the content format requirements and necessary parameter information, the deviation information is marked and an abnormal reminder message is generated.

[0014] Furthermore, the step of extracting key information from the correction recognition result by natural language processing technology according to the structural content of the environmental evaluation form also includes: Extracting key information from the modified recognition result of each form by natural language processing technology, and converting the key information corresponding to each form into a feature vector of the modified form; Select several eigenvectors as initial cluster centers; Calculate the distance from each eigenvector to all cluster centers and assign it to the nearest cluster center; Calculate the mean of all feature vectors in each cluster and update the cluster center position; Perform iteration, the iteration method is as follows: Recalculate the distance from each eigenvector to the updated cluster center; Reassign each eigenvector to the nearest cluster center; Recalculate the mean of all feature vectors in each cluster and update the cluster center; Continue iterating until the positions of the cluster centers remain stable for two consecutive iterations.

[0015] Furthermore, the iteration is continued until the position of the cluster center remains stable in two consecutive iterations, and the determination method is as follows: After selecting the initial cluster centers and performing the first cluster assignment, the sum of squared errors within the initial clusters is calculated; After updating the cluster center, reassign each eigenvector to the nearest cluster center and calculate the new intra-cluster sum of squared errors; Compare the change in the sum of squared errors within the cluster between the current iteration and the previous iteration; If the change in the sum of squared errors within the cluster is less than the preset threshold, or reaches the set maximum number of iterations, the algorithm is considered to have converged and the iteration can be stopped.

[0016] The calculation method of the intra-cluster error sum of squares is as follows: Through the feature vector and cluster center, the sum of squared errors within the cluster is calculated through the objective function to accurately extract the evaluation items in the environmental evaluation form and cluster similar environmental evaluation forms together. The objective function is as follows:

[0017] in, J is the sum of squared errors within the cluster, K is the number of clusters, υ ί is the eigenvector of the ιth form, μκ is the center of the kth cluster, C κ is the set of all eigenvectors in the kth cluster.

[0018] Furthermore, the feature vectors used as the initial cluster centers may randomly select K feature vectors as the initial cluster centers, and the number of the selected feature vectors K is the initial number of clusters K.

[0019] Another aspect disclosed in the present application provides an intelligent recognition system for environmental assessment forms, the system comprising: an information acquisition module, the information acquisition module being used to use an image acquisition device to collect image information of environmental assessment forms according to preset clarity requirements and construct an environmental assessment form image library; a preprocessing module, the preprocessing module being used to preprocess images in the environmental assessment form image library, including denoising and edge detection; an image segmentation module, the image segmentation module being used to use image segmentation technology to segment the preprocessed environmental assessment form image according to preset segmentation requirements and to annotate according to attributes corresponding to the segmentation results; a segmentation and recognition module, the segmentation and recognition module being used to use optical character recognition technology to sequentially identify the segmentation results of the environmental assessment form to obtain recognition results; an evaluation and correction module, the evaluation and correction module being used to evaluate and correct the recognition results based on attribute annotation to obtain corrected recognition results; an information extraction module, the information extraction module being used to extract key information from the corrected recognition results according to the structural content of the environmental assessment form through natural language processing technology, and to store the key information in a structured form based on the structural content.

[0020] One or more technical solutions provided in this application have at least the following technical effects or advantages: The above-mentioned intelligent recognition method for environmental evaluation forms uses image acquisition equipment to accurately capture the image information of environmental evaluation forms according to the established clarity standard, and systematically organizes these images into an image library. Subsequently, the form images in the image library are preprocessed. Through denoising, various interference factors in the image are effectively removed; at the same time, the edge detection technology is used to accurately define the various components of the form, laying a solid foundation for subsequent processing. After that, the image segmentation technology is used to divide the preprocessed form image into different logical areas according to the preset segmentation rules. Each area is accurately labeled according to its content attributes to provide clear guidance for subsequent recognition work. On this basis, optical character recognition technology is used to accurately recognize the text of the segmented form area, and the text information in the image is converted into computer-processable text data, providing raw materials for subsequent data analysis. In order to ensure the accuracy of the recognition results, the recognition results are further strictly evaluated and corrected based on the previous attribute annotation information. By comparing the annotation information with the recognition results, possible errors are discovered and corrected in a timely manner to ensure the accuracy of the data. Finally, natural language processing technology is used to extract key information from the corrected recognition results based on the structural characteristics of the environmental evaluation form. This information is stored in a structured form to facilitate subsequent data analysis and processing. The entire process achieves efficient and accurate processing of environmental assessment forms, providing strong technical support for the smooth implementation of environmental protection work.

[0021] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0023] Figure 1 A schematic diagram of a flow chart of an intelligent identification method for an environmental evaluation form in one embodiment; Figure 2 This is an architecture diagram of an intelligent recognition system for an environmental evaluation form in one embodiment.

[0024] Explanation of the accompanying drawings: information acquisition module 1, preprocessing module 2, image segmentation module 3, segmentation and recognition module 4, evaluation and correction module 5, information extraction module 6. DETAILED DESCRIPTION

[0025] The embodiments of the present application provide a method and system for intelligently identifying an environmental evaluation form, thereby solving the technical problem of insufficient recognition accuracy and efficiency due to the diversity of form content.

