A Radar Image Calculation Method Based on Image Recognition

Abstract

This invention discloses a radar image plotting method based on image recognition, specifically including: acquiring radar track images and user-drawn track segments and target information; constructing an initial track data sequence and recording the start, end, and length; generating key points from the track data, performing topological semantic annotation and hierarchical encoding; inputting the image and key points into an improved RegNet network to extract graphic, numerical, alphanumeric, and Chinese character features to generate multi-branch representations; selecting a multi-resolution map to match the track position based on the track length and range; performing normalization and fusion on the multi-branch features to generate a unified plotting vector and associating it with map coordinates; executing a comparison algorithm to identify error points, offsets, and symbol errors; and generating an assessment output based on the plotting accuracy. This invention achieves high-precision, multi-element automatic plotting through key point hierarchical encoding and multi-branch feature fusion.

Description

Technical Field

[0001] This invention relates to the field of radar image plotting technology, and in particular to a radar image plotting method based on image recognition. Background Technology

[0002] In radar monitoring and navigation systems, target trajectory plotting refers to visualizing the target motion trajectory information acquired by radar to assist in situational awareness, navigation monitoring, and combat decision-making. However, existing trajectory plotting methods mostly rely on manual drawing or simple line connection algorithms, which suffer from problems such as low recognition efficiency, large error accumulation, and inconsistent symbol labeling.

[0003] Furthermore, radar images typically contain multiple types of elements, such as numbers, letters, symbols, and Chinese characters. Traditional methods have limited capabilities in multi-element recognition and processing, making it difficult to achieve automated and high-precision plotting results. Especially in scenarios with complex tracks, multiple intersecting targets, or overlapping tracks, manual plotting is prone to omissions or misjudgments, affecting the accuracy of subsequent analysis and evaluation.

[0004] Meanwhile, existing methods lack systematic algorithms for the automatic identification and semantic annotation of track length, turning points, segmentation points, and key points, and cannot perform structured encoding of key nodes, resulting in low efficiency in data post-processing, comparison, and assessment. Therefore, there is an urgent need for an automatic radar image plotting method that combines image recognition, key point generation and hierarchical encoding, multi-branch feature extraction, and multi-resolution map matching to achieve automatic track plotting, accurate key point identification, multi-element feature fusion, and automatic generation of assessment output. This would improve plotting accuracy, operational efficiency, and standardization, meeting the application needs of modern radar monitoring, track analysis, and training assessment. Summary of the Invention

[0005] One objective of this invention is to propose a radar image plotting method based on image recognition. This invention introduces a hierarchical coding of key points of the track and a multi-branch feature fusion mechanism to perform semantic annotation of key points, feature extraction of graphic, number, letter and Chinese character elements, and multi-resolution map linkage matching on radar target track images, construct a unified plotting feature vector, and generate plotting accuracy evaluation results. It has the advantages of high plotting accuracy, complete multi-element recognition and automated evaluation output.

[0006] A radar image rendering method based on image recognition according to an embodiment of the present invention includes the following steps: Collect radar target trajectory images, obtain user-drawn trajectory segments and related target information, construct the initial trajectory data sequence, and record the start point, end point, and length of the trajectory segments; A key point generation algorithm is executed on the initial data sequence of the track. Based on the topological features of the track, the key points are semantically labeled and hierarchically encoded to form a set of hierarchically encoded key points. The improved RegNet multi-output branch network is input with the trajectory image and key point set to extract features from graphic, number, letter and Chinese character elements respectively and generate multi-branch feature representations. A multi-resolution map and range linkage algorithm is adopted to select the corresponding resolution map layer according to the track length and range setting, so as to realize the linkage display of track and map and the matching of position; Normalization and feature fusion are performed on the multi-branch feature representation to generate a unified plotting feature vector. The unified plotting feature vector is then associated with map coordinates and range information to form a complete plotting data structure. A comparison algorithm is executed on the plotted data structure to compare the plotted content with the standard answer point by point, identify error points, offsets and symbol errors, and generate plotted accuracy data. The assessment output is generated based on the accuracy data of the plotting, including the recognition results of graphic, number, letter, and Chinese character elements and a deviation analysis report.

[0007] Optionally, the construction of the initial data sequence for the trajectory, recording the start point, end point, and length of the line segment, specifically includes: Read the radar target trajectory image, extract the set of trajectory pixels of the trajectory line segments drawn by the user, perform connected region scanning on the set of trajectory pixels of the trajectory line segments, and generate a sequence of trajectory line segment numbers; Perform endpoint detection on the set of trajectory pixels corresponding to each track segment number, extract the position coordinates where the pixel coordinate change direction is interrupted, and generate the starting point coordinates and ending point coordinates of the line segment. Perform coordinate position association between the starting coordinates and ending coordinates of the line segment to generate a line segment coordinate pair sequence. Read the target number, target type identifier, target heading angle value and target speed value from the radar target information, and write the target number, target type identifier, target heading angle value and target speed value into the corresponding line segment coordinate pair position to generate a target associated line segment record. Perform pixel distance calculation on the line segment coordinate pair sequence, count the number of consecutive pixels between the start coordinate and end coordinate of the line segment, and generate the line segment length value. Read the radar image range parameters and image scale parameters, perform scale conversion on the line segment length values, and generate the actual track length values. The initial data sequence of the trajectory is formed by arranging the fields in the following order: line segment number, line segment start coordinates, line segment end coordinates, line segment length value, actual track length value, target number, target type identifier, target heading angle value, and target speed value.

[0008] Optionally, the step of performing a track key point generation algorithm on the initial track data sequence includes the following steps: Read the route records in the initial data sequence of the track, establish a track segment buffer area according to the segment number, write the starting coordinates and ending coordinates of the segment into the endpoint field, and write the segment length value into the length field. Within the track segment buffer area, read two adjacent track segments in order of segment number, and calculate the difference in horizontal coordinates, the difference in vertical coordinates, and the distance between endpoints between the endpoint coordinates of the previous track segment and the starting coordinates of the next track segment. The endpoint spacing value is compared with the preset connection distance threshold. If the endpoint spacing value does not exceed the preset connection distance threshold, the coordinates of the end point of the previous track segment and the coordinates of the start point of the next track segment are merged into the connection point coordinates and written into the connection point buffer. For each track segment, the direction vector is calculated by performing a direction vector calculation on the coordinates of the starting point and the ending point of the segment. The angle between the direction vectors of two adjacent track segments is calculated to generate the track turning angle value. The trajectory turning angle value is compared with the preset turning angle threshold. When the trajectory turning angle value reaches the preset turning angle threshold, the coordinates of the connection point or the coordinates of the adjacent endpoints are written into the turning point buffer. Read the line segment length value from the length field, compare the line segment length value with the preset segment length threshold, and when the line segment length value reaches the preset segment length threshold, perform coordinate interpolation between the starting coordinates and ending coordinates of the line segment according to the preset segment spacing to generate a sequence of segment point coordinates; Perform coordinate deduplication on the connection point buffer, turning point buffer, and segment point coordinate sequence, retaining a key point record with a coordinate difference of zero, and deleting duplicate key point records; The key point coordinate records are arranged according to the fields of line segment number, key point coordinate, key point type identifier, and key point sequence number to form a set of track key points.

