Diopter parameter analysis model optimization method and system

Abstract

The invention relates to the technical field of image pattern recognition, in particular to a diopter parameter analysis model optimization method and system, and the method comprises the following steps: collecting image sequence recognition region features, constructing edge segments, judging the structural continuity, executing concurrent recognition and trajectory analysis, and outputting a diopter parameter analysis optimization graph. According to the method, through constructing curvature section and gray vector distribution in an image sequence, extracting continuous features of a boundary structure, screening stable graph frames in combination with inter-frame direction consistency, recording corresponding boundaries by using a dual-channel recognition rhythm alignment mode, establishing a structure difference value and generating an error trajectory graph; the stable nodes are identified based on the offset trend, and the key graph segment is backtracked, so that the restoration coherence and response consistency of the image structure are enhanced, and the expression ability of the graph result to the change characteristics of the crystalline lens and the accuracy of parameter extraction are improved.

Description

Technical Field

[0001] This invention relates to the field of image pattern recognition technology, and in particular to a method and system for optimizing a refractive power parameter analysis model. Background Technology

[0002] Image pattern recognition technology involves the detection, analysis, and classification of targets, features, or patterns in static images or dynamic videos. Its core aspects include image data acquisition and preprocessing, feature extraction, feature matching, pattern training, and classification. It is a crucial interdisciplinary field at the intersection of computer vision, artificial intelligence, and machine learning. This area is widely applied in industries such as security monitoring, medical diagnosis, autonomous driving, and intelligent manufacturing. The structure, texture, color, and motion information in images are converted into feature vectors that can be used for pattern analysis, and then trained models are used for recognition. In this field, the choice of training methods, image annotation mechanisms, verification mechanisms, and recognition algorithms directly affects the accuracy and stability of the pattern recognition system. Traditional refractive parameter analysis model optimization methods refer to techniques that use neural network models to process ophthalmic images to analyze the refractive power changes reflected in the lens region. These methods primarily focus on improving the model's ability to analyze changes in refractive parameters during training. Traditional methods typically construct training and validation sets based on fixed image datasets, performing unified verification after training, lacking a real-time cross-validation mechanism. During the training process, the image order remains unchanged, and the training model processes the dataset sequentially in a single thread. The model iteration process lacks real-time supervision and dynamic adjustment capabilities.

[0003] Existing technologies lack a mechanism for recognizing dynamic changes in boundary curvature within lens image sequences during refractive power parameter analysis. Image processing also lacks the ability to analyze inter-frame grayscale changes and structural continuity. The model training phase neglects screening criteria for edge direction jumps and image frame stability. Image recognition results often accumulate errors due to structural blurring or boundary breaks. There is a lack of differentiated recognition methods targeting the internal response characteristics of image frames. Image results fail to reflect the continuous response characteristics of the lens in the time dimension. The processing flow does not provide time-series comparison methods for multi-channel data, resulting in weak perception of spatial offset trends in parameter recognition. This leads to significant limitations in the accuracy of refractive power change trend recognition and image structure restoration. Summary of the Invention

[0004] To achieve the above objectives, the present invention adopts the following technical solution: a method for optimizing a refractive power parameter analysis model, comprising the following steps: S1: Acquire a sequence of lens images, identify lens regions based on pixel position relationships, perform grayscale change direction calculation in each frame, construct continuous edge segments through grayscale difference, and integrate edge segments to generate a curvature graphic structure; S2: Perform inter-frame structure comparison on the curvature graphic structure, perform continuity judgment based on the coverage area of ​​edge connected segments in adjacent frames, exclude frame graphics with interrupted edge segments, and output a stable set of frame graphics. S3: Divide the stable frame graphic set into two types of graphic data according to the parity of the frame index, with odd-numbered frame graphics as the first type and even-numbered frame graphics as the second type. These are respectively assigned to the main channel and the secondary channel for concurrent operation. By loading frame by frame, structural recognition processing is performed to generate contour boundary graphics. S4: Based on the contour boundary graphic, calculate the structural coordinate difference, establish a trajectory path based on the edge position difference generated in the two channels of the same frame graphic, and construct an error trajectory graphic; S5: Compare the trajectory direction consistency of the error trajectory graphics, identify structural nodes with stable trajectory changes, extract the lens boundary region graphics corresponding to the stable nodes at the time points from the identified contour boundary graphics, and output the refractive power parameter analysis and optimization graphics.

[0005] As a further embodiment of the present invention, the curvature graphic structure includes grayscale gradient distribution, boundary curvature features, and region marker vectors; the stable frame graphic set includes orientation-consistent frames, edge-closure frames, and frames with continuous inter-frame angles; the contour boundary graphic includes main channel edge data, secondary channel edge data, and temporal alignment information; the error trajectory graphic includes offset distance vectors, path direction lines, and node connection sequences; and the diopter parameter analysis and optimization graphic includes response node features, boundary structure fragments, and offset trend patterns.

[0006] As a further aspect of the present invention, the frame-by-frame loading and structural recognition processing refers to performing structural feature analysis and recognition on each frame of the image sequence in turn.

[0007] As a further aspect of the present invention, the structural nodes with stable trajectory changes refer to edge structural feature points in the error trajectory graph that exhibit consistent directional changes and continuous values ​​greater than a preset threshold.