[0026] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0027] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0028] Embodiment 1 like Figure 1 As shown, the present application provides an intelligent recognition method for an environmental evaluation form, the method comprising: Using image acquisition equipment to collect image information of the environmental assessment form according to preset clarity requirements, and constructing an environmental assessment form image library; The environmental assessment form is an important tool for systematically evaluating and predicting the possible environmental impact of a project. It records in detail various environmental information of the project, including the impact on the atmosphere, water, soil, biodiversity, and other aspects, as well as the impact on human health and social economy.

[0029] In the embodiment of the present application, the system terminal uses an image acquisition device to collect image information of the environmental assessment form according to a preset clarity standard. During this process, the device will ensure that the collected image is clear and complete in detail for subsequent processing and analysis. The collected image information will be organized and stored in a special image library to construct a complete environmental assessment form image collection. This step provides the necessary image material and data basis for subsequent form recognition and processing.

[0030] Preprocessing the images in the environmental assessment form image library, including denoising and edge detection; In one embodiment, in order to improve the quality of the environmental assessment form image, the system terminal performs a series of preprocessing operations on each form image in the image library. First, the denoising technology is used to eliminate interference factors such as noise and spots in the image to make the image clearer and purer. Subsequently, edge detection is performed to determine whether there is a problem with the collected image information. If there is a problem, the system terminal will store the image with the problem, and generate a re-collection instruction for these problematic images according to the image number, so that the image acquisition device can re-collect these images. After that, the edge detection step is repeated until all the images are correct, and the detected images are stored in the environmental assessment form image library. After these preprocessing steps, the quality of the form image will be significantly improved, laying a solid foundation for subsequent recognition work.

[0031] Furthermore, the present application provides preprocessing of images in the environmental assessment form image library, including denoising and edge detection, and the method further includes: Performing denoising processing on the images in the environmental assessment form image library; Use the denoised images to perform edge detection, extract images that do not meet the edge detection requirements, and build an abnormal image library; Preferably, in order to optimize the recognition effect of the environmental assessment form image, the system terminal first performs denoising on each form image in the image library. This step is intended to remove noise and interference in the image to make the image content clearer and more accurate. The image quality after denoising has been significantly improved, laying a good foundation for subsequent edge detection work. Subsequently, the denoised image is used for edge detection. Edge detection is a key step in identifying the boundaries of different regions in the image. The system terminal performs edge positioning and refinement, non-maximum suppression and edge tracking, and post-processing steps through a preset edge detection algorithm to complete the edge detection of the processed image. Afterwards, the results of the edge detection are visually and algorithmically evaluated to identify images with blurred edges, broken edges, or detected edges that do not match the actual form content. Then, images whose edge detection results do not meet the requirements are extracted from the image library and stored in a special abnormal image library. The establishment of this abnormal image library helps to further analyze and process these images, as well as re-collect and pre-process them, to improve the overall recognition accuracy of the environmental assessment form image. In summary, by denoising and edge detection of environmental assessment form images and combining them with the establishment of an abnormal image library, the image recognition effect can be better optimized and the accuracy and efficiency of environmental assessment form data processing can be improved.

[0032] Confirming the number of the acquired image based on the abnormal image library generates a re-acquisition instruction, sends it to the image acquisition device, and establishes a mapping relationship between the re-acquisition image and the abnormal image; The edge detection of the re-collected images is performed based on the abnormal image library until the preset requirements are met, and all images that meet the preset requirements are stored in the environmental assessment form image library.

[0033] Preferably, the system terminal confirms the image number that needs to be recaptured based on the image information in the abnormal image library, and generates a corresponding recapture instruction. This instruction will be sent to the image acquisition device to instruct the device to reacquire these images. At the same time, in order to ensure the consistency and traceability of the data, a mapping relationship between the recaptured images and the abnormal images is established. In this way, in the subsequent processing and analysis, it can be clearly known which images are recaptured and their corresponding original abnormal images. After the recapture is completed, the system terminal will perform edge detection on these images again. This process will be repeated until all the recaptured images meet the preset edge detection requirements. Once the images meet the requirements, these images are stored in the environmental assessment form image library for subsequent use. Through this process, it can be ensured that the image quality in the environmental assessment form image library is effectively improved, providing accurate and reliable data support for subsequent environmental assessment work.

[0034] Use image segmentation technology to segment the preprocessed environmental assessment form image according to preset segmentation requirements, and mark it according to the attributes corresponding to the segmentation results; In one embodiment, the quality and clarity of the pre-processed environmental assessment form image have been significantly improved, laying a good foundation for subsequent segmentation and annotation work. The system terminal will use image segmentation technology to further process these images. In the segmentation process of the environmental assessment form image, the system terminal accurately divides the different parts in the form according to the preset segmentation requirements, such as features such as titles and fields, and annotates the relevant attributes. After the segmentation is completed, each area is annotated according to the attributes corresponding to the segmentation results. This helps the system terminal to better understand the structure and content of the form. Through image segmentation and annotation, the environmental assessment form image can be converted into structured data, which is convenient for subsequent automatic recognition and extraction. This can not only improve the efficiency of environmental assessment work, but also improve the accuracy of data.

[0035] Furthermore, the present application provides the method of segmenting the preprocessed environmental assessment form image according to preset segmentation requirements by using the image segmentation technology, and marking the image according to the attributes corresponding to the segmentation results, and the method also includes: According to the recognition task in the form recognition request, the preset segmentation requirements are set, including title, field, and data; Based on the preset segmentation requirement, generating the attribute annotation; The preset segmentation requirement is added to the image segmentation technology to generate a segmentation execution module, the pre-processed environment assessment form image is segmented, and the corresponding segmentation result is annotated based on the attribute annotation.

