Skin prick test image data processing and wheal feature extraction method and system

Abstract

The application discloses a kind of skin prick test image data processing and wheal feature extraction method and system, it is related to image extraction technical field.This kind of skin prick test image data processing and wheal feature extraction method and system, comprising: S1, obtains prick image acquisition data and carries out image normalization and standardization processing;S2, carry out point scale distortion analysis, carry out directly measurable point set division and to be corrected point set generation;S3, to the corrected point set carries out surface-perspective joint scale correction analysis, carries out point level pixel conversion ratio generation and point level correction image block generation;S4, carry out wheal feature data output, review mark generation and trace archiving.Solve the existing method multiple global scale conversion, it is difficult to correct the point level pixel scale difference caused by forearm surface and shooting inclination, leading to the problem of scale misalignment in different position wheal feature measurement.

Description

Technical Field

[0001] This invention relates to the field of image extraction technology, specifically to a method and system for processing skin prick test image data and extracting wheal features. Background Technology

[0002] Skin prick testing is a common method for observing surface reactions in clinical allergen screening. It typically involves setting multiple test points on the skin surface and observing the local skin reaction after a certain period, providing a basis for subsequent result recording. With the development of medical information technology and image acquisition equipment, the recording methods for skin prick test data are gradually evolving from simple manual observation, manual measurement, and paper registration towards image storage, electronic recording, and computer-aided measurement. Current technologies often use cameras, mobile terminals, or dedicated imaging equipment to acquire images of the test area, and employ image processing techniques such as image enhancement, color correction, edge detection, region segmentation, contour fitting, and size conversion to visualize and extract parameters from the local reaction areas formed on the skin surface. Simultaneously, artificial intelligence technologies such as image recognition, semantic segmentation, and object detection are increasingly being applied to medical and surface image analysis scenarios to improve the automation of image region recognition, contour extraction, and feature quantification. Overall, skin prick test image data processing is developing towards standardized acquisition, automated extraction, digital storage, and traceable management.

[0003] For example, the invention patent with announcement number CN109948476B discloses a computer vision-based facial skin detection system and its implementation method. The method includes: guiding the user to a designated position in front of the camera via video, acquiring a facial photo of a preset size and transmitting it to the server; the server preprocesses the facial photo, performs facial key point detection based on dlib, performs image matting and smoothing on the cropped real-time image, performs image feature decomposition using wavelet analysis, converts the feature image to Lab space, and performs skin problem detection and classification based on deep neural networks to finally obtain a comprehensive facial skin score; it also discloses a facial skin detection system including a front-end system, a server system, and a mobile terminal system, used to realize image acquisition, preprocessing, classification detection, result display, and historical record storage.

[0004] For example, the invention patent with announcement number CN118262117B discloses a semi-supervised medical image semantic segmentation method based on hybrid enhancement and cross-EMA. The method includes: acquiring an original 3D medical image containing labeled and unlabeled data; preprocessing the original 3D medical image to obtain enhanced labeled and unlabeled data; mixing the enhanced labeled and unlabeled data to obtain mixed data; inputting the enhanced unlabeled data into a teacher network to obtain mask prediction and using it as pseudo-label; inputting the enhanced labeled data, unlabeled data, and mixed data into a student network to obtain probability prediction, and performing supervised learning based on the real label, pseudo label, and mixed label respectively; updating the teacher network parameters through cross-EMA to achieve semi-supervised semantic segmentation of the medical image region.

[0005] However, existing skin image detection and medical image segmentation methods mostly focus on image region recognition, image feature classification, or semantic segmentation output. When it comes to quantitative measurement of the detection area on the body surface, they usually still rely on a uniform scale conversion or global geometric correction method for the entire image. For skin prick test images, the detection area is located on the curved surface of the forearm, and the actual acquisition process is easily affected by the shooting angle, imaging distance, local perspective distortion, and differences in the spatial position of different points, resulting in inconsistent pixel scales at different prick points within the same image. Existing methods lack local scale distortion discrimination and point-level pixel conversion correction mechanisms for each prick point, making it difficult to ensure the consistency of phage feature measurement at different locations under the conditions of the curved surface of the forearm and shooting tilt, thus affecting the stable extraction of image feature data such as the longest diameter, shortest diameter, area, and perimeter of the phage.