[0009] Optionally, the step of semantically annotating and hierarchically encoding key points based on track topology features to form a hierarchically encoded key point set specifically includes: Read the line segment number, key point coordinate record, key point type identifier and key point sequence number from the track key point set, and establish a key point topology buffer according to the line segment number; Read the coordinate records of key points under the same line segment number in the key point topology buffer area, arrange them according to the key point numbering order, and generate the key point sequence within the line segment. Read the sequence of key points within the line segment corresponding to the adjacent line segment number, perform coordinate difference calculation on the coordinate record of the last key point of the previous line segment and the coordinate record of the first key point of the next line segment, and generate a line segment connection relationship identifier. Write the line segment connection relationship identifiers into the key point topology cache area to form a track topology feature that includes endpoint relationships, intra-line segment order relationships, and inter-line segment connection relationships; Read the topological features of the flight path and the key point type identifiers, write start and end semantic identifiers to the endpoint coordinate records, write connection semantic identifiers to the connection point coordinate records, write turning semantic identifiers to the turning point coordinate records, and write segment semantic identifiers to the segment point coordinate records to form semantically labeled key point records. The first layer of coding is generated according to the line segment number in the track topology feature, the second layer of coding is generated according to the key point sequence number, and the third layer of coding is generated according to the semantic identifier. The first layer of coding, the second layer of coding, and the third layer of coding are written into the semantic annotation key point record. The semantically labeled key point records are sorted by field, arranged according to line segment number, key point coordinate record, key point sequence number, line segment connection relationship identifier, semantic identifier, first-level code, second-level code and third-level code, forming a hierarchical coded key point set.

[0010] Optionally, the step of inputting the trajectory image and keypoint set into the improved RegNet multi-output branch network to extract features from graphic, number, letter, and Chinese character elements and generate multi-branch feature representations includes the following steps: Read the flight track image and the hierarchical encoded key point set, and map the key point coordinates to the flight track image to form a key point localization image; A fixed-size local region is cropped from the key point localization image centered on the key point coordinates. Grayscale conversion, edge detection and contour extraction are performed on the local region to generate local contour information. Based on local contour information, the local area is divided into candidate areas of line segments, strokes or symbols, and the endpoints and forking positions are marked to form a set of candidate areas for plotting elements. The candidate regions of plotted elements are written into the shared feature extraction layer of the improved RegNet multi-output branch network, and convolution calculation, normalization calculation and activation mapping are performed to generate a shared feature map. The shared feature map is input into the graphic branch, number branch, letter branch and Chinese character branch respectively. The graphic branch extracts the line contour and turning features to generate graphic feature representation. The digit branch performs feature extraction on the stroke endpoints and arrangement order of candidate digit regions to generate digit feature representations; The letter branch performs feature extraction on the stroke direction, closed structure, and character proportion of the letter region to generate letter feature representations; The Chinese character branch performs feature extraction on the horizontal and vertical strokes, left-falling and right-falling strokes, and component positions of the Chinese character region to generate Chinese character feature representations; The graphic feature representation, numerical feature representation, letter feature representation, and Chinese character feature representation are arranged and combined according to the key point order to generate a multi-branch feature representation.

[0011] Optionally, the step of employing a multi-resolution map and range linkage algorithm, selecting the corresponding resolution map layer based on the track length and range settings to achieve linked display of the track and map and location matching, specifically includes: Read the set of key points and track length information, calculate the overall track span, determine the applicable map resolution level based on the track span and range parameters, and generate a resolution selection list; Map the coordinates of key points on the flight path to the pixel coordinates of the candidate resolution map layer, perform preliminary position matching through the coordinate transformation matrix, and re-select the nearest resolution map for matching and fine-tuning for the mapped points that fall outside the map boundary; Read the map grid information on the finally selected map resolution layer, calculate the offset between the track key points and the map reference points, and adjust the key point coordinates until they are aligned with the map grid.

[0012] Optionally, the step of performing normalization processing and feature fusion on the multi-branch feature representation to generate a unified plotted feature vector includes the following steps: Read the multi-branch feature representations of graphic branches, number branches, letter branches and Chinese character branches, perform normalization processing on each branch feature, including mean standardization and variance normalization, and generate a standardized feature matrix; Cross-branch alignment is performed on the standardized feature matrix, aligning the corresponding positions of the features in each branch according to the key point order, and concatenating and weighting the feature dimensions to generate a preliminary fused feature vector; The initial fused feature vectors are subjected to nonlinear mapping and linear combination operations to adjust the feature distribution range and relative weights, thereby generating a unified plotted feature vector.

[0013] Optionally, the comparison algorithm performed on the plotted data structure specifically includes: Read the unified plotted feature vector and the corresponding key point coordinates, map the key point coordinates to the corresponding positions in the standard answer image, and establish the plotted point matching relationship; For each matching point, calculate the horizontal coordinate deviation, vertical coordinate deviation, and angular deviation, and compare whether the symbol categories are consistent, mark the error points, and generate the deviation value. The error points and deviation values ​​are arranged according to the order of key points, line segment numbers and element types. The deviations are accumulated and statistically analyzed to generate error summary information. The statistical results of deviations, the location of error points, and information on symbol errors are summarized to form data for evaluating the accuracy of plotting.