[0036] Preferably, the image segmentation technology is used to segment the pre-processed environmental assessment form image according to the preset segmentation requirements, and the segmentation result is marked according to the attributes corresponding to the segmentation result. The attribute marking refers to the marking of the category and meaning of the image area obtained after segmentation, including title marking, field marking, and data marking. The system terminal first sets the preset segmentation requirements according to the specific task requirements in the form recognition request, and these requirements mainly cover the key parts such as the title, field and data of the form. These requirements provide clear guidance for the subsequent image segmentation to ensure that each important area in the form can be accurately segmented. Subsequently, based on these preset segmentation requirements, corresponding attribute annotations are generated. These annotations are actually a kind of category and meaning marking of the image area obtained after segmentation, which helps to understand the specific content represented by each area. Afterwards, the system terminal uses the integrated preset segmentation requirements as input parameters and embeds them into the preset image segmentation algorithm to form a segmentation execution module. Then the pre-processed environmental assessment form image is input into the segmentation execution module. The segmentation execution module automatically segments the image according to the preset segmentation requirements and divides the form into multiple areas or objects. Then, according to the attribute annotations generated previously, a corresponding label is assigned to each area in the segmentation result to identify the functions and contents of different areas. Finally, the segmentation result with attribute annotations is output in the form of a data structure. This output result not only contains the segmented image area, but also the attribute annotation information of each area, which is convenient for subsequent information extraction and processing. Through this series of operations, the system terminal can achieve accurate segmentation and annotation of the environmental evaluation form image, laying a solid foundation for subsequent data extraction and analysis. This can not only improve the efficiency and accuracy of form recognition, but also greatly improve the overall work efficiency.

[0037] The segmentation results of the environmental assessment form are sequentially recognized using optical character recognition technology to obtain recognition results; In one embodiment, the main function of optical character recognition (OCR) technology is to convert text information in an image into a text format that can be edited and processed by a computer. In the environmental assessment form processing, the system terminal first uses OCR technology to scan and recognize the form image, extract the text information therein and convert it into text form. After converting to text information, the system terminal combines the previous segmentation results of the form image to perform operations. This segmentation result provides a division of different areas in the form, such as titles, fields, data, etc. Based on these segmentation results, more accurate and targeted operations can be performed on the converted text information. Specifically, the system terminal matches and corresponds the recognized text information with the corresponding form area according to the label in the segmentation result. In this way, the value of each form field can be accurately extracted, or the text in a specific area can be further analyzed and processed, and a recognition result can be generated. In summary, in this way, the system terminal can efficiently convert the text information in the environmental assessment form into a text format that can be processed by a computer, and accurately operate the converted text information based on the segmentation results, providing a convenient and accurate basis for subsequent data analysis and processing.

[0038] Furthermore, the present application provides the method of sequentially identifying the segmentation results of the environmental evaluation form using the optical character recognition technology to obtain the recognition results, and the method also includes: Converting the environmental assessment form into text information; Performing content recognition on the text information according to the segmentation result to obtain a recognition result; Based on the attribute annotation of the environmental evaluation form, the character structure of each segmentation result is determined, and the recognition result is converted according to the character structure to obtain the recognition result.

[0039] Preferably, the system terminal first performs optical character recognition on the entire form image. The optical character engine scans the entire image, analyzes the pixel pattern therein, and recognizes all visible text. After the optical character engine recognizes the text, it converts it into computer-editable text information, which provides a basis for subsequent operations. Subsequently, the system terminal performs content recognition on the converted text information based on the results of the previous segmentation of the form image. Since the segmentation result has divided the form into different areas, the system terminal can perform targeted content recognition for each area to ensure that the information of each field or title is accurately extracted. In the process of content recognition, the system terminal matches the identifier in the annotation with the corresponding information in the segmentation result to determine the attribute annotation corresponding to each segmentation result. Based on the matched attribute annotation, the system terminal determines the character structure of each segmentation result. The character structure describes the arrangement, font, size and other characteristics of the text in the area. For example, some fields may require fixed-length numbers, while other fields may allow text containing special characters or spaces. With the character structure information, the system terminal converts the recognition results. The purpose of the conversion is to make the recognition results meet the character structure requirements of the form. This includes adjusting the length, format, font, etc. of the text. For example, if a field requires a fixed-length number, but the length of the number in the recognition result does not match, padding or truncation is required. Finally, the converted recognition result is output to obtain the final recognition result. This result exists in the form of text and contains the key information in the environmental assessment form, which facilitates subsequent data analysis and processing. In summary, after converting the environmental assessment form into text information, the content is recognized by combining the segmentation results and attribute annotations, and the recognition results are converted according to the character structure, and finally an accurate recognition result is obtained. This process improves the automation of form processing and ensures the accuracy and completeness of the information.

