An automatic annotation processing method and system for image data

Abstract

The present invention relates to the field of image processing technology, and discloses an automatic annotation processing method and system for image data. The method includes: preprocessing the image to be processed and analyzing its content complexity, and selecting an initial annotation model for annotation. After annotation, obtain the confidence score, compare it with the qualified confidence, and determine whether to adjust the complexity threshold. If adjustment is required, reselect the annotation model based on the new complexity threshold. If the second selected model is different, re-annotate and obtain the second confidence; if the same, directly mark the image. By comparing the first and second confidences, further adjust the complexity threshold. If the model is different again, make adjustments and obtain the third confidence. Finally, compare the third confidence with the qualified confidence to decide whether to output the annotation result or continue to mark the image. This dynamic adjustment mechanism improves the annotation accuracy and system adaptability.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular, to a method and system for automatically annotating and processing image data. Background Art

[0002] With the continuous development of artificial intelligence technology, the automatic annotation technology of image data has been widely applied in many fields such as medical imaging, autonomous driving, and security monitoring. The automatic annotation technology can automatically identify and annotate objects, regions, and related features in large-scale image data by using machine learning and deep learning algorithms, greatly improving the data processing efficiency. However, existing automatic annotation methods usually have some problems, mainly reflected in the balance between annotation accuracy and efficiency, the processing of complex image content, and the adaptability and self-adaptive ability of the annotation model.

[0003] Currently, image annotation systems face an important problem that the accuracy of the annotation model often decreases when the complexity of image content is relatively high. This phenomenon is particularly obvious when dealing with images in complex scenarios or with high detail requirements. Existing methods usually rely on a single annotation model, and these models may not be able to adapt to changes in image content with different complexities, resulting in a low confidence level of the annotation results, which in turn affects subsequent processing and decision-making.

[0004] In addition, existing image annotation systems often lack a flexible adjustment mechanism when dealing with the selection of the annotation model. Once the annotation model is selected, it is difficult to dynamically optimize and adjust according to the actual annotation effect. Especially when the confidence level of the annotation result is lower than the preset qualified standard, how to automatically adjust the complexity threshold and select an appropriate annotation model to improve the accuracy of the annotation result is still an urgent problem to be solved. Summary of the Invention

[0005] To solve the above problems, the present invention proposes a method and system for automatically annotating and processing image data.

[0006] On the one hand, a method for automatically annotating and processing image data proposed by the present invention includes:

[0007] Obtaining the to-be-processed image after preprocessing, analyzing the content complexity of each frame of the to-be-processed image to obtain the content complexity; setting a complexity threshold, and first selecting an annotation model according to the relationship between the content complexity and the complexity threshold, and annotating the to-be-processed image through the selected initial annotation model;

[0008] After the annotation is completed, obtaining the confidence score of the annotation result, denoted as the first confidence, and judging whether to adjust the complexity threshold to obtain a corrected complexity threshold according to the first confidence and the preset qualified confidence;

[0009] If it is necessary to adjust the complexity threshold, select the annotation model again according to the relationship between the corrected complexity threshold and the content complexity after the adjustment. When the second selection is different from the annotation model selected for the first time, re-annotate the image to be processed, and obtain the confidence score after the annotation is completed, denoted as the second confidence. When the second selection is the same as the annotation model selected for the first time, mark the image to be processed; compare the first confidence with the second confidence, and determine whether to adjust the corrected complexity threshold according to the relationship between the first confidence and the second confidence;

[0010] When the second selection is different from the annotation model selected for the first time, adjust the corrected complexity threshold, and select the annotation model again for annotation to obtain the third confidence. Compare the third confidence with the qualified confidence. When the third confidence is less than the qualified confidence, mark the image to be processed. When the third confidence is greater than or equal to the qualified confidence, output the annotation result.

[0011] Further, the preprocessing includes:

[0012] Convert the image to be processed into a standard format, then perform denoising, and rotate, translate and scale the image to increase the training data of the annotation model.

[0013] Further, the content complexity analysis includes:

[0014] Calculate the comprehensive complexity according to edge detection, texture analysis and brightness / contrast distribution analysis; among them, obtain the number of edge pixels of each frame of the image to be processed according to edge detection, and calculate the first complexity; calculate the texture entropy of each frame of the image to be detected according to the gray-level co-occurrence matrix to obtain the second complexity; obtain the third complexity according to the brightness change value of each frame of the image to be detected;

[0015] Assign weights to the first complexity, the second complexity and the third complexity and perform weighted summation to obtain the content complexity of each frame of the image to be detected.