[0006] Therefore, in order to address the above problems, there is an urgent need for methods and systems for processing skin prick test image data and extracting wheal features. Summary of the Invention

[0007] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method and system for processing skin prick test image data and extracting wheal features. This solves the problem that existing methods often use global scale ratio conversion, which makes it difficult to correct for point-level pixel scale differences caused by the curvature of the forearm and shooting tilt, resulting in scale inaccuracies in the measurement of wheal features at different locations.

[0008] Technical solution To achieve the above objectives, the present invention provides the following technical solution: a method for processing skin prick test image data and extracting wheal features, comprising: S1, acquiring prick test image acquisition data, and performing image normalization and standardization processing on the prick test image acquisition data; S2, performing point scale distortion analysis based on the prick test image acquisition data, and dividing the set of directly measurable points and generating the set of points to be corrected based on the point scale distortion analysis results; S3, performing surface-perspective joint scale correction analysis on the set of points to be corrected, and generating point-level pixel conversion ratios and point-level corrected image blocks based on the surface-perspective joint scale correction analysis results; S4, performing wheal feature output credibility analysis on the point-level corrected image blocks, and outputting wheal feature data, generating verification marks, and archiving traceability based on the wheal feature output credibility analysis results.

[0009] Furthermore, the specific process for acquiring puncture image acquisition data is as follows: In the skin puncture test image acquisition stage, the forearm detection area is placed within the imaging range of the acquisition device to acquire puncture image acquisition data. The puncture image acquisition data includes: original skin puncture detection image, millimeter-level calibration scale image, point mark image, forearm outer contour image, shooting distance data from the acquisition device to the forearm detection area, acquisition device pitch angle data, acquisition device roll angle data, and image acquisition timestamp data.

[0010] Furthermore, the specific process of image normalization and standardization of the puncture image acquisition data is as follows: Based on the millimeter-level calibration scale image, the pixel positions of each scale line are identified, and a scale coordinate set and an initial global pixel conversion ratio are established according to the actual distance between adjacent scales; using the marker contour in the puncture image as the recognition object, the center coordinates corresponding to each puncture point are extracted to form an initial coordinate set; the forearm outer contour image is processed for contour boundary extraction and detection area principal axis positioning to obtain the forearm contour coordinate set and detection area principal axis direction; the original skin puncture detection image is processed for exposure equalization, noise smoothing, reflection suppression, and hair weakening to obtain a standardized detection image; the shooting distance data from the acquisition device to the forearm detection area, the acquisition device pitch angle data, the acquisition device roll angle data, and the image acquisition timestamp data are processed for numerical format unification and timestamp binding; the puncture image acquisition data is standardized using a zero-mean unit variance normalization algorithm; and the standardized detection image is normalized using a maximum-minimum value normalization algorithm.

[0011] Furthermore, the specific process of performing point-scale distortion analysis based on point-splatter image acquisition data is as follows: The initial coordinate set of the points, the forearm contour coordinate set, and the principal axis direction of the detection area are calculated by normalizing the distance from the point to the principal axis to obtain the first... The deviation of the curved surface position of each puncture point; the first point is obtained by normalizing the perspective projection distance using the initial coordinate set of the point, the scale coordinate set, and the shooting distance data from the acquisition device to the forearm detection area. The perspective position deviation of each point; the initial coordinate set of the point and the scale coordinate set are calculated by counting the number of nearest neighbor scale marks and normalizing the distance inversely to obtain the first point. The scale neighbor support of the i-th puncture point is calculated; the tangent of the pitch angle data of the acquisition device is calculated, and the absolute value is added to obtain the pitch angle distortion term; the tangent of the roll angle data of the acquisition device is calculated, and the absolute value is added to obtain the roll angle distortion term; the surface position deviation and perspective position deviation of the i-th puncture point are added to obtain the surface deviation enhancement term and perspective deviation enhancement term; the scale neighbor support of the i-th puncture point is added to obtain the scale support suppression term; the product of the pitch angle distortion term, roll angle distortion term, surface deviation enhancement term, and perspective deviation enhancement term is calculated, divided by the scale support suppression term, and the arctangent of the result is taken to obtain the point scale distortion value of the i-th puncture point.