[0040] Based on the attribute annotation, the recognition result is evaluated and corrected to obtain a corrected recognition result; In one embodiment, the process of evaluating and correcting the recognition results based on attribute annotation mainly uses the known form attribute annotation information to check and correct the text results recognized by the optical character recognition technology. The purpose of this step is to improve the accuracy and completeness of the recognition results and ensure that the converted text is consistent with the original form content. First, the system terminal evaluates the recognition results by comparing the content format requirements, necessary parameters and other information defined in the attribute annotation. If it is found that the recognition results do not match the annotation information, such as format errors, missing data or garbled characters, correction is required. The correction process includes a variety of methods. For simple format errors or spelling errors, they can be processed by automatic replacement. For complex situations, such as missing data or garbled characters, it is necessary to combine the correction algorithm to repair. After completing the evaluation and correction, the corrected recognition results will be obtained. These results have been strictly checked and adjusted, and are more in line with the attribute requirements of the original form, providing a reliable basis for subsequent data processing and analysis. In summary, evaluating and correcting the recognition results based on attribute annotation is a key step to ensure the quality of conversion. Through this step, the accuracy of the recognition results can be improved, errors and omissions can be reduced, and high-quality text data can be provided for the subsequent processing of the environmental assessment form.

[0041] Furthermore, the present application provides for evaluating and correcting the recognition result based on attribute annotation to obtain a corrected recognition result, and the method further includes: Based on the attribute annotation, determine the content format requirements and necessary parameter information of the segmentation result; According to the content format requirements and necessary parameter information, the recognition results are traversed and matched to determine abnormal matching results; Preferably, based on attribute annotation, the system terminal first determines the content format requirements and necessary parameter information of each segmentation result in the environmental assessment form. This usually includes the data type, format specification, length limit, and specific identifier that each segmented area should contain. This information is crucial for the subsequent recognition result verification. Once these format requirements and parameter information are available, the recognition results can be traversed and matched. This means that the system terminal will check each data item in the recognition result one by one and compare it with the corresponding content format requirements and parameter information. The purpose of this is to find out those abnormal matching results that do not meet the preset requirements. Abnormal matching results include data with incorrect format, data with missing necessary parameters, or data containing invalid characters. Through traversal matching, these abnormal results can be accurately located to provide a basis for subsequent data correction. In summary, it is an effective data verification process to determine the content format requirements and necessary parameter information based on attribute annotation, and then traverse and match the recognition results. This can help the system terminal to promptly discover and correct abnormal data in the recognition results, ensure the accuracy and completeness of the data, and provide reliable data support for subsequent environmental assessment work.

[0042] When the abnormal matching result does not exist, a mapping relationship between the recognition result and the form is established and exported to the recognition database; When the abnormal matching result exists, the deviation information is obtained, and the deviation information is supplemented by using the attribute tolerance to obtain a corrected recognition result; for the deviation information exceeding the attribute tolerance, a correction recognition constraint condition is generated, and the constraint condition is added to the optical character recognition technology, and a re-recognition module is generated for recognition, and the recognition result that meets the requirements is stored in the recognition database.

[0043] Preferably, the attribute tolerance refers to the extent to which the data is allowed to deviate from the preset standard and still be considered valid, which is a preset value. When the recognition result of the optical character recognition technology completely matches the content format requirements and necessary parameter information of the form attribute annotation, and there is no abnormality, the system terminal successfully establishes a mapping relationship between the recognition result and the form. This means that each recognized data item can accurately correspond to the corresponding position in the form, ensuring the accuracy and integrity of the data. Subsequently, these mapping relationships and their corresponding recognition results will be exported and stored in the recognition database for subsequent data analysis and processing. However, if an abnormal result is found during the matching process, that is, the recognition result does not meet the preset format requirements or parameter information, the system terminal needs to take further measures. First, obtain these deviation information and use the attribute tolerance to supplement or adjust them. By using this tolerance, some minor deviations can be corrected to obtain a corrected recognition result. However, if some deviation information exceeds the range of the attribute tolerance, the system terminal cannot rely on simple modification to make corrections. At this time, correction recognition constraints are generated, which specifically describe which types of deviations are unacceptable and require optical character recognition technology to be re-recognized. These constraints are added to the optical character recognition technology to generate a re-recognition module, which is specifically used to process data items that failed the initial recognition. Through re-recognition, more accurate recognition results that meet the requirements are obtained and stored in the recognition database. In summary, this process is a process of continuously optimizing and improving the recognition results. By establishing a mapping relationship, using attribute tolerance for correction, and generating a re-recognition module, it can be ensured that the data finally stored in the recognition database is accurate, complete, and reliable.

[0044] Furthermore, the present application provides adding the constraint condition to the optical character recognition technology to generate a re-recognition module for recognition, and the method further includes: Determine the image number of the environmental evaluation form based on the deviation information, and traverse from the abnormal image; Determine whether there is an abnormal graph according to the traversal result, and if so, locate the deviation information of the abnormal graph; Optionally, when the system terminal finds deviation information in the result of identifying the environmental evaluation form, the specific form image number is first determined based on the deviation information. This is because the form is usually composed of multiple images, each of which may contain different information areas. By determining the image number, the image with the problem can be more accurately located. Subsequently, traversal is started from the determined abnormal image. The purpose of traversal is to carefully check every part of the image to find out the specific location or area that may cause the recognition deviation. During the traversal process, the system terminal will determine whether there is an abnormal image. Abnormal images refer to images that contain obvious recognition errors or format problems. Once an abnormal image is found, the deviation information will be further located, and the text, lines, symbols and other elements in the image will be accurately analyzed and compared to determine the specific location where the deviation information appears. In summary, this process is to determine the image number, traverse the abnormal image, determine the existence of the abnormal image, and finally locate the location of the deviation information. This helps to identify the problem more accurately and provides an important basis for subsequent data correction and image optimization.