[0016] Further, the first complexity, the second complexity and the third complexity and the content complexity are obtained through the following steps:

[0017] The first complexity is the ratio of the number of edge pixels of each frame of the image to be processed to the total number of pixels; the second complexity is the texture entropy of each frame of the image to be processed; the third complexity is the difference between the brightness of the current frame of the image to be processed and the brightness of the previous adjacent frame of the image to be processed;

[0018] The content complexity satisfies the following relationship:

[0019] ;

[0020] Among them, is the content complexity, is the first complexity, is the second complexity, is the third complexity, and and are the first influence coefficient, the second influence coefficient and the third influence coefficient respectively, and .

[0021] Furthermore, when initially selecting a labeling model according to the relationship between the content complexity and the complexity threshold, it includes: setting multiple levels of the complexity threshold, and selecting a labeling model according to the relationship between several complexity thresholds and the content complexity;

[0022] Predetermine a first-level complexity threshold and a second-level complexity threshold, and the first-level complexity threshold is less than the second-level complexity threshold; when the content complexity is less than the first-level complexity threshold, select a lightweight labeling model, when the content complexity is greater than or equal to the second-level complexity threshold, select a high-precision labeling model, otherwise select a medium-complexity labeling model.

[0023] Furthermore, when determining whether to adjust the complexity threshold to obtain a corrected complexity threshold according to the first confidence level and a preset qualified confidence level, it includes:

[0024] Compare the first confidence level with the qualified confidence level; when the first confidence level is greater than or equal to the qualified confidence level, do not adjust the complexity threshold and output the labeling result, when the first confidence level is less than the qualified confidence level, determine that the complexity threshold needs to be adjusted; at this time, obtain the labeling model determined by the initial selection, when the labeling model is a high-precision labeling model, determine that it does not meet the adjustment conditions and mark the image to be processed, when the labeling model is not a high-precision labeling model, obtain the complexity threshold on the right side of the content complexity, adjust the complexity threshold, and the corrected complexity threshold obtained after adjustment satisfies the following relationship:

[0025] ;

[0026] Among them, is the corrected complexity threshold, is the complexity threshold before correction, is the qualified confidence level, is the first confidence level, is the unit adjustment amount.

[0027] Furthermore, the unit adjustment amount is obtained through the following relationship:

[0028] ;

[0029] wherein, is the unit adjustment amount, is the first-level complexity threshold, is the second-level complexity threshold.

[0030] Further, when determining whether to adjust the corrected complexity threshold according to the relationship between the first confidence level and the second confidence level, it includes:

[0031] When the first confidence level is greater than the second confidence level, it is determined that the corrected complexity threshold needs to be adjusted;

[0032] When the first confidence level is less than or equal to the second confidence level, it is determined that the corrected complexity threshold is not adjusted.

[0033] Further, when adjusting the corrected complexity threshold, the adjusted corrected complexity threshold is denoted as the final complexity threshold. When adjusting the final complexity threshold, the complexity threshold on the right side of the content complexity is obtained, and the corrected complexity threshold is adjusted. The adjusted final complexity threshold satisfies:

[0034] ;

[0035] wherein, is the final complexity threshold, is the second confidence level.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0037] Traditional image data annotation methods often use fixed annotation models. When facing image content with high complexity, the annotation results may be inaccurate, especially when dealing with images with high detail requirements or complex backgrounds. This technical solution dynamically selects an annotation model according to the content complexity of the image, ensuring that the most suitable annotation model can be selected for images with different complexities, thereby improving the accuracy of annotation. At the same time, by adjusting the complexity threshold, the annotation model can better adapt to complex scenarios and enhance the adaptive ability of the system.

[0038] Through multi-level analysis of the image content complexity (including multi-faceted information such as edges, textures, and brightness), this solution can accurately evaluate the complexity of each frame of the image and select a lightweight, medium-complexity, or high-precision annotation model based on this complexity. This way of selecting models on demand effectively avoids the use of inappropriate models, reduces the error rate when dealing with complex scenarios, and improves the calculation efficiency.

[0039] During the annotation process, if the confidence level of the preliminary annotation result is low, existing systems often cannot automatically adjust to improve accuracy. By introducing a mechanism to dynamically adjust the complexity threshold, this solution can automatically make adjustments when the annotation result does not meet expectations. By reselecting the annotation model and determining whether to further adjust the complexity threshold based on the new annotation result, this self-optimizing mechanism ensures that the image annotation system can flexibly handle different situations and avoid a decline in annotation accuracy.

[0040] The "dynamic annotation model selection" and "complexity threshold adjustment" in this solution can automatically complete the annotation task in most cases and automatically optimize through the adjustment mechanism when the annotation does not meet expectations. This automated mechanism can effectively reduce manual intervention. At the same time, when the confidence level is low, the system can also improve the quality of the annotation result through manual assistance. This mode of combining automation and manual collaboration greatly improves the working efficiency of the annotation system and reduces labor costs.