[0008] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Acquire a sequence of lens images, aggregate pixels with a reflection intensity higher than the preset lens grayscale reference value in the central block of each frame image, obtain the boundary coordinates and perform closure fitting processing to generate a lens boundary coordinate set; S102: Based on the lens boundary coordinate set, extract the gray value matrix within the boundary, calculate the normal direction of the boundary point, compare the included angle between adjacent normals and record continuous segments where the direction change exceeds the preset gray direction change threshold, and generate a gray direction abrupt change segment sequence. S103: Call the gray-scale direction abrupt change segment sequence, extract the start and end coordinates of the segment, calculate the corresponding arc length, included angle and radius, construct the position and curvature mapping vector, integrate all segment mapping results, and generate curvature graphic structure.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Call the grayscale vector sequence of each frame in the curvature graphic structure, extract the vector direction values ​​of adjacent positions, and perform angle calculation on the vector direction of the same position in each pair of consecutive frames. Filter the coordinate segments whose angle values ​​change more than a preset fluctuation threshold in consecutive frames, and generate a sequence of direction change segments. S202: Based on the direction change segment sequence, frequency statistics are performed on the corresponding frames of the sequence segment to determine whether the number of regions with sudden direction changes in each frame exceeds the preset direction change frequency threshold, and the corresponding graphic frame index sequence is recorded. Graphic frames with jump directions are removed to obtain an abnormal frame exclusion sequence. S203: Invoke the abnormal frame exclusion sequence, perform closure detection on the edge trajectories in the remaining frame graphics, calculate the spatial distance between the start and end boundary points and determine whether it is less than the preset boundary closure threshold, and perform index aggregation to generate a stable frame graphics set.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Divide the stable frame graphic set into two types of graphic data according to the parity of the frame index. Perform index splitting for each group of graphic data and establish a corresponding data frame sequence identifier. Associate one group of data frame sequence identifiers with the main channel loading queue and the other group of data frame sequence identifiers with the secondary channel loading queue. Perform channel mapping to generate a dual-channel graphic data sequence. S302: Based on the dual-channel graphic data sequence, obtain the timestamp information carried by the data frame, perform timestamp difference calculation on the data frames of the main channel and the secondary channel, determine whether the difference is less than the preset channel alignment threshold, and perform synchronization marking on the data frames that meet the conditions to obtain the channel synchronization frame sequence. S303: Call the channel synchronization frame sequence, extract the edge coordinate point set of the corresponding frame of the main channel and the secondary channel in each recognition cycle, perform position registration on the edge coordinates of the two channels and perform closed boundary judgment, record the registered contour trajectory and perform structured integration to establish contour boundary graphics.

[0011] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Based on the outline boundary graphics, perform a corresponding search on the edge point coordinates of the same numbered graphics of the main channel and the secondary channel, calculate the structural coordinate difference of the corresponding points in the same coordinate system, and arrange the differences in point order to generate a structural coordinate difference sequence. S402: Call the structure coordinate difference sequence, calculate the Euclidean distance for each pair of edge point differences and construct a direction vector, connect the vectors corresponding to the same numbered graphic in point order to form a path line, and record the timestamp sequence index to obtain a set of distance offset path lines; S403: Based on the set of distance offset path lines, perform vector connection on the continuous path lines in chronological order, perform aggregation and structured integration on all path lines, establish a continuous relationship in time and space, and generate an error trajectory graph.

[0012] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Based on the error trajectory graph, extract the vector sequence of each path line, perform point-by-point difference calculation on the direction angle and amplitude change value of the vector, identify continuous segments with change amplitude less than the preset direction stability threshold and consistent direction offset, mark the start and end positions, and generate a stable offset graph node set. S502: Call the stable offset graphic node set, locate the corresponding contour boundary graphic frame in the main channel and the secondary channel according to the time index of each graphic node, extract the corresponding frame graphic area at each node time point and perform synchronous interception, recombine the frame fragments according to the node index sequence, and obtain the node frame combined structure graphic. S503: Based on the node frame combination structure diagram, perform a unified normalization transformation on the edge coordinate points of the main channel and secondary channel regions, and perform morphological aggregation and curvature adjustment processing on the normalized boundary structure matrix to generate a diopter parameter analysis and optimization diagram.

[0013] A refractive power parameter analysis model optimization system, comprising: The curvature structure extraction module is used to implement S1: acquiring a sequence of lens images, identifying lens regions based on pixel position relationships, performing grayscale change direction calculations in each frame of the image, constructing continuous edge segments through grayscale differences, and integrating edge segments to generate a curvature graphic structure; The stable frame filtering module is used to implement S2: perform inter-frame structure comparison on the curvature graphic structure, perform continuity judgment based on the coverage area of ​​edge connected segments in adjacent frames, exclude frame graphics with interrupted edge segments, and output a set of stable frame graphics. The dual-channel contour recognition module is used to implement S3: the stable frame graphic set is divided into two types of graphic data according to the parity of the frame index, with odd-numbered frame graphics as the first type and even-numbered frame graphics as the second type, which are respectively assigned to the main channel and the secondary channel for concurrent operation. By loading frame by frame and performing structural recognition processing, contour boundary graphics are generated. The error trajectory construction module is used to implement S4: based on the contour boundary graphic, calculate the structural coordinate difference, establish a trajectory path based on the edge position difference generated in two channels of the same frame graphic, and construct the error trajectory graphic; The structural node identification and parameter analysis module is used to implement S5: compare the trajectory direction consistency of the error trajectory graphics, identify structural nodes with stable trajectory changes, extract the lens boundary region graphics corresponding to the stable nodes at the time points from the identified contour boundary graphics, and output the refractive parameter analysis and optimization graphics.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by constructing curvature segments and grayscale vector distribution in an image sequence, continuous features of the boundary structure are extracted. Stable graphic frames are selected by combining inter-frame directional consistency. The corresponding boundaries are recorded using a dual-channel recognition rhythm alignment method. Structural differences are established and error trajectory graphics are generated. Stable nodes are identified based on offset trends and key graphic segments are traced back. This enhances the reconstruction coherence and response consistency of the image structure, improves the ability of the graphic results to express the changes in the lens and the accuracy of parameter extraction. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention; Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent.

