Image generation-based stevioside product impurity detection method

Abstract

The invention discloses a stevioside product impurity detection method based on image generation, and relates to the technical field of food quality safety detection, and the method comprises the following steps: S100, carrying out the overall analysis of historical generated images formed in a continuous image generation process according to a time evolution relation, and obtaining an overall analysis result; and distinguishing and extracting a morphological stable region and a hidden drift region which are formed along with time accumulation, and constructing a time bearing substrate. According to the method, a time bearing substrate is constructed, and a risk pointing line and a rhythm updating mechanism are introduced, so that a reference expression is kept in an adjustable evolution state in a time dimension, and abnormal long-term solidification is avoided; and meanwhile, the participation weight of the reference image is periodically collected and released in combination with the production rhythm, so that high-risk impurities are continuously separated from a reference expression range, and the stability, consistency and long-term reliability of stevioside impurity detection in continuous production are improved.

Description

Technical Field

[0001] This invention relates to the field of food quality and safety testing technology, specifically to a method for detecting impurities in steviol glycoside products based on image generation. Background Technology

[0002] Image-based steviol glycoside impurity detection refers to the process of collecting images of product samples during steviol glycoside production or inspection. This involves constructing an analytical foundation centered on visual inspection and introducing image generation technology to represent and complete the sample morphology under different conditions. Impurity morphology, distribution characteristics, and manifestation conditions that are originally difficult to consistently present are enhanced and reconstructed at the image level. Based on this, the generated images are compared and analyzed to identify foreign objects, abnormal particles, or non-target components in the product through visual inspection. This method does not rely on a single static image for judgment but uses visual inspection as the main line. Through the diverse representation of generated images, it amplifies the differences in morphology, texture, and spatial distribution between impurities and normal steviol glycoside crystals, making impurity characteristics clearer and more distinguishable. This approach is suitable for improving the stability and consistency of visual inspection of steviol glycoside product impurities in complex production environments.

[0003] The existing technology has the following shortcomings: Under current technological conditions, image-generated steviol glycoside impurity detection typically involves superimposing multiple rounds of image generation results to form a reference image, which is then repeatedly used as the basis for judgment in subsequent detection processes. However, when a small number of high-risk impurities are not accurately distinguished during early image generation and are instead incorrectly absorbed into the generated results and solidified as part of the reference image, this bias expression will be continuously inherited with the repeated reuse of the reference image. In this case, subsequent detection processes will compare new samples based on the already shifted reference expression, causing similar high-risk impurities to be repeatedly ignored in time-continuous detection scenarios, failing to be identified and removed in a timely manner. This leads to the continuous accumulation and spread of impurity risks across multiple production batches and detection cycles.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide an image-based method for detecting impurities in steviol glycoside products, thereby addressing the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for detecting impurities in steviol glycoside products based on image generation, comprising the following steps: S100: The historical generated images formed during the continuous image generation process are analyzed as a whole according to the temporal evolution relationship. The morphologically stable region and the hidden drift region formed over time are distinguished and extracted to construct a time-bearing basis for characterizing the evolutionary state of the reference expression. S200, based on the time-bearing base, compares and analyzes newly entered images in the image generation process with morphologically stable regions, identifies impurity signs that show a deviation trend in the current stage and have been absorbed in the historical generation process, and extends the impurity signs along the time evolution direction to form a risk pointing line; S300, based on the risk pointing line, rearranges and controls the order of image composition in the subsequent image generation process, so that the image content corresponding to the risk pointing line is presented forward in the generation process and forms a counterbalancing relationship with the morphologically stable area, thereby limiting the abnormal content from being incorporated into the reference expression again. S400, based on the hedging relationship, rhythmically decomposes the update rhythm of the reference image, so that the reference representation presents a segmented breathing change state in the time dimension, thereby weakening the long-term dominant role of the single generation result on the reference representation. S500, based on segmented breathing change states, periodically compresses and releases the participation weight of the reference image in the image generation process according to the production rhythm, so that high-risk impurities continuously leave the reference expression range as time progresses, thereby suppressing the cumulative amplification of impurity risks in the time dimension.

[0007] Preferably, step S100 includes: Historical generated images generated during the continuous image generation process are uniformly collected and a time series relationship is constructed according to the order in which the images were generated. Based on the time series relationship, the morphological change trend presented in the historical generated images is compared and analyzed. Image regions that appear repeatedly in multiple consecutive time nodes are extracted as morphological stable regions, and image regions that show gradual change characteristics in the time series are extracted as hidden drift regions. Based on the time dimension, the distribution of morphologically stable regions and hidden drift regions in historically generated images is uniformly processed, so that the reference representation states at different time stages form a continuous and connected hierarchical structure. Based on unified carrying capacity processing, a time-carrying base is constructed to characterize the evolutionary state of the reference expression. The morphologically stable regions and hidden drift regions formed at different time stages are incorporated into the time-carrying base for continuous recording and updating.

