Intelligent recognition and automatic input method and system for report data based on OCR

Abstract

This invention belongs to the field of image recognition and automated processing technology, and provides a method and system for intelligent recognition and automatic input of report data based on OCR. The method includes obtaining the average gradient magnitude of the report image to be processed and generating an image sharpness anomaly index; predicting the recognition accuracy benefit score at different accuracy levels based on the relationship curve between historical OCR accuracy levels and recognition accuracy, according to the sharpness anomaly index; simultaneously generating resource slack based on the estimated number of tasks and processing time in the OCR processing queue; finally, establishing an optimization model by combining the accuracy benefit score and resource slack, and dynamically selecting the optimal OCR accuracy level for recognition and input. This invention achieves a balance between recognition quality and processing efficiency, and has the advantages of dynamically optimizing OCR recognition accuracy and processing efficiency, improving resource utilization and recognition accuracy.

Description

Technical Field

[0001] This invention belongs to the field of image recognition and automated processing technology, and particularly relates to a method and system for intelligent recognition and automatic input of report data based on OCR. Background Technology

[0002] In the fields of office automation and data processing, optical character recognition (OCR) technology is widely used for the digitization and automatic entry of paper documents such as reports and invoices. With the increasing complexity of business scenarios, the requirements for recognition accuracy and processing speed vary significantly across different application environments. While high-precision recognition modes can improve recognition quality, they also significantly increase computational resource consumption and processing time latency. Especially in batch processing scenarios, this resource overhead can lead to a decrease in system response efficiency. Achieving a dynamic adaptive balance between recognition quality and processing efficiency has become a core challenge in current technological development.

[0003] Existing OCR systems generally employ preset fixed accuracy levels or rely on manual intervention to select recognition modes in practical applications. Some systems attempt to integrate image sharpness detection functions, but when blurry images are detected, they only perform simple image enhancement operations or directly interrupt the recognition process. Regarding resource scheduling mechanisms, most systems use a basic first-in-first-out queue management approach, failing to organically link the real-time load status of system resources with recognition quality requirements. This approach has significant limitations: for high-resolution report images, the system still forces the use of a high-precision mode, resulting in redundant consumption of computing resources; while for images with insufficient sharpness, the system lacks dynamic adjustment capabilities, leading to recognition accuracy that fails to meet actual needs. Furthermore, resource scheduling strategies and recognition quality optimization are independent. When the system is under high load, it cannot automatically reduce accuracy based on queue pressure to ensure timely response; conversely, during idle periods, it fails to proactively increase accuracy to fully utilize idle resources, ultimately affecting overall processing performance and user experience. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for intelligent recognition and automatic entry of report data based on OCR, thus solving the aforementioned problems.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligent recognition and automatic entry of report data based on OCR, which specifically includes:

[0006] Obtain the average gradient magnitude of the report image to be processed, and generate the sharpness anomaly index of the report image to be processed based on the average gradient magnitude of the report image to be processed.

[0007] Obtain the accuracy and corresponding image sharpness anomaly index of the report images recognized by all historical OCR accuracy levels. Based on the accuracy and corresponding image sharpness anomaly index of the report images recognized by the OCR accuracy level, generate image sharpness-recognition accuracy relationship curves under different OCR accuracy levels.

[0008] Based on the image sharpness-recognition accuracy relationship curves under different OCR accuracy levels, a recognition accuracy prediction model is established according to the sharpness anomaly index of the report image to be processed, and a recognition accuracy benefit score for the report image to be processed is generated.

[0009] Obtain the number of tasks waiting in the OCR processing queue and the estimated processing time for each task. Based on these parameters, generate OCR resource slack.

[0010] Based on the image recognition accuracy and efficiency score of the report to be processed and the slackness of OCR resources, an OCR accuracy level optimization model is established to generate the optimal OCR accuracy level.

[0011] Based on the optimal OCR accuracy level, the report data of the report image to be processed is intelligently recognized and automatically entered.

[0012] Based on the above technical solutions, the present invention also provides the following optional technical solutions:

[0013] A further technical solution: The method for obtaining the average gradient magnitude of the report image to be processed is as follows: convert the report image to be processed into a grayscale image, then use the Sobel operator to calculate the horizontal and vertical gradients of the image to obtain the gradient magnitude of each pixel, and calculate the average gradient magnitude of the entire image based on the gradient magnitude of each pixel as the average gradient magnitude of the report image to be processed.

[0014] Further technical solution: The method for generating the image sharpness anomaly index of the report to be processed specifically includes:

[0015] Through the formula: ;

[0016] Generate an image sharpness anomaly index for the report to be processed. ;

[0017] In the formula, This represents the average gradient magnitude of the report image to be processed. This represents the average gradient magnitude reference value for a normal resolution image.

[0018] Further technical solution: The method for generating the image recognition accuracy and efficiency score of the report to be processed specifically includes:

[0019] Based on the image sharpness-recognition accuracy relationship curves under different OCR accuracy levels, the estimated recognition accuracy of the report image to be processed is obtained according to the sharpness anomaly index of the report image to be processed.

[0020] Based on the estimated accuracy of the image recognition of the report to be processed, a recognition accuracy prediction model is established, and a benefit score for the recognition accuracy of the report to be processed is generated.

[0021] Further technical solution: The expression of the recognition accuracy prediction model is specifically as follows: ;

[0022] In the expression, This represents the benefit score of the report image recognition accuracy when using OCR accuracy level i. This represents the estimated accuracy of the report image recognition when using OCR accuracy level i. This represents the threshold for the accuracy of image recognition in the report.

[0023] Further technical solution: The method for generating the OCR resource slack specifically includes:

[0024] Based on the number of tasks waiting in the OCR processing queue and the estimated processing time for each task, an estimated queuing time for the OCR processing queue is generated. The estimated queuing time refers to the estimated time required to complete all tasks in the OCR processing queue.

[0025] Through the formula: ;

[0026] Generate OCR resource slack ;

[0027] In the formula, This refers to the OCR resource slackness at OCR accuracy level i. This represents the estimated queuing time for OCR processing at OCR accuracy level i. This represents the threshold for the queuing time in the OCR processing queue.

[0028] Further technical solutions: The specific method for generating the optimal OCR accuracy level includes:

[0029] Based on the image recognition accuracy and efficiency score of the report to be processed and the slackness of OCR resources, a comprehensive score for OCR accuracy level is generated.