[0020] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] Please see Figure 1 This invention provides a method for optimizing a refractive power parameter analysis model, comprising the following steps: S1: Acquire a sequence of lens images. For the high-reflectivity grayscale region located in the central block of each frame image, perform contour closure processing, extract lens boundary segments, and mark the curvature change segments between the vector start and end points in the image matrix by comparing the grayscale gradient directions inside the boundary. Integrate all segments to generate a curvature graphic structure. S2: Call the grayscale vector sequence of each frame in the curvature graphic structure, perform vector direction comparison between adjacent frames, identify edge segments with the same angle change and position distribution, remove graphic frames with frequent changes in direction and abrupt changes in direction, retain only graphic frames with continuous direction and closed edges, and output a set of stable frame graphics. S3: Divide the stable frame image set into two types of image data according to the parity of the frame index, and import them into the image recognition process of the main channel and the secondary channel respectively. Control the loading rhythm of the two channels by timestamp alignment. Record the contour boundary images generated by the main channel and the secondary channel in each recognition cycle to establish the contour boundary images. S4: Based on the contour boundary graphics, calculate the structural coordinate difference of the contour boundary graphics output by the main channel and the secondary channel, construct the distance offset vector based on the edge point sequence of each pair of graphics with the same number, connect the continuous vectors in time order, aggregate all path line sets, and output the error trajectory graphics; S5: Analyze the direction and fluctuation of the path lines in the error trajectory graph, identify the stable graphic nodes with consistent offset trends in the trajectory, trace back the frame graphic segments corresponding to the node time points in the contour boundary graphics of the main channel and secondary channel, combine them to form the lens response structure graphic, and output the refractive parameter analysis and optimization graphic.

[0023] The curvature graphic structure includes grayscale gradient distribution, boundary curvature features, and region marker vectors. The stable frame graphic set includes orientation-consistent frames, edge-closure frames, and frames with continuous inter-frame angles. The contour boundary graphic includes main channel edge data, secondary channel edge data, and temporal alignment information. The error trajectory graphic includes offset distance vectors, path direction lines, and node connection sequences. The diopter parameter analysis and optimization graphic includes response node features, boundary structure fragments, and offset trend patterns.

[0024] Please see Figure 2 The specific steps of S1 are as follows: S101: Acquire a sequence of lens images, aggregate pixels with a reflection intensity higher than the preset lens grayscale reference value in the central block of each frame image, obtain the boundary coordinates and perform closure fitting processing to generate a lens boundary coordinate set; The device is equipped with an integrated light source and camera module. During image acquisition, the device is stably worn on the user's head, with the user's gaze directed towards a fixed area. Under stable or off ambient light conditions, the infrared light source is activated to illuminate the eye, causing the lens area to produce a reflective feature with obvious grayscale contrast. The camera module continuously acquires image sequences at a fixed frequency. After each frame is acquired, the central area is cropped into a standard window, which is a 256×256 pixel area in the center of the image. The grayscale values ​​of all pixels within this area are extracted, and the current lens grayscale reference value is set based on the arithmetic mean of the grayscale median sequence of the central area of ​​the past 10 frames. In practical scenarios, for example, if the median grayscale values ​​of the first 10 frames are 110, 112, 114, 108, 111, 113, 115, 112, 109, and 114, the corresponding average is 111.8. Therefore, the grayscale baseline value for the lens is set to 112 (which can be rounded up according to actual application requirements). Then, all pixels in the central region of the current image frame are judged, and pixels with grayscale values ​​greater than this baseline value are aggregated into a set of high-reflectivity pixels. During the aggregation process, the principles of row and column continuity and spatial adjacency are used to connect the pixel regions. Domains are grouped into the same aggregation unit. After multiple units are extracted, they are sorted by area from largest to smallest. The unit with the largest area is selected as the main lens region. Boundary coordinates are extracted from the edge contour of this region, and the sequential relationship of the boundary points is analyzed point by point. The discrete edge points are fitted to form a closed contour. During the fitting process, the boundary point set is used as input, and spline interpolation is used to construct a continuous curve to fill in the boundary gaps and discontinuities. Finally, the boundary point set is output as the boundary coordinate set of the lens in the current image frame. This set consists of sequentially arranged coordinate pairs, covering the complete lens edge contour region.

[0025] S102: Based on the lens boundary coordinate set, extract the gray value matrix within the boundary, calculate the normal direction of the boundary point, compare the angle between adjacent normals and record continuous segments where the direction change exceeds the preset gray direction change threshold, and generate a sequence of gray direction abrupt change segments. For each boundary point, a corresponding local gray-level neighborhood is established in the image. Pixel gray values ​​within a 3×3 area are selected, and the gray-level differences in the horizontal and vertical directions of this region are extracted based on their relative positions. The direction of gray-level change at the point is determined by jointly considering the horizontal and vertical gray-level change directions, thus obtaining the normal direction of that point. After extracting the normal directions of all boundary points, the angle between the normal directions of adjacent boundary points is calculated, and the result is recorded as an angle. When the angle between a pair of normals exceeds a set direction change threshold (set to 10 degrees), the point is identified as a point of abrupt change in gray-level direction. In engineering, this value can be set empirically based on image clarity and edge transition characteristics. During the sequential traversal of the entire boundary coordinate set, once it is found that the angle between the normals of multiple consecutive points exceeds the threshold, the boundary points of that segment are classified as a sudden change segment, and the boundary coordinates from the start point to the end point of the segment are recorded. When processing a frame of image, multiple such sudden change segments can be identified. For example, if four segments are identified in an image, each containing about eight boundary points, these four segments are recorded as start and end coordinate sets for subsequent curvature analysis processing. This sequence is the gray-scale direction sudden change segment sequence, which will be called later for boundary configuration feature analysis.

[0026] S103: Call the sequence of gray-scale direction change segments, extract the start and end coordinates of the segments, calculate the corresponding arc length, included angle and radius, construct the position and curvature mapping vector, integrate all segment mapping results, and generate curvature graphic structure; The starting and ending coordinates of each segment are extracted, and the boundary points within each segment are traversed sequentially. The arc length of these points is calculated, i.e., the spatial distance between adjacent points is measured point by point and accumulated to obtain the total length of the entire segment. This process uses standard Euclidean distance between points and sums them sequentially. After obtaining the arc length of the boundary line segment, the geometric relationship of the boundary points is used to estimate the corresponding circular arc model. Based on the vector relationship formed between the starting point, ending point, and the assumed center of the circle, the included angle of the arc corresponding to the segment is calculated. Simultaneously, the arc length is divided by the included angle to obtain the corresponding radius of curvature. The radius represents the curvature of the lens boundary segment. The starting and ending coordinates of the segment are used to construct its position marker in the entire boundary. The spatial position, radius of curvature, included angle, arc length and other parameters of each segment are integrated into a set of vectors to form the curvature mapping structure of the segment. The mapping results of all segments are merged to form the overall curvature graphic structure corresponding to a frame of image. This structure reflects the local deformation distribution characteristics of the lens edge and can be used as an input reference for the model in the subsequent training stage to model the correlation between the trend of refractive power change and the edge geometric features.

