Quantitative analysis method for detectability of technical institution

Abstract

According to the quantitative analysis method for the detectability of the technical institution, intelligent recognition of multi-format documents is achieved through self-adaptive layout analysis and an OCR engine, the data cleaning and automatic correction technology is combined, the data collection accuracy is remarkably improved (the manual intervention requirement is reduced), and complex layouts and unstructured data are effectively processed. By adopting intelligent data analysis and structured processing, automatic extraction of key fields is realized based on the analysis capability of a Transform model, and a structured data storage system is established through dynamic modeling, so that the limitation of traditional template matching is solved, and the generalization capability and structured efficiency of data processing are improved. A scientific ability evaluation model is established, outlier rejection and normal distribution modeling are adopted to optimize data quality, objective and quantitative ability evaluation is realized in combination with a Z-ratio score comparison mechanism, subjective deviation is eliminated, and it is ensured that an evaluation result meets statistical specification requirements.

Description

Technical Field

[0001] This invention relates to the field of proficiency testing technology for technical institutions, and more specifically, to a method for quantitative analysis of the proficiency of technical institutions. Background Technology

[0002] Proficiency testing for technical institutions (such as interlaboratory comparisons and measurement audits) is an important means of assessing the technical capabilities and quality levels of technical institutions (such as testing laboratories and calibration laboratories). It is not only a "check-up" of technical capabilities, but also a key measure to ensure quality, maintain reputation, and comply with regulatory requirements. Through objective and comparable methods, it helps institutions continuously improve their technical capabilities, ultimately serving the public interest and industry development.

[0003] In the process of proficiency testing, the proficiency testing of quantitative indicators is usually carried out manually. The testing organization sends samples to the verification organization, the verification organization tests the samples, and then compares the test results with the test results of the testing organization. The implementation process is complicated, and unreasonable verification results often occur.

[0004] Currently, AI technology is booming, and large-scale model-based recognition and analysis technologies are advancing rapidly. Combining OCR recognition with large-scale model technology for quantitative analysis of testing capabilities in technical institutions can greatly improve the work efficiency of proficiency testing management departments, reduce human intervention in proficiency testing results, and make the proficiency testing process more transparent, ensuring fair and impartial results. Summary of the Invention

[0005] This invention addresses the technical problems existing in the prior art by providing a method for quantitative analysis of the testing capabilities of technical institutions, which can realize automatic quantitative analysis of the testing capabilities of technical institutions.

[0006] According to a first aspect of the present invention, a method for quantitative analysis of the detection capability of a technical institution is provided, comprising:

[0007] Obtain proficiency testing reports from each participating technical institution for testing the same type of samples;

[0008] For each technical institution, the test result table area in the proficiency testing report is located based on an adaptive layout analysis algorithm.

[0009] The detection result table area is used to perform character recognition based on a deep learning OCR engine, and the incorrectly recognized characters are corrected to obtain the character recognition result.

[0010] Based on a dedicated parsing model of the Transformer architecture, detection result data is identified from the character recognition results;

[0011] The test results data of each technical institution are normalized, and the robust median and standardized IQR of the normalized test results data are calculated.

[0012] Based on the robust median and the standardized IQR, outliers in the test results data are removed.

[0013] Based on the test results and benchmark values ​​of each technical institution, calculate the Z-score for each technical institution;

[0014] Based on the Z-score, the testing capabilities of each technical institution are evaluated, and the evaluation results of each technical institution's testing capabilities are obtained.

[0015] Based on the above technical solution, the present invention can also be improved as follows.

[0016] Optionally, locating the test result table area in the proficiency testing report based on the adaptive layout analysis algorithm includes:

[0017] The input proficiency testing report document image is denoised and tilt-corrected.

[0018] Based on prior knowledge of the layout, connected components that are not text in the document image are analyzed and extracted. Projection analysis is performed on the connected components, and based on the projection characteristics and the shape of the connected components, it is determined whether the connected components are graphics, tables, or text.