[0030] An OCR accuracy level optimization model is established based on the comprehensive score of OCR accuracy level to generate the optimal OCR accuracy level.

[0031] Further technical solutions: The specific method for generating the comprehensive score of OCR accuracy level includes:

[0032] Through the formula: ;

[0033] Generate a comprehensive score for OCR accuracy level. ;

[0034] In the formula, This represents the benefit score of the report image recognition accuracy when using OCR accuracy level i. This represents the OCR resource slackness at OCR accuracy level i. , All are weighting coefficients, and .

[0035] Further technical solution: The expression of the OCR accuracy level optimization model is as follows: ;

[0036] In the expression, This indicates the optimal OCR accuracy level. This represents the overall score of OCR accuracy level when the image to be processed is at OCR accuracy level i, and G represents the set of OCR accuracy levels.

[0037] An OCR-based intelligent report data recognition and automatic data entry system is used to execute the aforementioned OCR-based intelligent report data recognition and automatic data entry method, specifically including:

[0038] The sharpness analysis unit is used to obtain the average gradient magnitude of the report image to be processed, and to generate a sharpness anomaly index of the report image to be processed based on the average gradient magnitude of the report image to be processed.

[0039] The historical analysis unit is used to obtain the accuracy of the report images recognized by all historical OCR accuracy levels and the corresponding report image sharpness anomaly index. Based on the accuracy of the report images recognized by the OCR accuracy level and the corresponding report image sharpness anomaly index, it generates image sharpness-recognition accuracy relationship curves under different OCR accuracy levels.

[0040] The recognition accuracy analysis unit is used to establish a recognition accuracy prediction model based on the image sharpness-recognition accuracy relationship curve under different OCR accuracy levels, and generate a recognition accuracy benefit score for the image of the report to be processed based on the sharpness anomaly index of the report to be processed.

[0041] The resource slack analysis unit is used to obtain the number of tasks waiting in the OCR processing queue and the estimated time required to process the waiting tasks. Based on the number of tasks waiting in the OCR processing queue and the estimated time required to process the waiting tasks, the OCR resource slack is generated.

[0042] The optimal accuracy analysis unit is used to establish an OCR accuracy level optimization model based on the image recognition accuracy benefit score of the report to be processed and the OCR resource slackness, and generate the optimal OCR accuracy level.

[0043] The processing unit is used to intelligently recognize and automatically input report data from the report image to be processed based on the optimal OCR accuracy level.

[0044] This invention provides a method and system for intelligent recognition and automatic entry of report data based on OCR, which has the following advantages compared with the prior art:

[0045] This invention achieves a balance between recognition quality and processing efficiency by generating a sharpness anomaly index, a relationship curve, an accuracy benefit score, and a resource slack, and dynamically selecting the optimal OCR accuracy level. It has the advantages of dynamically optimizing OCR recognition accuracy and processing efficiency, improving resource utilization and recognition accuracy. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the OCR-based intelligent recognition and automatic data entry method for reports provided by the present invention.

[0047] Figure 2 This is a flowchart illustrating step S40 of the present invention.

[0048] Figure 3 This is a flowchart illustrating step S50 of the present invention.

[0049] Figure 4 This is a schematic diagram of the structure of the OCR-based intelligent recognition and automatic entry system for report data provided by the present invention.

[0050] Figure 5 This is a schematic diagram of the resource slackness analysis unit provided by the present invention.

[0051] Figure 6 A schematic diagram of the optimal accuracy analysis unit provided by the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0053] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0054] Please see Figure 1 This invention provides an embodiment of an OCR-based intelligent recognition and automatic data entry method for report data, which specifically includes the following steps:

[0055] Step S10: Obtain the average gradient magnitude of the report image to be processed, and generate the sharpness anomaly index of the report image to be processed based on the average gradient magnitude of the report image to be processed.

[0056] Step S20: Obtain the accuracy of the report images recognized by all historical OCR accuracy levels and the corresponding report image sharpness anomaly index. Based on the accuracy of the report images recognized by the OCR accuracy level and the corresponding report image sharpness anomaly index, generate the image sharpness-recognition accuracy relationship curve under different OCR accuracy levels.

[0057] Step S30: Based on the image sharpness-recognition accuracy relationship curve under different OCR accuracy levels, establish a recognition accuracy prediction model according to the image sharpness anomaly index of the report to be processed, and generate a recognition accuracy benefit score for the report to be processed.

[0058] Step S40: Obtain the number of tasks waiting in the OCR processing queue and the estimated processing time of the waiting tasks. Based on the number of tasks waiting in the OCR processing queue and the estimated processing time of the waiting tasks, generate OCR resource slack.

[0059] Step S50: Based on the image recognition accuracy benefit score and OCR resource slackness of the report to be processed, establish an OCR accuracy level optimization model to generate the optimal OCR accuracy level;

[0060] Step S60: Based on the optimal OCR accuracy level, intelligently recognize and automatically input the report data of the report image to be processed;

[0061] Among them, OCR technology, or Optical Character Recognition, is a technology that converts text in an image into editable text. This technology has wide applications in the digitization of paper documents such as reports and invoices, aiming to achieve automatic data recognition and entry.

[0062] The average gradient magnitude of a report image is an indicator of image sharpness. By calculating the average intensity of grayscale changes in pixels within an image, it can reflect the richness of detail and edge sharpness of the image. The larger the average gradient magnitude, the sharper the image is generally.

[0063] OCR accuracy level refers to the different recognition modes or algorithm configurations used by the OCR system during the recognition process; different accuracy levels usually correspond to different computational resource consumption and recognition accuracy levels. For example, high accuracy level may involve more complex image preprocessing, more refined character segmentation and more powerful recognition models.

[0064] The recognition accuracy benefit score measures the improvement in accuracy or benefit that can be obtained by recognizing the report image under a specific OCR accuracy level; the score takes into account the image's own sharpness characteristics and the recognition performance at different accuracy levels.

[0065] OCR resource slack is used to assess the current resource strain of the OCR system. This metric reflects whether the system has enough slack to handle recognition tasks with higher computational overhead when processing the current task queue, or whether it needs to reduce accuracy to speed up processing.

[0066] Specifically, in step S10, the average gradient magnitude of the report image to be processed is obtained, and a sharpness anomaly index for the report image to be processed is generated based on the average gradient magnitude. The average gradient magnitude can be obtained in various ways, for example, it can be calculated using the sharpness assessment module built into the image processing software, or the sharpness level of the image can be assessed manually and converted into a numerical value. When generating the sharpness anomaly index, a fixed sharpness threshold can be set, and the average gradient magnitude is compared with the threshold. If it exceeds the threshold, it is marked as an anomaly.