[0041] This technical solution has high scalability and can adapt to the image data annotation requirements in different fields. For example, in medical images, the complexity and fineness requirements of the images are relatively high. This solution can select a suitable annotation model according to the specific scenario and is also applicable in fields such as security monitoring and satellite remote sensing. The system can automatically adjust the annotation model for image data with different complexities, thus having cross-domain applicability.

[0042] By repeatedly evaluating and adjusting the confidence level of the annotation result, this technical solution can not only improve the accuracy of the annotation but also enhance the quality control ability of the system. At each step of the annotation process, the system tracks the confidence level and adjusts the complexity threshold and annotation model according to the actual annotation effect. Finally, when the confidence level reaches the qualified standard, the system outputs a high-quality annotation result. This can greatly reduce the negative impact brought by low-quality annotations and ensure the high quality of the annotated data.

[0043] When faced with complex and detailed image content, this solution can accurately evaluate the complexity and flexibly select a high-precision annotation model. During the annotation process, it can dynamically adjust the annotation strategy and complexity threshold to ensure the processing effect of high-complexity images. This method can effectively handle the image data in complex scenarios and provide accurate annotation information for subsequent processing, improving the efficiency of the entire image analysis system.

[0044] On the other hand, the present invention also proposes an automatic annotation processing system for image data, including:

[0045] The first module is configured to obtain the image to be processed after preprocessing, analyze the content complexity of each frame of the image to be processed to obtain the content complexity; set a complexity threshold, and first select a labeling model according to the relationship between the content complexity and the complexity threshold, and label the image to be processed through the selected initial labeling model;

[0046] The second module is configured to, after the labeling is completed, obtain the confidence score of the labeling result, denoted as the first confidence, and determine whether to adjust the complexity threshold to obtain a corrected complexity threshold according to the first confidence and the preset qualified confidence;

[0047] The third module is configured to, if the complexity threshold needs to be adjusted, secondarily select a labeling model according to the relationship between the adjusted corrected complexity threshold and the content complexity. When the second selection is different from the labeling model selected for the first time, relabel the image to be processed, and obtain the confidence score after the labeling is completed, denoted as the second confidence. When the second selection is the same as the labeling model selected for the first time, mark the image to be processed; compare the first confidence with the second confidence, and determine whether to adjust the corrected complexity threshold according to the relationship between the first confidence and the second confidence;

[0048] The fourth module is configured to, when the second selection is different from the labeling model selected for the first time, adjust the corrected complexity threshold, and select a labeling model again for labeling to obtain the third confidence. Compare the third confidence with the qualified confidence. When the third confidence is less than the qualified confidence, mark the image to be processed. When the third confidence is greater than or equal to the qualified confidence, output the labeling result.

[0049] It can be understood that an image data automatic labeling processing system and method of the present invention have the same beneficial effects, which will not be elaborated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0051] Figure 1 is a flowchart of an image data automatic labeling processing method provided by an embodiment of the present invention.

[0052] Figure 2 is a functional framework diagram of an image data automatic labeling processing system provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0053] Exemplary embodiments disclosed in the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art. It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the drawings and in combination with the embodiments.

[0054] Referring to Figure 1 as shown, an embodiment of the present invention provides an automatic annotation processing method for image data, including:

[0055] S1: Obtain the to-be-processed image after preprocessing, analyze the content complexity of each frame of the to-be-processed image to obtain the content complexity; set a complexity threshold, and for the first time select an annotation model according to the relationship between the content complexity and the complexity threshold, and annotate the to-be-processed image through the selected initial annotation model;

[0056] S2: After the annotation is completed, obtain the confidence score of the annotation result, denoted as the first confidence, and judge whether to adjust the complexity threshold to obtain a corrected complexity threshold according to the first confidence and the preset qualified confidence;

[0057] S3: If it is necessary to adjust the complexity threshold, select an annotation model for the second time according to the relationship between the adjusted corrected complexity threshold and the content complexity. When the second selection is different from the first selected annotation model, re-annotate the to-be-processed image, and obtain the confidence score after the annotation is completed, denoted as the second confidence. When the second selection is the same as the first selected annotation model, mark the to-be-processed image; compare the first confidence with the second confidence, and judge whether to adjust the corrected complexity threshold according to the relationship between the first confidence and the second confidence;

[0058] S4: When the second selection is different from the first selected annotation model, adjust the corrected complexity threshold, and select an annotation model again for annotation to obtain a third confidence. Compare the third confidence with the qualified confidence. When the third confidence is less than the qualified confidence, mark the to-be-processed image. When the third confidence is greater than or equal to the qualified confidence, output the annotation result.

[0059] It should be noted that the to-be-detected images are from image data such as drone aerial photography, satellite remote sensing, medical images, autonomous driving data, and industrial inspection.