[0045] Adding the located abnormal image to the recognition list, performing recognition through the re-recognition module, and obtaining an auxiliary recognition result; The auxiliary recognition result is fused with the re-recognition result, and the recognition result is corrected using the fusion information that meets the content format requirements and necessary parameter information to obtain the corrected recognition result.

[0046] Optionally, once the system terminal locates the abnormal images in the environmental assessment form, these images will be added to the recognition list. The purpose of this is to perform further recognition processing on these specific images to correct the errors that may exist during the initial recognition. Subsequently, these abnormal images are recognized using the previously generated re-recognition module. This re-recognition module is built based on the deviation information and correction recognition constraints found in the previous recognition process, so it is more focused on dealing with situations that may cause recognition errors. Through the processing of the re-recognition module, auxiliary recognition results for these abnormal images are obtained. After obtaining the auxiliary recognition results, the system terminal compares and matches the auxiliary recognition results with the data items in the previous recognition results one by one. For data items in the same position, if their contents are consistent or very close, one of them is selected as the final result; if there are significant differences, the auxiliary recognition result is selected as the final result. During the fusion process, the fusion results need to be screened according to the preset content format requirements and necessary parameter information. This means that the system terminal will check whether each data item conforms to the expected format and whether it contains all the necessary parameter information. Data items that do not meet the requirements will be marked as abnormal. After information screening, the original recognition results are corrected using the fusion information that meets the requirements. This includes replacing incorrect data items, supplementing missing information, etc. The purpose of correction is to ensure that the final recognition result is both accurate and meets the requirements of the form. In summary, this process is a cycle of recognition, fusion and correction. The auxiliary recognition result is obtained through the re-recognition module, and then fused with the original result. Finally, the recognition result is corrected using the fusion information, thereby obtaining a more accurate and reliable corrected recognition result.

[0047] Furthermore, the present application provides that before storing the recognition results that meet the requirements in the recognition database, the method also includes: When the re-identification result cannot meet the content format requirements and necessary parameter information, the deviation information is marked and an abnormal reminder message is generated.

[0048] Optionally, when the re-identification result cannot meet the content format requirements of the environmental assessment form or the necessary parameter information is missing, it is necessary to identify the deviation information and generate abnormal reminder information. This means that during the identification process, if the system terminal finds that some data items do not meet the preset format specifications or parameter requirements, or some key information is missing, it will specially mark these deviation information. These identifiers help to quickly locate the problem and understand which parts need to be focused on and corrected. At the same time, the system terminal will generate abnormal reminder information based on these identifiers. These reminder information will be presented to the user in an easy-to-understand way, clearly indicating which data items have problems and possible causes or effects. In this way, users can quickly understand the problems in the identification results and take appropriate measures to correct or supplement them. In summary, identifying deviation information and generating abnormal reminder information are important steps to ensure the accuracy and completeness of identification results. It helps to find and solve problems in a timely manner and improve the efficiency and accuracy of data processing.

[0049] Embodiment 2: This embodiment is a further improvement on the basis of the first embodiment. In order to facilitate the search and selection of classified storage of similar forms with the same or similar contents, so as to improve the search efficiency of the sorted forms, the inventor achieves the above technical effects through the method disclosed in this embodiment: The step of extracting key information from the correction recognition result by natural language processing technology according to the structural content of the environmental evaluation form also includes: Extracting key information from the modified recognition result of each form by natural language processing technology, and converting the key information corresponding to each form into a feature vector of the modified form; Select several eigenvectors as initial cluster centers; Calculate the distance from each eigenvector to all cluster centers and assign it to the nearest cluster center; Calculate the mean of all feature vectors in each cluster and update the cluster center position; Perform iteration, the iteration method is as follows: Recalculate the distance from each eigenvector to the updated cluster center; Reassign each eigenvector to the nearest cluster center; Recalculate the mean of all feature vectors in each cluster and update the cluster center; Continue iterating until the positions of the cluster centers remain stable for two consecutive iterations.

[0050] The iteration is continued until the position of the cluster center remains stable in two consecutive iterations, and the determination method is as follows: After selecting the initial cluster centers and performing the first cluster assignment, the sum of squared errors within the initial clusters is calculated; After updating the cluster center, reassign each eigenvector to the nearest cluster center and calculate the new intra-cluster sum of squared errors; Compare the change in the sum of squared errors within the cluster between the current iteration and the previous iteration; If the change in the sum of squared errors within the cluster is less than the preset threshold, or reaches the set maximum number of iterations, the algorithm is considered to have converged and the iteration can be stopped.

[0051] The calculation method of the intra-cluster error sum of squares is as follows: Through the feature vector and cluster center, the sum of squared errors within the cluster is calculated through the objective function to accurately extract the evaluation items in the environmental evaluation form and cluster similar environmental evaluation forms together. The objective function is as follows:

[0052] in, J is the sum of squared errors within the cluster, K is the number of clusters, υ ί is the eigenvector of the ιth form, μκ is the center of the kth cluster, C κ is the set of all eigenvectors in the kth cluster.

