Legend curve matching method and system for power amplifier typical characteristic diagram

Abstract

The application discloses a legend curve matching method and system for a typical characteristic diagram of a power amplifier, and the method comprises the following steps: extracting multi-dimensional features of each legend item and each candidate curve respectively, and generating a legend feature set and a curve feature set; performing constraint matching on the legend items and the candidate curves based on the legend feature set and the curve feature set; constructing a multi-dimensional joint matching representation based on the pairing relationship between each group of legend items and candidate curves; calculating adaptive weights of each dimension according to image quality factors corresponding to each feature dimension, and then generating a joint matching score matrix between the legend items and the candidate curves; taking a candidate legend item set and a candidate curve set as nodes on two sides of a two-part graph respectively, taking joint matching scores as edge weights, and constructing a legend-curve two-part graph matching model; solving the legend-curve two-part graph matching model to generate a globally optimal matching matrix; calculating matching confidence of each pairing relationship, and outputting a matching result containing an identity label.

Description

Technical Field

[0001] This invention relates to the field of radio frequency device test data processing technology, and in particular to a method and system for matching legend curves of typical characteristic plots of power amplifiers. Background Technology

[0002] Typical characteristic graphs of power amplifiers are used to characterize the electrical performance variations of devices under different frequency, power, or bias conditions. Common curves include output power, gain, power-added efficiency, and S-parameter curves. These graphs typically consist of a plotting area, coordinate axes, scales, legend entries, parameter text, and multiple curves. For test data processing and result verification, the core task lies not only in extracting curve outlines from the graph but also in establishing a stable one-to-one correspondence between each curve and its corresponding legend entry, parameter name, and test conditions. Only after completing this mapping can subsequent parameter comparisons, anomaly localization, and trend analysis have clear engineering significance.

[0003] The closest existing technology to this invention typically employs a processing flow of "drawing area localization + OCR recognition of legend text + color / line type similarity calculation + fixed weight scoring + local optimal allocation". This type of scheme first separates the drawing area and legend area from the typical feature map, extracting candidate curves and candidate legend items; then, it calculates the color difference between legend color blocks and curve pixels, and the similarity between legend line type and curve local texture, and combines the legend text recognition results to form a comprehensive score; finally, according to the maximum score principle, a certain legend item is assigned to the candidate curve with the highest score, or, in the event of a conflict, it is corrected using rules such as nearest neighbor replacement or manual threshold backoff.

[0004] The aforementioned closest existing technology has at least the following obvious defects: First, the color, line type, and text features in its comprehensive score usually use fixed weights. When the image has color distortion, compression noise, broken lines, or blurred text, the fixed weights cannot adaptively adjust with changes in image quality, which can easily lead to a certain distortion feature having an excessive impact on the final result. Second, its matching process is usually based on a local greedy allocation of a single legend item or a single curve. When multiple curves have similar colors, local intersections, or closely arranged legends, a chain of mismatches can easily occur, where the previous round of allocation occupies the best candidate, and the next round can only be forced to select the second-best candidate. Third, it mostly makes judgments based on the appearance features of the image, lacking constraints on the consistency of the curve trend after physical mapping, and also lacking quantitative output of the uncertainty of the result. Therefore, in scenarios where similar curves coexist, it often directly outputs incorrect identity labels, making it difficult to provide an effective basis for subsequent manual review. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for legend curve matching of typical characteristic plots of power amplifiers, so as to solve the problems of strong dependence on fixed rules and fixed weights, insufficient global consistency in complex scenarios, and insufficient expression of uncertainty in the results in the legend matching process of the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] The first aspect is a method for matching legend curves for typical characteristic plots of power amplifiers, which includes the following steps:

[0008] S1. Collect and preprocess the typical characteristic diagrams of the original power amplifier to generate a set of candidate curves and a set of candidate diagram examples;

[0009] S2. Extract the multi-dimensional features of each legend item and each candidate curve to generate the legend feature set and the curve feature set; the multi-dimensional features include color features, line type features, spatial location features, text semantic features and trend features;

[0010] S3. Constraint matching is performed on legend items and candidate curves based on legend feature set and curve feature set; a multi-dimensional joint matching representation is constructed based on the pairing relationship between each set of legend items and candidate curves; adaptive weights of each dimension are calculated according to the image quality factor corresponding to each feature dimension, and then a joint matching score matrix between legend items and candidate curves is generated.

[0011] S4. Using the candidate legend item set and candidate curve set as the two sides of the bipartite graph respectively, and the joint matching score as the edge weight, construct a legend-curve bipartite graph matching model; solve the legend-curve bipartite graph matching model to generate the globally optimal matching matrix;

[0012] S5. Based on the global optimal matching matrix, calculate the matching confidence of each pairing relationship and output the matching results including identity tags.