[0014] Optionally, the step of generating assessment output based on the accuracy data of the plotting, including the recognition results of graphic, number, letter, and Chinese character elements and a deviation analysis report, specifically includes: Read the plotting accuracy evaluation data and error summary information, and classify and organize the error point coordinates, deviation values ​​and symbol categories according to the key point order and line segment number; Recognition points are extracted for graphic elements, number elements, letter elements, and Chinese character elements according to element type. The correct recognition rate, deviation statistics, and symbol consistency are calculated for each type of element. The deviation data is categorized into tables and lists by element type. The coordinates of the error points, the deviation values, and the error symbol categories are recorded one by one to form a deviation analysis dataset. The identification points and deviation data are sorted and combined according to element type, line segment number, and key point order to generate plotting assessment output, including the identification results and corresponding deviation information of graphic, number, letter, and Chinese character elements.

[0015] The beneficial effects of this invention are: 1. By using hierarchical coding and semantic annotation of key points in the track, endpoints, turning points, connection points and segmentation points can be accurately identified, achieving structured expression of key nodes, reducing human error and improving the accuracy of radar track plotting.

[0016] 2. The improved RegNet multi-output branch network can extract features from graphic, number, letter and Chinese character elements respectively to form multi-branch feature representations, realize the automatic recognition and fusion of different types of symbols, and improve the integrity of the symbol plotting.

[0017] 3. By adopting a track length and range linkage algorithm, key points are mapped to the optimal map resolution to achieve precise matching between the track and the map grid, ensuring the consistency and spatial accuracy of track visualization.

[0018] 4. Perform normalization processing and feature fusion on multi-branch feature representations to generate unified plotted feature vectors, realize cross-element information integration, and provide standardized input for subsequent comparison and assessment. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a radar image rendering method based on image recognition proposed in this invention. Figure 2 This is a schematic diagram of the key point hierarchical encoding of a radar image rendering method based on image recognition proposed in this invention; Figure 3This is a schematic diagram illustrating the multi-branch feature extraction and fusion of a radar image rendering method based on image recognition proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figure 1-3 A radar image rendering method based on image recognition includes the following steps: Collect radar target trajectory images, obtain user-drawn trajectory segments and related target information, construct the initial trajectory data sequence, and record the start point, end point, and length of the trajectory segments; A key point generation algorithm is executed on the initial data sequence of the track. Based on the topological features of the track, the key points are semantically labeled and hierarchically encoded to form a set of hierarchically encoded key points. The improved RegNet multi-output branch network is input with the trajectory image and key point set to extract features from graphic, number, letter and Chinese character elements respectively and generate multi-branch feature representations. A multi-resolution map and range linkage algorithm is adopted to select the corresponding resolution map layer according to the track length and range setting, so as to realize the linkage display of track and map and the matching of position; Normalization and feature fusion are performed on the multi-branch feature representation to generate a unified plotting feature vector. The unified plotting feature vector is then associated with map coordinates and range information to form a complete plotting data structure. A comparison algorithm is executed on the plotted data structure to compare the plotted content with the standard answer point by point, identify error points, offsets and symbol errors, and generate plotted accuracy data. The assessment output is generated based on the accuracy data of the plotting, including the recognition results of graphic, number, letter, and Chinese character elements and a deviation analysis report.

[0022] In this embodiment, constructing the initial data sequence of the trajectory and recording the start point, end point, and length of the line segment specifically includes: Read the radar target trajectory image, extract the set of trajectory pixels of the trajectory line segments drawn by the user, perform connected region scanning on the set of trajectory pixels of the trajectory line segments, and generate a sequence of trajectory line segment numbers; Perform endpoint detection on the set of trajectory pixels corresponding to each track segment number, extract the position coordinates where the pixel coordinate change direction is interrupted, and generate the starting point coordinates and ending point coordinates of the line segment. Perform coordinate position association between the starting coordinates and ending coordinates of the line segment to generate a line segment coordinate pair sequence. Read the target number, target type identifier, target heading angle value and target speed value from the radar target information, and write the target number, target type identifier, target heading angle value and target speed value into the corresponding line segment coordinate pair position to generate a target associated line segment record. Perform pixel distance calculation on the line segment coordinate pair sequence, count the number of consecutive pixels between the start coordinate and end coordinate of the line segment, and generate the line segment length value. Read the radar image range parameters and image scale parameters, perform scale conversion on the line segment length values, and generate the actual track length values. The initial data sequence of the trajectory is formed by arranging the fields in the following order: line segment number, line segment start coordinates, line segment end coordinates, line segment length value, actual track length value, target number, target type identifier, target heading angle value, and target speed value.

[0023] Specifically, performing a connected component scan on the set of trajectory pixels of the flight path segments to generate a sequence of flight path segment numbers includes: Read the pixel set of the track segment, determine the neighborhood based on the spatial position of the pixel in the image, and group interconnected pixels into the same connected region; assign a number to each connected region to form a track segment number sequence for subsequent endpoint, turning point, and length calculations; for isolated pixels or noise points, remove pixel sets with small areas according to the minimum region threshold; establish a correspondence between the connected region numbers and the original pixel set to generate a traceable line segment pixel index.

[0024] Specifically, endpoint detection for the set of trajectory pixels corresponding to each track segment number includes: Read the pixel set corresponding to the track segment number, and scan the position changes according to the order of the pixels in the sequence; identify the pixels where the direction of pixel coordinate change is discontinuous, and record them as the endpoints of the line segment; extract the start coordinates and end coordinates of each line segment to form an endpoint coordinate record set; establish a mapping relationship between the endpoint coordinates and the corresponding track segment number to generate endpoint data that can be processed later.

[0025] Specifically, the process of associating the coordinates of the starting and ending points of a line segment to generate a sequence of line segment coordinate pairs includes: Read the start and end coordinates of each line segment and sort them by line segment number; Pair the start and end coordinates of each line segment to form line segment coordinate pairs; Record the corresponding track line segment number for each line segment coordinate pair; Organize the line segment coordinate pair sequence into an index structure that can be used for target information writing and subsequent calculations; The fields are arranged in the following order: line segment number, line segment start coordinates, line segment end coordinates, line segment length, actual track length, target number, target type identifier, target heading angle, and target speed. Read the segment number, start coordinates, and end coordinates, and sort them in numerical order; read the segment length value and the actual track length value, and match them with the corresponding segment number; read the target number, target type identifier, target heading angle value, and target speed value, and establish a correspondence with the corresponding segment; combine the segment number, start coordinates, end coordinates, segment length, actual track length, target number, target type identifier, target heading angle, and target speed in the order of the fields to generate the initial track data sequence.