[0008] Preferably, in the process of constructing the time-bearing substrate, the original generation state of the historical generated images is maintained, so that the morphologically stable region and the hidden drift region are incorporated into the time-bearing substrate in a continuous mapping manner in the time series, and the evolution trajectory of the reference expression is carried by the time dimension, so that the reference expression is transformed from a static expression into an overall state with time evolution characteristics.

[0009] Preferably, step S200 includes: Based on the time-bearing substrate, newly entering images in the image generation process are connected to the end time node of the time-bearing substrate in the order of entry, so that the newly entering images in the image generation process maintain a continuous temporal evolution relationship with the historically generated images. Based on the time-bearing basis, the newly entered image in the image generation process is compared and analyzed with the morphologically stable region. Deviation trend candidate regions that deviate from the morphologically stable region are extracted and written into the evolution record of the time-bearing basis. Based on the time-bearing basis, the historical evolution trajectory of the deviation trend candidate region and the hidden drift region is correlated to identify the impurity signs absorbed in the historical generation process. Based on the extension of impurity signs along the direction of time evolution, risk pointing lines are formed in the time-bearing substrate to characterize the risk evolution path.

[0010] Preferably, during the formation of the risk indicator line, the spatial position of the impurity signs in the time-bearing substrate is continuously mapped to the morphological boundary, so that the risk indicator line maintains a correspondence with the morphologically stable region in the direction of temporal evolution, and the deviation direction and deviation range of the impurity signs relative to the morphologically stable region are recorded, so as to enhance the indicative role of the risk indicator line in the evolutionary state of the reference expression.

[0011] Preferably, step S300 includes: Based on the time-bearing basis, the images to be entered into the reference representation update sequence during the subsequent image generation process are associated with the risk pointing line, so that the relationship between the images to be generated and the risk pointing line in the direction of temporal evolution is indexed; Based on the association results, the order of image composition in the subsequent image generation process is rearranged so that the image content associated with the risk direction line is presented forward in the generation process. Based on the rearrangement of the image composition order, the risk pointing line presented in the front is made to participate in the construction of reference expression in the same time period as the image content and the morphological stable area, thereby forming a hedging relationship. After the hedging relationship is formed, the rearrangement and regulation of the image composition order continues to act on the subsequent image generation process, so that the directional effect of the risk pointing line in the time-bearing substrate remains continuous, thereby limiting the reintegration of abnormal content into the reference expression.

[0012] Preferably, during the image composition order rearrangement process, the forward shift of the image content corresponding to the risk indicator line in the time-bearing base continuously spans multiple time segments, and maintains a temporal juxtaposition relationship with the morphologically stable area in each time segment. This ensures that the image content corresponding to the risk indicator line is continuously in a counterbalancing state during the evolution of the reference expression, thereby enhancing the restriction effect on the reintegration of abnormal content into the reference expression.

[0013] Preferably, step S400 includes: Based on the hedging relationship formed by the risk indicator line and the stable region, the continuous image generation process is rhythmically divided in the time dimension, so that the update process of the reference image is split into interconnected time update segments. Based on time update segments, the update method of the reference representation in each time update segment is rhythmically controlled so that the reference representation absorbs and generates images within the segment and maintains a stable state when switching segments, forming a segmented breathing change state. Based on the segmented breathing change state, the hedging relationship between the risk direction line and the stable morphology zone is embedded into each time update segment, so that the hedging relationship can act independently in each time update segment. Based on the segmented respiratory change state, the rhythmic update method is continuously applied to the subsequent image generation process to weaken the long-term dominant role of a single generation result on the reference representation.

[0014] Preferably, under segmented breathing change state, the reference expression formed in each time update segment is carried as a whole as a historical expression and participates in the evolution constraint of subsequent time update segments, so that the reference expression maintains the continuous correlation between stages in the process of time advancement, while limiting the image generated in any time update segment from continuously dominating the reference expression.

[0015] Preferably, step S500 includes: Based on the segmented breathing change state, the continuous production process is divided into interconnected production stages according to the production rhythm, so that the corresponding reference expression update segment of each production stage is updated. Based on the production stage, the participation weight of the reference image in the image generation process is compressed in each production stage, so that the guiding role of the reference image on the newly generated image is reduced in the corresponding stage. Based on the segmented breathing change state, the participation weight of the reference image is released after the production stage switch, so that the reference image can re-participate in the image generation process within a limited time period. By alternately compressing and releasing the participation weights during continuous production, high-risk impurities are continuously removed from the reference expression range over time.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention introduces a time-based foundation, a risk indicator line, and rhythmic control of update frequency. This prevents the reference expression from evolving based on the continuous superposition of single generation results. Instead, it forms a distinguishable and controllable evolutionary state within the time dimension, effectively preventing early anomalies from being inherited and solidified in subsequent generation processes. By maintaining the opposing relationship between deviation and stability trends within the reference expression, impurity-related expressions remain identifiable and controllable, improving the stability of steviol glycoside impurity detection in continuous production scenarios and providing a consistent basis for judgment across different time stages.