[0013] Preferably, S1 includes:

[0014] Obtain the original typical characteristic map of the power amplifier; perform preprocessing operations on the original typical characteristic map of the power amplifier, including orientation correction, resolution normalization, color consistency correction, noise suppression, plotting area localization, legend area localization, and non-curve element stripping; in the preprocessed image, extract candidate curve objects and candidate legend objects through connected component analysis; for any connected region, calculate its trajectory continuity, edge consistency, internal color stability, and background noise interference intensity, and obtain a weighted candidate retention score; when the candidate retention score is greater than the preset retention threshold, the connected region is retained as a candidate curve object.

[0015] Preferably, S2 includes:

[0016] S21. Extract legend color features, legend line type features, legend text semantic features, and legend position features for each legend item; among them, the legend color features are represented by a comprehensive color mean vector and color covariance matrix; the legend line type features are represented by solid line length distribution, dashed line interval distribution, and local endpoint statistics; the legend text semantic features are represented by the semantic vector of the encoded legend text; and the legend position features are represented by the center coordinates of the legend box and its sorting index.

[0017] S22. Extract curve color features, curve line shape features, curve position features, and curve trend features for each candidate curve. Among them, curve color features are used to characterize the main color component and color dispersion of curve pixels; curve line shape features are used to characterize the continuous length distribution and interval pattern of the curve; curve position features are used to characterize the geometric positional relationship of the candidate curve relative to the drawing area and legend area; curve trend features are used to characterize the local rise and fall trend, peak and valley positions, and relative high and low relationships at key frequency points of the curve in the physical data space.

[0018] Preferably, S3 includes:

[0019] S31. For any legend item and candidate curve, perform color matching and initial screening of the candidate curve based on color features; generate a color candidate set corresponding to each legend item;

[0020] S32. For candidate curves in the color candidate set, perform line type consistency judgment and elimination based on line type features, and output the line type constraint candidate set;

[0021] S33. For candidate curves in the linear constraint candidate set, perform spatial adjacency judgment and retention based on spatial location features, and output the spatial constraint candidate set;

[0022] S34. For the candidate curves of the legend items, perform text constraint determination based on text semantic features and generate text constraint scores;

[0023] S35. When multiple candidate curves corresponding to the same legend item cannot be effectively distinguished in terms of color features, line type features, and spatial location features, or when the difference in text constraint scores between the best candidate curve and the second best candidate curve is lower than the preset difference threshold, auxiliary trend determination is performed based on trend features, and a trend constraint candidate set is output.

[0024] As a preferred method, color matching and preliminary screening of candidate curves based on color features includes: calculating the color distance between the legend color features and the curve color features; if the color distance is less than or equal to the preliminary color screening threshold, the candidate curve is considered to have passed the preliminary color screening of the legend item, and the candidate curve is included in the color candidate set corresponding to the legend item.

[0025] The line type consistency judgment and elimination based on line type features includes: comparing the legend line type features and the curve line type features to obtain a line type consistency score; if the line type consistency score is less than the line type consistency threshold, the candidate curve is eliminated from the candidate curve set of the legend item.

[0026] The spatial adjacency determination and retention based on spatial location features includes: establishing a spatial adjacency consistency score based on the relative position of the legend item in the legend area and the representative position of the candidate curve in the drawing area; if the spatial adjacency consistency score is lower than a preset spatial consistency threshold, the comprehensive prior weight corresponding to the candidate curve is reduced; if the spatial adjacency consistency score is higher than the preset spatial consistency threshold, the candidate curve is retained.

[0027] As a preferred method, performing text constraint determination based on text semantic features includes:

[0028] Semantic encoding is performed on the text content of the legend items to obtain the legend text semantic vector. Construct a semantic constraint vector for the curve by combining coordinate axis text, unit text, and adjacent curve annotation information. ; Calculate legend items With candidate curve Textual constraint score :

[0029] ;

[0030] in, Candidate curve The semantic vectors of the coordinate axes in the current coordinate system. Candidate curve The corresponding unit compatibility vector, Candidate curve The associated test condition semantic vector, to Indicates the weight of the text constraint item. Represents the cosine similarity function. This represents the hit determination function; if the candidate curve If the semantics of the coordinate axes or test conditions are incompatible with the parameter names or test condition keywords contained in the legend text, the text constraint score will be reduced. .

[0031] As a preferred method, auxiliary trend determination based on trend characteristics includes: selecting candidate curves based on coordinate axis scales and unit information. The pixel coordinates are mapped to the physical data space to obtain the candidate curve. physical trajectory The mapping relationship is represented as:

[0032] ;

[0033] in, and These represent the x and y coordinates of the candidate curve pixels, respectively. and These represent the mapped physical x and y coordinates, and These represent the scaling coefficients for the horizontal and vertical axes, respectively. and These represent the corresponding offsets.

[0034] Based on physical trajectory Construct candidate curves trend residual Similarity to trends Trend residual Used to characterize the deviation between the candidate curve and the reference fitted trajectory; if legend items Based on the existing parameter-condition prototype trajectory, candidate curves are further determined by trend similarity. The degree of consistency with the prototype trajectory.