[0060] "The method of secondarily selecting the annotation model according to the relationship between the adjusted revised complexity threshold and the content complexity is the same as the method of initially selecting the annotation model. The difference lies in replacing the original complexity threshold with the adjusted revised complexity threshold, and then determining to select any one of the lightweight annotation model, the medium-complexity annotation model, and the high-precision annotation model."

[0061] It should be noted that the technical solution has the following beneficial effects:

[0062] Adaptive annotation model selection: Dynamically match a suitable annotation model through content complexity analysis, improving annotation efficiency and accuracy.

[0063] Optimizing the utilization of computing resources: Using a low-computation model for lightweight images and a high-precision model for high-complexity images to achieve reasonable allocation of computing resources.

[0064] Dynamically adjusting the complexity threshold: Based on the annotation confidence feedback, adjust the complexity threshold to adapt to different image features and improve annotation stability.

[0065] Multi-round confidence verification mechanism: Ensure the accuracy of the final annotation result through multiple confidence comparisons and model adjustments, reduce misannotations, and improve the quality of automated annotation.

[0066] Reducing manual intervention: Reduce the need for manual review through an intelligent adjustment mechanism, improving the autonomy and reliability of automatic annotation.

[0067] Wide adaptability: Applicable to image data with different content complexities, enhancing the generalization ability in various application scenarios during annotation.

[0068] In some embodiments of the present application, the preprocessing includes:

[0069] Convert the image to be processed into a standard format, then perform denoising, and rotate, translate, and scale the image to increase the training data of the annotation model.

[0070] It should be noted that different data sources may use different image formats and encoding methods. Converting to a unified standard format (such as JPEG, PNG, or TIFF) can improve processing compatibility and ensure that the annotation model can correctly parse the input data. Images may be affected by noise interference, such as sensor noise, compression artifacts, or environmental interference. Reducing noise through filtering (such as Gaussian filtering, median filtering, or wavelet denoising) helps improve the accuracy of feature extraction such as edge detection and texture analysis.

[0071] In some embodiments of the present application, the content complexity analysis includes:

[0072] Calculate the comprehensive complexity based on edge detection, texture analysis, and brightness / contrast distribution analysis; among them, obtain the number of edge pixels of each frame of the image to be processed according to edge detection, and calculate the first complexity; calculate the texture entropy of each frame of the image to be detected according to the gray-level co-occurrence matrix to obtain the second complexity; obtain the third complexity according to the brightness change value of each frame of the image to be detected.

[0073] Assign weights to the first complexity, the second complexity, and the third complexity and sum them up weighted to obtain the content complexity of each frame of the image to be detected.

[0074] It should be noted that

[0075] In some embodiments of the present application, the first complexity, the second complexity, and the third complexity and the content complexity are obtained through the following steps:

[0076] The first complexity is the ratio of the number of edge pixels of each frame of the image to be processed to the total number of pixels; the second complexity is the texture entropy of each frame of the image to be processed; the third complexity is the difference between the brightness of the current frame of the image to be processed and the brightness of the previous adjacent frame of the image to be processed.

[0077] The content complexity satisfies the following relationship:

[0078] ;

[0079] Among them, is the content complexity, is the first complexity, is the second complexity, is the third complexity, , and are the first influence coefficient, the second influence coefficient, and the third influence coefficient respectively, and .

[0080] It should be noted that traditional image analysis usually only relies on a single feature, such as edge information or brightness distribution, while this method combines edge detection, texture analysis, and brightness change to calculate the comprehensive complexity, enabling a more comprehensive quantitative evaluation of the content features of the image. This multi-dimensional calculation method makes the evaluation of image complexity more accurate, helps to select a suitable model in the annotation process, and improves the annotation efficiency and accuracy.

[0081] Using edge detection to calculate the first complexity can identify the structural information of the image and is applicable to scenarios where the target boundary is clear; the texture entropy calculated by the gray-level co-occurrence matrix as the second complexity is applicable to images with rich textures or many details; calculating the third complexity based on brightness change helps to analyze images with large changes in lighting conditions. Through the fusion of the three complexities, this method can be applicable to different types of images and improve the adaptability to complex environments.

[0082] By setting weights for three levels of complexity and performing a weighted sum, this method can adjust the weight parameters according to different task requirements. For example, when processing industrial inspection images, the weight of edge detection can be increased, while in remote sensing image processing, the influence of texture analysis can be enhanced. This flexibility enables the method to maintain good generalization ability in multiple fields.

[0083] Since the content complexity of different images varies, directly using a unified annotation model may lead to a decrease in annotation accuracy. Through the calculation of comprehensive complexity, this method ensures that lightweight models are selected for simple images and high-precision models are selected for complex images, avoiding waste of computing resources while reducing misjudgments caused by overfitting for low-complexity images and missed annotations caused by insufficient model capabilities for high-complexity images, thus significantly improving the overall annotation accuracy.