[0017] This invention combines the participation weight of the reference image in the image generation process with the production rhythm, implementing periodic compression and release. This allows high-risk impurities to continuously escape the reference representation range over time, preventing the gradual accumulation and amplification of impurity risks across multiple production stages. This method endows the reference representation with self-mitigation capabilities, continuously suppressing the covert integration of anomalous content during long-term operation. This improves the adaptability of steviol glycoside product impurity detection to complex production fluctuations, ensuring the reliability and sustainability of the detection process under long-term operating conditions. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0019] Figure 1 This is a flowchart of the image-based method for detecting impurities in steviol glycoside products according to the present invention. Detailed Implementation

[0020] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0021] This invention provides, for example Figure 1 The image-based method for detecting impurities in steviol glycoside products, as shown, includes the following steps: S100: The historical generated images formed during the continuous image generation process are analyzed as a whole according to the temporal evolution relationship. The morphologically stable region and the hidden drift region formed over time are distinguished and extracted to construct a time-bearing basis for characterizing the evolutionary state of the reference expression. To address the issue of long-term shifts in reference representations due to early anomalies during continuous image generation, this study focuses on the temporal evolution characteristics of historical generated images. It performs holistic and temporally correlated processing on these historical images, transforming the formation process of reference representations from a single, cumulative result into an evolutionary state that can be continuously carried and distinguished. This lays the foundation for subsequent identification and control of impurity risks. The specific implementation steps are as follows: In the continuous image generation process, historical images of the same steviol glycoside product generated at different times are uniformly collected and constructed according to the order of image generation, forming a complete time series relationship. This ensures that the historical generated images form a continuous and non-jumping arrangement structure in the time dimension. During this process, no historical generated images are filtered or deleted; instead, their original generation state is maintained, allowing them to fully reflect the natural evolution trajectory of the reference expression at different time stages. In this way, subtle morphological changes caused by variations in production conditions, raw material fluctuations, or differences in the generation environment during the continuous image generation process are preserved in the time series, providing a sufficient information basis for subsequent morphological differentiation.

[0022] After constructing the temporal sequence relationship of the historical generated images, a comparative analysis of the morphological change trends presented in the historical generated images is performed based on the time dimension. Image regions that repeatedly appear across multiple consecutive time points, exhibit limited morphological changes, and have relatively stable spatial distribution are categorized as morphologically stable regions. Simultaneously, image regions exhibiting slow displacement, gradual deformation, or blurred boundary expansion in the time series, but not easily identified as anomalies at a single time point, are distinguished as hidden drift regions. This distinction based on temporal evolution clearly separates morphologically stable regions from hidden drift regions within the same historical generated image set, avoiding the misinterpretation of long-accumulated offset representations as stable reference content.

[0023] After distinguishing between morphologically stable regions and hidden drift regions, the distribution of these two types of regions in the time series is uniformly processed. Using time as the main thread, the morphologically stable regions and hidden drift regions corresponding to each time node are mapped to the same carrying space according to their generation order. This allows the reference representation states at different time stages to form a continuously connected hierarchical structure within this carrying space. In this process, the morphologically stable region is used to represent the basic morphology of the reference representation that maintains consistency over time, while the hidden drift region is used to represent the potential evolutionary trajectory of the reference representation that gradually shifts over time. Thus, the reference representation is no longer represented as a single static image, but is transformed into a holistic state with temporal thickness and evolutionary direction.

[0024] Building upon the existing data processing, a temporal data carrier basis is constructed to characterize the evolutionary state of the reference representation. Morphologically stable regions and hidden drift regions from different time stages in historically generated images are incorporated into this temporal data carrier basis for continuous recording and updating. This allows the reference representation to possess traceable, distinguishable, and extensible evolutionary characteristics over time. By constructing the temporal data carrier basis, the formation process of the reference representation during continuous image generation is transformed from simple superposition into a layered evolutionary process, thus providing a stable and consistent foundation for subsequent comparative analysis of newly generated images based on the temporal data carrier basis.

[0025] S200, based on the time-bearing base, compares and analyzes newly entered images in the image generation process with morphologically stable regions, identifies impurity signs that show a deviation trend in the current stage and have been absorbed in the historical generation process, and extends the impurity signs along the time evolution direction to form a risk pointing line; To prevent newly entered images from absorbing high-risk impurities into normal form when the reference representation has already formed a bias, a constructed temporal basis is used as the temporal evolution carrier of the reference representation. A direct comparison relationship is established between the newly entered images and the morphologically stable regions in the temporal basis. During the comparison process, the historical offset trajectory reflected by the hidden drift regions carried by the temporal basis is traced, thereby identifying impurity signs that show a deviation trend in the current stage and have been absorbed in the historical generation process. The impurity signs are then extended along the temporal evolution direction to form a risk indicator line. The specific implementation steps are as follows: Based on a temporal carrying base, images newly entering the image generation process are temporally located and associated. These images are then connected to the final time node of the temporal carrying base in their order of entry, ensuring a continuous temporal evolution relationship between them and the historical image sequence within the base. In this process, the temporal carrying base includes both morphologically stable regions extracted from historical images and hidden drift regions differentiated from them. Therefore, newly entering images simultaneously acquire a reference basis with the morphologically stable regions and a temporal tracing basis with the hidden drift regions upon connection. To ensure consistency, newly entering images are connected to the temporal carrying base using the same appearance scale and spatial alignment as historical images, ensuring their spatial distribution corresponds to the spatial location of the morphologically stable regions. This allows subsequent comparative analysis to be conducted under the same reference coordinate system. Through this connection method, newly entering images are not viewed in isolation but are incorporated into a continuous chain of reference evolution, providing temporally continuous comparison conditions for identifying deviation trends.