[0019] Optionally, the character recognition of the detection result table area based on the deep learning OCR engine includes:

[0020] In the encoding stage, feature extraction of the input image is performed based on the multi-layer convolutional and pooling layers in the Convolutional Neural Network (CNN). The feature map output by the CNN is converted into feature sequence data based on the Convolutional Recurrent Neural Network (CRNN), which is then input into the Recurrent Neural Network (RNN) for decoding.

[0021] After the recurrent neural network (RNN), add one or more fully connected layers to map the features to the space of character categories;

[0022] In the decoding stage, an attention mechanism is introduced into the RNN to dynamically calculate the context vector at each time step. The feature sequence data output by the CRNN is used as the input of the attention mechanism. The attention mechanism calculates attention weights based on the current state of the decoder and all hidden states of the encoder, performs a weighted summation of the encoder features to obtain the context vector, and combines it with the current state of the decoder to generate the prediction of the next character.

[0023] Optionally, the dedicated parsing model based on the Transformer architecture includes an encoder and a decoder, wherein the encoder includes a visual Transformer model and an image Transformer model, and the decoder includes a language Transformer model;

[0024] The detection result data is identified from the character recognition results, including:

[0025] The character recognition result is input as a text image into the visual Transformer model. The visual Transformer model adjusts the text image to a specific size and divides it into multiple image blocks. Each image block is input into the image Transformer model.

[0026] The image Transformer model generates image embeddings for each image block through multi-head attention blocks and feedforward blocks. Each image embedding is input into the language Transformer model to generate a text sequence corresponding to each image embedding, i.e., key fields. The key fields include detection items, units, and measurement values. The measurement values ​​are used as detection result data.

[0027] Optionally, after identifying the detection result data from the character recognition result, the method further includes:

[0028] The detection results data of each identified technical institution are stored in a structured database;

[0029] And outliers were removed from the test results data of all technical institutions.

[0030] Optionally, the normalization processing of the test results data from each technical institution includes:

[0031] Based on the Box-Cox transform algorithm, the detection results data of each technical institution are normalized. The expression of the Box-Cox transform algorithm is as follows: = ;

[0032] Where y is the original data, λ is the transformation parameter, which is usually estimated by maximizing the log-likelihood function; y(λ) is the transformed data.

[0033] Optionally, the calculation of the robust median and standardized IQR of the normally normalized detection result data includes:

[0034] Calculate the robust median Q2 of the test results data from all technical institutions. Based on the robust median Q2, divide the data into upper and lower quartiles and calculate the lower quartile Q1 and the upper quartile Q3 respectively.

[0035] Based on the lower quartile Q1 and the upper quartile Q3, calculate the standardized IQR: IQR = Q3 - Q1;

[0036] The process of removing outliers from the test results data based on the robust median and the standardized IQR includes:

[0037] Remove data points that are below Q1-k×(Q3-Q1) and above Q3+k×(Q3-Q1) to prune the data, where k is a constant;

[0038] Repeat the data trimming step until convergence or the maximum number of iterations is reached to obtain the trimmed detection results data;

[0039] Data points that are below Q1-1.5×IQR or above Q3+1.5×IQR are considered outliers and removed.

[0040] Optionally, the step of calculating the Z-score for each technical institution based on its test results data and benchmark values ​​includes:

[0041] The difference for each technical institution is: ;

[0042] in, For the first The test results data from various technical institutions As the baseline value, For the first The difference between the technical institutions;

[0043] Calculate the percentage relative difference for each technical institution. The calculation formula is: ;

[0044] Standard deviation of computational ability assessment The calculation formula is: ;

[0045] in, For all valid The mean of n is the number of technical institutions participating in the testing, where n is a positive integer and n≥3;

[0046] The Z-score for each technical institution is expressed as: .

[0047] Optionally, the assessment of the detection capabilities of each technical institution based on the Z-ratio score includes: ;

[0048] in, Represents absolute value. For the evaluation results.

[0049] Optionally, the assessment of the detection capabilities of each technical institution based on the Z-ratio score includes:

[0050] Based on the testing capability assessment results of each technical institution, an evaluation report is automatically generated using the evaluation report template of the ISO / IEC 17043 standard.