[0084] Feature analysis such as brightness change and texture entropy can effectively reduce the influence of interference factors such as illumination change and noise, reduce misjudgments caused by environmental factors, and improve the stability of complexity calculation. At the same time, this method can adapt to image inputs of different resolutions and qualities, maintain consistent performance under different shooting conditions, and improve the reliability of the image data annotation process.

[0085] By automatically calculating the image complexity and dynamically adjusting the selection of the annotation model, this method reduces manual participation and improves the automation degree of image data processing. This is particularly important for the annotation tasks of large-scale data sets, which can effectively reduce labor costs while ensuring the consistency of annotation quality.

[0086] In some embodiments of the present application, when first selecting an annotation model according to the relationship between the content complexity and the complexity threshold, it includes: setting multiple levels of complexity thresholds, and selecting an annotation model according to the relationship between several complexity thresholds and the content complexity;

[0087] A primary complexity threshold and a secondary complexity threshold are preset, and the primary complexity threshold is less than the secondary complexity threshold; when the content complexity is less than the primary complexity threshold, a lightweight annotation model is selected, when the content complexity is greater than or equal to the secondary complexity threshold, a high-precision annotation model is selected, otherwise a medium-complexity annotation model is selected.

[0088] It should be noted that the annotation models include: a lightweight annotation model, a medium-complexity annotation model, and a high-precision annotation model;

[0089] Each annotation model includes at least one set of models capable of automatically annotating and processing image data. In this embodiment, when the current model determines which level of model it is, it is divided according to the computational overhead of the model.

[0090] It should be noted that by setting multiple complexity thresholds, lightweight models are adopted for simple images to reduce computational overhead and improve processing speed, while high-precision models are used for complex images to ensure labeling quality, thereby improving the overall computational efficiency while ensuring labeling accuracy.

[0091] Multilevel complexity thresholds are used to distinguish image types, making the labeling model match the actual complexity of the images, avoiding overfitting caused by using overly complex models for low-complexity images, and also avoiding mislabeling due to insufficient model capabilities for high-complexity images, thus improving the overall labeling accuracy.

[0092] This method is applicable to different types of image scenarios. The complexity thresholds can be adjusted according to task requirements to adapt to different data distributions, improving generality and robustness, and enhancing the model's adaptability to diverse data.

[0093] By setting multiple complexity thresholds, the labeling model can be automatically selected without manual intervention, thereby reducing labor costs and improving the intelligence level of the image labeling process, which helps to efficiently execute large-scale automatic labeling tasks.

[0094] In some embodiments of the present application, when determining whether to adjust the complexity threshold to obtain a corrected complexity threshold based on the first confidence level and a preset qualified confidence level, it includes:

[0095] Compare the first confidence level with the qualified confidence level; when the first confidence level is greater than or equal to the qualified confidence level, do not adjust the complexity threshold and output the labeling result. When the first confidence level is less than the qualified confidence level, it is determined that the complexity threshold needs to be adjusted; at this time, obtain the labeling model determined by the first selection. When the labeling model is a high-precision labeling model, it is determined that it does not meet the adjustment conditions, and the image to be processed is marked. When the labeling model is not a high-precision labeling model, obtain the complexity threshold on the right side of the content complexity, and adjust the complexity threshold. The corrected complexity threshold obtained after adjustment satisfies the following relationship:

[0096] ;

[0097] Wherein, is the corrected complexity threshold, is the complexity threshold before correction, is the qualified confidence level, is the first confidence level, is the unit adjustment amount.

[0098] It should be noted that through the confidence feedback mechanism, the complexity threshold is dynamically adjusted to make the model selection more accurate, avoid the selection of unreasonable annotation models caused by a fixed threshold, and ensure more accurate annotation results. The complexity threshold is only adjusted when the confidence is lower than the qualified standard, and the model is reselected when the conditions are met, so as to reduce unnecessary model switching, reduce the computational overhead, and improve the processing efficiency. This method adaptively adjusts the complexity threshold according to the actual annotation confidence of the image, can flexibly handle different image types, improve the adaptability to diverse data, and enhance the robustness in complex environments. The complexity threshold is only adjusted in the case of a non-high-precision model and a low confidence, and the unit adjustment amount is used to control the change range to ensure that the adjustment process is stable and controllable, and avoid the instability caused by excessive adjustment. By automatically judging whether the complexity threshold needs to be adjusted and performing reasonable dynamic optimization, this method can reduce manual intervention, improve the intelligence and automation level of the annotation process, and make it more suitable for large-scale image data processing tasks.

[0099] In some embodiments of the present application, the unit adjustment amount is obtained through the following relationship:

[0100] ;

[0101] where is the unit adjustment amount, is the first-level complexity threshold, is the second-level complexity threshold.