[0027] Please see Figure 3 The specific steps of S2 are as follows: S201: Call the grayscale vector sequence of each frame in the curvature graphic structure, extract the vector direction values ​​of adjacent positions, and perform angle calculation on the vector direction of the same position in each pair of consecutive frames. Filter the coordinate segments whose angle values ​​change more than the preset fluctuation threshold in consecutive frames, and generate a sequence of direction change segments. First, the corresponding vector sequence needs to be read frame by frame, and the position coordinates and direction values ​​of each vector need to be extracted to establish an ordered structure of the vector sequence within the frame. The direction value of each vector is marked with an angle, and it forms a direction difference pair with the direction of the vectors at adjacent positions. According to the arrangement order of the image edge trajectory, the direction values ​​of adjacent vectors are compared pairwise. For example, if the direction of the first vector is 62 degrees and the second is 67 degrees, the direction difference is 5 degrees. And so on, to construct the direction difference sequence of the frame. This sequence records the direction fluctuation of each adjacent vector in the current frame. Then, the direction values ​​of the same position are extracted and compared between frames. That is, for each pair of vectors with the same coordinate position in consecutive frames, the direction values ​​in frame 1 and frame 2 are recorded respectively, and the angle difference between the two is used as the fluctuation index. For example, if the direction of a certain position in frame t is 70 degrees, and the direction of the same position in frame t+1 is 83 degrees, the direction angle fluctuation is 13 degrees. If the angle difference of the position changes more than a specified amplitude in multiple frames, for example, the direction is 68, 80 and 92 degrees in three consecutive frames, then it is determined that there is a significant direction change trend of the coordinate. The minimum standard for directional angle fluctuation in consecutive frames is set to 10 degrees. As long as any consecutive coordinate segment exhibits such fluctuation in multiple frames, the segment is recorded in the directional change segment sequence, and its start and end coordinate positions in the frame and the frame index in which it appears are marked simultaneously. Finally, this process will select a set of coordinate segments with directional fluctuation behavior from all frames, which will serve as the basis for subsequent anomaly judgment, and thus obtain the directional change segment sequence.

[0028] S202: Based on the sequence of direction change segments, frequency statistics are performed on the corresponding frames of the sequence segments to determine whether the number of regions with sudden direction changes in each frame exceeds the preset frequency threshold for sudden direction changes, and the corresponding graphic frame index sequence is recorded. Graphic frames with jump directions are removed to obtain the abnormal frame exclusion sequence. Frequency statistics of the relevant graphic frames are required. Specifically, the graphic frame index corresponding to each record in the segment sequence is extracted first. Then, the number of directional change coordinate segments contained in each frame is aggregated according to the frame number. The cumulative number of change segments in each frame is counted, and the number of times a position marked as a directional change in each frame is counted. For example, if there are 5 different positions in a frame where directional change occurs, the statistical result is 5. This value reflects the density of directionally unstable regions in the current frame. The statistical result is then compared with a set directional change frequency threshold. This threshold is empirically set based on the average number of changes and fluctuation range of all graphic frames, usually set as the average change frequency plus two units of standard deviation. In the test sample, if the average is 3 segments and the standard deviation is 1, the frequency threshold can be set to 5 segments. It is then determined whether the number of changes in each frame exceeds 5. If it does, the frame is marked as an abnormal frame and its index is recorded. Each frame is judged independently during the statistical process, unaffected by preceding or following frames. All frame indices that meet the change frequency limit are summarized to form an abnormal frame exclusion sequence. This sequence will be used as a filtering condition for subsequent boundary quality detection to exclude graphic frames with poor stability due to drastic changes in edge structure.

[0029] S203: Call the abnormal frame exclusion sequence, perform closure detection on the edge trajectories in the remaining frame graphics, calculate the spatial distance between the start and end boundary points and determine whether it is less than the preset boundary closure threshold, and perform index aggregation to generate a set of stable frame graphics; Edge trajectory closure detection is performed on the remaining image frames. The processing flow sequentially reads the edge point sequence in each frame's image structure. First, the coordinates of the start and end points of the edge are identified, and the distance between the two points in the image coordinate space is measured. This distance reflects whether the edge forms a complete closed loop. The boundary closure threshold is determined based on the image resolution and the expected edge continuity. For example, when the image size is 512×512, the threshold is set to 10 pixels. If the spatial distance between the start and end points is less than this threshold, the image trajectory is considered to form a closed loop; if it exceeds this threshold, the image boundary is considered not closed and does not meet the geometric integrity requirements. For each... The frame graph structure executes the same judgment logic, recording the frame index that meets the closure condition into the qualified list, and otherwise discarding it. In actual operation, if the start and end points of a frame are located at (50, 130) and (57, 136) respectively, the distance between the two points is about 9 pixels, which is lower than the 10-pixel threshold, and it is judged as a closed frame. The frame index is retained. Finally, after traversing all frames, the qualified graph selection is completed, and the set of all frame indexes that meet the closure standard is output to form a stable frame graph set. This set contains all the graph frame data after abnormal fluctuation rejection and geometric closure detection, which can be directly used as a stable graph data source for subsequent model training samples.