[0026] After a newly generated image is integrated into the time-bound substrate, it is compared with the morphologically stable region. A comparative analysis is established, focusing on the stable crystal morphology, stable texture structure, and stable spatial distribution represented by the morphologically stable region. This ensures a layer-by-layer comparison between the morphological boundaries, grain size, texture orientation, crystal arrangement direction, and intergranular distribution of the corresponding regions in the newly generated image and the morphologically stable region. In this comparison, the morphologically stable region serves as a stable benchmark, characterizing the consistent appearance of normal steviol glycoside crystals throughout continuous image generation. The newly generated image, representing the current stage of appearance, reflects the actual morphological presentation under the current production rhythm and sample condition. When a newly generated image exhibits morphological changes inconsistent with the morphologically stable region in certain areas, these areas are not directly defined as anomalous. Instead, they are recorded as deviation candidate regions in the evolution record at the end of the time-bound substrate. This ensures that the deviation candidate regions occupy a clear temporal position within the time-bound substrate and maintain a comparative association with the morphologically stable region, thus providing a foundation for further identification based on historical absorption features. In this way, the formation process of deviation from the trend is incorporated into the temporal evolution link of the time-bearing base, avoiding unstable judgments based solely on differences in a single image.

[0027] After forming candidate regions for deviation trends, a historical absorption correlation analysis is performed on these regions based on a time-bearing substrate. This establishes a retrospective relationship between the candidate regions and hidden drift regions in the time-bearing substrate, identifying impurity signs that exhibit deviation trends in the current stage and were absorbed during historical generation. Specifically, the evolution trajectory of hidden drift regions at historical time points is unfolded using the time-bearing substrate. The candidate regions for deviation trends are correlated with the evolution trajectory of hidden drift regions in terms of spatial location, morphological outline, texture graininess, and distribution sparsity. When a candidate region for deviation trends exhibits a deviation trend in a morphologically stable region and a hidden drift expression with a continuous evolutionary relationship can be found in an earlier time point of the time-bearing substrate, the candidate region for deviation trend is identified as an impurity sign. The term "impurity indications" here emphasizes two meanings: First, impurity indications currently exhibit a deviation from the morphologically stable region, distinguishing them from normal crystal morphology. Second, impurity indications can be traced back to their absorption into the reference representation during the historical evolution of the substrate. That is, impurity indications existed as hidden drift regions in the historical generated image and gradually integrated into the reference representation over time, causing a latent shift in the reference representation. Through this temporal tracing and spatial correspondence, the identification of impurity indications relies not only on the current deviation trend but also on the historical absorption evolutionary path. This allows high-risk impurities that have already been absorbed to be pulled out of the reference representation and transformed into actionable risk objects, providing clear guidance for the subsequent reintegration of confinement anomalies into the reference representation.

[0028] After impurity signs are identified, they are extended along the temporal evolution direction to form risk indicator lines. These risk indicator lines provide continuous indication within the temporal substrate and can penetrate the evolutionary state of the reference representation. Specifically, the spatial location and morphological boundary of the impurity signs in the newly generated image are mapped to the end time node of the temporal substrate. Starting from this mapping, the risk indicator lines are continuously extended along the temporal evolution direction of the substrate. The evolution trajectory of the hidden drift zone corresponding to the impurity signs at historical time nodes is linked with the deviation trend of the impurity signs at the current stage, forming a continuous risk trajectory expression from early absorption to current deviation. The extension process of the risk indicator lines maintains a correlation with the morphologically stable region. That is, the risk indicator lines simultaneously record their deviation direction and deviation range relative to the morphologically stable region during extension. This allows the risk indicator lines to not only represent the temporal source and evolutionary path of the impurity signs but also the spatial opposition between the impurity signs and the stable reference representation. By forming a risk pointing line at the end of the time-bearing substrate, the time-bearing substrate is further expanded from the evolution state of the carrying reference representation to the evolution state of the carrying reference representation and the risk evolution pointing. This allows the subsequent image generation process to rearrange and regulate the generation order based on the risk pointing line, and the content corresponding to the risk pointing line is presented forward in the generation process to form a counterbalancing relationship. In this way, in the detection scenario with continuous time, high-risk impurities are continuously suppressed from being absorbed into the reference representation again and causing risk accumulation and diffusion.