[0051] A trend analysis chart is generated based on the testing capability assessment results of each technical institution at different time periods;

[0052] And a heat map is drawn based on the deviation of the test results data of each technical institution from the benchmark value.

[0053] According to a second aspect of the present invention, a quantitative analysis system for the detection capability of technical institutions is provided, comprising:

[0054] The acquisition module is used to acquire proficiency testing reports from each participating technical institution after testing the same type of samples.

[0055] The recognition module is used to locate the detection result table area in the proficiency testing report for each technical institution based on an adaptive layout analysis algorithm; to perform character recognition on the detection result table area based on a deep learning OCR engine, and to correct erroneous characters to obtain character recognition results; and to identify the detection result data from the character recognition results based on a dedicated parsing model of the Transformer architecture.

[0056] The calculation module is used to normalize the test result data of each technical institution and calculate the robust median and standardized IQR of the normalized test result data; and to calculate the Z-score of each technical institution based on the test result data and benchmark value of each technical institution.

[0057] The elimination module is used to eliminate outliers in the test result data based on the robust median and the standardized IQR.

[0058] The evaluation module is used to assess the testing capabilities of each technical institution based on the Z-score, and obtain the testing capability evaluation results for each technical institution.

[0059] According to a third aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the processor is configured to execute a computer management program stored in the memory to implement the steps of a method for quantitative analysis of the detection capabilities of a technical institution.

[0060] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer management program is stored, wherein the computer management program, when executed by a processor, implements the steps of a method for quantitative analysis of the detection capabilities of a technical institution.

[0061] This invention provides a quantitative analysis method for the testing capabilities of technical institutions. It achieves intelligent recognition of multi-format documents through adaptive layout analysis and an OCR engine, combined with automatic data cleaning and correction technology, significantly improving data collection accuracy (reducing the need for manual intervention) and effectively handling complex layouts and unstructured data. It employs intelligent data analysis and structured processing, using the parsing capabilities of the Transformer model to automatically extract key fields. A structured data storage system is established through dynamic modeling, overcoming the limitations of traditional template matching and improving the generalization ability and structuring efficiency of data processing. A scientific capability assessment model is established, employing outlier removal and normal distribution modeling to optimize data quality. Combined with a Z-ratio score comparison mechanism, it achieves objective and quantitative capability assessment, eliminating subjective bias and ensuring that the assessment results meet statistical standards. Attached Figure Description

[0062] Figure 1 A flowchart of a method for quantitative analysis of the detection capability of a technical institution is provided in one embodiment of the present invention;

[0063] Figure 2 An overall architecture diagram of a method for quantitative analysis of the detection capability of a technical institution provided in one embodiment of the present invention;

[0064] Figure 3 This is a schematic diagram of the network structure of an OCR engine;

[0065] Figure 4 This is a schematic diagram of the structure of a quantitative analysis system for the detection capability of a technical institution, provided in one embodiment of the present invention;

[0066] Figure 5 A schematic diagram of the hardware structure of a possible electronic device provided by the present invention;

[0067] Figure 6 This is a schematic diagram of the hardware structure of a possible computer-readable storage medium provided by the present invention. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined with each other to form feasible technical solutions. Such combinations are not constrained by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0069] Figure 1 A flowchart of a method for quantitative analysis of the detection capability of a technical institution according to an embodiment of the present invention is shown, as follows: Figure 1 and Figure 2 As shown, the method includes the following steps:

[0070] Step 1: Obtain the proficiency testing reports from each participating technical institution for the same type of samples.

[0071] Understandably, verification bodies may send the same type of samples to multiple testing organizations simultaneously. After receiving the samples, these organizations will conduct tests on the same type of samples and obtain proficiency testing reports. These proficiency testing reports can be presented in various formats, such as PDFs, images, or scanned copies.

[0072] Step 2: For each technical institution, locate the test result table area in the proficiency testing report based on the adaptive layout analysis algorithm.