[0012] Furthermore, the specific process of dividing the set of directly measurable points and generating the set of points to be corrected based on the point scale distortion analysis results is as follows: Real-time comparison of point scale distortion values ​​and point scale distortion thresholds: When the point scale distortion value is less than the point scale distortion threshold, the puncture points are included in the set of directly measurable points, and the initial global pixel conversion ratio is used as the basic conversion ratio for the puncture points, and then the surface-perspective joint scale correction analysis is entered; When the point scale distortion value is greater than or equal to the point scale distortion threshold, point scale correction markers are generated, the puncture points are included in the set of points to be corrected, and the point scale distortion value, surface position deviation, perspective position deviation, and scale adjacent support are sent to the surface-perspective joint scale correction analysis.

[0013] Furthermore, the specific process of performing surface-perspective joint scale correction analysis on the set of points to be corrected is as follows: The surface development residual of the i-th point is obtained by approximately unfolding the forearm contour coordinate set, the principal axis direction of the detection area, and the initial coordinate set of the points using a thin-plate spline; the ratio of the point-level pixel conversion ratio of the i-th point to the initial global pixel conversion ratio is calculated, and the natural logarithm and absolute value of the comparison value are taken, followed by a negative exponent to obtain the scale deviation suppression term; the point-level scale distortion value of the i-th point is calculated, along with the surface development value of the i-th point. The sum of residuals is calculated, and the sum is taken as an inverse hyperbolic sine plus one, then the reciprocal is taken to obtain the distortion residual suppression term. The product of the scale deviation suppression term and the distortion residual suppression term is calculated to obtain the scale correction bearing value of the i-th dot puncture point. Among them, the product of the initial global pixel conversion ratio and the scale neighbor support of the i-th dot puncture point plus one is calculated, and then divided by the sum of the dot scale distortion value of the i-th dot puncture point, the surface position deviation of the i-th dot puncture point, the perspective position deviation of the i-th dot puncture point, and one, to obtain the dot-level pixel conversion ratio of the i-th dot puncture point.

[0014] Furthermore, the specific process for generating point-level pixel conversion ratios and point-level corrected image blocks based on the surface-perspective joint scale correction analysis results is as follows: Real-time comparison of scale correction tolerance value and scale correction tolerance threshold: When the scale correction tolerance value is greater than or equal to the scale correction tolerance threshold, the point-level pixel conversion ratio is used to perform perspective correction on the point image window of the corresponding point-puncture point based on the shooting distance data from the acquisition device to the forearm detection area, the pitch angle data of the acquisition device, the roll angle data of the acquisition device, and the projection offset relationship between the initial coordinates of the point and the scale coordinate set, generating point-level corrected image blocks. The point-level pixel conversion ratio, surface unfolding residual, and point-level corrected image blocks are then sent to the wind cluster feature output confidence score. Analysis: When the scale correction acceptance value is less than the scale correction acceptance threshold, the window of the corresponding point image is expanded by using n times the distance between adjacent point positions as the window side length increment. The adjacent scale coordinates in the scale scale coordinate set and the adjacent forearm contour coordinates in the forearm contour coordinate set are rematched. If the recalculated scale correction acceptance value is still less than the scale correction acceptance threshold, a scale correction verification mark is generated, and the corresponding point position is prohibited from outputting wind cluster feature data using the initial global pixel conversion ratio. The standardized detection image, initial coordinates of the point position, scale distortion value of the point position, and scale correction record corresponding to the point position are created and archived to the scale correction verification database. Point positions with scale correction verification marks are not included in the output point set.