[0067] In step S20, the accuracy rate and corresponding image sharpness anomaly index of the report images recognized by all historical OCR accuracy levels are obtained, and image sharpness-recognition accuracy relationship curves are generated based on these data. Historical data can be extracted from system logs or accumulated by periodically performing recognition tests on images of different sharpness and accuracy levels. The relationship curve can be generated by data fitting, for example, by plotting a scatter plot and manually connecting the points to form a curve, or by constructing it using a simple linear interpolation method.

[0068] In step S30, based on the image sharpness-recognition accuracy relationship curves under different OCR accuracy levels, a recognition accuracy prediction model is established according to the sharpness anomaly index of the report image to be processed, and a benefit score for the recognition accuracy of the report image to be processed is generated. The recognition accuracy prediction model can be a simple lookup table mechanism, which finds the corresponding accuracy value on the curve based on the sharpness anomaly index. The benefit score can be generated by comparing the predicted accuracy with a benchmark accuracy. For example, the predicted accuracy can be directly used as the benefit score, or a fixed minimum accuracy requirement can be set, with only the accuracy exceeding the requirement being included in the benefit score.

[0069] In step S40, the number of tasks waiting in the OCR processing queue and the estimated processing time for each task are obtained, and OCR resource slack is generated based on this information. The number of waiting tasks and the estimated processing time can be provided by the system scheduling module, for example, by counting the number of task entries in the current queue and the average processing time of each task. The resource slack can be generated as a simple reciprocal relationship, for example, the more tasks, the lower the slack; or by manually setting a resource stress level and mapping it to the corresponding level based on the number of tasks and the estimated processing time.

[0070] In step S50, an OCR accuracy level optimization model is established based on the image recognition accuracy benefit score and OCR resource slackness of the report to be processed, and the optimal OCR accuracy level is generated. The optimization model can be a simple decision tree, which presets different accuracy levels based on different combinations of benefit score and slackness. For example, high accuracy is selected when the benefit score is high and the slackness is high, and low accuracy is selected when the benefit score is low and the slackness is low.

[0071] In step S60, the report data of the report image to be processed is intelligently recognized and automatically entered according to the optimal OCR accuracy level. The recognition and entry process can be directly executed by the OCR engine. For example, the image data and the selected accuracy level parameter can be passed to the OCR engine, which will then complete the recognition and output the result to the specified data entry interface.

[0072] This invention first obtains the average gradient magnitude of the report image to be processed, generating an image sharpness anomaly index. Based on the relationship curve between historical OCR accuracy levels and recognition accuracy, it predicts the recognition accuracy benefit score for different accuracy levels according to the sharpness anomaly index. Simultaneously, it generates resource slack based on the estimated number of tasks and processing time in the OCR processing queue. Finally, it establishes an optimization model by combining the accuracy benefit score and resource slack, dynamically selecting the optimal OCR accuracy level for recognition and input, thus achieving a dynamic balance between recognition quality and processing efficiency, and improving the overall performance and resource utilization of the OCR system.

[0073] Furthermore, the method for obtaining the average gradient magnitude of the report image to be processed is as follows: convert the report image to be processed into a grayscale image, then use the Sobel operator to calculate the horizontal and vertical gradients of the image to obtain the gradient magnitude of each pixel, and calculate the average gradient magnitude of the entire image based on the gradient magnitude of each pixel as the average gradient magnitude of the report image to be processed; in this embodiment, calculating the average gradient magnitude of the report image to be processed using the Sobel operator is a prior art, and will not be elaborated here.

[0074] Preferably, the present invention further proposes a method for generating the image sharpness anomaly index of the report to be processed, specifically including:

[0075] Through the formula: ;

[0076] Generate an image sharpness anomaly index for the report to be processed. ;

[0077] In the formula, This represents the average gradient magnitude of the report image to be processed. This represents the average gradient magnitude reference value for a normal resolution image;

[0078] in, The sharpness anomaly index indicates that the higher the image sharpness is compared to the normal level, with a larger value indicating a sharper image than normal. This index is calculated by taking into account the positive difference between the image's average gradient magnitude and a normal sharpness reference value, and normalizing this difference to provide a relative and comparable sharpness assessment metric. One implementation method is to directly substitute the average gradient magnitude of the image to be processed into a preset formula for calculation after obtaining it.

[0079] The average gradient magnitude reference value for normal-resolution images can be obtained through statistical analysis of a large number of standard-resolution report images, such as by taking their average or median. Alternatively, the system can dynamically adjust... The value can be adaptively adjusted based on the overall quality distribution of the current batch of reports, or calibrated according to user-defined sharpness standards, to ensure... It can more accurately reflect situations where image clarity is higher than normal in the current application scenario.

[0080] Gradient magnitude is an indicator of the richness of image detail and edge information. A larger gradient magnitude indicates richer edge information, generally meaning a sharper image. The average gradient magnitude is the average of the gradient magnitudes of all pixels in the entire image, comprehensively reflecting the overall image sharpness. One implementation method is to calculate the image gradient during image preprocessing using edge detection operators (such as Sobel, Prewitt, Roberts, etc.), and then average the gradient magnitudes of all pixels to obtain the average gradient magnitude. .

[0081] This application's solution addresses the challenge of accurately assessing the portion of a report image's sharpness that exceeds normal levels by introducing a quantified sharpness anomaly index. Specifically, it calculates the anomaly using the difference between the average gradient magnitude of the report image and a preset reference value for the average gradient magnitude of a normal-sharpness image. When the average gradient magnitude of the image to be processed is significantly higher than the preset reference value for the average gradient magnitude of a normal-resolution image, it indicates that the image resolution exceeds the normal level. This will display a positive value, and the larger the value, the greater the degree to which the sharpness exceeds the normal level. (Through...) The design ensures that an anomaly index is generated only when image sharpness exceeds a normal reference value, preventing situations with normal or below-normal sharpness from being misjudged as "abnormal" as measured by this index. Subsequently, this difference is... Normalization is performed to make This becomes a dimensionless relative indicator, facilitating comparison and application across different images or scenarios. This quantitative evaluation method provides reliable input for subsequent recognition accuracy prediction models, enabling the system to more accurately determine the impact of excessive image clarity on OCR recognition results, thus laying the foundation for optimizing OCR accuracy levels.