[0073] Understandably, for each proficiency testing report from a technical institution, an adaptive layout analysis algorithm is used to locate the table area of ​​the detection results. Before layout analysis, the input proficiency testing report document image is first denoised and tilted to ensure document image quality. Using complexity analysis algorithms, connected components in the document image are searched. Combined with prior layout knowledge, connected components that are not identical to text are analyzed and extracted. Then, projection analysis is performed on these connected components. Based on the projection features and shape of the connected components, it is determined whether the part is a graphic, a table, or text. The projection features refer to whether the text within the connected components is continuous or discontinuous; for example, text in a table is usually arranged in rows.

[0074] By searching for connected components in a document image, the detection result table area in the recognition capability verification test report is located from the document image.

[0075] Step 3: Based on the deep learning OCR engine, perform character recognition on the detection result table area, correct the incorrectly recognized characters, and obtain the character recognition result.

[0076] Understandably, for the detection result table region identified from the document image in step 2, character recognition is performed on the detection result table region. See [link / reference] Figure 3 This invention employs a deep learning OCR engine (CRNN + Attention mechanism) to achieve character recognition. In the encoding stage, features are extracted from the input image using multiple convolutional and pooling layers in a Convolutional Neural Network (CNN). A Recurrent Convolutional Neural Network (CRNN) converts the feature map output by the CNN into a feature sequence, which is then input into a Recurrent Neural Network (RNN) for decoding. After the RNN, one or more fully connected layers are added to map the features to a character category space. In the decoding stage, an Attention mechanism is introduced into the RNN to dynamically calculate the context vector at each time step. The feature sequence data output by the CRNN is used as input to the Attention mechanism. The Attention mechanism calculates attention weights based on the current state of the decoder and all hidden states of the encoder, performs a weighted summation of the encoder features to obtain the context vector, and combines it with the current state of the decoder to generate a prediction for the next character. A data cleaning module automatically corrects erroneous data.

[0077] Step 4: Based on the dedicated parsing model of the Transformer architecture, identify the detection result data from the character recognition results.

[0078] Understandably, for the recognized character results, this step identifies key fields from the character recognition results. Specifically, based on the Transformer architecture's dedicated parsing model, it automatically identifies key fields such as detection items, units, and measurement values, among which the identified detection values ​​are the detection result data.

[0079] The Transformer architecture-specific parsing model includes an encoder and a decoder. The encoder includes a visual Transformer model and an image Transformer model, and the decoder includes a language Transformer model.

[0080] The detection result data is identified from the character recognition results, including:

[0081] The character recognition result is input as a text image into the visual Transformer model. The visual Transformer model adjusts the text image to a specific size and divides it into multiple image blocks. Each image block is input into the image Transformer model.

[0082] The image Transformer model generates image embeddings for each image block through multi-head attention blocks and feedforward blocks. Each image embedding is input into the language Transformer model to generate a text sequence corresponding to each image embedding, i.e., key fields. The key fields include detection items, units, and measurement values. The measurement values ​​are used as detection result data.

[0083] The structured data (including detection result data) of each technical institution identified by the Transformer architecture's dedicated parsing model is stored in a structured database.

[0084] Outliers and outliers were removed from the test results data of each identified technical institution. Data results with obvious errors, such as incorrect units, decimal point errors, calculation errors, or misreporting as other proficiency testing items, should be removed from the dataset and processed separately. These results will not be included in outlier testing or robust statistical analysis.

[0085] Step 5: Normalize the test results data of each technical institution, and calculate the robust median and standardized IQR of the normalized test results data.

[0086] Step 6: Based on the robust median and the standardized IQR, remove outliers from the test result data.

[0087] Understandably, for the detection results data after removing outliers, normalization processing is performed. In this embodiment of the invention, the Box-Cox transform is used to normalize the experimental results data of each technical institution. The Box-Cox transform is a commonly used data transformation method to convert non-normally distributed data into an approximately normal distribution. It improves the symmetry and normality of the data by introducing a parameter λ to perform a power transformation on the data. The formula for the Box-Cox transform is: = ;

[0088] Where y is the original data, λ is the transformation parameter, which is usually estimated by maximizing the log-likelihood function; y(λ) is the data after normalization transformation.

[0089] The Box-Cox transformation requires all data to be positive. If the data contains negative numbers or zero, it needs to be shifted (e.g., by adding a constant) to make it positive.