[0035] Preferably, S3 includes:

[0036] For the legend item With candidate curve Candidate relationships between them, construct joint representation vectors :

[0037] ;

[0038] in, Indicates color matching response, Indicates a linear matching response. Indicates spatial matching response, Indicates a text constraint response. Indicates the trend constraint response;

[0039] Based on the color quality, line quality, character recognition quality, spatial stability, and trend reliability corresponding to the current candidate relationship, calculate the quality factor. , And further calculate adaptive weights , :

[0040] ;

[0041] Where Ω represents the set of feature categories, Indicates the first Sensitivity parameters of class features, , Indicates the first Class features in candidate relations ( , The quality factor under ( )

[0042] Based on the joint representation vector and adaptive weights , Calculate the legend item With candidate curve Joint matching score matrix between :

[0043] .

[0044] As a preferred embodiment, S5 includes:

[0045] S51. Based on the global optimal matching matrix, determine the optimal legend item corresponding to each candidate curve, and generate the identity label of the candidate curve from the parameter name and test conditions in the legend item.

[0046] S52. For each candidate curve, construct a matching probability distribution based on the matching score distribution of the legend items; calculate the matching confidence based on the matching probability distribution and the trend residual of the candidate curve.

[0047] Secondly, a legend matching system for typical characteristic curves of power amplifiers includes:

[0048] The curve space extraction module is used to collect and preprocess the typical characteristic diagrams of the original power amplifier, and generate a set of candidate curves and a set of candidate diagram examples.

[0049] The feature extraction module is used to extract multi-dimensional features of each legend item and each candidate curve, and generate legend feature set and curve feature set; multi-dimensional features include color features, line type features, spatial location features, text semantic features and trend features;

[0050] The joint representation generation module is used to perform constraint matching on legend items and candidate curves based on legend feature sets and curve feature sets; and to construct multi-dimensional joint matching representations based on the pairing relationship between each set of legend items and candidate curves.

[0051] The quality-aware weight generation module is used to calculate the adaptive weights of each dimension based on the image quality factors corresponding to each feature dimension, and then generate a joint matching score matrix between legend items and candidate curves.

[0052] The global consistency matching module is used to construct a legend-curve bipartite graph matching model by using the candidate legend item set and the candidate curve set as the two sides of the bipartite graph, respectively, and the joint matching score as the edge weight; the legend-curve bipartite graph matching model is solved to generate the globally optimal matching matrix;

[0053] The output module is used to calculate the matching confidence of each pairing relationship based on the global optimal matching matrix and output the matching results including identity tags.

[0054] The legend curve matching system for typical characteristic plots of power amplifiers is used to implement the legend curve matching method and steps for typical characteristic plots of power amplifiers as described in the first aspect.

[0055] Compared with the prior art, the beneficial effects of the present invention are reflected in:

[0056] 1. Unlike traditional techniques that employ "color threshold matching, OCR recognition, and local nearest neighbor correction," this invention combines legend item-candidate curve joint representation, quality-aware adaptive weighting, and global consistency optimization. For any legend item and candidate curve, it constructs color features, line type features, text semantic features, relative position features, and trend features after physical mapping. A joint matching score matrix is ​​established within a unified feature space, allowing the contribution of different features to the matching result to dynamically change with image quality. Furthermore, it elevates the allocation relationship between legend items and curves from local rule selection to global consistency optimization. Therefore, even in scenarios where multiple curves have similar colors, local intersections, blurred text, or broken lines, it can effectively reduce mismatch rates and improve the stability of identity assignment. This invention does not simply combine color discrimination, OCR, and graph matching in parallel; instead, it couples a "quality factor-driven joint representation weighting" with a "global optimization objective with conflict penalty," resulting in a clear structural improvement and significant technical effects.

[0057] 2. Unlike traditional techniques that directly output identity labels after obtaining the highest-scoring candidate or use only a single threshold to mark anomalies, this invention employs a combined approach of physical space trend constraints and uncertainty-coupled evaluation. After the candidate curve completes the mapping from pixel coordinates to physical coordinates, a trend residual is constructed first, and then a confidence function is established by combining the matching score distribution. This invention can not only output the final identity label but also quantitatively distinguish between "high-confidence matching results" and "results requiring manual verification." Therefore, in complex scenarios with similar colors, similar trends, or strong abbreviations in legend text, it can effectively suppress mislabeling and enhance the interpretability of the results. This secondary inventive point is not simply adding an empirical threshold to the traditional maximum score method, but rather jointly modeling the score interval, distribution entropy, and physical space trend residual, making the result credibility computationally calculable, interpretable, and verifiable. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the legend matching process in the typical characteristic diagram of the power amplifier of the present invention.

[0059] Figure 2 The diagram below illustrates the typical characteristics of the power amplifier of this invention, including the matching input, preprocessing, candidate extraction, and initial screening process.

[0060] Figure 3 This is a schematic diagram of the multi-feature fusion matching process for the typical characteristics of the power amplifier of the present invention.