[0053] Specifically, this application obtains the text content of the key information of each form and uses natural language processing technology (such as BERT) to convert the text content of each form into a feature vector. Specifically, by extracting the text content of the key information of each form, ί , and then use the BERT model to transform the text T ί Convert to vector representation υ; where υ is the eigenvector of the form; Initial cluster center: Select K eigenvectors as the initial cluster center; Assign data points: that is, obtain the feature vector set and the current cluster center set, calculate the Euclidean distance from each feature vector υ to each cluster center, and assign each feature vector to the nearest cluster center; Update cluster center: calculate the mean of all feature vectors in each cluster and update the cluster center position; Specifically, suppose that after one iteration, we have assigned all data points (forms) to the nearest cluster center. Then we can obtain several clusters, each containing several data points; For each cluster, calculate the mean of all data points in the cluster. This mean will become the new cluster center; Iteration: Repeat the steps of assigning data points and updating cluster centers until the cluster centers remain stable and no longer change significantly (that is, if the position changes of all cluster centers are less than the preset threshold x) or the maximum number of iterations is reached, that is, convergence is achieved.

[0054] That is, after each update of the cluster center, the sum of squared errors within the cluster assigned by the current cluster is calculated by the above formula, and the changes in the sum of squared errors within the cluster between the current iteration and the previous iteration are compared to evaluate the current clustering effect. If the change in the sum of squared errors within the cluster is less than the preset threshold or reaches the set maximum number of iterations, the algorithm is considered to have converged and the iteration can be stopped. This method allows it to iterate continuously to minimize the sum of squared errors and ensure that the data points within each cluster are as close as possible in the feature space, thereby achieving intelligent recognition and management of environmental evaluation forms.

[0055] The specific steps are as follows: Initialize SSE: After selecting the initial cluster centers and performing the first cluster assignment, the sum of squared errors within the initial clusters is calculated; Start iteration: Enter an iterative loop until the algorithm converges or reaches the maximum number of iterations.

[0056] Update cluster center: According to the current cluster assignment, the mean of all feature vectors in each cluster is calculated and the cluster center is updated.

[0057] Reassign data points: Calculate the distance of each eigenvector to all new cluster centers and assign it to the nearest cluster center.

[0058] Compute the new within-cluster sum of squared errors for each cluster generated in each iteration: That is, the square of the Euclidean distance from each eigenvector to the center of its new cluster is calculated, and all eigenvectors are summed to obtain the new intra-cluster sum of squared errors; Compare the changes in the within-cluster sum of squared errors: Determine the difference between the new intra-cluster sum of squared errors and the previous intra-cluster sum of squared errors; Determine convergence: If the difference in the sum of squared errors within the cluster is less than a preset threshold (for example, 0.001), the algorithm is considered to have converged and the iteration is exited.

[0059] Otherwise, continue to the next iteration; End iteration: When the convergence condition is met or the maximum number of iterations is reached, the iteration is stopped and the final cluster allocation result is output; Furthermore, K eigenvectors can be randomly selected as the initial cluster centers; In the initial cluster center selection stage, K points need to be randomly selected from the data set as the initial cluster centers. These initial cluster centers will be continuously updated in subsequent iterations until the convergence condition is reached.

[0060] Preferably, the feature vector K may be selected by the following method: selecting the K value according to the domain knowledge and experience of the user; That is, some representative data points can be selected as the initial cluster centers based on domain knowledge; or when the data distribution is obvious, eigenvectors of different regions can be selected as the initial cluster centers to ensure that the initial cluster centers cover the overall distribution of the data. Compared with randomly assigning eigenvectors, the method of autonomously determining eigenvectors can reduce the number of iterations, so as to achieve the technical effect of improving the convergence speed and the stability of the results.

[0061] In summary, the embodiments of the present application have at least the following technical effects: The embodiment of the present application uses an image acquisition device to collect image information of an environmental evaluation form according to a preset clarity requirement, and constructs an image library. Subsequently, the images in the image library are preprocessed. The image segmentation technology is then used to segment the preprocessed image according to the preset segmentation requirements, and the attributes corresponding to the segmentation results are annotated. After that, the segmentation results of the environmental evaluation form are sequentially recognized using optical character recognition technology, the form content is converted into text information, and the recognition result is formatted according to the attribute annotation to obtain the final recognition result. Then, the recognition result is evaluated and corrected based on the attribute annotation. By comparing the recognition result with the preset format requirements and necessary parameter information, the abnormal matching result is found. For the case where there is no abnormality, a mapping relationship between the recognition result and the form is established and exported to the recognition database. For the case where there is an abnormality, the attribute tolerance is used to supplement the deviation information or generate a correction recognition constraint condition, and the optical character recognition technology is added for re-recognition to obtain a corrected recognition result. Finally, the key information in the corrected recognition result is extracted by natural language processing technology, and stored in a structured form according to the structural content of the environmental evaluation form. Before storage, if the recognition result still fails to meet the format requirements or lacks necessary parameters, the deviation information will be marked and an abnormal reminder will be generated. These technical effects jointly solve the technical problems of insufficient recognition accuracy and efficiency due to the diversity of form content, ensuring that environmental assessment work is carried out efficiently and accurately.