[0090] Choice of λ value: The value of λ is usually estimated by maximizing the log-likelihood function. If λ is close to 0, the Box-Cox transform is close to the logarithmic transform.

[0091] Normality verification: The data after normalization may still not perfectly conform to a normal distribution. Therefore, it is necessary to verify the normality through histograms, QQ plots or statistical tests. If the verification fails, the test result data should be removed.

[0092] Calculate the robust median and standardized IQR for the test results data from all technical institutions. Robust Median Calculation: Calculate the median Q2 of the data. Prune extreme values ​​by dividing the data into upper and lower quartiles based on the median, calculating the lower quartile Q1 and upper quartile Q3 respectively. Then remove data points below Q1-k×(Q3-Q1) and above Q3+k×(Q3-Q1) (where k is a constant, typically 1.5). Repeat the above steps for the pruned data until convergence or the maximum number of iterations is reached.

[0093] Based on the data pruned using a robust median, the lower quartile Q1 and upper quartile Q3 are recalculated, and the standardized IQR is calculated: IQR = Q3 - Q1. Data points below Q1 - 1.5 × IQR or above Q3 + 1.5 × IQR are considered outliers. The IQR is standardized to a form with a mean of 0 and a standard deviation of 1 for comparisons across different datasets.

[0094] Step 7: Calculate the Z-score for each technical institution based on the test results data and benchmark values ​​of each technical institution.

[0095] The difference between each technical institution and its benchmark value is calculated based on the test results data and benchmark values ​​of each institution: ;

[0096] in, For the first The test results data from various technical institutions As the baseline value, For the first The difference between the technical institutions;

[0097] Calculate the percentage relative difference for each technical institution. ;

[0098] Standard deviation of computational ability assessment The calculation formula is: ;

[0099] in, For all valid The mean of n is the number of technical institutions participating in the testing, where n is a positive integer and n≥3.

[0100] The Z-score for each technical institution is expressed as: .

[0101] Furthermore, based on the test results and benchmark values ​​of each technical institution, the deviation degree of each institution can be calculated to generate a deviation heatmap. In this embodiment of the invention, seaborn or matplotlib is used to draw the heatmap.

[0102] Step 8: Evaluate the testing capabilities of each technical institution based on the Z-score.

[0103] Understandably, the testing capability of each technical institution is assessed based on its Z-score. Specifically, |Z|≤2 indicates satisfactory, 2<|Z|<3 indicates questionable, and |Z|≥3 indicates unsatisfactory.

[0104] ;

[0105] After assessing the testing capabilities of each technical institution, the assessment results of each institution's testing capabilities are obtained.

[0106] By leveraging the document polishing capabilities of large-scale models and combining the results of proficiency testing quantitative analysis, an evaluation report is automatically generated based on the evaluation report template of the ISO / IEC 17043 standard.

[0107] Preferably, the graphical generation capability of the large model is utilized to generate a multi-dimensional trend analysis based on the current detection capability assessment results and the historical detection capability assessment results.

[0108] See Figure 4 The present invention provides a quantitative analysis system for the detection capability of technical institutions according to an embodiment of the present invention, the system comprising:

[0109] Module 401 is used to acquire proficiency testing reports from each participating technical institution after testing the same type of samples.

[0110] The recognition module 402 is used to locate the detection result table area in the proficiency testing report for each technical institution based on an adaptive layout analysis algorithm; perform character recognition on the detection result table area based on a deep learning OCR engine, and correct erroneous characters to obtain character recognition results; and identify the detection result data from the character recognition results based on a dedicated parsing model of the Transformer architecture.

[0111] The calculation module 403 is used to normalize the test result data of each technical institution and calculate the robust median and standardized IQR of the normalized test result data; and to calculate the Z-score of each technical institution based on the test result data and benchmark value of each technical institution.

[0112] The elimination module 404 is used to eliminate outliers in the test result data based on the robust median and the standardized IQR.

[0113] The evaluation module 405 is used to evaluate the testing capabilities of each technical institution based on the Z-ratio score, and obtain the testing capability evaluation results for each technical institution.