[0061] Figure 4 This is a schematic diagram of the global consistency optimization process for matching legends in the typical characteristic diagram of the power amplifier of the present invention.

[0062] Figure 5 This is a schematic diagram illustrating the integrated processing of curve space extraction, curve recognition and correction, data space physical mapping, and legend semantic matching and identity assignment of the present invention. Detailed Implementation

[0063] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0064] For consistency throughout the text, in this specific implementation, the "typical characteristic diagram of the original power amplifier" will be uniformly referred to as... The preprocessed images are uniformly denoted as The set of candidate legend items is uniformly denoted as The candidate curve set is uniformly denoted as Legend item With candidate curve The joint matching score between them is denoted as .

[0065] Example 1:

[0066] like Figures 1 to 4 The method for matching legend curves for typical characteristic plots of a power amplifier, as shown, includes the following steps:

[0067] S1. Collect and preprocess typical characteristic diagrams of the power amplifier to generate a set of candidate curves and a set of candidate diagram examples; such as... Figure 2 As shown, it specifically includes:

[0068] S11. Collect typical characteristic diagrams of the original power amplifier;

[0069] Specifically, obtain the original typical characteristic diagram of the power amplifier, which includes a plotting area, coordinate axes, scales, legend area, and multiple curves. The original typical characteristic diagram of the power amplifier can be obtained from a graph exported from the test software, a screenshot of the test report, or a formatted test image. The input for this sub-step is the original image. The output is the image to be preprocessed.

[0070] S12. Preprocessing includes: performing orientation correction, resolution normalization, color consistency correction, noise suppression, plotting area positioning, legend area positioning, and non-curved element removal on the original power amplifier typical characteristic map to obtain the preprocessed image. Subsequently, in the preprocessed image Extract candidate curve objects and candidate legend item objects.

[0071] To improve the reliability of candidate curves, for any connected region u, a candidate curve is constructed that retains its score. :

[0072] ;

[0073] in, Represents connected regions trajectory continuity, Represents connected regions Consistency with the curved edge model, Represents connected regions Internal color stability This represents the intensity of background noise interference in the connected region u. to This represents the corresponding weight parameter. When... Greater than or equal to the candidate retention threshold At that time, connect the regions Retain it as a candidate curve object.

[0074] This sub-step outputs a set of candidate curves. and candidate legend item set The candidate curve set The candidate legend item set serves as input for subsequent curve feature extraction. As input for subsequent feature extraction of the legend.

[0075] S2. Extract the multi-dimensional features of each legend item and each candidate curve respectively, and generate the legend feature set and the curve feature set;

[0076] like Figure 3 As shown, features such as color, text, line style, and position of the legend items and candidate curves are extracted. The input for this step is the set of candidate curves. and candidate legend item set The output consists of a legend feature set and a curve feature set.

[0077] S21. For the candidate legend item set Each legend item in Extracting legend color features Legendary line type characteristics Semantic features of illustration text and the location features of the legend Among them, the legend's color features It can be represented by a combination of color mean vector and color covariance matrix, with the following line type characteristics. The distribution of solid line lengths, the distribution of dashed line intervals, and local endpoint statistics can be used to represent the semantic features of the legend text. The legend text can be represented by a semantic vector after encoding, and the legend's positional features can be used. The coordinates of the center of the legend box and its sorting index can be used to represent it.

[0078] S22, For the candidate curve set Each candidate curve in Extracting curve color features Curve shape characteristics Curve position characteristics and the characteristics of the curve trend Among them, the curve color feature Used to characterize the primary color component and color dispersion of curve pixels, curve line type characteristics. Used to characterize the continuous length distribution and interval patterns of a curve, and the positional characteristics of the curve. Used to characterize the geometric positional relationship of candidate curves relative to the plotting area and legend area, and the curve trend characteristics. It is used to characterize the local rise and fall trend of the curve in the physical data space, the position of the peak and valley, and the relative high and low relationship at key frequency points.

[0079] S3. Constraint matching is performed on legend items and candidate curves based on legend feature set and curve feature set; a multi-dimensional joint matching representation is constructed based on the pairing relationship between each set of legend items and candidate curves; adaptive weights of each dimension are calculated according to the image quality factor corresponding to each feature dimension, and then a joint matching score matrix between legend items and candidate curves is generated.

[0080] S31. Perform initial color screening and candidate set establishment.

[0081] For any legend item With candidate curve Calculate the color distance between the legend color feature and the curve color feature. :

[0082] ;

[0083] in, Legend item The color mean vector, Candidate curve The color mean vector, and These represent the corresponding color covariance matrices. This represents the regularization term used to ensure the invertibility of a matrix.

[0084] like Less than or equal to the initial color screening threshold Then the candidate curve is considered Through the legend items Initial color screening, and inclusion of it in the legend item. Corresponding color candidate set The output of this sub-step is the color candidate set corresponding to each legend item. This provides input for subsequent line type consistency judgment.

[0085] S32, Perform a line type consistency check.