[0030] Please see Figure 4 The specific steps of S3 are as follows: S301: Divide the stable frame graphic set into two types of graphic data according to the parity of the frame index. Perform index splitting for each group of graphic data and establish a corresponding data frame sequence identifier. Associate one group of data frame sequence identifiers with the main channel loading queue and the other group of data frame sequence identifiers with the secondary channel loading queue. Perform channel mapping to generate a dual-channel graphic data sequence. First, the index and image path information of all graphic frames in the set need to be read. Then, a parity grouping operation is performed according to the frame number, that is, graphic frames with odd indices are assigned to the first group, and graphic frames with even indices are assigned to the second group, forming two independent graphic data sets. Then, index splitting is performed on each graphic set, extracting the number, path, timestamp, and other information of each frame graphic and assembling them into a data frame sequence structure, generating a first sequence identifier list and a second sequence identifier list respectively. For example, if the index of the first group of graphic frames is 1, 3, 5, 7, and the index of the second group of graphic frames is 2, 4, 6, 8, then the data frame sequence identifiers are two sets {1, 3, 5, 7} and {2, 4, 6, 8}. Then, the first sequence... The identifiers are assigned to the main channel loading queue and pushed onto the stack in sequence, marking them as the main channel reading objects. The second sequence identifiers are assigned to the secondary channel loading queue, pushed onto the stack in sequence, and marked as the secondary channel reading objects. In the loading mechanism, the main channel is scheduled first, while the secondary channel is scheduled with a delay. The allocation of the two channels uses separate memory address blocks to cache the processing data of the graphics frames. The channel mapping process uses a synchronous mapping table structure to record the one-to-one frame index relationship between the main channel and the secondary channel. After all graphics frames are loaded, the corresponding frames in the two loading queues are matched sequentially according to the original acquisition time, and aligned to form a unified dual-channel graphics data sequence, which serves as the input data structure for subsequent synchronous processing.

[0031] S302: Based on the dual-channel graphic data sequence, obtain the timestamp information carried by the data frame, perform timestamp difference calculation on the data frames of the main channel and the secondary channel, determine whether the difference is less than the preset channel alignment threshold, and perform synchronization marking on the data frames that meet the conditions to obtain the channel synchronization frame sequence. For each image frame in each channel, the timestamp information is read. This timestamp is a time identifier recorded by the device's internal real-time clock at the time of image acquisition, in milliseconds. The format is usually UNIX timestamp format. For example, the timestamp of the first frame of the main channel is 1671234567890, and the timestamp of the first frame of the secondary channel is 1671234567896, with a difference of 6 milliseconds. Subsequently, the timestamp difference is calculated for data frames at the same sequence index position in each pair of main and secondary channels. The set channel alignment threshold is based on the acquisition frame rate and the allowable time error range. For example, if the frame rate is 30... At fps, the interval between each frame is approximately 33 milliseconds. When the allowable error is no more than 20%, the alignment threshold can be set to 7 milliseconds. Therefore, the timestamp difference needs to be determined whether it is less than or equal to this set value. If it is less than the set value, the pair of frames is marked as channel synchronization frames, and the correspondence between its main channel frame number and secondary channel frame number is recorded. If it exceeds the set value, the pair of frames is not marked for synchronization and is skipped directly without participating in subsequent position registration processing. Following the above operation, synchronization judgment is performed on all frame pairs, and finally the selection of valid frame pairs is completed. Data frame pairs that meet the conditions are combined to form a channel synchronization frame sequence for structural alignment and contour integration operations.

[0032] S303: Call the channel synchronization frame sequence, extract the edge coordinate point set of the corresponding frame of the main channel and the secondary channel in each recognition cycle, perform position registration on the edge coordinates of the two channels and perform closed boundary judgment, record the registered contour trajectory and perform structured integration to establish contour boundary graphics; The corresponding image frames are extracted from the main channel and the secondary channel, and their edge coordinate point sets are obtained. The edge points of each image frame are directly accessed through the previously established boundary coordinate set. Typically, each coordinate set contains hundreds of consecutive coordinate points. Then, a registration operation is performed on the edge coordinate points of the image frames from the two channels. The registration operation first performs coordinate system normalization, mapping the edge coordinates under different channels to the same pixel reference frame. For example, the coordinate range of the main channel is normalized to 0 to 1, and the coordinates of the secondary channel are also normalized to make their position scale consistent. Then, the coordinates are aligned according to the starting point of the main channel edge. For each main channel edge point, the nearest secondary channel edge point is found. The system performs positional matching on the edge points, constructs a list of edge point pairs, and records the distance difference between all corresponding points. Edge point pairs with a distance difference exceeding 10 pixels are discarded, and paired data with a registration accuracy within 10 pixels are retained. After registration, the system performs closure checks on the start and end points of the main channel edge and the end point of the secondary channel edge after registration, determining whether the distance between the two endpoints is less than the set closure threshold. If the threshold is met, the contour registration result within this recognition cycle is recorded, and the complete set of registered edge points is integrated and saved in the form of a structured list. Finally, a contour boundary graphic under channel synchronization is established for subsequent multi-frame trend analysis.

[0033] Please see Figure 5The specific steps of S4 are as follows: S401: Based on the contour boundary graphics, perform a corresponding retrieval of the edge point coordinates of the same numbered graphics in the main channel and the secondary channel, calculate the structural coordinate difference of the corresponding points in the same coordinate system, and arrange the differences in point order to generate a structural coordinate difference sequence. In processing images with the same frame number in both the main and secondary channels, it is necessary to first extract the edge coordinate point set corresponding to the same frame number in both channels. The extraction of the two point sets maintains the same point order, that is, the edge point sequence of the main channel is P1={p11, p12, ..., p1...} n} and the secondary channel edge point sequence P2={p21, p22, ..., p2 n In the diagram, the i-th point should represent the same edge position in both sets. The coordinates of each pair of points with the same number in the two sets are retrieved. For example, in frame 25, the coordinates of the edge point p11 of the main channel are (122, 140), and the corresponding point p21 of the secondary channel is (127, 138). After aligning the two points in a unified coordinate system, the differences between them in the x-axis and y-axis directions are calculated, which are +5 and -2, respectively. This structural coordinate difference is recorded as a pair of difference values ​​(5, -2). The same difference is extracted for each pair of points with the same number. Finally, all difference pairs are arranged by point number to form the structural coordinate difference sequence under the graphic frame. This sequence can be used to analyze the degree of spatial offset of the edge contour structure between channels and to provide direct input for subsequent path line construction and error aggregation.