[0082] Through the above technical solution, this application provides a standardized and quantitative method for generating an image sharpness anomaly index for reports to be processed. This index accurately reflects the degree to which image sharpness exceeds normal levels, providing reliable input for subsequent recognition accuracy prediction models. This enables the system to more accurately assess the potential impact of excessive image sharpness on OCR recognition results, thereby improving the reliability of recognition accuracy prediction and providing a solid data foundation for intelligent optimization of OCR accuracy levels. Ultimately, this helps to improve the overall efficiency and accuracy of intelligent recognition and automatic data entry for reports.

[0083] For preferred options, please refer to [link / reference]. Figure 2 The present invention further proposes a method for generating the accuracy and efficiency score of the image recognition in the report to be processed, specifically including:

[0084] Step S41: Based on the image sharpness-recognition accuracy relationship curve under different OCR accuracy levels, obtain the estimated recognition accuracy of the report image to be processed according to the sharpness anomaly index of the report image to be processed.

[0085] Step S42: Establish a recognition accuracy prediction model based on the estimated accuracy of the image recognition of the report to be processed, and generate a benefit score for the recognition accuracy of the image to be processed.

[0086] In step S41, the image sharpness-recognition accuracy relationship curves at different OCR accuracy levels are obtained beforehand through historical data analysis. These curves depict the functional relationship between image sharpness (usually characterized by a sharpness anomaly index) and OCR recognition accuracy at different OCR accuracy levels. This curve can be a discrete set of data points or a continuous function curve obtained through fitting. Its purpose is to provide a benchmark for predicting the possible recognition accuracy at different OCR accuracy levels under specific sharpness conditions. The image sharpness anomaly index of the report to be processed is a quantitative indicator that measures the deviation of the image sharpness from the normal level; a higher value generally indicates worse image sharpness.

[0087] In step S41, the index is used as input to locate or calculate the corresponding estimated recognition accuracy in the aforementioned relationship curve graph. Obtaining the estimated recognition accuracy of the report image to be processed is the core output of step S41, representing the expected accuracy achievable when using a specific OCR accuracy level under the current sharpness conditions of the report image to be processed. This estimated accuracy can be obtained by looking up a table on the relationship curve graph, interpolation calculation, or derivation through a preset mathematical model. For example, the corresponding point can be found on the curve graph based on the sharpness anomaly index, and its ordinate value can be read as the estimated accuracy; or, if the curve graph is represented by a function, the sharpness anomaly index can be substituted into the function for calculation. In step S42, the estimated recognition accuracy of the report image to be processed is the output of step S41 and serves as the input to step S42, providing a preliminary quantification of image recognition performance. Establishing a recognition accuracy prediction model aims to transform the estimated recognition accuracy into a more decision-making benefit score. The model can be a simple mapping function, such as comparing the estimated accuracy with a baseline accuracy to calculate the relative improvement or decrease; or it can be a more complex statistical or machine learning model that comprehensively considers multiple factors to evaluate the "benefit" of recognition accuracy. Its role is to transform a purely numerical accuracy rate into a quantitative indicator that can guide subsequent decisions. Generating a report image recognition accuracy benefit score is the final output of step S42. It is a comprehensive score reflecting the "value" or "benefit" of recognizing the report image at a specific OCR accuracy level. This score can be a dimensionless numerical value used for comprehensive comparison with other factors (such as resource slack) to select the optimal OCR accuracy level in subsequent steps.

[0088] The proposed solution refines the generation process of the recognition accuracy benefit score for the report image to be processed into two interrelated steps, thereby achieving precise quantification of the recognition accuracy benefit. First, in step S41, the system utilizes pre-established image sharpness-recognition accuracy relationship curves under different OCR accuracy levels, combined with the sharpness anomaly index of the current report image to be processed, to obtain the estimated recognition accuracy of the image under different OCR accuracy levels. This process ensures that the estimated accuracy is objectively evaluated based on historical experience data and current image characteristics. Subsequently, in step S42, the system uses the estimated recognition accuracy obtained in step S41 as input, and transforms it into a recognition accuracy benefit score with decision-guiding significance by establishing a recognition accuracy prediction model. This step-by-step processing method transforms the evaluation of recognition accuracy from a single numerical value into a benefit indicator that reflects its potential value and contribution to the overall system. In this way, this application can more accurately evaluate the recognition performance of different OCR accuracy levels under specific image conditions, providing a more solid and quantitative basis for the subsequent selection of the optimal OCR accuracy level, thereby avoiding decision-making bias caused by coarse accuracy evaluation and improving the overall efficiency and accuracy of intelligent recognition and automatic entry of report data.

[0089] The above technical solution refines and standardizes the process of generating an accuracy score for the reported image recognition. This refinement makes the evaluation of recognition accuracy no longer vague, but effectively transforms the historical relationship between image clarity and recognition accuracy, as well as the estimated accuracy, into a quantifiable benefit score through explicit steps S41 and S42. This significantly improves the accuracy and reliability of recognition accuracy evaluation, providing more solid data support for subsequent selection of the optimal OCR accuracy level. Therefore, in the entire intelligent report data recognition and automatic entry method, it can more accurately predict the performance of different OCR accuracy levels, thereby optimizing OCR resource allocation, improving overall recognition accuracy and processing efficiency, and avoiding resource waste or recognition errors caused by inaccurate accuracy evaluation.

[0090] Preferably, the present invention further proposes the following expression for the recognition accuracy prediction model: ;

[0091] In the expression, This represents the benefit score of the report image recognition accuracy when using OCR accuracy level i. This represents the estimated accuracy of the report image recognition when using OCR accuracy level i. This represents the threshold for the accuracy of image recognition in the report.

[0092] This expression defines how to calculate the benefit score of the recognition accuracy of the report image to be processed. Its purpose is to transform the recognition accuracy at different OCR accuracy levels into a standardized and comparable benefit index. This expression quantifies the benefit or loss brought about by the recognition accuracy by calculating the relative deviation of the estimated accuracy of the report image recognition relative to the report image recognition accuracy threshold.

[0093] The image recognition accuracy benefit score is a quantitative indicator that measures the benefit or value brought by the image recognition accuracy at a specific OCR accuracy level i. Its value can be positive, negative, or zero. A positive value indicates that the recognition accuracy is higher than the threshold, bringing positive benefits; a negative value indicates that it is lower than the threshold, bringing negative benefits; and a zero value indicates that it is equal to the threshold. This score is an important input parameter for the subsequent OCR accuracy level optimization model, used to comprehensively evaluate the advantages and disadvantages of different accuracy levels.