[0086] For color candidate set Any candidate curve in Compare the legend items With candidate curve Linear statistical characteristics between them yield a linear consistency score. :

[0087] ;

[0088] in, Legend item The linear length histogram features. Candidate curve The linear length histogram features. Legend item Interval statistics, Candidate curve Interval statistics, and These represent the normalization coefficients for linear distribution differences and interval differences, respectively.

[0089] like Less than the linear consistency threshold Then the candidate curve From the legend items Remove from the candidate set; otherwise, retain the candidate curve. Then it proceeds to determine spatial adjacency. The input for this step is the color candidate set. The output is the candidate set after linear constraints. .

[0090] S33, Perform spatial adjacency relationship judgment.

[0091] According to the legend item The relative position in the legend area, and the candidate curve Establish spatial adjacency consistency scores at representative locations within the plotting area. :

[0092] ;

[0093] in, Legend item The normalized position vector in the legend region, Candidate curve The representative position vector in the drawing area. Represents the spatial distance normalization coefficient. This indicates the consistency factor between the legend item order and the relative order of the candidate curves. When the order of the legend items is consistent with the relative distribution order of the candidate curves in the plotting area, Take the larger value; otherwise, take the smaller value.

[0094] like Below the spatial consistency threshold This reduces the candidate curve. The corresponding comprehensive prior weights; if If the value exceeds the threshold, the candidate relation is retained and proceeds to the text constraint determination stage. The input for this step is the candidate set after linear constraints. The output is a spatially constrained candidate set. .

[0095] S34. Perform text constraint judgment.

[0096] For legend items The text content is semantically encoded to obtain the semantic vector of the legend text. Furthermore, a semantic constraint vector for the curve is constructed by combining coordinate axis text, unit text, and adjacent curve annotation information. .

[0097] Calculate legend items With candidate curve Textual constraint score :

[0098] ;

[0099] in, Candidate curve The semantic vectors of the coordinate axes in the current coordinate system. Candidate curve The corresponding unit compatibility vector, Candidate curve The associated test condition semantic vector, to Indicates the weight of the text constraint item. Represents the cosine similarity function. This indicates the hit determination function. If the legend text contains parameter names or test condition keywords, and the candidate curves... If the coordinate axis semantics or test condition semantics are incompatible with it, then reduce the speed. The output of this step is the candidate set after textual constraints. .

[0100] S35. Perform auxiliary trend determination when necessary.

[0101] When the same legend item The corresponding multiple candidate curves are still indistinguishable in terms of color, line type, and spatial location, or the score difference between the best candidate and the second-best candidate is lower than the preset difference threshold. At that time, auxiliary trend determination is triggered.

[0102] Based on the coordinate axis scale and unit information, candidate curves are selected. The pixel coordinates are mapped to the physical data space to obtain the candidate curve. physical trajectory The mapping relationship can be represented as:

[0103] ;

[0104] in, and These represent the x and y coordinates of the candidate curve pixels, respectively. and These represent the mapped physical x and y coordinates, and These represent the scaling coefficients for the horizontal and vertical axes, respectively. and These represent the corresponding offsets.

[0105] Based on physical trajectory Construct candidate curves trend residual Similarity to trends Among them, the trend residual Used to characterize the deviation between the candidate curve and the reference fitted trajectory:

[0106] ;

[0107] If the legend item If we can map this to an existing parameter-condition prototype trajectory, we can further determine candidate curves based on trend similarity. The degree of consistency with the prototype trajectory. The output of this step is the candidate set after trend constraints. .

[0108] S36, such as Figure 3 As shown, a joint representation is generated and the quality-aware weights are calculated.

[0109] For the legend item With candidate curve Candidate relationships between them are used to construct a multi-dimensional joint representation vector. :

[0110] ;

[0111] in, Indicates color matching response, Indicates a linear matching response. Indicates spatial matching response, Indicates a text constraint response. This indicates a trend constraint response.

[0112] For candidate relations For the k-th feature, first extract the original quality index corresponding to this feature. And through the normalization function Mapping it to the [0,1] interval yields the quality factor. Among them, the original quality index of color features is determined by the separation and contrast between legend color and curve color; the original quality index of line type features is determined by line continuity and texture clarity; the original quality index of text features is determined by OCR recognition confidence; the original quality index of spatial features is determined by the volatility of the relative positions of legend and curve; and the original quality index of trend features is determined by trend similarity. Adaptive weights are then calculated based on the quality factors corresponding to each type of feature. :

[0113] ;

[0114] Where Ω represents the set of feature categories, Indicates the first Sensitivity parameters of class features, , Indicates the first Class features in candidate relations ( , The quality factor under ().

[0115] Based on the joint representation vector and adaptive weights , Calculate the legend item With candidate curve Joint matching score between :

[0116] (15)

[0117] The input for this step is specifically the candidate set after trend constraints. The output includes the color matching response, line type matching response, spatial matching response, text constraint response, trend constraint response, and quality factor corresponding to each candidate legend item and candidate curve pair; the output is the legend item. With each candidate curve Joint matching score matrix between .