[0102] It should be noted that by setting the unit adjustment amount k to 1 / 3 of the difference between the two complexity thresholds, smooth adjustment is achieved, avoiding too large or too small adjustment amplitude, and improving the accuracy of complexity threshold regulation.

[0103] Adopting a hierarchical adjustment strategy, only 1 / 3 of the range is adjusted each time to ensure a moderate change in the threshold and avoid instability caused by parameter mutation.

[0104] By uniformly distributed adjustment instead of jumping to extreme values at one time, the optimization convergence speed can be accelerated, making the model selection process of the annotation more efficient.

[0105] Adopting a calculation method based on the first-level and second-level complexity thresholds makes this adjustment method applicable to different image complexity situations, improving the generality and adaptability.

[0106] Only simple numerical calculations are required to determine the unit adjustment amount, reducing unnecessary complex operations, reducing the consumption of computing resources, and improving the processing efficiency.

[0107] In some embodiments of the present application, when judging whether to adjust the corrected complexity threshold according to the relationship between the first confidence and the second confidence, it includes:

[0108] When the first confidence level is greater than the second confidence level, it is determined that the correction complexity threshold needs to be adjusted;

[0109] When the first confidence level is less than or equal to the second confidence level, it is determined that the correction complexity threshold is not adjusted.

[0110] It should be noted that by comparing the first confidence level and the second confidence level, it is ensured that the corrected complexity threshold will not reduce the reliability of the annotation result and improve the accuracy of the final annotation.

[0111] The complexity threshold is adjusted only when the first confidence level is greater than the second confidence level, reducing ineffective adjustment operations and improving stability and computational efficiency.

[0112] By dynamically adjusting the correction complexity threshold, the selection of the annotation model is more adapted to the actual complexity of the image, improving the model adaptability and reducing the misjudgment rate.

[0113] Adjustment is made only when the annotation quality fails to improve, avoiding excessive calculation, reducing computational resource consumption, and improving operating efficiency.

[0114] By real-time monitoring of the change in confidence level and dynamically adjusting the complexity threshold, it can adapt to different image scenarios and improve the universality of annotation.

[0115] In some embodiments of the present application, when the correction complexity threshold is adjusted, the adjusted correction complexity threshold is denoted as the final complexity threshold. When adjusting the final complexity threshold, the complexity threshold on the right side of the content complexity is obtained, and the correction complexity threshold is adjusted. The final complexity threshold obtained after adjustment satisfies:

[0116] ;

[0117] Wherein, is the final complexity threshold, is the second confidence level.

[0118] It should be noted that by calculating the final complexity threshold, the adjusted complexity threshold is more in line with the actual situation of the image content, improving the applicability of the annotation model. Adjusting the complexity threshold in combination with the confidence level P2 makes the final threshold better match the annotation reliability of the model and reduces the occurrence of low-confidence annotation results. Dynamically adjusting according to the complexity threshold on the right side of the image content complexity ensures that the change in image complexity can be reasonably responded to and improves the adaptive ability of annotation. Using a reasonable adjustment formula ensures that the adjustment of the complexity threshold is neither too large nor too small, preventing the stability of the annotation model from being affected by excessive parameter changes. Through targeted adjustment of the complexity threshold, unnecessary calculations are reduced.

[0119] Refer to Figure 2As shown in the figure, an embodiment of the present invention also provides an automatic annotation processing system for image data, including:

[0120] A first module, configured to obtain the to-be-processed image after preprocessing, analyze the content complexity of each frame of the to-be-processed image to obtain the content complexity; set a complexity threshold, and for the first time select a annotation model according to the relationship between the content complexity and the complexity threshold, and annotate the to-be-processed image through the selected initial annotation model;

[0121] A second module, configured to, after the annotation is completed, obtain the confidence score of the annotation result, denoted as the first confidence, and judge whether to adjust the complexity threshold to obtain a corrected complexity threshold according to the first confidence and the preset qualified confidence;

[0122] A third module, configured to, if it is necessary to adjust the complexity threshold, select a annotation model for the second time according to the relationship between the adjusted corrected complexity threshold and the content complexity. When the second selection is different from the first selected annotation model, re-annotate the to-be-processed image, and obtain the confidence score after the annotation is completed, denoted as the second confidence. When the second selection is the same as the first selected annotation model, mark the to-be-processed image; compare the first confidence with the second confidence, and judge whether to adjust the corrected complexity threshold according to the relationship between the first confidence and the second confidence;

[0123] A fourth module, configured to, when the second selection is different from the first selected annotation model, adjust the corrected complexity threshold, and select a annotation model again for annotation, obtain the third confidence, and compare the third confidence with the qualified confidence. When the third confidence is less than the qualified confidence, mark the to-be-processed image. When the third confidence is greater than or equal to the qualified confidence, output the annotation result.