[0029] S300, based on the risk pointing line, rearranges and controls the order of image composition in the subsequent image generation process, so that the image content corresponding to the risk pointing line is presented forward in the generation process and forms a counterbalancing relationship with the morphologically stable area, thereby limiting the abnormal content from being incorporated into the reference expression again. To prevent anomalous content already identified as impurities from being smoothly absorbed back into the reference representation during subsequent image generation, the order of image composition in the subsequent image generation process is specifically rearranged and controlled based on the formed risk indicator lines. This causes the image content corresponding to the risk indicator lines to be actively shifted forward during generation, changing from passive superposition to active presentation, and forming a continuous counterbalancing relationship with the morphologically stable region in the temporal dimension. This restricts the integration path of anomalous content without disrupting the overall continuity of the reference representation. The specific implementation steps are as follows: Based on the risk indicator lines already formed in the temporal matrix and continuously pointing towards the evolution direction of the reference expression, images about to enter the reference expression update sequence in subsequent image generation processes undergo pre-association processing. The spatial positional relationship and evolutionary direction relationship between each image to be generated and the risk indicator lines in the temporal matrix are correspondingly indexed, ensuring that the degree of association between each image and the risk indicator lines is clear before entering the generation process. In this process, the spatial range covered by the risk indicator lines and their trajectory extending along the temporal evolution direction are used as reference coordinates for controlling the image composition order, indicating which image content has a continuous evolutionary relationship with historical impurity signs and which image content mainly corresponds to stable expression regions of morphologically stable areas. Through this association method, the subsequent image generation process no longer proceeds according to a single temporal order or natural generation order, but introduces a structural sorting basis based on the risk indicator lines, providing a clear regulatory foundation for subsequent order rearrangement.

[0030] After the subsequent images are associated with the risk indicator lines, the order of image composition in the subsequent image generation process is rearranged and controlled, so that image content with a high degree of correlation with the risk indicator lines is presented earlier in the generation process. Specifically, risk-related image content that was originally located in the later part of the generation process and was easily covered by the expression of the morphologically stable region is adjusted to the earlier part of the generation process, so that it can enter the temporal carrying base first and participate in the construction of the reference expression when the reference expression is updated. Through this forward presentation method, the image content corresponding to the risk indicator lines gains a higher exposure position in the temporal dimension and is no longer diluted by a large number of subsequent stable morphological expressions, thereby avoiding the rapid smoothing and reabsorption of abnormal content into the reference expression during the generation process. At the same time, image content that is highly consistent with the morphologically stable region is kept in the relatively later part of the generation process, so that it enters the reference expression update process after the forward-moved risk content, reserving space for the formation of hedging relationships later.

[0031] After the image content corresponding to the risk indicator line is presented forward, the forward-presented risk-related image content and the morphologically stable region are temporally juxtaposed, creating a counterbalancing relationship between the two during the reference representation update process. Specifically, within the same time frame, the image content corresponding to the risk indicator line and the stable representation represented by the morphologically stable region simultaneously participate in the construction of the reference representation. This allows the reference representation to simultaneously carry both deviation trends and stable trends within that time frame, thus creating a state of mutual restraint in the temporal dimension. Under this counterbalancing relationship, the morphologically stable region no longer unilaterally dominates the update direction of the reference representation, and the image content corresponding to the risk indicator line is no longer passively integrated into the stable representation, but rather continues to exist in the evolutionary state of the reference representation in a contradictory evolutionary manner. By temporally juxtaposing rather than spatially eliminating the content, the abnormal content maintains its visibility and independence in the reference representation, thereby limiting the abnormal content from being hidden and integrated into the morphologically stable region again.

[0032] Based on the formation and stable existence of the hedging relationship, the rearrangement and regulation results of the image composition order are continuously applied to the subsequent image generation process. This ensures the continuity of the directional effect of the risk indicator line within the temporal basis, and that each batch of subsequently generated images is constrained by both the forward presentation and the hedging relationship when entering the reference representation update process. Through this continuous sequential rearrangement regulation, the reference representation remains in a dynamic equilibrium state throughout the temporal progression. This means that while maintaining the normal appearance of steviol glycoside crystals through morphologically stable regions, the forward presentation of content corresponding to the risk indicator line continuously exposes abnormal trends. This, in turn, limits the reabsorption and solidification of abnormal content into part of the reference representation during the continuous image generation process. Through the synergistic effect of the aforementioned sequential rearrangement regulation and the hedging relationship, the evolutionary path of the reference representation is actively guided within the temporal dimension, preventing early anomalies or historical shifts from being continuously amplified in subsequent generation processes. This provides a stable and extensible foundation for further control over the reference representation update rhythm.