[0026] In this embodiment, a track key point generation algorithm is performed on the initial track data sequence, including the following steps: Read the route records in the initial data sequence of the track, establish a track segment buffer area according to the segment number, write the starting coordinates and ending coordinates of the segment into the endpoint field, and write the segment length value into the length field. Within the track segment buffer area, read two adjacent track segments in order of segment number, and calculate the difference in horizontal coordinates, the difference in vertical coordinates, and the distance between endpoints between the endpoint coordinates of the previous track segment and the starting coordinates of the next track segment. The endpoint spacing value is compared with the preset connection distance threshold. If the endpoint spacing value does not exceed the preset connection distance threshold, the coordinates of the end point of the previous track segment and the coordinates of the start point of the next track segment are merged into the connection point coordinates and written into the connection point buffer. For each track segment, the direction vector is calculated by performing a direction vector calculation on the coordinates of the starting point and the ending point of the segment. The angle between the direction vectors of two adjacent track segments is calculated to generate the track turning angle value. The trajectory turning angle value is compared with the preset turning angle threshold. When the trajectory turning angle value reaches the preset turning angle threshold, the coordinates of the connection point or the coordinates of the adjacent endpoints are written into the turning point buffer. Read the line segment length value from the length field, compare the line segment length value with the preset segment length threshold, and when the line segment length value reaches the preset segment length threshold, perform coordinate interpolation between the starting coordinates and ending coordinates of the line segment according to the preset segment spacing to generate a sequence of segment point coordinates; Perform coordinate deduplication on the connection point buffer, turning point buffer, and segment point coordinate sequence, retaining a key point record with a coordinate difference of zero, and deleting duplicate key point records; The key point coordinate records are arranged according to the fields of line segment number, key point coordinate, key point type identifier, and key point sequence number to form a set of track key points.

[0027] Specifically, the calculation of the direction vector for the starting and ending coordinates of each track segment includes: Read the coordinates of the starting point and ending point of the line segment, and calculate the difference between the starting point and ending point coordinates in the horizontal and vertical directions; combine the horizontal and vertical differences to form the line segment direction vector.

[0028] Specifically, the calculation of the included angle between the direction vectors of two adjacent track segments to generate the track turning angle includes: Read the direction vectors of the previous and next track segments, calculate the angle between the two vectors according to the vector coordinates, use the calculated angle as the track turning angle value, and establish a mapping between the turning angle value and the corresponding adjacent track segment number to form a data sequence that can be used for key point annotation. Specifically, generating a sequence of segment point coordinates by performing coordinate interpolation between the starting and ending coordinates of a line segment according to a preset segment spacing includes: Read the coordinates of the start and end points of the line segment, obtain the preset segment spacing value, and calculate the total length of the line segment; calculate the number of segment points based on the line segment length and segment spacing, calculate the horizontal and vertical coordinates of each segment point at uniform intervals between the start and end points, organize the segment point coordinates according to the line segment number and sequence number, and generate a data sequence that can be used for the key point set.

[0029] Specifically, the coordinate deduplication process is performed on the connection point buffer, the turning point buffer, and the segment point coordinate sequence, retaining only one key point record with a coordinate difference of zero, and deleting duplicate key point records. Read the connection point buffer, turning point buffer, and segment point coordinate sequence, sort them by coordinate value, identify points with identical horizontal and vertical coordinates as duplicate points, retain the first record in each group of duplicate points, delete other duplicate records, generate a deduplicated key point coordinate sequence, organize the deduplicated coordinate sequence by line segment number and sequential number, and use it to form the final track key point set.

[0030] In this embodiment, semantic annotation and hierarchical encoding of key points based on track topology features are performed to form a hierarchically encoded key point set, specifically including: Read the line segment number, key point coordinate record, key point type identifier and key point sequence number from the track key point set, and establish a key point topology buffer according to the line segment number; Read the coordinate records of key points under the same line segment number in the key point topology buffer area, arrange them according to the key point numbering order, and generate the key point sequence within the line segment. Read the sequence of key points within the line segment corresponding to the adjacent line segment number, perform coordinate difference calculation on the coordinate record of the last key point of the previous line segment and the coordinate record of the first key point of the next line segment, and generate a line segment connection relationship identifier. Write the line segment connection relationship identifiers into the key point topology cache area to form a track topology feature that includes endpoint relationships, intra-line segment order relationships, and inter-line segment connection relationships; Read the topological features of the flight path and the key point type identifiers, write start and end semantic identifiers to the endpoint coordinate records, write connection semantic identifiers to the connection point coordinate records, write turning semantic identifiers to the turning point coordinate records, and write segment semantic identifiers to the segment point coordinate records to form semantically labeled key point records. The first layer of coding is generated according to the line segment number in the track topology feature, the second layer of coding is generated according to the key point sequence number, and the third layer of coding is generated according to the semantic identifier. The first layer of coding, the second layer of coding, and the third layer of coding are written into the semantic annotation key point record. The semantically labeled key point records are sorted by field, arranged according to line segment number, key point coordinate record, key point sequence number, line segment connection relationship identifier, semantic identifier, first-level code, second-level code and third-level code, forming a hierarchical coded key point set.

[0031] Specifically, the process involves generating a first-layer code based on the line segment numbers in the track topology features, a second-layer code based on the key point sequence numbers, and a third-layer code based on semantic identifiers. The first, second, and third-layer codes are then written into the semantically labeled key point record. Read all keypoint records and their corresponding track segment numbers, keypoint sequence numbers, and semantic identifiers; sort the keypoint records in ascending order by track segment number, group records with the same track segment number into the same group, assign a unique first-level code to each group, and write it into the first-level code field of the keypoint record; arrange the keypoint records in ascending order by keypoint sequence number, generate a second-level code for each record, and write it into the second-level code field of the keypoint record; generate a third-level code for each record according to its semantic identifier, and write it into the third-level code field of the keypoint record; organize the keypoint record order to generate a semantically annotated keypoint set containing the three-level codes, which will be used for subsequent multi-branch feature extraction and plotting processing.