[0124] It should be noted that the beneficial effects of the above technical solutions include the following aspects:

[0125] Select a suitable annotation model through content complexity analysis to make the initial annotation results more in line with the characteristics of the images, reducing mislabeling and missing labels. Adopt multiple rounds of confidence verification to ensure that the final output annotation results have high confidence and improve the reliability of the overall annotation. Set a complexity threshold and dynamically adjust the complexity threshold based on the actual annotation confidence to avoid the problem of insufficient generalization caused by a fixed threshold. Select models of different levels according to the image complexity during the annotation process to ensure that there is neither excessive calculation nor errors caused by insufficient model capabilities. Combine multiple adjustment mechanisms to gradually optimize the complexity threshold by comparing the confidence levels, enabling the system to adapt to different types of image data. Adopt a hierarchical screening mechanism, giving priority to using lightweight models to process simple images and reducing the call frequency of high-precision models, thereby optimizing computing resources. Only perform secondary or tertiary annotation when necessary to avoid ineffective repeated calculations and improve the operating efficiency of the entire system. By gradually adjusting the complexity threshold, reduce the calculation instability caused by large-scale parameter adjustments and improve the system response speed. Through multiple rounds of confidence evaluation, achieve adaptive optimization of the annotation process, reduce manual inspection and intervention, and improve the automation level of image data processing. Adopt an intelligent threshold adjustment strategy, enabling the system to automatically adjust according to the data characteristics without the need for manual setting of the complexity range for each image. Through content complexity analysis, it can adapt to different categories of images and improve the versatility of the annotation system. Adopt a multi-level annotation model to ensure that the appropriate annotation strategy matches the level of detail of different images, avoiding problems of over-annotation or under-annotation. By comprehensively evaluating the first confidence level, the second confidence level, and the third confidence level, make the final output annotation results have higher reliability and consistency. Adopt multi-level confidence comparison to prevent error accumulation caused by incorrect selection of the initial annotation model. By gradually adjusting the complexity threshold, ensure that the system can self-correct, reduce the result deviation caused by single annotation mistakes, and improve the stability of the overall system.

[0126] This solution improves the accuracy, efficiency, and intelligence of image automatic annotation through strategies such as multi-level preprocessing, dynamic model selection, confidence evaluation, and adaptive adjustment of the complexity threshold. It not only reduces the computing cost but also enhances the system adaptability, is applicable to various image processing scenarios, and can be widely used in fields such as medical imaging, remote sensing data, and intelligent monitoring.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: It is still possible to modify the specific implementation manners of the present invention or make equivalent replacements, and any modification or equivalent replacement that does not deviate from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. An automatic annotation processing method for image data, characterized in that, Including: Obtain the image to be processed after preprocessing, analyze the content complexity of each frame of the image to be processed, and obtain the content complexity; Set a complexity threshold, first select a labeling model according to the relationship between the content complexity and the complexity threshold, and label the image to be processed through the selected initial labeling model; After the labeling is completed, obtain the confidence score of the labeling result, denoted as the first confidence, and determine whether to adjust the complexity threshold to obtain a corrected complexity threshold according to the relationship between the first confidence and the preset qualified confidence; If it is necessary to adjust the complexity threshold, secondarily select a labeling model according to the relationship between the corrected complexity threshold after the adjustment and the content complexity. When the second selection is different from the labeling model selected for the first time, relabel the image to be processed, and obtain the confidence score after the labeling is completed, denoted as the second confidence. When the second selection is the same as the labeling model selected for the first time, mark the image to be processed; Compare the first confidence with the second confidence, and determine whether to adjust the corrected complexity threshold according to the relationship between the first confidence and the second confidence; When the second selection is different from the labeling model selected for the first time, adjust the corrected complexity threshold, and select a labeling model again for labeling to obtain the third confidence. Compare the third confidence with the qualified confidence. When the third confidence is less than the qualified confidence, mark the image to be processed. When the third confidence is greater than or equal to the qualified confidence, output the labeling result.

2. The automatic annotation processing method for image data according to claim 1, wherein The preprocessing includes: Convert the image to be processed into a standard format, then perform denoising, and rotate, translate, and scale the image to increase the training data of the labeling model.

3. The automatic annotation processing method for image data according to claim 2, wherein The content complexity analysis includes: Calculate the comprehensive complexity according to edge detection, texture analysis, and brightness / contrast distribution analysis; among them, obtain the number of edge pixels of each frame of the image to be processed according to edge detection, and calculate the first complexity; calculate the texture entropy of each frame of the image to be detected according to the gray-level co-occurrence matrix to obtain the second complexity; obtain the third complexity according to the brightness change value of each frame of the image to be detected; Assign weights to the first complexity, the second complexity, and the third complexity and perform weighted summation to obtain the content complexity of each frame of the image to be detected.