[0015] Furthermore, the specific process of performing credibility analysis on wind cluster feature output for point-level corrected image blocks is as follows: The point-level corrected image blocks are processed through connected component contour extraction, boundary closure ratio calculation, and ellipse fitting residual calculation to obtain the first... The windblown boundary closure support of each point puncture site; the reflective areas, hair-occluded areas, marked ink areas, and shadow areas in the point-level corrected image block are obtained by extracting the brightness saturation area, detecting fine line structures, and statistically analyzing the color anomaly areas. The residual amount of non-target interference at each puncture point is calculated. The product of the wind mass boundary closure support at the i-th puncture point and the scale correction bearing value at the i-th puncture point is calculated. The negative exponent of the product is then subtracted by one to obtain the boundary correction coordination term. The negative exponent of the surface unfolding residual at the i-th puncture point is then calculated to obtain the surface residual suppression term. The inverse hyperbolic sine of the residual amount of non-target interference at the i-th puncture point is then added by one to obtain the interference residual suppression term. The product of the boundary correction coordination term and the surface residual suppression term is calculated and then divided by the interference residual suppression term to obtain the wind mass feature output confidence value at the i-th puncture point.

[0016] Furthermore, the specific process of outputting wind mass feature data, generating verification markers, and tracing archives based on the credibility analysis results of wind mass feature output is as follows: Real-time comparison of the credibility value and the credibility threshold of the wind mass feature output: When the credibility value of the wind mass feature output is greater than or equal to the credibility threshold, the wind mass contour is extracted based on point-level corrected image blocks, and the longest diameter, shortest diameter, average diameter, area, and perimeter of the wind mass are calculated according to the point-level pixel conversion ratio, generating wind mass feature data corresponding to the point puncture points; when the credibility value of the wind mass feature output is greater than or equal to the credibility threshold, the wind mass contour is extracted based on the point-level corrected image blocks, and the longest diameter, shortest diameter, average diameter, area, and perimeter of the wind mass are calculated, generating wind mass feature data corresponding to the point puncture points; when the credibility value of the wind mass feature output is greater than or equal to the credibility threshold, the wind mass contour is extracted based on the point-level corrected image blocks, and the longest diameter, shortest diameter, average diameter, area, and perimeter of the wind mass are calculated based on the point-level pixel conversion ratio, generating wind mass feature data corresponding to the point puncture points; when the credibility value of the wind mass feature output is greater than or equal to the credibility threshold, the wind mass contour is extracted based on the point-level corrected image blocks, and ... When the value is less than the confidence threshold of the wind cluster feature output, a wind cluster feature verification mark is generated. The point-level corrected image block, wind cluster outline, point-level pixel conversion ratio, non-target interference residue, and surface unfolding residual corresponding to the point puncture point are retained. The wind cluster feature data corresponding to the point puncture point is not written into the final output result, and the corresponding point puncture point is removed from the output point set. The wind cluster feature data, point-level pixel conversion ratio, wind cluster feature output confidence value, scale correction verification mark, and wind cluster feature verification mark are created and archived into the wind cluster feature extraction result database.

[0017] The second aspect of this invention provides a system for processing skin prick test image data and extracting wheal features, comprising: an acquisition and preprocessing module for acquiring prick test image acquisition data and performing image normalization and standardization processing on the prick test image acquisition data; a point scale distortion discrimination module for performing point scale distortion analysis based on the prick test image acquisition data, dividing the set of directly measurable points and generating the set of points to be corrected based on the point scale distortion analysis results; a surface-perspective joint scale correction module for performing surface-perspective joint scale correction analysis on the set of points to be corrected, generating point-level pixel conversion ratios and point-level corrected image blocks based on the surface-perspective joint scale correction analysis results; and a wheal feature extraction and reliable output module for performing reliable output analysis of wheal features on the point-level corrected image blocks, outputting wheal feature data, generating verification marks, and archiving traceability based on the reliable output analysis results.