[0033] S400, based on the hedging relationship, rhythmically decomposes the update rhythm of the reference image, so that the reference representation presents a segmented breathing change state in the time dimension, thereby weakening the long-term dominant role of the single generation result on the reference representation. To weaken the long-term dominance of a single generated result on the reference representation over time during continuous image generation, after the risk indicator line has been moved forward and its relationship with the morphologically stable region has been balanced, the update rhythm of the reference image is rhythmically broken down. This prevents the reference representation from being updated in a continuous, superimposed manner, instead exhibiting a segmented, breathing change over time. This allows the reference representation to possess a slowed and adjustable evolutionary characteristic during continuous generation, preventing anomalous content or localized expressions from dominating over time. The specific implementation steps are as follows: Based on the established hedging relationship, the continuous image generation process is rhythmically divided along the time dimension, breaking down the update process of the reference image into multiple interconnected but independent time update segments. In this process, each time update segment corresponds to a certain number of subsequently generated images, ensuring that the reference representation only absorbs a limited range of generated results within each time update segment, rather than continuously and uninterruptedly superimposing all generated results into a single reference image. This rhythmic division creates clear stage boundaries for the reference representation along the time dimension, allowing the hedging relationship formed by the forward-shifting risk indicator line to be preserved and function independently within each time update segment, preventing gradual dilution by a large amount of subsequently generated content during long-term continuous updates.

[0034] After dividing the time update segments, the update method of the reference images within each time update segment is rhythmically controlled, causing the reference representation to exhibit an alternating state between different time update segments. In specific implementation, when the reference representation is within a certain time update segment, only the generated images within the current segment are allowed to participate in the construction of the reference representation, while the reference representation formed in the previous time update segment is maintained in a relatively stable historical state and does not directly participate in the update and superposition within the current segment. In this way, the reference representation exhibits a breathing-like change characteristic as time progresses, that is, it gradually absorbs new generated content within a certain time update segment, maintains brief stability when switching segments, and then enters the update process of the next time update segment. This forms a segmented breathing change state in the overall time dimension, preventing the reference representation from being dominated by the generation results of a single stage for a long time due to continuous superposition.

[0035] Based on the segmented breathing changes in the reference expression, the hedging relationship formed by the aforementioned risk indicator line and the morphologically stable region is embedded into each time update segment, allowing the hedging relationship to independently exert a suppressive effect within each time update segment. Specifically, within each time update segment, the image content corresponding to the risk indicator line and the stable expression corresponding to the morphologically stable region simultaneously participate in the construction of the reference expression within that segment, ensuring that the reference expression maintains a state where deviation trends and stable trends coexist within the segment. When a time update segment ends and enters the next, the hedging state formed in the previous segment is carried over as a whole and retained as historical expression, rather than being directly broken up or re-overlaid. Through this method of intra-segment hedging and inter-segment carrying, the risk indicator line continuously constrains the reference expression in the time dimension, preventing abnormal content from being locally smoothed in a certain time update segment and then regaining dominance in subsequent segments.

[0036] After completing the rhythmic decomposition of the reference image update rhythm and forming a stable segmented breathing change state, this update method is continuously applied to the subsequent image generation process, ensuring that the reference representation maintains its rhythmic update characteristics throughout the entire time progression. Through this continuous rhythmic decomposition, no single generated result or generated content within a single time segment can exert a long-term dominant influence on the reference representation, thus maintaining a dynamic equilibrium state of the reference representation over the overall time dimension. In this state, the morphologically stable region can continuously provide a stable reference foundation, the risk indicator line can continuously expose potential abnormal trends, and the reference representation, through segmented breathing changes, continuously releases the expression pressure accumulated within a single segment, making it difficult for abnormal content to continuously accumulate and solidify into part of the reference representation over time. Through this method, the update process of the reference image is transformed from continuous superposition to rhythmic evolution, laying a stable and coherent foundation for subsequently introducing dynamic contraction and expansion control and further controlling the participation weight of the reference image in the generation process.

[0037] S500, based on the segmented breathing change state, periodically compresses and releases the participation weight of the reference image in the image generation process according to the production rhythm, so that high-risk impurities continuously leave the reference expression range as time progresses, thereby suppressing the cumulative amplification of impurity risk in the time dimension. Based on the segmented breathing state formed by the rhythmic decomposition of the reference representation through update rhythm, a dynamic control method adapted to the production rhythm is introduced. This method periodically compresses and releases the participation weight of the reference image during image generation, ensuring that the reference representation is not always in a high-participation state as time progresses, but rather exhibits orderly changes in participation intensity at different production stages. This allows high-risk impurities to gradually escape the reference representation's range during the continuous generation timeline, preventing the accumulation and amplification of impurity risks over time. The specific implementation steps are as follows: Based on the established segmented breathing change state, the continuous production process is divided into several interconnected production stages according to the actual production rhythm, so that each production stage corresponds to several reference expression update segments in the time dimension. In this process, the segmented breathing change state, as the basic time structure of the reference expression, is used to carry the change information of the production rhythm, so that the update of the reference expression does not operate independently of the actual production process. By introducing the production rhythm into the time structure of the reference expression, the participation role of the reference image in different production stages has stage-specific differences, providing a clear time reference framework for the subsequent periodic compression and release of participation weights.