4. The automatic annotation processing method for image data according to claim 3, wherein The first complexity, the second complexity, and the third complexity and the content complexity are obtained through the following steps: The first complexity is the ratio of the number of edge pixels of each frame of the image to be processed to the total number of pixels; the second complexity is the texture entropy of each frame of the image to be processed; the third complexity is the difference between the brightness of the current frame of the image to be processed and the brightness of the previous adjacent frame of the image to be processed; The content complexity satisfies the following relationship: ； Among them, is the content complexity, is the first complexity, is the second complexity, is the third complexity, 、 and are the first influence coefficient, the second influence coefficient and the third influence coefficient respectively, and .

5. The automatic annotation processing method for image data according to claim 4, characterized in that When first selecting a labeling model according to the relationship between the content complexity and the complexity threshold, it includes: setting multiple levels of the complexity threshold, and selecting a labeling model according to the relationship between several complexity thresholds and the content complexity; A first-level complexity threshold and a second-level complexity threshold are preset, and the first-level complexity threshold is less than the second-level complexity threshold; when the content complexity is less than the first-level complexity threshold, a lightweight annotation model is selected; when the content complexity is greater than or equal to the second-level complexity threshold, a high-precision annotation model is selected; otherwise, a medium-complexity annotation model is selected.

6. The automatic annotation processing method for image data according to claim 5, wherein, When judging whether to adjust the complexity threshold to obtain a corrected complexity threshold according to the first confidence level and a preset qualified confidence level, it includes: Comparing the first confidence level with the qualified confidence level; when the first confidence level is greater than or equal to the qualified confidence level, the complexity threshold is not adjusted, and the annotation result is output; when the first confidence level is less than the qualified confidence level, it is judged that the complexity threshold needs to be adjusted; at this time, the annotation model determined by the first selection is obtained. When the annotation model is a high-precision annotation model, it is judged that the adjustment condition is not met, and the image to be processed is marked. When the annotation model is not a high-precision annotation model, the complexity threshold on the right side of the content complexity is obtained, and the complexity threshold is adjusted. The corrected complexity threshold obtained after adjustment satisfies the following relationship: ； Among them, is the correction complexity threshold, is the complexity threshold before correction, is the qualified confidence level, is the first confidence level, is the unit adjustment amount.

7. The automatic annotation processing method for image data according to claim 6, wherein The unit adjustment amount is obtained through the following relationship: ； Among them, is the unit adjustment amount, is the first-level complexity threshold, is the second-level complexity threshold.

8. The automatic annotation processing method for image data according to claim 7, wherein When judging whether to adjust the corrected complexity threshold according to the relationship between the first confidence level and the second confidence level, it includes: When the first confidence level is greater than the second confidence level, it is judged that the corrected complexity threshold needs to be adjusted; When the first confidence level is less than or equal to the second confidence level, it is judged that the corrected complexity threshold is not adjusted.

9. The automatic annotation processing method for image data according to claim 8, wherein, When the corrected complexity threshold is adjusted, the adjusted corrected complexity threshold is recorded as the final complexity threshold. When adjusting the final complexity threshold, the complexity threshold on the right side of the content complexity is obtained, and the corrected complexity threshold is adjusted. The final complexity threshold obtained after adjustment satisfies: ； Among them, is the final complexity threshold, is the second confidence level.

10. An image data automatic annotation processing system for implementing the image data automatic annotation processing method according to any one of claims 1-9, characterized in that, It includes: A first module configured to obtain the image to be processed after preprocessing, analyze the content complexity of each frame of the image to be processed, and obtain the content complexity; Set the complexity threshold, first select an annotation model according to the relationship between the content complexity and the complexity threshold, and annotate the image to be processed through the selected initial annotation model; A second module configured to, after the annotation is completed, obtain the confidence score of the annotation result, record it as the first confidence level, and judge whether to adjust the complexity threshold to obtain a corrected complexity threshold according to the first confidence level and a preset qualified confidence level; A third module configured to, if the complexity threshold needs to be adjusted, second select an annotation model according to the relationship between the adjusted corrected complexity threshold and the content complexity. When the second selection is different from the annotation model selected for the first time, re-annotate the image to be processed, and obtain the confidence score after the annotation is completed, record it as the second confidence level. When the second selection is the same as the annotation model selected for the first time, mark the image to be processed; Compare the first confidence level with the second confidence level, and determine whether to adjust the correction complexity threshold according to the relationship between the first confidence level and the second confidence level; A fourth module, configured to adjust the correction complexity threshold when the second selection is different from the first selection of the annotation model, and select the annotation model again for annotation to obtain a third confidence level, compare the third confidence level with the qualified confidence level, and when the third confidence level is less than the qualified confidence level, mark the image to be processed, and when the third confidence level is greater than or equal to the qualified confidence level, output the annotation result.

Images