[0034] S402: Call the structural coordinate difference sequence, calculate the Euclidean distance for each pair of edge point differences and construct a direction vector, connect the vectors corresponding to the same numbered graphic in point order to form a path line, and record the timestamp sequence index to obtain the distance offset path line set; For each pair of difference coordinate points, distance and direction extraction operations need to be performed. Specifically, this involves calculating the absolute displacement of the two coordinate points using their difference information in two-dimensional space. That is, for each pair of coordinate differences (x_diff, y_diff), the Euclidean distance between the two points is obtained by taking the square root of the sum of squares. This distance value represents the positional offset of the main channel and the secondary channel at that point. Simultaneously, combined with the definition of direction, the vector pointing from the origin to the difference point can be used to construct the edge change direction vector. This direction vector not only indicates the magnitude of the difference but also records the trend of the offset. For example, if the difference pair is (+4, -3), the direction points to the lower right. The direction vector is an arrow pointing from the origin to the fourth quadrant. All calculated direction vectors are then connected sequentially according to the edge points to form a continuous path structure. The consistency of the original edge point sequence is maintained between each vector segment to ensure that the spatial correspondence between the path line and the boundary structure is not disrupted. At the same time, the original image timestamp is added to the path structure of each frame as identification information, so that each path line has both spatial offset trajectory and temporal sequence attributes. After all graphic frames are processed in this way, a set of distance offset path lines numbered by timestamp and with spatial offset information is generated. This set can be used for subsequent continuous aggregation and error trajectory analysis.

[0035] S403: Based on the set of distance offset path lines, perform vector connection on continuous path lines in chronological order, perform aggregation and structured integration on all path lines, establish a continuous relationship between time and space, and generate an error trajectory graph; To establish temporal and spatial continuity, all pathlines in the set need to be sorted by timestamp in ascending order to ensure that subsequent connection operations strictly follow the acquisition time sequence of the image frame sequence. After sorting, each pathline is connected to its next pathline. Specifically, the endpoint of the current pathline's ending direction vector is connected to the starting point of the next pathline's starting direction vector, forming a continuous trajectory segment. If the distance between the endpoint and starting point of two pathlines is less than a set continuity tolerance threshold, the connection is valid and included in the trajectory aggregation process. This threshold is usually based on the image... For example, the frame rate and edge contour change rate can be set. If a maximum abrupt change distance of 3 pixels is allowed under the condition of 25 frames per second, the threshold is set to 3. After performing the same operation on all path lines in sequence, several continuous trajectory segments are formed. These trajectory segments are aggregated, that is, the trajectories within the continuous time period are merged into a longer path unit according to the start and end time. At the same time, all direction vectors and position offset values ​​within it are structurally integrated to generate a trajectory structure containing multiple time periods. Finally, a set of error trajectory graphics with complete time series and continuous spatial distribution is obtained. This graphic serves as the core output content for analyzing the structural offset trend between channels.

[0036] Please see Figure 6 The specific steps of S5 are as follows: S501: Based on the error trajectory graph, extract the vector sequence of each path line, perform point-by-point difference calculation on the direction angle and amplitude change value of the vector, identify continuous segments with change amplitude less than the preset direction stability threshold and consistent direction offset, mark the start and end positions, and generate a stable offset graph node set. The process involves sequentially reading the constructed vector sequence along each path, extracting the orientation angle and amplitude value for each vector node. The orientation angle is recorded as a polar angle, and the amplitude value is the spatial offset length corresponding to that vector. Then, a point-by-point difference processing is performed on the sequence, calculating the orientation angle difference and amplitude difference between the i-th and (i+1)-th vectors. The absolute value of the orientation angle difference is used to determine the directional stability between the two vectors. If this difference is lower than a preset directional stability threshold, the two vectors are considered to have similar directional characteristics. This threshold is generally set based on resolution and angular noise. In stable image edge regions, the directional stability threshold can be set between 5 and 8 degrees. For example, setting it to 6 degrees would result in three consecutive vectors with orientation angles of 72, 74, and 75 degrees. If the segment satisfies the condition that the directional difference is less than 6 degrees, it is judged as a segment with consistent direction. At the same time, it is necessary to determine whether these vectors have the same offset trend, that is, whether the change in direction angle is continuous in the same direction. If it rises from 72 to 74 and then to 75, it is rising in the same direction and satisfies the condition of consistent direction. If there is a reversal in the middle, it does not satisfy the rule. Record the start index and end index of the continuous vector segment that satisfies these two conditions. For example, if the vectors from the 5th to the 9th satisfy the rule, it is recorded as segment (5, 9). Traverse the entire path line vector sequence and record the index of all segments that satisfy the conditions in this way. Finally, a stable offset graphic node set is generated. This node set is a structure set composed of the start and end position indices of multiple continuous direction stable segments, which is used for the next step of synchronous segment extraction of image frames.

[0037] S502: Call the stable offset graphic node set, locate the corresponding contour boundary graphic frame in the main channel and the secondary channel according to the time index of each graphic node, extract the corresponding frame graphic area at each node time point and perform synchronous interception, recombine the frame fragments according to the node index sequence, and obtain the node frame combined structure graphic. The time index of each node is parsed and located to its specific frame number in the main and secondary channel image sequences. Index matching is performed using the timestamp as the key to ensure that time-consistent graphic frames are extracted from both the main and secondary channels. For each node's corresponding frame number, its contour boundary graphic structure in the image sequence is extracted, and a region cropping operation is performed on that frame image. The region range is determined based on the spatial range of the node's position vector. For example, if the vector segment corresponding to a node covers the coordinate range of (100, 150) - (200, 250) in the image, then the main and secondary channel images... Synchronous cropping is performed in this area, with the cropping area maintaining a consistent size and extracted using the same coordinate frame. This ensures that the main channel and secondary channel frame segments are completely spatially corresponding. All frame segments are named and stored using node index numbers as the naming rule. Subsequently, all frame segments are recombined in ascending order of node numbers to generate a set of frame segment combination graphics that match in both timing and structure. This structure organizes the graphic frames in the main channel and secondary channel under stable offset states into a unified structure, recording their node numbers, corresponding timestamps, image paths, and edge region position ranges to form a node frame combination structure graphic.