[0038] After dividing the production process into stages, the participation weight of the reference image in the image generation process is compressed within each stage, thus actively weakening the guiding role of the reference image in the newly generated image. In practice, when the reference representation is at the beginning or transitional stage of a production process, by reducing the participation intensity of the reference image, the newly generated image reflects the morphological characteristics of the current sample more accurately in appearance, rather than being overly constrained by historical reference representations. This compression of participation weights makes it less likely that potentially high-risk impurities will be masked or smoothed out by historical reference representations in the newly generated image, gradually distancing themselves from the reference representation over time and creating conditions for high-risk impurities to detach from the reference representation.

[0039] After participating in weighted compression for a period of time and completing the corresponding production stage, the participation weight of the reference image is released in conjunction with the segmented breathing change state. This allows the reference image to re-participate in the image generation process in subsequent production stages, but this participation is influenced by the compression state of the previous stage and no longer forms a continuous dominant force. During the release process, the reference image participates in the generation process in a phased recovery manner, guiding the newly generated image only within a limited time segment, and then entering a relatively slow-release state after the segment ends. Through the alternating effects of compression and release in the time dimension, the reference representation exhibits periodic expansion and contraction changes throughout the production process, rather than maintaining a single participation intensity for a long time. This avoids the generation result of a certain historical stage continuously dominating the reference representation in multiple production stages.

[0040] By repeatedly compressing and releasing the participation weight of the reference image during continuous production, high-risk impurities gradually escape from the reference representation range over time and maintain an opposing relationship with the morphologically stable region under the constraint of segmented breathing changes. Because the participation intensity of the reference image changes continuously at different production stages, high-risk impurities are difficult to be fixedly absorbed as part of the reference representation at a certain stage. Even if they enter the reference representation boundary in a short period, they will be gradually eliminated in subsequent participation weight compression stages. Through this dynamic compression and release control method combined with the production rhythm, the diffusion path of impurity risk in the time dimension is continuously interrupted, thereby suppressing the trend of impurity risk accumulating and amplifying in multiple production stages and continuous generation, ensuring that the reference representation maintains its ability to repel high-risk impurities throughout long-term operation.

[0041] This invention introduces a time-based foundation, a risk indicator line, and rhythmic control of update frequency. This prevents the reference expression from evolving based on the continuous superposition of single generation results. Instead, it forms a distinguishable and controllable evolutionary state within the time dimension, effectively preventing early anomalies from being inherited and solidified in subsequent generation processes. By maintaining the opposing relationship between deviation and stability trends within the reference expression, impurity-related expressions remain identifiable and controllable, improving the stability of steviol glycoside impurity detection in continuous production scenarios and providing a consistent basis for judgment across different time stages.

[0042] This invention combines the participation weight of the reference image in the image generation process with the production rhythm, implementing periodic compression and release. This allows high-risk impurities to continuously escape the reference representation range over time, preventing the gradual accumulation and amplification of impurity risks across multiple production stages. This method endows the reference representation with self-mitigation capabilities, continuously suppressing the covert integration of anomalous content during long-term operation. This improves the adaptability of steviol glycoside product impurity detection to complex production fluctuations, ensuring the reliability and sustainability of the detection process under long-term operating conditions.

[0043] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for detecting impurities in steviol glycoside products based on image generation, characterized in that, Includes the following steps: S100: The historical generated images formed during the continuous image generation process are analyzed as a whole according to the temporal evolution relationship, and the morphologically stable region and the hidden drift region formed over time are distinguished and extracted to construct a time-bearing basis. S200, based on the time-bearing base, compares and analyzes newly entered images in the image generation process with morphologically stable regions, identifies impurity signs that show a deviation trend in the current stage and have been absorbed in the historical generation process, and extends the impurity signs along the time evolution direction to form a risk pointing line; S300, based on the risk indicator line, rearranges and controls the order of image composition in the subsequent image generation process, so that the image content corresponding to the risk indicator line is presented forward in the generation process and forms a counteracting relationship with the morphologically stable region. S400, based on the hedging relationship, rhythmically decomposes the update rhythm of the reference image, so that the reference expression presents a segmented breathing change state in the time dimension. S500, based on segmented breathing changes, periodically compresses and releases the participation weight of the reference image in the image generation process according to the production rhythm, so that high-risk impurities continuously leave the reference expression range as time progresses.

2. The method for detecting impurities in steviol glycoside products based on image generation according to claim 1, characterized in that, Step S100 includes: Historical generated images generated during the continuous image generation process are uniformly collected and a time series relationship is constructed according to the order in which the images were generated. Based on the time series relationship, the morphological change trend presented in the historical generated images is compared and analyzed. Image regions that appear repeatedly in multiple consecutive time nodes are extracted as morphological stable regions, and image regions that show gradual change characteristics in the time series are extracted as hidden drift regions. Based on the time dimension, the distribution of morphologically stable regions and hidden drift regions in historically generated images is uniformly processed, so that the reference representation states at different time stages form a continuous and connected hierarchical structure. Based on unified carrying capacity processing, a time-carrying base is constructed to characterize the evolutionary state of the reference expression. The morphologically stable regions and hidden drift regions formed at different time stages are incorporated into the time-carrying base for continuous recording and updating.