[0114] It is understood that the quantitative analysis system for the testing capability of technical institutions provided by the present invention corresponds to the quantitative analysis method for the testing capability of technical institutions provided in the foregoing embodiments. The relevant technical features of the quantitative analysis system for the testing capability of technical institutions can be referred to the relevant technical features of the quantitative analysis method for the testing capability of technical institutions, and will not be repeated here.

[0115] Please see Figure 5 , Figure 5 This is a schematic diagram illustrating an embodiment of the electronic device provided in this invention. For example... Figure 5 As shown, an embodiment of the present invention provides an electronic device, including a memory 510, a processor 520, and a computer program 511 stored in the memory 510 and executable on the processor 520. When the processor 520 executes the computer program 511, it implements the steps of a method for quantitative analysis of the detection capabilities of technical institutions.

[0116] Please see Figure 6 , Figure 6 This is a schematic diagram illustrating an embodiment of a computer-readable storage medium provided by the present invention. (See diagram below.) Figure 6 As shown, this embodiment provides a computer-readable storage medium 600, on which a computer program 611 is stored. When the computer program 611 is executed by a processor, it implements the steps of a method for quantitative analysis of the detection capabilities of technical institutions.

[0117] This invention provides a method for quantitative analysis of the testing capabilities of technical institutions. By constructing an intelligent and automated technical system, it significantly improves the efficiency and accuracy of data processing, achieving standardization and scientification of the entire proficiency testing process. Adaptive layout analysis and an OCR engine enable rapid recognition of complex documents. Combined with automatic data cleaning and correction technology, it effectively solves the problem of collecting unstructured data (such as images and scanned documents), significantly reducing the need for manual intervention and ensuring data quality. The analytical capabilities based on the Transformer model overcome the limitations of traditional template matching, automatically identifying dynamic keyword segments and constructing a structured database, adapting to the flexible processing needs of multi-scenario and multi-format data, and improving the generalization ability of data analysis. Outlier removal and normal distribution modeling optimize data distribution, combined with a Z-score comparison mechanism to achieve objective quantitative evaluation, eliminate subjective bias, ensure that the proficiency assessment results conform to statistical standards, and provide a reliable basis for inter-laboratory comparisons. This invention integrates computer vision, deep learning, and statistical modeling technologies to achieve unmanned processing throughout the entire chain from raw data collection to standardized report generation. It improves proficiency testing efficiency by more than 80%, controls the error rate to ≤1.5%, and empowers technology institutions to continuously optimize through trend analysis. It is innovative, practical, and compliant.

[0118] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0119] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0120] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0121] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0122] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0123] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0124] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for quantitatively analyzing the detection ability of a technical organization, characterized by, The method comprises the following steps: obtaining a capability verification test report of each participating technical institution after detecting the same type of sample; for each technical institution, locating a detection result table area in the capability verification test report based on an adaptive layout analysis algorithm; performing character recognition on the detection result table area based on a deep learning OCR engine, and correcting the recognized characters to obtain character recognition results; identifying detection result data from the character recognition results based on a special parsing model of a Transformer architecture; performing normalization processing on the detection result data of each technical institution, and calculating the robust median and standardized IQR of the normalized detection result data; based on the robust median and the standardized IQR, removing outliers in the detection result data; calculating the Z-score of each technical institution according to the detection result data and the reference value of each technical institution; based on the Z-score, evaluating the detection capability of each technical institution to obtain the detection capability evaluation result of each technical institution.

2. The method of claim 1, wherein the method comprises: The method comprises the following steps: performing denoising processing and tilt correction on the input capability verification test report document image; combining layout prior knowledge, analyzing and extracting the connected domain in the document image that is abnormal in the text part, performing projection analysis on the connected domain, and analyzing the connected domain as a graph, table or text according to the projection features and the shape of the connected domain.