[0032] In this embodiment, the improved RegNet multi-output branch network is input with the trajectory image and key point set to extract features from graphic, number, letter and Chinese character elements to generate multi-branch feature representations, including the following steps: Read the flight track image and the hierarchical encoded key point set, and map the key point coordinates to the flight track image to form a key point localization image; A fixed-size local region is cropped from the key point localization image centered on the key point coordinates. Grayscale conversion, edge detection and contour extraction are performed on the local region to generate local contour information. Based on local contour information, the local area is divided into candidate areas of line segments, strokes or symbols, and the endpoints and forking positions are marked to form a set of candidate areas for plotting elements. The candidate regions of plotted elements are written into the shared feature extraction layer of the improved RegNet multi-output branch network, and convolution calculation, normalization calculation and activation mapping are performed to generate a shared feature map. The shared feature map is input into the graphic branch, number branch, letter branch and Chinese character branch respectively. The graphic branch extracts the line contour and turning features to generate graphic feature representation. The digit branch performs feature extraction on the stroke endpoints and arrangement order of candidate digit regions to generate digit feature representations; The letter branch performs feature extraction on the stroke direction, closed structure, and character proportion of the letter region to generate letter feature representations; The Chinese character branch performs feature extraction on the horizontal and vertical strokes, left-falling and right-falling strokes, and component positions of the Chinese character region to generate Chinese character feature representations; The graphic feature representation, numerical feature representation, letter feature representation, and Chinese character feature representation are arranged and combined according to the key point order to generate a multi-branch feature representation.

[0033] Specifically, performing grayscale conversion, edge detection, and contour extraction on local regions to generate local contour information includes: Read the local region image data obtained by cropping the center of key points; convert the local region color image to grayscale image to form a grayscale pixel matrix; perform edge detection operation on the grayscale image, calculate the gradient value and direction of each pixel, and mark the positions with obvious grayscale changes; perform binarization processing on the edge detection results to separate edge pixels from non-edge pixels and generate a preliminary edge map; perform a contour tracking algorithm on the edge map to identify continuous pixel paths to form contour lines; record and sort the coordinates of the contour lines, and store the contour endpoints, bifurcation points, and closed paths into the local contour information set respectively; organize the local contour information to generate a sequence that can be used for subsequent plotting element candidate region division, including contour coordinates, pixel index, and contour type; repeat the processing of all local regions to form a complete local contour information set.

[0034] Specifically, the process of dividing a local region into candidate regions for line segments, strokes, or symbols based on local contour information, marking endpoints and forking positions, and forming a set of candidate regions for plotted elements includes: The process involves: reading a sequence of local contour information, including contour coordinates, pixel indices, and contour type; performing pixel continuity analysis on the contour lines to divide continuous pixel segments into independent line segments; calculating the change in direction vectors of adjacent pixels within each line segment, and using the position where the direction change exceeds a preset threshold as the cutting point; dividing the line segment into pixel sub-segments at the cutting point, determining the spacing and direction continuity between adjacent sub-segments, and merging sub-segments that meet the conditions to form a complete stroke candidate region; identifying closed contours or combined contours, recording them as symbol candidate regions, and generating a unique number; marking the endpoints and bifurcation points of each line segment, stroke, or symbol region, and saving the corresponding pixel coordinates; organizing all line segment, stroke, and symbol candidate regions and their endpoint and bifurcation point information to generate a set of plotting element candidate regions; establishing a mapping relationship according to the key point order and region index to form an input data structure that can be used for multi-branch feature extraction and plotting analysis; and repeating the processing of all local regions to form a complete set of plotting element candidate regions.

[0035] Specifically, writing the candidate regions of plotted elements into the shared feature extraction layer of the improved RegNet multi-output branch network includes: The shared feature extraction layer is an improved front-end of the RegNet network, used to uniformly encode all candidate regions of plotted elements into a general feature map, which is then used by each branch to extract specific element features.

[0036] Specifically, the digit branch extracts features from the stroke endpoints and arrangement order of the candidate digit regions to generate digit feature representations, including: Read the candidate digit region image and the corresponding stroke endpoint coordinates, organize the pixel coordinate sequence according to the stroke arrangement order in the image, calculate the direction vector, length and endpoint position of each stroke; extract the connection relationship and spatial relative position between strokes, combine the stroke features to form digit region feature vectors, write the feature vector of each digit region into the digit branch output for multi-branch feature fusion, repeat the processing of all candidate digit regions, and generate a complete digit feature representation set.

[0037] Specifically, the letter branch performs feature extraction on the stroke direction, closed structure, and character proportion of the letter region, generating letter feature representations including: Read the candidate letter region image and the corresponding stroke coordinate sequence, organize the pixel sequence according to the stroke arrangement order in the letter region, calculate the direction vector and length of each stroke, identify the contour coordinates of closed paths or closed structures in the letter region, calculate the aspect ratio and character ratio information of the letter region, combine the stroke direction, closed structure and ratio features to form the letter region feature vector, write the feature vector of each letter region into the letter branch output, repeat the processing of all candidate letter regions, and generate a complete set of letter feature representations.

[0038] Specifically, the Chinese character branch performs feature extraction on the horizontal and vertical strokes, left-falling and right-falling strokes, and component positions of the Chinese character region, generating a Chinese character feature representation, including: Read the candidate Chinese character region image and the corresponding stroke coordinate sequence, organize the pixel sequence according to the arrangement order of the strokes in the Chinese character region, identify the stroke types such as horizontal, vertical, left-falling, and right-falling strokes, record the starting and ending coordinates, analyze the spatial position relationship and intersection between each stroke, extract the relative position and composition information of the Chinese character components, combine the stroke type, starting and ending coordinates and component position features to form the Chinese character region feature vector, write the feature vector of each Chinese character region into the Chinese character branch output, repeat the processing of all candidate Chinese character regions, and generate a complete Chinese character feature representation set.

[0039] In this embodiment, a multi-resolution map and range linkage algorithm is adopted. Based on the track length and range settings, the corresponding resolution map layer is selected to achieve linked display of the track and map, and location matching. Specifically, this includes: Read the set of key points and track length information, calculate the overall track span, determine the applicable map resolution level based on the track span and range parameters, and generate a resolution selection list; Map the coordinates of key points on the flight path to the pixel coordinates of the candidate resolution map layer, perform preliminary position matching through the coordinate transformation matrix, and re-select the nearest resolution map for matching and fine-tuning for the mapped points that fall outside the map boundary; Read the map grid information on the finally selected map resolution layer, calculate the offset between the track key points and the map reference points, and adjust the key point coordinates until they are aligned with the map grid.

[0040] Specifically, calculating the overall track span includes: Read the coordinates of all key points in the track key point set; determine the minimum and maximum x-coordinates of all key points; determine the minimum and maximum y-coordinates of all key points; calculate the lateral span: the maximum x-coordinate minus the minimum x-coordinate; calculate the longitudinal span: the maximum y-coordinate minus the minimum y-coordinate; record the lateral and longitudinal spans to generate overall track span data.