3. The method for detecting impurities in steviol glycoside products based on image generation according to claim 2, characterized in that, In the process of constructing the time-bearing substrate, the original generation state of the historical generated images is maintained, so that the morphologically stable region and the hidden drift region are incorporated into the time-bearing substrate in a continuous mapping manner in the time series, and the evolution trajectory of the reference expression is carried through the time dimension, so that the reference expression is transformed from a static expression into an overall state with time evolution characteristics.

4. The method for detecting impurities in steviol glycoside products based on image generation according to claim 2, characterized in that, Step S200 includes: Based on the time-bearing substrate, newly entering images in the image generation process are connected to the end time node of the time-bearing substrate in the order of entry, so that the newly entering images in the image generation process maintain a continuous temporal evolution relationship with the historically generated images. Based on the time-bearing basis, the newly entered image in the image generation process is compared and analyzed with the morphologically stable region. Deviation trend candidate regions that deviate from the morphologically stable region are extracted and written into the evolution record of the time-bearing basis. Based on the time-bearing basis, the historical evolution trajectory of the deviation trend candidate region and the hidden drift region is correlated to identify the impurity signs absorbed in the historical generation process. Based on the extension of impurity signs along the direction of time evolution, risk pointing lines are formed in the time-bearing substrate to characterize the risk evolution path.

5. The method for detecting impurities in steviol glycoside products based on image generation according to claim 4, characterized in that, In the process of forming the risk indicator line, the spatial position of the impurity signs in the time-bearing substrate is continuously mapped to the morphological boundary, so that the risk indicator line maintains a correspondence with the morphologically stable region in the direction of temporal evolution, and the deviation direction and deviation range of the impurity signs relative to the morphologically stable region are recorded, so as to enhance the indicative role of the risk indicator line in the evolutionary state of the reference expression.

6. The method for detecting impurities in steviol glycoside products based on image generation according to claim 4, characterized in that, Step S300 includes: Based on the time-bearing basis, the images to be entered into the reference representation update sequence during the subsequent image generation process are associated with the risk pointing line, so that the relationship between the images to be generated and the risk pointing line in the direction of temporal evolution is indexed; Based on the association results, the order of image composition in the subsequent image generation process is rearranged so that the image content associated with the risk direction line is presented forward in the generation process. Based on the rearrangement of the image composition order, the risk pointing line presented in the front is made to participate in the construction of reference expression in the same time period as the image content and the morphological stable area, thereby forming a hedging relationship. After the hedging relationship is formed, the rearrangement and regulation of the image composition order continues to act on the subsequent image generation process, so that the directional effect of the risk pointing line in the time-bearing substrate remains continuous, thereby limiting the reintegration of abnormal content into the reference expression.

7. The method for detecting impurities in steviol glycoside products based on image generation according to claim 6, characterized in that, During the process of rearranging the order of image composition, the forward shift of the image content corresponding to the risk pointing line in the time-bearing base continuously spans multiple time segments, and maintains a temporal juxtaposition relationship with the morphologically stable area in each time segment, so that the image content corresponding to the risk pointing line is continuously in a hedging state during the evolution of the reference expression.

8. The method for detecting impurities in steviol glycoside products based on image generation according to claim 6, characterized in that, Step S400 includes: Based on the hedging relationship formed by the risk indicator line and the stable region, the continuous image generation process is rhythmically divided in the time dimension, so that the update process of the reference image is split into interconnected time update segments. Based on time update segments, the update method of the reference representation in each time update segment is rhythmically controlled so that the reference representation absorbs and generates images within the segment and maintains a stable state when switching segments, forming a segmented breathing change state. Based on the segmented breathing change state, the hedging relationship between the risk direction line and the stable morphology zone is embedded into each time update segment, so that the hedging relationship can act independently in each time update segment. Based on the segmented respiratory change state, the rhythmic update method is continuously applied to the subsequent image generation process to weaken the long-term dominant role of a single generation result on the reference representation.

9. The method for detecting impurities in steviol glycoside products based on image generation according to claim 8, characterized in that, Under segmented breathing change state, the reference representation formed in each time update segment is carried as a whole as a historical representation and participates in the evolution constraint of subsequent time update segments, so that the reference representation maintains the continuous correlation between stages in the process of time advancement, while limiting the image generated in any time update segment from continuously dominating the reference representation.

10. The method for detecting impurities in steviol glycoside products based on image generation according to claim 8, characterized in that, Step S500 includes: Based on the segmented breathing change state, the continuous production process is divided into interconnected production stages according to the production rhythm, so that the corresponding reference expression update segment of each production stage is updated. Based on the production stage, the participation weight of the reference image in the image generation process is compressed in each production stage, so that the guiding role of the reference image on the newly generated image is reduced in the corresponding stage. Based on the segmented breathing change state, the participation weight of the reference image is released after the production stage switch, so that the reference image can re-participate in the image generation process within a limited time period. By alternately compressing and releasing the participation weights during continuous production, high-risk impurities are continuously removed from the reference expression range over time.

Images