3. The method of claim 1, wherein the method is used to quantify the ability of a technology institution to detect. The method comprises the following steps: in the encoding stage, performing feature extraction on the input image based on the multi-layer convolutional layer and the pooling layer in the convolutional neural network CNN, converting the feature map output by the CNN into feature sequence data based on the convolutional recurrent neural network CRNN, inputting the feature sequence data into the recurrent neural network RNN, and performing decoding based on the recurrent neural network RNN; after the recurrent neural network RNN, one or more fully connected layers are added to map the features to the space of character categories; wherein, in the decoding stage, an Attention mechanism is introduced into the recurrent neural network RNN to dynamically calculate the context vector of each time step, the feature sequence data output by the CRNN is input into the Attention mechanism, the Attention mechanism calculates the attention weight according to the current state of the decoder and all hidden states of the encoder, weights and sums the features of the encoder to obtain the context vector, and combines the context vector with the current state of the decoder to generate the prediction of the next character.

4. The method of claim 1, wherein the method is used to quantify the ability of a technology institution to detect. The special parsing model of the Transformer architecture comprises an encoder and a decoder, the encoder comprises a visual Transformer model and an image Transformer model, and the decoder comprises a language Transformer model; the detection result data is identified from the character recognition results by the following steps: The character recognition result is input into the visual Transformer model as a text image, the visual Transformer model adjusts the text image to a specific size and divides it into multiple image blocks, and each image block is input into the image Transformer model; The image Transformer model generates image embeddings of each image block through multi-head attention blocks and feedforward blocks, and each image embedding is input into the language Transformer model to generate a corresponding text sequence, i.e., a key field, which includes a detection item, a unit, and a measurement value, and the measurement value is taken as detection result data.

5. The method of claim 1, wherein the method is used to quantify the ability of a technology institution to detect. After identifying the detection result data from the character recognition result, the method further includes: storing the detection result data of each technical institution into a structured database; and removing outliers in the detection result data of all technical institutions.

6. The method of claim 1, wherein the method further comprises: The normalization processing of the detection result data of each technical institution includes: based on the Box-Cox transformation algorithm, the detection result data of each technical institution is normalized, wherein the expression of the Box-Cox transformation algorithm is: = ； where y is the original data, λ is the transformation parameter, and is usually estimated by maximizing the log-likelihood function; y(λ) is the transformed data.

7. The method of claim 1, wherein the method further comprises: The calculation of the robust median and the standardized IQR of the normalized detection result data includes: calculating the robust median Q2 of the detection result data of all technical institutions, dividing the data into two parts according to the robust median Q2, and calculating the lower quartile Q1 and the upper quartile Q3 respectively; based on the lower quartile Q1 and the upper quartile Q3, the standardized IQR is calculated: IQR=Q3-Q1; Based on the robust median and the standardized IQR, the method further includes: removing data points below Q1-k×(Q3-Q1) and above Q3+k×(Q3-Q1), and pruning the data, where k is a constant; repeating the data pruning step until convergence or reaching the maximum number of iterations to obtain pruned detection result data; and regarding data points below Q1-1.5×IQR or above Q3+1.5×IQR as outliers, and removing the outliers.

8. The method of claim 1, wherein the method further comprises: The calculation of the Z-score of each technical institution according to the detection result data and the reference value of each technical institution includes: the difference of each technical institution is: ； wherein, is the detection result data of the technical institution, is the reference value, is the difference value of the technical institution; calculating the percent relative difference for each technical agency ; The standard deviation of the assessment criteria of the computing power The formula is: ； wherein, is the mean value of all valid values, n is the number of technical institutions participating in the detection, n is a positive integer, and n≥3; the Z-score of each technical institution is expressed as: 。 9. The method of claim 1, wherein the method is used to quantify the ability of a technology institution to detect. The evaluation of the detection capability of each technical institution based on the Z-score includes: ； wherein denotes the absolute value, is the evaluation result.

10. The method of claim 1, wherein the method is used to quantify the ability of a technology institution to detect. After evaluating the detection capability of each technical institution based on the Z-score, the method further includes: generating an evaluation report based on the evaluation report template of the ISO / IEC 17043 standard according to the detection capability evaluation results of each technical institution; generating a trend analysis graph according to the detection capability evaluation results of each technical institution in different time periods; and a heat map is drawn based on the deviation of the detection result data of each technical organization from the reference value.

Images