[0041] Specifically, the applicable map resolution level is determined based on the track span and range parameters, and a resolution selection list is generated, including: Read the lateral and longitudinal span values ​​of the track key point set; read the radar image range parameters, including the minimum and maximum range; calculate the proportion of the track span to the range to determine the candidate resolution range; sort the candidate resolutions from largest to smallest according to the actual ground distance corresponding to the pixels, and record the corresponding map layer index and pixel size for each resolution; generate a resolution selection list, recording the pixel size, map layer index and corresponding track coverage ratio for each resolution level; output the resolution selection list.

[0042] Specifically, the process of mapping the coordinates of key points on the flight path to pixel coordinates in the candidate resolution map layer, performing preliminary position matching through a coordinate transformation matrix, and re-selecting neighboring resolution maps for matching and fine-tuning for mapped points falling outside the map boundaries includes: Read the set of keypoint coordinates of the flight path; read the number of rows and columns of pixels and the map coverage of the candidate resolution map layer, including the minimum and maximum horizontal and vertical coordinate values; calculate the actual ground distance corresponding to each pixel in the horizontal and vertical directions of the map; for each keypoint coordinate, first calculate its horizontal offset within the map range = keypoint horizontal coordinate - minimum map horizontal coordinate, and vertical offset = keypoint vertical coordinate - minimum map vertical coordinate; divide the horizontal offset by the horizontal span and the vertical offset by the vertical span to obtain the relative position ratio of the keypoint within the map range; multiply the horizontal ratio by the number of map pixel columns and the vertical ratio by the number of map pixel rows to obtain the corresponding map pixel coordinates; organize the mapping relationship between the horizontal and vertical pixel coordinates into a coordinate transformation matrix; use the coordinate transformation matrix to batch map all keypoints of the flight path to the map pixel coordinate system to generate preliminary position matching results; check whether the mapped pixel coordinates fall within the map boundary range; for keypoints falling outside the boundary, select an adjacent resolution map layer, remap, and fine-tune the coordinates to make them fall into the effective map area; record the final map layer index and pixel coordinates corresponding to each keypoint to form a matching result set; output the keypoint mapping results.

[0043] In this embodiment, the normalization and feature fusion processes performed on the multi-branch feature representation to generate a unified plotted feature vector include the following steps: Read the multi-branch feature representations of graphic branches, number branches, letter branches and Chinese character branches, perform normalization processing on each branch feature, including mean standardization and variance normalization, and generate a standardized feature matrix; Cross-branch alignment is performed on the standardized feature matrix, aligning the corresponding positions of the features in each branch according to the key point order, and concatenating and weighting the feature dimensions to generate a preliminary fused feature vector; The initial fused feature vectors are subjected to nonlinear mapping and linear combination operations to adjust the feature distribution range and relative weights, thereby generating a unified plotted feature vector.

[0044] In this embodiment, a comparison algorithm is performed on the plotted data structure. Specifically, the comparison algorithm for the plotted data structure includes: Read the unified plotted feature vector and the corresponding key point coordinates, map the key point coordinates to the corresponding positions in the standard answer image, and establish the plotted point matching relationship; For each matching point, calculate the horizontal coordinate deviation, vertical coordinate deviation, and angular deviation, and compare whether the symbol categories are consistent, mark the error points, and generate the deviation value. The error points and deviation values ​​are arranged according to the order of key points, line segment numbers and element types. The deviations are accumulated and statistically analyzed to generate error summary information. The statistical results of deviations, the location of error points, and information on symbol errors are summarized to form data for evaluating the accuracy of plotting.

[0045] In this embodiment, the assessment output is generated based on the accuracy data of the plotting, including the recognition results of graphic, number, letter, and Chinese character elements, as well as a deviation analysis report, specifically including: Read the plotting accuracy evaluation data and error summary information, and classify and organize the error point coordinates, deviation values ​​and symbol categories according to the key point order and line segment number; Recognition points are extracted for graphic elements, number elements, letter elements, and Chinese character elements according to element type. The correct recognition rate, deviation statistics, and symbol consistency are calculated for each type of element. The deviation data is categorized into tables and lists by element type. The coordinates of the error points, the deviation values, and the error symbol categories are recorded one by one to form a deviation analysis dataset. The identification points and deviation data are sorted and combined according to element type, line segment number, and key point order to generate plotting assessment output, including the identification results and corresponding deviation information of graphic, number, letter, and Chinese character elements.

[0046] Specifically, the identification points for graphic elements, numerical elements, letter elements, and Chinese character elements are extracted according to element type. For each element type, the correct recognition rate, deviation statistics, and symbol consistency are calculated, including: Read the accuracy evaluation data for plotting, including the coordinates of all key points, deviation values, and symbol category information; classify the key point records into graphic elements, numerical elements, letter elements, and Chinese character elements according to element type; count the total number of recognized points and the number of correctly matched points within each element category, and calculate the correct recognition ratio; read the lateral deviation, vertical deviation, and angular deviation of each matched point, and perform numerical statistics, including the maximum, minimum, and average values; check whether the symbol category of the matched points is consistent with the standard answer, and count the number of consistent and inconsistent points; organize the correct recognition ratio, deviation statistics, and symbol consistency of each element category into a structured dataset; output the recognition point statistics data in order of element type for generating tables or subsequent assessment analysis; repeat the processing for all element types to generate a complete set of element recognition and deviation statistics results; Example 1: To verify the effectiveness of this invention, it was applied to a real-world simulation environment for radar track plotting. In this environment, the radar system monitors multiple moving targets, generates raw track images, and the user manually annotates track segments to aid analysis. Traditional methods suffer from high plotting errors, inaccurate key point identification, confusion between different types of elements, and difficulties in handling multi-target intersection scenarios, severely impacting track visualization and subsequent analysis. This invention achieves automatic plotting through hierarchical key point encoding and multi-branch feature fusion, improving both accuracy and efficiency.

[0047] In practical applications, the process begins by acquiring radar target track images, obtaining user-drawn track segments and target information, and constructing an initial track data sequence, including the start point, end point, and length of each segment. Subsequently, a track keypoint generation algorithm is executed to identify endpoints, turning points, connection points, and segmentation points. Semantic annotation and hierarchical encoding are then performed based on track topological features to form a structured keypoint set. The track image and keypoint set are input into an improved RegNet multi-output branch network, extracting features from graphic, numerical, alphanumeric, and Chinese character elements to generate multi-branch feature representations. Next, a multi-resolution map is selected for matching based on the track length and radar range, enabling synchronized display of track and map positions. Finally, the multi-branch features are normalized and fused to generate a unified plotted feature vector. A comparison algorithm is then executed to identify error points, offsets, and sign errors, ultimately generating the final assessment output.