[0062] Embodiment three: Based on the same inventive concept as the method for intelligently identifying an environmental evaluation form in the aforementioned embodiment, Figure 2 As shown, the present application provides an environmental assessment form intelligent recognition system, the system comprising: Information collection module 1: The information collection module 1 is used to collect image information of the environmental evaluation form according to preset clarity requirements using an image collection device, and to construct an environmental evaluation form image library; Preprocessing module 2: the preprocessing module 2 is used to preprocess the images in the environmental assessment form image library, including denoising and edge detection; Image segmentation module 3: The image segmentation module 3 is used to segment the pre-processed environment assessment form image according to preset segmentation requirements using image segmentation technology, and to mark the attributes corresponding to the segmentation results; Segmentation and recognition module 4: the segmentation and recognition module 4 is used to sequentially recognize the segmentation results of the environmental assessment form using optical character recognition technology to obtain recognition results; Evaluation and correction module 5: the evaluation and correction module 5 is used to evaluate and correct the recognition result based on the attribute labeling to obtain a corrected recognition result; Information extraction module 6: the information extraction module 6 is used to extract key information in the modified recognition result according to the structural content of the environmental evaluation form through natural language processing technology, and store it in a structured form based on the structural content.

[0063] Furthermore, the preprocessing module 2 is used to perform the following method: Performing denoising processing on the images in the environmental assessment form image library; Use the denoised images to perform edge detection, extract images that do not meet the edge detection requirements, and build an abnormal image library; Confirming the number of the acquired image based on the abnormal image library generates a re-acquisition instruction, sends it to the image acquisition device, and establishes a mapping relationship between the re-acquisition image and the abnormal image; The edge detection of the re-collected images is performed based on the abnormal image library until the preset requirements are met, and all images that meet the preset requirements are stored in the environmental assessment form image library.

[0064] Furthermore, the image segmentation module 3 is used to perform the following method: According to the recognition task in the form recognition request, the preset segmentation requirements are set, including title, field, and data; Based on the preset segmentation requirement, generating the attribute annotation; The preset segmentation requirement is added to the image segmentation technology to generate a segmentation execution module, the pre-processed environment assessment form image is segmented, and the corresponding segmentation result is annotated based on the attribute annotation.

[0065] Furthermore, the segmentation and recognition module 4 is used to perform the following method: Converting the environmental assessment form into text information; Performing content recognition on the text information according to the segmentation result to obtain a recognition result; Based on the attribute annotation of the environmental evaluation form, the character structure of each segmentation result is determined, and the recognition result is converted according to the character structure to obtain the recognition result.

[0066] Furthermore, the evaluation and correction module 5 is used to perform the following method: Based on the attribute annotation, determine the content format requirements and necessary parameter information of the segmentation result; According to the content format requirements and necessary parameter information, the recognition results are traversed and matched to determine abnormal matching results; When the abnormal matching result does not exist, a mapping relationship between the recognition result and the form is established and exported to the recognition database; When the abnormal matching result exists, the deviation information is obtained, and the deviation information is supplemented by using the attribute tolerance to obtain a corrected recognition result; for the deviation information exceeding the attribute tolerance, a correction recognition constraint condition is generated, and the constraint condition is added to the optical character recognition technology, and a re-recognition module is generated for recognition, and the recognition result that meets the requirements is stored in the recognition database.

[0067] Furthermore, the evaluation and correction module 5 is used to perform the following method: Determine the image number of the environmental evaluation form based on the deviation information, and traverse from the abnormal image; Determine whether there is an abnormal graph according to the traversal result, and if so, locate the deviation information of the abnormal graph; Adding the located abnormal image to the recognition list, performing recognition through the re-recognition module, and obtaining an auxiliary recognition result; The auxiliary recognition result is fused with the re-recognition result, and the recognition result is corrected using the fusion information that meets the content format requirements and necessary parameter information to obtain the corrected recognition result.

[0068] Furthermore, the evaluation and correction module 5 is used to perform the following method: When the re-identification result cannot meet the content format requirements and necessary parameter information, the deviation information is marked and an abnormal reminder message is generated.

[0069] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. Other embodiments are within the scope of the attached claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0070] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0071] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, those skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A method for intelligent recognition of environmental evaluation forms, characterized in that: include: Using image acquisition equipment to collect image information of the environmental assessment form according to preset clarity requirements, and constructing an environmental assessment form image library; Preprocessing the images in the environmental assessment form image library, including denoising and edge detection; Use image segmentation technology to segment the preprocessed environmental assessment form image according to preset segmentation requirements, and annotate according to the attributes corresponding to the segmentation results; The segmentation results of the environmental assessment form are sequentially recognized using optical character recognition technology to obtain recognition results; Based on the attribute annotation, the recognition result is evaluated and corrected to obtain a corrected recognition result; According to the structural content of the environmental evaluation form, key information in the modified identification result is extracted through natural language processing technology, and stored in a structured form based on the structural content.

2. The method according to claim 1, characterized in that Preprocessing the images in the environmental assessment form image library, including denoising and edge detection, includes: Performing denoising processing on the images in the environmental assessment form image library; Use the denoised images to perform edge detection, extract images that do not meet the edge detection requirements, and build an abnormal image library; Confirming the number of the acquired image based on the abnormal image library generates a re-acquisition instruction, sends it to the image acquisition device, and establishes a mapping relationship between the re-acquisition image and the abnormal image; The edge detection of the re-collected images is performed based on the abnormal image library until the preset requirements are met, and all images that meet the preset requirements are stored in the environmental assessment form image library.

3. The method according to claim 1, characterized in that The method of using image segmentation technology to segment the preprocessed environmental assessment form image according to preset segmentation requirements and marking the image according to the attributes corresponding to the segmentation results includes: According to the recognition task in the form recognition request, the preset segmentation requirements are set, including title, field, and data; Based on the preset segmentation requirement, generating the attribute annotation; The preset segmentation requirement is added to the image segmentation technology to generate a segmentation execution module, the pre-processed environment assessment form image is segmented, and the corresponding segmentation result is annotated based on the attribute annotation.