[0038] S503: Based on the node frame combination structure diagram, perform a unified normalization transformation on the edge coordinate points of the main channel and secondary channel regions, perform morphological aggregation and curvature adjustment processing on the normalized boundary structure matrix, and generate a diopter parameter analysis and optimization diagram. First, the coordinate point sets of the boundary structures of the two channels are read, and the original image size of the corresponding regions is mapped to a normalized size. For example, if the original coordinate range is (100, 100) - (200, 200), it is mapped to the standard coordinate system (0, 0) - (1, 1). After performing normalization transformation on all coordinate points, a boundary structure matrix at a unified scale is obtained. Then, structure aggregation is performed on the normalized coordinates. First, the boundary points of the primary and secondary channels are paired according to the principle of proximity. The edge points in the same region of the two channels are merged into a group of boundary nodes. Then, curvature adjustment is performed on the aggregated boundary. The processing steps are to calculate the local curvature value of the curve segment according to the boundary point order, identify the points with excessive curvature changes and perform... The smoothing operation uses a fixed-length neighborhood sliding window to perform a weighted average of adjacent curvature values. The length of the sliding window during the adjustment process depends on the boundary point density, which is set to a range of 5 points under the condition of a point density of 0.5 pixels / point. Finally, the adjusted boundary curvature change result is reprojected back into a normalized coordinate system and mapped to the original optical refractive model structure layer according to the coordinate correspondence. This layer is constructed based on the eye structure model and is divided into optical structures such as the lens refraction area, the lens projection area, and the pupil opening boundary by region. After mapping the boundary structure, it is superimposed on the corresponding optical layer according to the structural partition, and a complete refractive power parameter analysis and optimization graph is output. This graph is used for subsequent refractive accommodation function deduction and lens matching evaluation.

[0039] Please see Figure 7 A refractive power parameter analysis model optimization system, comprising: The curvature structure extraction module is used to implement S1: acquiring a sequence of lens images, identifying lens regions based on pixel position relationships, performing grayscale change direction calculations in each frame of the image, constructing continuous edge segments through grayscale differences, and integrating edge segments to generate a curvature graphic structure; The stable frame filtering module is used to implement S2: perform inter-frame structure comparison on the curvature graphic structure, perform continuity judgment based on the coverage area of ​​edge connected segments in adjacent frames, exclude frame graphics with interrupted edge segments, and output a set of stable frame graphics. The dual-channel contour recognition module is used to implement S3: the stable frame graphic set is divided into two types of graphic data according to the parity of the frame index, with odd-numbered frame graphics as the first type and even-numbered frame graphics as the second type. They are respectively assigned to the main channel and the secondary channel for concurrent operation. By loading frame by frame, the structure recognition process is executed to generate contour boundary graphics. The error trajectory construction module is used to implement S4: based on the contour boundary graphics, calculate the structural coordinate difference, establish the trajectory path based on the edge position difference generated in two channels of the same frame graphics, and construct the error trajectory graphics; The structural node identification and parameter analysis module is used to implement S5: compare the consistency of trajectory direction of error trajectory graphics, identify structural nodes with stable trajectory changes, extract the lens boundary region graphics corresponding to the stable nodes at the time points from the identified contour boundary graphics, and output the refractive parameter analysis and optimization graphics.

[0040] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for optimizing a refractive power parameter analysis model, characterized in that, The method comprises the following steps: S1: collecting a lens image sequence, identifying a lens region based on a pixel position relationship, performing a gray level change direction calculation in each frame of image, constructing a continuous edge segment through a gray level difference, and integrating the edge segment to generate a curvature graph structure; S2: performing inter-frame structure comparison on the curvature graph structure, performing continuity judgment based on the coverage area of the edge connected segment in adjacent frames, excluding the frame graph with interrupted edge segments, and outputting a stable frame graph set; S3: dividing the stable frame graph set into two types of graph data according to the frame index parity, wherein the odd frame graph is the first type and the even frame graph is the second type, and respectively assigning them to the main channel and the secondary channel for concurrent operation, performing structure recognition processing through frame-by-frame loading, and generating a contour boundary graph; S4: based on the contour boundary graph, calculating a structure coordinate difference value, establishing a trajectory path based on the edge position difference value generated in the two channels for the same frame graph, and constructing an error trajectory graph; S5: comparing the trajectory direction consistency of the error trajectory graph, identifying a structure node with stable trajectory change, extracting a lens boundary region graph corresponding to the stable node at the time point from the contour boundary graph, and outputting a refractive parameter analysis optimization graph.

2. The method of claim 1, wherein, The curvature graph structure comprises a gray level gradient distribution, a boundary curvature feature, and a region marker vector. The stable frame graph set comprises a direction consistency frame, an edge closure frame, and an inter-frame angle continuity frame. The contour boundary graph comprises main channel edge data, secondary channel edge data, and time sequence alignment information. The error trajectory graph comprises an offset distance vector, a path direction line segment, and a node connection sequence. The refractive parameter analysis optimization graph comprises a response node feature, a boundary structure segment, and an offset trend pattern.

3. The method of claim 1, wherein, The frame-by-frame loading and execution of the structure recognition processing refers to sequentially performing structure feature analysis and recognition on each frame of graph in the image sequence.

4. The method of claim 1, wherein, The structure node with stable trajectory change refers to an edge structure feature point in the error trajectory graph that exhibits consistent direction change and a continuity value greater than a preset threshold.