[0094] The estimated accuracy of the report image recognition is the system's predicted accuracy for recognizing the current report image when using OCR accuracy level i. This estimated value is usually derived from historical data and factors such as image sharpness through an image sharpness-recognition accuracy curve. It is the fundamental data for evaluating recognition performance at a specific OCR accuracy level.

[0095] The report image recognition accuracy threshold represents the minimum acceptable standard or expectation of the system or user for report image recognition accuracy. It serves as a baseline for judging whether the current recognition accuracy is "good" or "poor," and to what extent. This threshold can be determined based on actual business needs, industry standards, historical average accuracy, or user-defined settings. For example, 95% can be set as the minimum acceptable accuracy; anything below this value is considered risky or requires additional manual intervention.

[0096] This application's solution standardizes and quantifies the calculation process of the benefit score for the recognition accuracy of the report image by introducing a specific expression for the recognition accuracy prediction model. Specifically, after obtaining the sharpness anomaly index of the report image to be processed, the system derives the corresponding estimated recognition accuracy for each possible OCR accuracy level i based on a pre-established image sharpness-recognition accuracy relationship curve for different OCR accuracy levels. Subsequently, this estimated accuracy and a preset report image recognition accuracy threshold are substituted into the expression. Through this calculation method, the benefit score for the recognition accuracy of the report image to be processed can clearly reflect the improvement or decrease in recognition accuracy relative to the benchmark threshold at a specific OCR accuracy level i, thereby transforming the abstract "accuracy" into a meaningful "benefit" or "loss". This quantification provides a unified and comparable input for subsequent OCR accuracy level optimization, ensuring that decisions are made based on objective data when comprehensively considering recognition accuracy and resource slack, avoiding evaluation bias caused by ambiguous model definitions.

[0097] The above technical solution clarifies the specific expression of the recognition accuracy prediction model, solving the problem of objectively comparing and effectively evaluating the recognition accuracy benefit scores under different OCR accuracy levels when a unified quantitative standard is lacking. This expression compares the estimated recognition accuracy with a preset accuracy threshold to generate a standardized and quantifiable benefit score. This allows the system to more accurately assess the potential benefits or risks of different OCR accuracy levels in terms of recognition accuracy, providing a solid data foundation for subsequent OCR accuracy level optimization models. Therefore, when comprehensively considering the recognition accuracy benefit and OCR resource slack, more scientific and reasonable decisions can be made, effectively improving the overall efficiency and accuracy of intelligent recognition and automatic data entry in reports.

[0098] Preferably, the present invention further proposes a method for generating the OCR resource slackness, specifically including:

[0099] Based on the number of tasks waiting in the OCR processing queue and the estimated processing time for each task, an estimated queuing time for the OCR processing queue is generated. The estimated queuing time refers to the estimated time required to complete all tasks in the OCR processing queue.

[0100] Through the formula: ;

[0101] Generate OCR resource slack ;

[0102] In the formula, This refers to the OCR resource slackness at OCR accuracy level i. This represents the estimated queuing time for OCR processing at OCR accuracy level i. This represents the threshold for the queuing time in the OCR processing queue;

[0103] The number of tasks waiting in the OCR processing queue refers to the total number of report image recognition tasks that have not yet been processed in the current OCR system task queue. This number can be obtained by querying the status of pending tasks in the task scheduling system, message queue service, or database, for example, by reading the number of messages currently to be consumed from the message queue. The estimated processing time for waiting tasks refers to the estimated time required for the OCR system to process a single or batch of waiting tasks. This estimate can be based on statistical analysis of historical processing data, such as calculating the average processing time for different types of report images at different OCR accuracy levels; or it can be dynamically estimated by monitoring the CPU, GPU utilization, and memory usage of the OCR server in real time, combined with the complexity of the tasks. The estimated queuing time for the OCR processing queue refers to the total estimated time required to complete all currently waiting tasks in the OCR processing queue. This value is usually calculated by summing or averaging the number of waiting tasks in the OCR processing queue with the estimated processing time for each waiting task.

[0104] OCR resource slack is a quantitative indicator used to measure how "relaxed" the processing capacity of an OCR system is relative to its acceptable queuing time threshold at a specific OCR accuracy level i. The larger the value, the more abundant the system resources. OCR accuracy level i represents a specific recognition mode or configuration used by the OCR system when recognizing report data, such as "high-precision mode", "standard mode", "fast mode", etc.

[0105] The OCR processing queue queuing time threshold is a preset upper limit for an acceptable OCR processing queue queuing time. This threshold can be set according to actual business needs, user experience requirements, or system performance indicators.

[0106] This application's solution introduces OCR resource slack to comprehensively consider the processing capacity and task load of the OCR system, thereby optimizing the selection of OCR accuracy levels. Specifically, firstly, the system obtains the number of tasks waiting in the current OCR processing queue and the estimated processing time for each task. Based on this information, the system can calculate the total time required to complete all waiting tasks under the current load, i.e., the estimated OCR processing queue queuing time. This estimated queuing time reflects the current busyness of the OCR system. Subsequently, by comparing this estimated queuing time with a preset OCR processing queue queuing time threshold, the OCR resource slack is calculated using a specific formula. This slack measures the system's "margin" for processing tasks at different OCR accuracy levels. When the OCR resource slack is high, it indicates that the system resources are relatively abundant, and a higher accuracy OCR level can be considered to improve recognition accuracy; conversely, when the slack is low or even negative, it indicates that the system resources are strained, and a lower accuracy OCR level may need to be selected to ensure processing efficiency. In this way, the solution of this application incorporates the resource status of the OCR system into the decision-making process for the optimal OCR accuracy level, enabling the system to effectively manage and utilize computing resources while ensuring recognition accuracy, and avoiding efficiency decline caused by resource bottlenecks.