[0048] Table 1: Comparison of Key Data in Radar Track Plotting

[0049] This table details a comparison of key performance indicators between the present invention and traditional methods in radar track plotting tasks. The table includes 50 targets, a total track length of 12,000 pixels, 320 track segments, and 1,050 key points, used to evaluate the accuracy of endpoint, turning point, connection point, and segmentation point recognition. The present invention achieves 99.2% accuracy in endpoint recognition, 7.1 percentage points higher than the traditional method; the accuracy rate for turning point recognition is 97.6%, compared to only 89.5% for the traditional method; and the accuracy rates for connection point and segmentation point recognition both exceed 98%, while the traditional method is below 91%. Multi-branch feature extraction shows recognition accuracy rates of 98.9%, 98.4%, 97.8%, and 97.5% for graphics, numbers, letters, and Chinese characters, respectively, while the traditional method is between 88% and 92%. The error point percentage is only 0.8%, the average offset is 1.3 pixels, and the symbol error rate is 0.7%, while the traditional method's are 5.6%, 5.2 pixels, and 4.3%, respectively. These data clearly demonstrate that the present invention significantly improves plotting accuracy, key point identification, and symbol recognition in scenarios with multiple targets, multiple types of elements, and complex flight paths. It also boasts a high degree of automation, significantly enhancing the clarity of flight path visualization and the reliability of assessment. Furthermore, the present invention maintains high accuracy even with multiple target intersections and overlapping flight paths, avoiding the plotting omissions, identification errors, and inefficiencies often encountered by traditional methods in complex scenarios. This provides a stable and reliable data foundation and operational reference for flight path analysis, situation assessment, and training evaluation, making the overall system more operable and practical in real-world applications.

[0050] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A radar image rendering method based on image recognition, characterized in that, Includes the following steps: Collect radar target trajectory images, obtain user-drawn trajectory segments and related target information, construct the initial trajectory data sequence, and record the start point, end point, and length of the trajectory segments; A key point generation algorithm is executed on the initial data sequence of the track. Based on the topological features of the track, the key points are semantically labeled and hierarchically encoded to form a hierarchically encoded key point set. The improved RegNet multi-output branch network is input with the trajectory image and key point set to extract features from graphic, number, letter and Chinese character elements respectively and generate multi-branch feature representations. A multi-resolution map and range linkage algorithm is adopted to select the corresponding resolution map layer according to the track length and range setting, so as to realize the linkage display of track and map and the matching of position; Normalization and feature fusion are performed on the multi-branch feature representation to generate a unified plotting feature vector. The unified plotting feature vector is then associated with map coordinates and range information to form a complete plotting data structure. A comparison algorithm is executed on the plotted data structure to compare the plotted content with the standard answer point by point, identify error points, offsets and symbol errors, and generate plotted accuracy data. The assessment output is generated based on the accuracy data of the plotting, including the recognition results of graphic, number, letter, and Chinese character elements and a deviation analysis report.

2. The radar image rendering method based on image recognition according to claim 1, characterized in that, The construction of the initial data sequence for the trajectory, recording the start and end points and length of the line segments, specifically includes: Read the radar target trajectory image, extract the set of trajectory pixels of the trajectory line segments drawn by the user, perform connected region scanning on the set of trajectory pixels of the trajectory line segments, and generate a sequence of trajectory line segment numbers; Perform endpoint detection on the set of trajectory pixels corresponding to each track segment number, extract the position coordinates where the pixel coordinate change direction is interrupted, and generate the starting point coordinates and ending point coordinates of the line segment. Perform coordinate position association between the starting coordinates and ending coordinates of the line segment to generate a line segment coordinate pair sequence. Read the target number, target type identifier, target heading angle value and target speed value from the radar target information, and write the target number, target type identifier, target heading angle value and target speed value into the corresponding line segment coordinate pair position to generate a target associated line segment record. Perform pixel distance calculation on the line segment coordinate pair sequence, count the number of consecutive pixels between the start coordinate and end coordinate of the line segment, and generate the line segment length value. Read the radar image range parameters and image scale parameters, perform scale conversion on the line segment length values, and generate the actual track length values. The initial data sequence of the trajectory is formed by arranging the fields in the following order: line segment number, line segment start coordinates, line segment end coordinates, line segment length value, actual track length value, target number, target type identifier, target heading angle value, and target speed value.

3. The radar image rendering method based on image recognition according to claim 1, characterized in that, The algorithm for generating key points of the track on the initial data sequence includes the following steps: Read the route records in the initial data sequence of the track, establish a track segment buffer area according to the segment number, write the starting coordinates and ending coordinates of the segment into the endpoint field, and write the segment length value into the length field. Within the track segment buffer area, read two adjacent track segments in order of segment number, and calculate the difference in horizontal coordinates, the difference in vertical coordinates, and the distance between endpoints between the endpoint coordinates of the previous track segment and the starting coordinates of the next track segment. The endpoint spacing value is compared with the preset connection distance threshold. If the endpoint spacing value does not exceed the preset connection distance threshold, the coordinates of the end point of the previous track segment and the coordinates of the start point of the next track segment are merged into the connection point coordinates and written into the connection point buffer. For each track segment, the direction vector is calculated by performing a direction vector calculation on the coordinates of the starting point and the ending point of the segment. The angle between the direction vectors of two adjacent track segments is calculated to generate the track turning angle value. The trajectory turning angle value is compared with the preset turning angle threshold. When the trajectory turning angle value reaches the preset turning angle threshold, the coordinates of the connection point or the coordinates of the adjacent endpoints are written into the turning point buffer. Read the line segment length value from the length field, compare the line segment length value with the preset segment length threshold, and when the line segment length value reaches the preset segment length threshold, perform coordinate interpolation between the starting coordinates and ending coordinates of the line segment according to the preset segment spacing to generate a sequence of segment point coordinates; Perform coordinate deduplication on the connection point buffer, turning point buffer, and segment point coordinate sequence, retaining a key point record with a coordinate difference of zero, and deleting duplicate key point records; The key point coordinate records are arranged according to the fields of line segment number, key point coordinate, key point type identifier, and key point sequence number to form a set of track key points.