5. The method of claim 1, wherein, The specific steps of S1 are as follows: S101: collecting a lens image sequence, aggregating the pixels with a reflection intensity higher than a preset lens gray level reference value in each frame of image central block, obtaining boundary coordinates and performing closed fitting processing, and generating a lens boundary coordinate set; S102: based on the lens boundary coordinate set, extracting a boundary inner gray level value matrix, calculating a boundary point normal direction, comparing adjacent normal angles and recording a continuous segment with a direction change exceeding a preset gray level direction change threshold, and generating a gray level direction mutation segment sequence; S103: calling the gray level direction mutation segment sequence, extracting segment start and end coordinates, calculating corresponding arc length, angle and radius, and constructing a position and curvature mapping vector, integrating all segment mapping results, and generating a curvature graph structure.

6. The method of claim 1, wherein, The specific steps of S2 are as follows: S201: Call the gray vector sequence of each frame in the curvature pattern structure, extract the vector direction value of the adjacent position, and perform angle calculation on the same position vector direction in each pair of continuous frames, screen the coordinate section whose angle change value exceeds the preset fluctuation threshold in the continuous frames, and generate a direction change segment sequence; S202: Based on the direction change segment sequence, the frequency of the sequence segment corresponding frame is counted, it is judged whether the number of direction mutation appearing in each frame exceeds the preset direction mutation frequency threshold, and the corresponding pattern frame index sequence is recorded, the pattern frame with jump direction is eliminated, and an abnormal frame exclusion sequence is obtained; S203: Call the abnormal frame exclusion sequence, perform closedness detection on the edge track in the remaining frame pattern, calculate the spatial distance of the start and end boundary points, and judge whether it is less than the preset boundary closure threshold, and perform index aggregation, and generate a stable frame pattern set.

7. The method of claim 1, wherein, The specific steps of S3 are: S301: Divide the stable frame pattern set into two types of pattern data according to the odd and even of the frame index, perform index splitting for each group of pattern data and establish the corresponding data frame sequence identifier, associate one group of data frame sequence identifier to the main channel loading queue and the other group of data frame sequence identifier to the secondary channel loading queue, perform channel mapping, and generate a double-channel pattern data sequence; S302: Based on the double-channel pattern data sequence, the timestamp information carried by the data frame is obtained, the timestamp difference value of the data frame of the main channel and the secondary channel is calculated, it is judged whether the difference value is less than the preset channel alignment threshold, and the data frame satisfying the condition is marked synchronously, and a channel synchronization frame sequence is obtained; S303: Call the channel synchronization frame sequence, extract the edge coordinate point set of the corresponding frame of the main channel and the secondary channel in each recognition cycle, perform position registration on the edge coordinates of the two channels and perform closed boundary judgment, record the registered contour track and perform structured integration, and establish a contour boundary pattern.

8. The method of claim 1, wherein, The specific steps of S4 are: S401: Based on the contour boundary pattern, perform corresponding search on the edge point coordinates of the same numbered patterns of the main channel and the secondary channel, calculate the structural coordinate difference value of the corresponding points in the same coordinate system, and arrange the difference value according to the point sequence to generate a structural coordinate difference value sequence; S402: Call the structural coordinate difference value sequence, calculate the Euclidean distance for each pair of edge point difference value and construct a direction vector, connect the vectors corresponding to the same numbered patterns according to the point sequence to form a path line, and record the timestamp order index to obtain a distance offset path line set; S403: According to the distance offset path line set, the head and tail of the vector are connected in time sequence, the aggregation and structured integration of all path lines are performed, the time and space continuous relationship is established, and an error trajectory pattern is generated.

9. The method of claim 1, wherein, The specific steps of S5 are: S501: Based on the error trajectory pattern, extract the vector sequence of each path line, perform point-by-point difference calculation on the direction angle and amplitude change value of the vector, identify the continuous section with a change amplitude less than a preset direction stability threshold and a consistent direction offset, and mark the start and end positions to generate a stable offset pattern node set; S502: Call the stable offset pattern node set, locate the corresponding contour boundary pattern frames in the main channel and the secondary channel according to the time index of each pattern node, extract the corresponding frame pattern area at each node time point and perform synchronous interception, recombine the frame segments in the node index sequence, and obtain the node frame combined structure pattern; S503: According to the node frame combined structure pattern, uniformly normalize the edge coordinate points of the main channel and the secondary channel region, perform shape aggregation and curvature adjustment processing on the normalized boundary structure matrix, and generate a refractive parameter analysis optimization pattern.

10. A refraction parameter analysis model optimization system, comprising: The system is used to realize the refractive parameter analysis model optimization method of any one of claims 1-9, and the system comprises: The curvature structure extraction module is used to realize S1: collecting a lens image sequence, identifying a lens region based on a pixel position relationship, performing gray change direction calculation in each frame image, constructing continuous edge segments through gray difference, and integrating edge segments to generate a curvature pattern structure; The stable frame screening module is used to realize S2: frame structure comparison is performed on the curvature pattern structure, continuity judgment is performed based on the coverage area of the edge connected segments in adjacent frames, frames with interrupted edge segments are excluded, and a stable frame pattern set is output; The double-channel contour recognition module is used to realize S3: the stable frame pattern set is divided into two types of pattern data according to the frame index oddness, wherein the odd frame pattern is the first type and the even frame pattern is the second type, and is respectively assigned to the main channel and the secondary channel for concurrent operation, and structure recognition processing is performed through frame-by-frame loading to generate a contour boundary pattern; The error trajectory construction module is used to realize S4: based on the contour boundary pattern, the structure coordinate difference value is calculated, the edge position difference value generated in the two channels based on the same frame pattern is used to establish a trajectory path, and an error trajectory pattern is constructed; The structure node recognition and parameter analysis module is used to realize S5: the error trajectory pattern is compared for trajectory direction consistency, the structure node with stable trajectory change is identified, the lens boundary region pattern corresponding to the stable node at the time point is extracted from the identified contour boundary pattern, and a refractive parameter analysis optimization pattern is output.

Images