[0107] Through the above technical solution, this application effectively incorporates the resource load of the OCR system into the decision-making process for intelligent recognition and automatic data entry in reports. By quantifying OCR resource slack, the system no longer relies solely on image quality and recognition accuracy to select the OCR precision level, but can comprehensively consider the current task queue's workload and the system's processing capacity. This allows the system to intelligently adjust the OCR precision level when faced with a large number of pending tasks or resource constraints. For example, when resources are scarce, it prioritizes the faster low-precision mode to ensure overall processing efficiency and response speed; while when resources are abundant, it can prioritize the high-precision mode to improve recognition accuracy. This dynamic resource awareness and precision adjustment mechanism avoids system overload and task backlog caused by blindly pursuing high precision, significantly improving the OCR system's throughput and user experience, and achieving a balanced optimization between recognition accuracy and processing efficiency.

[0108] For preferred options, please refer to [link / reference]. Figure 3 The present invention further proposes a method for generating the optimal OCR accuracy level, specifically including:

[0109] Step S51: Generate a comprehensive score for OCR accuracy level based on the image recognition accuracy benefit score and OCR resource slackness of the report to be processed;

[0110] Step S52: Establish an OCR accuracy level optimization model based on the comprehensive score of OCR accuracy level, and generate the optimal OCR accuracy level;

[0111] Step S51 aims to integrate the two key indicators—image recognition accuracy benefit score and OCR resource slack—into a single, comprehensive indicator that can fully measure the merits of different OCR accuracy levels. Its purpose is to provide a unified evaluation standard for the subsequent optimization process. This comprehensive score can be generated by linearly combining the recognition accuracy benefit score and OCR resource slack using a weighted summation method, where the weighting coefficients can be set according to actual business needs or expert experience. Alternatively, nonlinear functions, such as product or exponential functions, can be used to fuse the two to better reflect their complex interaction, or machine learning methods such as fuzzy logic and neural networks can be used to learn and generate the comprehensive score based on historical data.

[0112] Step S52 aims to utilize the generated comprehensive score of OCR accuracy levels to determine the most suitable OCR accuracy level for the current report image to be processed through an optimization model. Its function is to automatically select the optimal OCR processing strategy to balance recognition performance and resource consumption. This optimization model can be established by traversing all possible OCR accuracy levels, calculating the comprehensive score for each level, and then selecting the level with the highest comprehensive score as the optimal OCR accuracy level. Alternatively, optimization algorithms, such as genetic algorithms or particle swarm optimization, can be used to search for the optimal solution within the set of OCR accuracy levels. Furthermore, it can be based on preset decision rules or lookup tables, directly mapping the comprehensive score to the optimal OCR accuracy level.

[0113] The proposed solution first integrates two indicators—the image recognition accuracy benefit score and the OCR resource slack—representing recognition performance and resource consumption respectively, into a comprehensive OCR accuracy level score in step S51. This transforms the multi-objective optimization problem into a single-objective optimization problem. Subsequently, in step S52, an OCR accuracy level optimization model is established based on this comprehensive score, and the optimal OCR accuracy level is selected. This step-by-step and quantitative processing method ensures that the selection of the OCR accuracy level is no longer a simple empirical judgment, but rather a comprehensive consideration of both recognition accuracy benefit and resource utilization efficiency. This allows for a more scientific and reasonable determination of the most suitable OCR accuracy level for the current task. In this way, the system can effectively balance recognition quality and processing cost, avoiding resource waste caused by blindly pursuing high accuracy or sacrificing recognition accuracy due to resource constraints.

[0114] Through the above technical solution, this application provides a more refined and intelligent OCR accuracy level selection mechanism. This mechanism introduces a comprehensive OCR accuracy level score and establishes an optimization model based on this score. This allows the system to comprehensively consider both recognition accuracy benefits and OCR resource slack when selecting an OCR accuracy level, thereby maximizing resource utilization efficiency while ensuring recognition quality. This effectively solves the problem of how to scientifically and rationally determine the OCR accuracy level to balance performance and cost in complex and ever-changing report data processing scenarios, thereby improving the overall efficiency and economy of intelligent report data recognition and automatic data entry methods.

[0115] Preferably, the present invention further proposes a method for generating the comprehensive score of OCR accuracy level, specifically including:

[0116] Through the formula: ;

[0117] Generate a comprehensive score for OCR accuracy level. ;

[0118] In the formula, This represents the benefit score of the report image recognition accuracy when using OCR accuracy level i. This represents the OCR resource slackness at OCR accuracy level i. , All are weighting coefficients, and ;

[0119] The OCR accuracy level comprehensive score is a quantitative indicator used to comprehensively evaluate the overall performance of a specific OCR accuracy level i. It aims to integrate multiple influencing factors (such as recognition accuracy benefits and resource utilization efficiency) into a single, comparable value, thereby providing a basis for subsequent decision-making regarding the optimal OCR accuracy level. This score can be generated using various mathematical models, such as weighted summation, product models, or more complex nonlinear functions.

[0120] The above formula is a linear weighted summation model used to calculate the comprehensive score for OCR accuracy level. This model assigns different weight coefficients to different evaluation indicators (i.e., the accuracy score of the image recognition of the report to be processed and the OCR resource slackness). , This is to reflect their relative importance in the overall evaluation.

[0121] Weighting coefficient , These are parameters used to adjust the relative importance of the image recognition accuracy score and OCR resource slack in the overall OCR accuracy score. These weighting coefficients are typically set based on actual business needs, system priorities, or expert experience. For example, if the system prioritizes recognition accuracy, higher weighting coefficients can be set. Value; if more emphasis is placed on resource utilization efficiency and processing speed, a higher value can be set. Value. The sum of the weighting coefficients equals 1 (i.e., The constraints ensure the normalization of the contributions of each factor, making the numerical range and interpretability of the comprehensive score more reasonable.

[0122] The proposed solution generates a comprehensive OCR accuracy score by weighted summing of the image recognition accuracy benefit score and OCR resource slack. Specifically, for each possible OCR accuracy level i, the system first obtains its corresponding recognition accuracy benefit score and OCR resource slack. Then, based on preset weighting coefficients... , Substituting these two scores into the formula above, a comprehensive score for accuracy level i is obtained. This comprehensive score quantitatively reflects the overall performance of the OCR accuracy level in balancing recognition accuracy and resource utilization efficiency. In this way, the system integrates two originally independent evaluation dimensions into a unified and comparable value, providing a clear and comprehensive input for the subsequent OCR accuracy level optimization model. This allows the optimization process to proceed on a balanced basis, effectively solving the problem of how to quantitatively integrate different indicators for optimal selection.