4. The radar image rendering method based on image recognition according to claim 1, characterized in that, The step of semantically annotating and hierarchically encoding key points based on track topology features to form a hierarchically encoded key point set specifically includes: Read the line segment number, key point coordinate record, key point type identifier and key point sequence number from the track key point set, and establish a key point topology buffer according to the line segment number; Read the coordinate records of key points under the same line segment number in the key point topology buffer area, arrange them according to the key point numbering order, and generate the key point sequence within the line segment. Read the sequence of key points within the line segment corresponding to the adjacent line segment number, perform coordinate difference calculation on the coordinate record of the last key point of the previous line segment and the coordinate record of the first key point of the next line segment, and generate a line segment connection relationship identifier. Write the line segment connection relationship identifiers into the key point topology cache area to form a track topology feature that includes endpoint relationships, intra-line segment order relationships, and inter-line segment connection relationships; Read the topological features of the flight path and the key point type identifiers, write start and end semantic identifiers to the endpoint coordinate records, write connection semantic identifiers to the connection point coordinate records, write turning semantic identifiers to the turning point coordinate records, and write segment semantic identifiers to the segment point coordinate records to form semantically labeled key point records. The first layer of coding is generated according to the line segment number in the track topology feature, the second layer of coding is generated according to the key point sequence number, and the third layer of coding is generated according to the semantic identifier. The first layer of coding, the second layer of coding, and the third layer of coding are written into the semantic annotation key point record. The semantically labeled key point records are sorted by field, arranged according to line segment number, key point coordinate record, key point sequence number, line segment connection relationship identifier, semantic identifier, first-level code, second-level code and third-level code, forming a hierarchical coded key point set.

5. The radar image rendering method based on image recognition according to claim 1, characterized in that, The process of inputting the trajectory image and key point set into the improved RegNet multi-output branch network, extracting features from graphic, number, letter, and Chinese character elements respectively, and generating multi-branch feature representations includes the following steps: Read the flight track image and the hierarchical encoded key point set, and map the key point coordinates to the flight track image to form a key point localization image; A fixed-size local region is cropped from the key point localization image centered on the key point coordinates. Grayscale conversion, edge detection and contour extraction are performed on the local region to generate local contour information. Based on local contour information, the local area is divided into candidate areas of line segments, strokes or symbols, and the endpoints and forking positions are marked to form a set of candidate areas for plotting elements. The candidate regions of plotted elements are written into the shared feature extraction layer of the improved RegNet multi-output branch network, and convolution calculation, normalization calculation and activation mapping are performed to generate a shared feature map. The shared feature map is input into the graphic branch, number branch, letter branch and Chinese character branch respectively. The graphic branch extracts the line contour and turning features to generate graphic feature representation. The digit branch performs feature extraction on the stroke endpoints and arrangement order of candidate digit regions to generate digit feature representations; The letter branch performs feature extraction on the stroke direction, closed structure, and character proportion of the letter region to generate letter feature representations; The Chinese character branch performs feature extraction on the horizontal and vertical strokes, left-falling and right-falling strokes, and component positions of the Chinese character region to generate Chinese character feature representations; The graphic feature representation, numerical feature representation, letter feature representation, and Chinese character feature representation are arranged and combined according to the key point order to generate a multi-branch feature representation.

6. The radar image rendering method based on image recognition according to claim 1, characterized in that, The aforementioned multi-resolution map and range linkage algorithm, which selects the corresponding resolution map layer based on the track length and range settings to achieve linked display of the track and map and location matching, specifically includes: Read the set of key points and track length information, calculate the overall track span, determine the applicable map resolution level based on the track span and range parameters, and generate a resolution selection list; Map the coordinates of key points on the flight path to the pixel coordinates of the candidate resolution map layer, perform preliminary position matching through the coordinate transformation matrix, and re-select the nearest resolution map for matching and fine-tuning for the mapped points that fall outside the map boundary; Read the map grid information on the finally selected map resolution layer, calculate the offset between the track key points and the map reference points, and adjust the key point coordinates until they are aligned with the map grid.

7. The radar image rendering method based on image recognition according to claim 1, characterized in that, The process of performing normalization and feature fusion on the multi-branch feature representation to generate a unified plotted feature vector includes the following steps: Read the multi-branch feature representations of graphic branches, number branches, letter branches and Chinese character branches, perform normalization processing on each branch feature, including mean standardization and variance normalization, and generate a standardized feature matrix; Cross-branch alignment is performed on the standardized feature matrix, aligning the corresponding positions of the features in each branch according to the key point order, and concatenating and weighting the feature dimensions to generate a preliminary fused feature vector; The initial fused feature vectors are subjected to nonlinear mapping and linear combination operations to adjust the feature distribution range and relative weights, thereby generating a unified plotted feature vector.

8. The radar image rendering method based on image recognition according to claim 1, characterized in that, The comparison algorithm performed on the plotted data structure specifically includes: Read the unified plotted feature vector and the corresponding key point coordinates, map the key point coordinates to the corresponding positions in the standard answer image, and establish the plotted point matching relationship; For each matching point, calculate the horizontal coordinate deviation, vertical coordinate deviation, and angular deviation, and compare whether the symbol categories are consistent, mark the error points, and generate the deviation value. The error points and deviation values ​​are arranged according to the order of key points, line segment numbers and element types. The deviations are accumulated and statistically analyzed to generate error summary information. The statistical results of deviations, the location of error points, and information on symbol errors are summarized to form data for evaluating the accuracy of plotting.

9. The radar image rendering method based on image recognition according to claim 1, characterized in that, The process of generating assessment outputs based on the accuracy data of the plotting, including the recognition results of graphic, numerical, letter, and Chinese character elements and a deviation analysis report, specifically includes: Read the plotting accuracy evaluation data and error summary information, and classify and organize the error point coordinates, deviation values ​​and symbol categories according to the key point order and line segment number; Recognition points are extracted for graphic elements, number elements, letter elements, and Chinese character elements according to element type. The correct recognition rate, deviation statistics, and symbol consistency are calculated for each type of element. The deviation data is categorized into tables and lists by element type. The coordinates of the error points, the deviation values, and the error symbol categories are recorded one by one to form a deviation analysis dataset. The identification points and deviation data are sorted and combined according to element type, line segment number, and key point order to generate plotting assessment output, including the identification results and corresponding deviation information of graphic, number, letter, and Chinese character elements.

Images