[0118] S4. Using the candidate legend item set and candidate curve set as the two sides of the bipartite graph, respectively, and the joint matching score as the edge weight, construct a legend-curve bipartite graph matching model; introduce a matching conflict penalty term into the legend-curve bipartite graph matching model, and solve for the globally consistent optimal matching relationship; such as Figure 4 As shown, it specifically includes:

[0119] S41, using the candidate legend item set As the set of left-hand nodes, with the set of candidate curves As the set of nodes on the right, to achieve joint matching scores As edge weights, construct a bipartite graph of legend items and candidate curves.

[0120] S42. Under the constraints that each legend item matches at most one candidate curve and each candidate curve matches at most one legend item, solve the global consistency optimization objective:

[0121] ;

[0122] in, Legend item With candidate curve Matching indicator variables between them Represents candidate matching relationships and) Conflict penalties when established simultaneously This represents the conflict suppression coefficient. After solving the optimization objective, the globally optimal matching matrix is ​​output. .

[0123] S5. Based on the globally optimal matching matrix, calculate the matching confidence of each pairing relationship and output the matching results. For example... Figure 4 As shown, it specifically includes:

[0124] S51. Based on the globally optimal matching matrix Determine each candidate curve The corresponding optimal legend item And by the legend item The parameter names and test conditions in the test generate candidate curves. Identity tags.

[0125] S52, For each candidate curve Construct a matching probability distribution based on the matching score distribution of the legend items. And calculate the matching confidence. :

[0126]

[0127]

[0128] in, Candidate curve At frequency sampling point Reference fitted trajectory at the location; Represents the trend residual of the candidate curve; Legend terms represent candidate curves The matching probability distribution; and Candidate curves are represented respectively. The corresponding best matching score and second-best matching score; This represents the score interval normalization coefficient; Represents the matching distribution entropy; This represents the entropy penalty coefficient; This represents the residual penalty coefficient for the trend. express Shape compression function; Candidate curve The matching confidence level.

[0129] S53. The output should include at least: candidate curve identifier, corresponding legend item identifier, identity label, joint matching score, matching confidence level, and a prompt indicating whether manual review is required. Below the confidence threshold At that time, the candidate curve Results marked as low confidence. Thus, Example 1 generates a typical characteristic plot of the original power amplifier. A complete method chain from input to identity label and confidence score output.

[0130] Example 2:

[0131] A legend curve matching system for typical characteristic plots of power amplifiers includes: a curve space extraction module, a feature extraction module, a joint characterization generation module, a quality-aware weight generation module, a global consistency matching module, and an output module.

[0132] The curve space extraction module is used to collect and preprocess the typical characteristic diagrams of the original power amplifier, and output the candidate curve set C and the candidate example set L.

[0133] The feature extraction module extracts the legend color features, legend line type features, legend text semantic features, legend position features, as well as the candidate curve color features, line type features, position features, and trend features.

[0134] The joint representation generation module is used to perform initial color screening, line type consistency judgment, spatial adjacency judgment, text constraint judgment, and auxiliary trend judgment, thereby generating a joint representation vector. ;

[0135] The quality-perceived weight generation module calculates adaptive weights. , This generates a joint matching score matrix between the legend items and the candidate curves;

[0136] The global consistency matching module is used to construct a bipartite graph of legend items and candidate curves and output the globally optimal matching matrix;

[0137] The output module is used to output the final identity label, matching confidence score, and verification prompts.

[0138] Example 3:

[0139] like Figure 5 As shown, this invention integrates curve space extraction from the original image, candidate curve identification and correction, physical mapping of the data space, and semantic matching and identity assignment of the legend into a unified processing link. Taking the typical characteristic diagram of the S22 power amplifier parameters as an example, the coordinate axes, scales, grid lines, legend boxes, and other non-curved elements in the original image are first located and stripped to obtain a set of candidate curves; then, local breakage correction, line type correction, and mapping of pixel coordinates to the physical data space are performed on the candidate curves to obtain the relative height, local rise and fall trends, and peak and valley distribution at key frequency points; finally, the physical semantic mapping and identity assignment between the legend items and the candidate curves are performed by combining the color, text, line type, and test condition information in the legend.

[0140] In this integrated processing, curve space extraction does not simply restore the entire image, but rather provides a more stable input of candidate curves for subsequent legend matching; curve recognition correction and data space physical mapping are used to improve the discriminability of candidate curve trends; legend semantic matching and identity assignment are used to accurately map parameter names or test conditions in the legend to the corresponding candidate curves. By integrating the aforementioned multiple technical steps into the same diagram, the continuous processing relationship of this invention from candidate curve generation to final identity label output can be more clearly demonstrated.