[0123] Through the above technical solution, this application provides a quantitative and flexible method to comprehensively evaluate the performance of different OCR accuracy levels. By introducing weighting coefficients, the system can dynamically adjust the importance of recognition accuracy benefits and resource slack in decision-making based on actual business needs and priorities, thereby generating a comprehensive score that fully reflects the quality of OCR accuracy levels. This allows subsequent OCR accuracy level optimization models to be developed under a unified standard that balances recognition quality and resource consumption, avoiding one-sided decisions that may be caused by a single indicator, and significantly improving the scientific rigor and adaptability of selecting the optimal OCR accuracy level.

[0124] Preferably, the present invention further proposes the following expression for the OCR accuracy level optimization model: ;

[0125] In the expression, This indicates the optimal OCR accuracy level. This represents the comprehensive score of OCR accuracy level when the report image to be processed uses OCR accuracy level i, and G represents the set of OCR accuracy levels;

[0126] in, This refers to the accuracy setting that, after evaluation, is determined to be the most suitable for recognizing and inputting the current report image among all available OCR accuracy levels. This optimal OCR accuracy level can be a specific OCR engine configuration parameter, such as a recognition mode (e.g., high-precision mode, balanced mode, fast mode, etc.), or it can be an identifier corresponding to a specific OCR algorithm or model.

[0127] G represents the set of OCR accuracy levels, which includes all OCR accuracy level options currently available in the system. This set can be predefined or dynamically generated based on system configuration or available OCR engines.

[0128] The above expression means that, from the set of OCR accuracy levels G, an accuracy level i is selected such that the comprehensive score of the OCR accuracy level corresponding to accuracy level i reaches the maximum value, and this accuracy level i is taken as the optimal OCR accuracy level. The implementation of this operation may include: iterating through each OCR accuracy level i in the set G, calculating its corresponding comprehensive OCR accuracy level score, then comparing all calculated comprehensive OCR accuracy level scores, and selecting the accuracy level corresponding to the maximum value as the optimal OCR accuracy level. Alternatively, the optimal OCR accuracy level can be directly determined based on the distribution characteristics of the comprehensive score through pre-set decision rules or lookup tables.

[0129] The solution in this application uses the above expression to take the comprehensive OCR accuracy level score, jointly determined by the accuracy benefit score of the report image recognition and the OCR resource slack, as the core indicator, and clarifies the method for determining the optimal OCR accuracy level. Specifically, for each possible OCR accuracy level i in the OCR accuracy level set G, the system first generates an image sharpness-recognition accuracy relationship curve based on the sharpness anomaly index of the report image to be processed and historical data, thereby obtaining the estimated accuracy of the report image recognition and calculating the accuracy benefit score of the report image recognition at accuracy level i. At the same time, the system also generates the OCR resource slack at accuracy level i based on the number of tasks waiting in the OCR processing queue and the estimated processing time of the waiting tasks. Subsequently, the accuracy benefit score of the report image recognition at accuracy level i and the OCR resource slack at accuracy level i are weighted and summed to obtain the comprehensive OCR accuracy level score of accuracy level i. After calculating the overall score for all OCR accuracy levels, the system selects the accuracy level that maximizes the overall score from all candidate accuracy levels, as the final optimal OCR accuracy level, as described above. This process ensures that the selection of the OCR accuracy level is dynamic and optimal, intelligently adjusting based on image quality and system resource availability, thereby maximizing resource utilization efficiency while maintaining recognition accuracy.

[0130] The above technical solution clarifies the mechanism for determining the optimal OCR accuracy level, enabling the system to intelligently select the most suitable OCR accuracy level for the current task and system state based on a quantified comprehensive score. This avoids suboptimal results that may result from manual experience-based judgment or fixed pattern selection, ensuring that the OCR processing achieves the best balance between recognition accuracy and resource utilization efficiency. Therefore, efficient and reliable intelligent recognition and automatic input of report data can be achieved under different image quality and system load conditions.

[0131] Please see Figure 4 In another embodiment of the present invention, an OCR-based intelligent recognition and automatic data entry system for report data is proposed. This system is used to execute the above-described OCR-based intelligent recognition and automatic data entry method for report data, specifically including:

[0132] The sharpness analysis unit 10 is used to obtain the average gradient magnitude of the report image to be processed, and generate the sharpness anomaly index of the report image to be processed based on the average gradient magnitude of the report image to be processed.

[0133] The historical analysis unit 20 is used to obtain the accuracy of the report images recognized by all historical OCR accuracy levels and the corresponding report image sharpness anomaly index. Based on the accuracy of the report images recognized by the OCR accuracy level and the corresponding report image sharpness anomaly index, it generates image sharpness-recognition accuracy relationship curves under different OCR accuracy levels.

[0134] The recognition accuracy analysis unit 30 is used to establish a recognition accuracy prediction model based on the image clarity-recognition accuracy relationship curve under different OCR accuracy levels, and generate a recognition accuracy benefit score for the report image to be processed based on the clarity anomaly index of the report image to be processed.

[0135] The resource slack analysis unit 40 is used to obtain the number of tasks waiting in the OCR processing queue and the estimated time required to process the waiting tasks, and to generate OCR resource slack based on the number of tasks waiting in the OCR processing queue and the estimated time required to process the waiting tasks.

[0136] The optimal accuracy analysis unit 50 is used to establish an OCR accuracy level optimization model based on the image recognition accuracy benefit score of the report to be processed and the OCR resource slackness, and generate the optimal OCR accuracy level.

[0137] The processing unit 60 is used to intelligently recognize and automatically input report data of the report image to be processed according to the optimal OCR accuracy level.

[0138] For preferred options, please refer to [link / reference]. Figure 5 The present invention further proposes that the resource slack analysis unit 40 specifically includes:

[0139] The accuracy prediction module 41 is used to obtain the image recognition prediction accuracy of the report to be processed based on the image sharpness-recognition accuracy relationship curve under different OCR accuracy levels and the image sharpness anomaly index of the report to be processed.

[0140] The benefit analysis module 42 is used to establish a recognition accuracy prediction model based on the estimated accuracy of the image recognition of the report to be processed, and generate a benefit score for the recognition accuracy of the report to be processed.

[0141] For preferred options, please refer to [link / reference]. Figure 6 The present invention further proposes that the optimal accuracy analysis unit 50 specifically includes:

[0142] The comprehensive analysis module 51 is used to generate a comprehensive score for OCR accuracy level based on the image recognition accuracy benefit score and OCR resource slackness of the report to be processed.