[0018] Beneficial effects The present invention has the following beneficial effects: (1) This invention performs point scale distortion analysis on each point puncture point, distinguishes between directly measurable points and points to be corrected, avoids using the same global scale ratio for conversion of all points, and thus achieves the effect of active identification of point-level scale differences, effectively solving the problem that pixel scale differences of different points are difficult to identify in the prior art.

[0019] (2) In this invention, by combining the forearm contour, the main axis direction of the detection area, the posture of the acquisition device and the scale relationship, the surface-perspective joint scale correction is performed on the point to be corrected, forming a point-level pixel conversion ratio, thereby achieving the effect of local scale adaptive conversion, effectively solving the problem of inaccurate wind feature measurement caused by the forearm surface and shooting tilt in the prior art.

[0020] (3) In this invention, point-level correction image blocks are generated from the point-splitting points, so that the wind cluster contour extraction is based on the locally corrected image, rather than directly relying on the original image or the whole image correction result. This achieves the effect of pre-reducing local imaging distortion before wind cluster feature extraction, effectively solving the problem of local perspective deformation affecting the stability of wind cluster contour extraction in the prior art.

[0021] (4) This invention incorporates the scale correction acceptance state, surface unfolding residual and wind mass boundary closure support into the credibility analysis of wind mass feature output, generates verification marks for points that do not meet the output conditions, and thus achieves the effect of credible access of wind mass feature data, effectively solving the problem of low credibility segmentation results directly participating in wind mass feature output in the prior art.

[0022] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0023] Figure 1 This is a flowchart of the skin prick test image data processing and wheal feature extraction method of the present invention; Figure 2 This is a structural diagram of the skin prick test image data processing and wheal feature extraction system of the present invention; Figure 3 This is a bar chart comparing the point-scale distortion value with the point-level pixel conversion ratio of the present invention. Figure 4 This is a diagram showing the relationship between the ternary components of distortion-residue-support and the scale correction bearing value of the present invention. Figure 5 This is a flowchart of the scale correction acceptance value discrimination and processing of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see Figures 1-5 This invention provides a technical solution: a method for processing skin prick test image data and extracting wheal features, comprising the following steps: S1, acquiring prick test image acquisition data, and performing image normalization and standardization processing on the prick test image acquisition data; S2, performing point scale distortion analysis based on the prick test image acquisition data, and dividing the set of directly measurable points and generating the set of points to be corrected according to the point scale distortion analysis results; S3, performing surface-perspective joint scale correction analysis on the set of points to be corrected, and generating point-level pixel conversion ratios and point-level corrected image blocks according to the surface-perspective joint scale correction analysis results; S4, performing wheal feature output credibility analysis on the point-level corrected image blocks, and outputting wheal feature data, generating verification marks, and archiving traceability according to the wheal feature output credibility analysis results.

[0026] Specifically, the process of acquiring prick test image data is as follows: During the skin prick test image acquisition stage, the forearm detection area is placed within the imaging range of the acquisition device. The image sensor of the acquisition device acquires the image of the detection area, and extracts the scale image corresponding to the millimeter-level calibration ruler, the point mark image, and the forearm outer contour image from the same acquired image. The attitude sensor of the acquisition device acquires the shooting attitude data, which includes the pitch angle data and roll angle data of the acquisition device. The ranging unit acquires the shooting distance data, which is the distance data from the acquisition device to the forearm detection area. The prick test points, point marks, positive control points, negative control points, and the millimeter-level calibration ruler are simultaneously entered into the image to acquire prick test image data. The prick test image data includes: the original skin prick test image, the millimeter-level calibration ruler scale image, the point mark image, the forearm outer contour image, the shooting distance data from the acquisition device to the forearm detection area, the pitch angle data of the acquisition device, the roll angle data of the acquisition device, and the image acquisition timestamp data.