[0141] exist Figure 5 In the illustrated link, the curve space extraction stage can further confirm the suspected curve region by combining the candidate curve retention score; the curve recognition and correction stage can combine local trajectory smoothing and direction constraints to avoid individual noise points or thin background lines being mistaken for curves; the physical mapping stage can use a linear mapping relationship to transform the pixel sequence into a data sequence with physical meaning; the identity assignment stage combines the semantics of the legend text, the features of the legend color blocks, and the features of the candidate curves for parameter identity output. For the parameter typical characteristic map, since the curves under different test conditions may have similar trends in some intervals, it is necessary to combine steps S3-S5 in Example 1 to jointly judge the trend differences, joint matching scores, and matching confidence, thereby improving the stability of identity assignment in scenarios where multiple curves coexist.

[0142] In summary, Example 1, as the main methodological approach, provides a framework for... Figures 1 to 4 The corresponding complete step chain from input to output is given in Example 2, which provides the system module structure corresponding to the step chain, and Example 3 provides... Figure 5 The detailed description of the integrated processing link shown makes the technical solution of the present invention more complete.

Claims

1. A method for matching legend curves in typical characteristic plots of power amplifiers, characterized in that, Includes the following steps: S1. Collect and preprocess the typical characteristic diagrams of the original power amplifier to generate a set of candidate curves and a set of candidate diagram examples; S2. Extract the multi-dimensional features of each legend item and each candidate curve respectively, and generate the legend feature set and the curve feature set; Multidimensional features include color features, line type features, spatial location features, textual semantic features, and trend features; S3. Constraint matching is performed on legend items and candidate curves based on legend feature sets and curve feature sets; a multi-dimensional joint matching representation is constructed based on the pairing relationship between each set of legend items and candidate curves; adaptive weights of each dimension are calculated according to the image quality factors corresponding to each feature dimension, thereby generating a joint matching score matrix between legend items and candidate curves; including: S31. For any legend item and candidate curve, perform color matching and initial screening of the candidate curve based on color features; generate a color candidate set corresponding to each legend item; S32. For candidate curves in the color candidate set, perform line type consistency judgment and elimination based on line type features, and output the line type constraint candidate set; S33. For candidate curves in the linear constraint candidate set, perform spatial adjacency judgment and retention based on spatial location features, and output the spatial constraint candidate set; S34. For the candidate curves of the legend items, perform text constraint determination based on text semantic features and generate text constraint scores; S35. When multiple candidate curves corresponding to the same legend item cannot be effectively distinguished in terms of color features, line type features, and spatial location features, or when the difference in text constraint scores between the best candidate curve and the second best candidate curve is lower than the preset difference threshold, auxiliary trend determination is performed based on trend features, and a trend constraint candidate set is output. S4. Using the candidate legend item set and candidate curve set as the two sides of the bipartite graph respectively, and the joint matching score as the edge weight, construct a legend-curve bipartite graph matching model; solve the legend-curve bipartite graph matching model to generate the globally optimal matching matrix; S5. Based on the global optimal matching matrix, calculate the matching confidence of each pairing relationship and output the matching results including identity tags.

2. The method for matching legend curves of typical characteristic plots of power amplifiers according to claim 1, characterized in that, S1 includes: Obtain the original typical characteristic map of the power amplifier; perform preprocessing operations on the original typical characteristic map of the power amplifier, including orientation correction, resolution normalization, color consistency correction, noise suppression, plotting area localization, legend area localization, and non-curve element stripping; in the preprocessed image, extract candidate curve objects and candidate legend objects through connected component analysis; for any connected region, calculate its trajectory continuity, edge consistency, internal color stability, and background noise interference intensity, and obtain a weighted candidate retention score; when the candidate retention score is greater than the preset retention threshold, the connected region is retained as a candidate curve object.

3. The method for matching legend curves of typical characteristic diagrams of power amplifiers according to claim 1, characterized in that, S2 include: S21. Extract legend color features, legend line type features, legend text semantic features, and legend position features for each legend item; among them, the legend color features are represented by a comprehensive color mean vector and color covariance matrix; the legend line type features are represented by solid line length distribution, dashed line interval distribution, and local endpoint statistics; the legend text semantic features are represented by the semantic vector of the encoded legend text; and the legend position features are represented by the center coordinates of the legend box and its sorting index. S22. Extract curve color features, curve line shape features, curve position features, and curve trend features for each candidate curve. Among them, curve color features are used to characterize the main color component and color dispersion of curve pixels; curve line shape features are used to characterize the continuous length distribution and interval pattern of the curve; curve position features are used to characterize the geometric positional relationship of the candidate curve relative to the drawing area and legend area; curve trend features are used to characterize the local rise and fall trend, peak and valley positions, and relative high and low relationships at key frequency points of the curve in the physical data space.