[0143] The result output module 52 is used to establish an OCR accuracy level optimization model based on the comprehensive score of OCR accuracy level and generate the optimal OCR accuracy level.

[0144] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent recognition and automatic entry of report data based on OCR, characterized in that, The method specifically includes: Obtain the average gradient magnitude of the report image to be processed, and generate the sharpness anomaly index of the report image to be processed based on the average gradient magnitude of the report image to be processed. Obtain the accuracy and corresponding image sharpness anomaly index of the report images recognized by all historical OCR accuracy levels. Based on the accuracy and corresponding image sharpness anomaly index of the report images recognized by the OCR accuracy level, generate image sharpness-recognition accuracy relationship curves under different OCR accuracy levels. Based on the image sharpness-recognition accuracy relationship curves under different OCR accuracy levels, a recognition accuracy prediction model is established according to the sharpness anomaly index of the report image to be processed, and a recognition accuracy benefit score for the report image to be processed is generated. Obtain the number of tasks waiting in the OCR processing queue and the estimated processing time for each task. Based on these parameters, generate OCR resource slack. Based on the image recognition accuracy and efficiency score of the report to be processed and the slackness of OCR resources, an OCR accuracy level optimization model is established to generate the optimal OCR accuracy level. Based on the optimal OCR accuracy level, the report data of the report image to be processed is intelligently recognized and automatically entered.

2. The method for intelligent recognition and automatic entry of report data based on OCR according to claim 1, characterized in that, The method for obtaining the average gradient magnitude of the report image to be processed is as follows: convert the report image to be processed into a grayscale image, then use the Sobel operator to calculate the horizontal and vertical gradients of the image to obtain the gradient magnitude of each pixel, and calculate the average gradient magnitude of the entire image based on the gradient magnitude of each pixel as the average gradient magnitude of the report image to be processed.

3. The method for intelligent recognition and automatic entry of report data based on OCR according to claim 1, characterized in that, The specific methods for generating the image sharpness anomaly index in the report to be processed include: Through the formula: ; Generate an image sharpness anomaly index for the report to be processed. ; In the formula, This represents the average gradient magnitude of the report image to be processed. This represents the average gradient magnitude reference value for a normal resolution image.

4. The method for intelligent recognition and automatic entry of report data based on OCR according to claim 1, characterized in that, The specific methods for generating the image recognition accuracy and efficiency score for the report to be processed include: Based on the image sharpness-recognition accuracy relationship curves under different OCR accuracy levels, the estimated recognition accuracy of the report image to be processed is obtained according to the sharpness anomaly index of the report image to be processed. Based on the estimated accuracy of the image recognition of the report to be processed, a recognition accuracy prediction model is established, and a benefit score for the recognition accuracy of the report to be processed is generated.

5. The method for intelligent recognition and automatic entry of report data based on OCR according to claim 4, characterized in that, The specific expression of the recognition accuracy prediction model is as follows: ; In the expression, This represents the benefit score of the report image recognition accuracy when using OCR accuracy level i. This represents the estimated accuracy of the report image recognition when using OCR accuracy level i. This represents the threshold for the accuracy of image recognition in the report.

6. The method for intelligent recognition and automatic entry of report data based on OCR according to claim 1, characterized in that, The specific methods for generating OCR resource slack include: Based on the number of tasks waiting in the OCR processing queue and the estimated processing time for each task, an estimated queuing time for the OCR processing queue is generated. The estimated queuing time refers to the estimated time required to complete all tasks in the OCR processing queue. Through the formula: ; OCR resource slack ; In the formula, This refers to the OCR resource slackness at OCR accuracy level i. This represents the estimated queuing time for OCR processing at OCR accuracy level i. This represents the threshold for the queuing time in the OCR processing queue.

7. The method for intelligent recognition and automatic entry of report data based on OCR according to claim 1, characterized in that, The specific methods for generating the optimal OCR accuracy level include: Based on the image recognition accuracy and efficiency score of the report to be processed and the slackness of OCR resources, a comprehensive score for OCR accuracy level is generated. An OCR accuracy level optimization model is established based on the comprehensive score of OCR accuracy level to generate the optimal OCR accuracy level.

8. The method for intelligent recognition and automatic entry of report data based on OCR according to claim 7, characterized in that, The specific methods for generating the comprehensive score for OCR accuracy level include: Through the formula: ; Generate a comprehensive score for OCR accuracy level. ; In the formula, This represents the benefit score of the report image recognition accuracy when using OCR accuracy level i. This represents the OCR resource slackness at OCR accuracy level i. , All are weighting coefficients, and .

9. The method for intelligent recognition and automatic entry of report data based on OCR according to claim 7, characterized in that, The specific expression for the OCR accuracy level optimization model is as follows: ; In the expression, This indicates the optimal OCR accuracy level. This represents the overall score of OCR accuracy level when the image to be processed is at OCR accuracy level i, and G represents the set of OCR accuracy levels.

10. An OCR-based intelligent recognition and automatic data entry system for report data, characterized in that: This system is used to execute the OCR-based intelligent recognition and automatic data entry method for report data as described in any one of claims 1-9, specifically including: The sharpness analysis unit is used to obtain the average gradient magnitude of the report image to be processed, and to generate a sharpness anomaly index of the report image to be processed based on the average gradient magnitude of the report image to be processed. The historical analysis unit is used to obtain the accuracy of the report images recognized by all historical OCR accuracy levels and the corresponding report image sharpness anomaly index. Based on the accuracy of the report images recognized by the OCR accuracy level and the corresponding report image sharpness anomaly index, it generates image sharpness-recognition accuracy relationship curves under different OCR accuracy levels. The recognition accuracy analysis unit is used to establish a recognition accuracy prediction model based on the image sharpness-recognition accuracy relationship curve under different OCR accuracy levels, and generate a recognition accuracy benefit score for the image of the report to be processed based on the sharpness anomaly index of the report to be processed. The resource slack analysis unit is used to obtain the number of tasks waiting in the OCR processing queue and the estimated time required to process the waiting tasks. Based on the number of tasks waiting in the OCR processing queue and the estimated time required to process the waiting tasks, the OCR resource slack is generated. The optimal accuracy analysis unit is used to establish an OCR accuracy level optimization model based on the image recognition accuracy benefit score of the report to be processed and the OCR resource slackness, and generate the optimal OCR accuracy level. The processing unit is used to intelligently recognize and automatically input report data from the report image to be processed based on the optimal OCR accuracy level.

Images