4. The method according to claim 1, characterized in that The step of sequentially recognizing the segmentation results of the environmental assessment form using the optical character recognition technology to obtain the recognition results includes: Converting the environmental assessment form into text information; Performing content recognition on the text information according to the segmentation result to obtain a recognition result; Based on the attribute annotation of the environmental evaluation form, the character structure of each segmentation result is determined, and the recognition result is converted according to the character structure to obtain the recognition result.

5. The method according to claim 1, characterized in that The step of evaluating and correcting the recognition result based on the attribute annotation to obtain a corrected recognition result includes: Based on the attribute annotation, determine the content format requirements and necessary parameter information of the segmentation result; According to the content format requirements and necessary parameter information, the recognition results are traversed and matched to determine abnormal matching results; When the abnormal matching result does not exist, a mapping relationship between the recognition result and the form is established and exported to the recognition database; When the abnormal matching result exists, the deviation information is obtained, and the deviation information is supplemented by using the attribute tolerance to obtain a corrected recognition result; for the deviation information exceeding the attribute tolerance, a correction recognition constraint condition is generated, and the constraint condition is added to the optical character recognition technology, and a re-recognition module is generated for recognition, and the recognition result that meets the requirements is stored in the recognition database.

6. The method according to claim 5, characterized in that Adding the constraint condition to the optical character recognition technology to generate a re-recognition module for recognition includes: Determine the image number of the environmental evaluation form based on the deviation information, and traverse from the abnormal image; Determine whether there is an abnormal graph according to the traversal result, and if so, locate the deviation information of the abnormal graph; Adding the located abnormal image to the recognition list, performing recognition through the re-recognition module, and obtaining an auxiliary recognition result; The auxiliary recognition result is fused with the re-recognition result, and the recognition result is corrected using the fusion information that meets the content format requirements and necessary parameter information to obtain the corrected recognition result.

7. The method according to claim 5, characterized in that Before storing the recognition results that meet the requirements in the recognition database, the method further includes: When the re-identification result cannot meet the content format requirements and necessary parameter information, the deviation information is marked and an abnormal reminder message is generated.

8. The method according to claim 1, characterized in that The step of extracting key information from the correction recognition result according to the structural content of the environmental evaluation form by natural language processing technology also includes: Extracting key information from the modified recognition result of each form by natural language processing technology, and converting the key information corresponding to each form into a feature vector of the modified form; Select several eigenvectors as initial cluster centers; Calculate the distance from each eigenvector to all cluster centers and assign it to the nearest cluster center; Calculate the mean of all feature vectors in each cluster and update the cluster center position; Perform iteration, the iteration method is as follows: Recalculate the distance from each eigenvector to the updated cluster center; Reassign each eigenvector to the nearest cluster center; Recalculate the mean of all feature vectors in each cluster and update the cluster center; Continue iterating until the positions of the cluster centers remain stable for two consecutive iterations.

9. The method according to claim 8, characterized in that The iteration continues until the position of the cluster center remains stable in two consecutive iterations. The determination method is as follows: After selecting the initial cluster centers and performing the first cluster assignment, the sum of squared errors within the initial clusters is calculated; After updating the cluster center, reassign each eigenvector to the nearest cluster center and calculate the new intra-cluster sum of squared errors; Compare the change in the sum of squared errors within the cluster between the current iteration and the previous iteration; If the change in the sum of squared errors within the cluster is less than the preset threshold, or reaches the set maximum number of iterations, the algorithm is considered to have converged and the iteration can be stopped; The calculation method of the intra-cluster error sum of squares is as follows: Through the feature vector and cluster center, the sum of squared errors within the cluster is calculated through the objective function to accurately extract the evaluation items in the environmental evaluation form and cluster similar environmental evaluation forms together. The objective function is as follows: ； in, J is the sum of squared errors within the cluster, K is the number of clusters, υ ί is the eigenvector of the ιth form, μκ is the center of the kth cluster, C κ is the set of all eigenvectors in the kth cluster.

10. An intelligent recognition system for environmental evaluation forms, characterized in that: A method for intelligently identifying an environmental assessment form according to any one of claims 1 to 9, comprising: Information collection module: using image collection equipment to collect image information of environmental evaluation forms according to preset clarity requirements, and building an environmental evaluation form image library; Preprocessing module: preprocessing the images in the environmental assessment form image library, including denoising and edge detection; Image segmentation module: Use image segmentation technology to segment the pre-processed environmental assessment form image according to the preset segmentation requirements, and mark it according to the attributes corresponding to the segmentation results; Segmentation and recognition module: Use optical character recognition technology to sequentially recognize the segmentation results of the environmental assessment form to obtain recognition results; Evaluation and correction module: based on attribute annotation, evaluates and corrects the recognition result to obtain a corrected recognition result; Information extraction module: according to the structure content of the environmental evaluation form, the key information in the correction recognition result is extracted through natural language processing technology, and stored in a structured form based on the structure content It is recommended to explain the meaning of parameter k separately Several eigenvectors are selected as initial cluster centers, and the number of eigenvectors is the number of clusters k, and a detailed explanation has been added in the specification.

Images