4. The method for matching legend curves for typical characteristic plots of power amplifiers according to claim 1, characterized in that, Color matching and initial screening of candidate curves based on color features includes: calculating the color distance between the legend color features and the curve color features; if the color distance is less than or equal to the initial color screening threshold, the candidate curve is considered to have passed the initial color screening of the legend item, and the candidate curve is included in the color candidate set corresponding to the legend item. The line type consistency judgment and elimination based on line type features includes: comparing the legend line type features and the curve line type features to obtain a line type consistency score; if the line type consistency score is less than the line type consistency threshold, the candidate curve is eliminated from the candidate curve set of the legend item. The spatial adjacency determination and retention based on spatial location features includes: establishing a spatial adjacency consistency score based on the relative position of the legend item in the legend area and the representative position of the candidate curve in the drawing area; if the spatial adjacency consistency score is lower than a preset spatial consistency threshold, the comprehensive prior weight corresponding to the candidate curve is reduced; if the spatial adjacency consistency score is higher than the preset spatial consistency threshold, the candidate curve is retained.

5. The method for matching legend curves of typical characteristic plots of a power amplifier according to claim 1, characterized in that, Performing text constraint determination based on text semantic features includes: Semantic encoding is performed on the text content of the legend items to obtain the legend text semantic vector. Construct a semantic constraint vector for the curve by combining coordinate axis text, unit text, and adjacent curve annotation information. ; Calculate legend items With candidate curve Textual constraint score : ； in, Candidate curve The semantic vectors of the coordinate axes in the current coordinate system. Candidate curve The corresponding unit compatibility vector, Candidate curve The associated test condition semantic vector, to Indicates the weight of the text constraint item. Represents the cosine similarity function. This represents the hit determination function; if the candidate curve If the semantics of the coordinate axes or test conditions are incompatible with the parameter names or test condition keywords contained in the legend text, the text constraint score will be reduced. .

6. The method for matching legend curves for typical characteristic plots of a power amplifier according to claim 1, characterized in that, Assisted trend determination based on trend characteristics includes: selecting candidate curves based on coordinate axis scales and unit information. The pixel coordinates are mapped to the physical data space to obtain the candidate curve. physical trajectory The mapping relationship is represented as: ； in, and These represent the x and y coordinates of the candidate curve pixels, respectively. and These represent the mapped physical x and y coordinates, and These represent the scaling coefficients for the horizontal and vertical axes, respectively. and These represent the corresponding offsets; Based on physical trajectory Construct candidate curves trend residual Similarity to trends Trend residual Used to characterize the deviation between the candidate curve and the reference fitted trajectory; if legend items Based on the existing parameter-condition prototype trajectory, candidate curves are further determined by trend similarity. The degree of consistency with the prototype trajectory.

7. The method for matching legend curves of typical characteristic plots of a power amplifier according to claim 1, characterized in that, S3 include: For the legend item With candidate curve Candidate relationships between them, construct joint representation vectors : ； in, Indicates color matching response, Indicates a linear matching response. Indicates spatial matching response, Indicates a text constraint response. Indicates the trend constraint response; Based on the color quality, line quality, character recognition quality, spatial stability, and trend reliability corresponding to the current candidate relationship, calculate the quality factor. , And further calculate adaptive weights , : ； Where Ω represents the set of feature categories, Indicates the first Sensitivity parameters of class features, , Indicates the first Class features in candidate relations ( , The quality factor under ( ) Based on the joint representation vector and adaptive weights , Calculate the legend item With candidate curve Joint matching score matrix between : 。 8. The method for matching legend curves of typical characteristic plots of a power amplifier according to claim 1, characterized in that, S5 include: S51. Based on the global optimal matching matrix, determine the optimal legend item corresponding to each candidate curve, and generate the identity label of the candidate curve from the parameter name and test conditions in the legend item. S52. For each candidate curve, construct a matching probability distribution based on the matching score distribution of the legend items; calculate the matching confidence based on the matching probability distribution and the trend residual of the candidate curve.

9. A legend curve matching system for typical characteristic diagrams of power amplifiers, characterized in that, include: The curve space extraction module is used to collect and preprocess the typical characteristic diagrams of the original power amplifier, and generate a set of candidate curves and a set of candidate diagram examples. The feature extraction module is used to extract multi-dimensional features of each legend item and each candidate curve, and generate legend feature set and curve feature set; Multidimensional features include color features, line type features, spatial location features, textual semantic features, and trend features; The joint representation generation module is used to perform constraint matching on legend items and candidate curves based on legend feature sets and curve feature sets; and to construct multi-dimensional joint matching representations based on the pairing relationship between each set of legend items and candidate curves. The quality-aware weight generation module is used to calculate the adaptive weights of each dimension based on the image quality factors corresponding to each feature dimension, and then generate a joint matching score matrix between legend items and candidate curves. The global consistency matching module is used to construct a legend-curve bipartite graph matching model by using the candidate legend item set and the candidate curve set as the two sides of the bipartite graph, respectively, and the joint matching score as the edge weight; the legend-curve bipartite graph matching model is solved to generate the globally optimal matching matrix; The output module is used to calculate the matching confidence of each pairing relationship based on the global optimal matching matrix and output the matching results including identity tags. The legend curve matching system for typical characteristic plots of power amplifiers is used to implement the legend curve matching method and steps for typical characteristic plots of power amplifiers as described in claim 1.

Images