[0027] This implementation scheme achieves synchronous acquisition of skin puncture test images, acquisition device posture status, shooting distance status, and image acquisition time. This enables the puncture response information, scale conversion information, and local geometric status within the forearm detection area to form a common acquisition basis, providing unified, complete, and traceable raw data support for subsequent initial coordinate positioning of points, forearm contour constraints, point scale distortion discrimination, surface-perspective joint scale correction, and reliable output of wind feature.

[0028] Specifically, the process of image normalization and standardization of the collected dot-matrix image data is as follows: Based on the millimeter-level calibration scale image, the pixel positions of each scale line are identified; abnormal pixel segments corresponding to broken or occluded scale lines are removed; and a set of scale coordinates and an initial global pixel conversion ratio are established according to the actual distance between adjacent scales. The initial global pixel conversion ratio represents the ratio of the actual distance between adjacent scales to the pixel distance between adjacent scale lines, in millimeters per pixel. This ratio is used to convert the pixel length in the image to the actual length and maintains the same unit as the subsequent dot-level pixel conversion ratio. Using the marker contour in the dot-matrix image as the recognition object, the center coordinates corresponding to each dot-matrix point are extracted through the geometric center localization method of the contour closed region, forming an initial set of dot-matrix coordinates. Contour boundary extraction and... The detection area's main axis is located by fitting the midline of the two side contour boundaries to determine the main axis direction, resulting in the forearm contour coordinate set and the detection area's main axis direction. The original skin prick detection image undergoes exposure equalization, noise smoothing, reflection suppression, and hair weakening to obtain a standardized detection image. Exposure equalization employs contrast-limited adaptive histogram equalization; noise smoothing uses bilateral filtering; reflection suppression uses brightness saturation region detection and neighborhood grayscale restoration; and hair weakening uses morphological fine line detection and local texture restoration to reduce interference from local bright spots, fine line occlusion, and background texture on subsequent wheal contour extraction. Positive and negative control points are included as components of the prick detection points in the initial coordinate set and processed according to the same point scale distortion discrimination, surface-perspective joint scale correction, and wheal feature extraction process.

[0029] The numerical format of the shooting distance data from the acquisition device to the forearm detection area, the pitch angle data of the acquisition device, the roll angle data of the acquisition device, and the image acquisition timestamp data are unified and timestamp bound. The data is bound to the corresponding image acquisition timestamp, so that the attitude data of the acquisition device and the standardized detection image form the same acquisition record. The point puncture image acquisition data is standardized by the zero mean unit variance standardization algorithm, so that coordinate data, distance data, and angle data of different dimensions are included in the same calculation scale. The standardized detection image is normalized by the maximum and minimum value normalization algorithm, so that the image pixel features are limited to a uniform numerical range, making the parameters calculated by the subsequent formulas dimensionless parameters.

[0030] In this implementation plan, this step unifies the scale reference, point location, forearm geometric boundary, acquisition posture, and image quality status in the skin prick test images, forming a set of scale coordinates, initial global pixel conversion ratio, initial point coordinate set, forearm contour coordinate set, principal axis direction of the detection area, and standardized detection image. This ensures that the original acquired data has a unified coordinate reference, a unified time reference, and a unified numerical scale, providing a stable, standardized, and computable data foundation for subsequent point scale distortion discrimination, surface-perspective joint scale correction, and reliable output of wind feature.

[0031] Specifically, the process of performing point-position scale distortion analysis based on point-puncture image acquisition data is as follows: The surface position deviation of the i-th point-puncture point is calculated by normalizing the initial coordinate set of the point-position, the forearm contour coordinate set, and the principal axis direction of the detection area using the distance from the point to the principal axis. The vertical distance from the initial coordinates of the i-th point-puncture point to the principal axis direction of the detection area is used as the deviation distance, and the contour width defined by the forearm contour coordinate set in the same cross-sectional direction is used as the normalization benchmark. The surface position deviation is the ratio between the deviation distance and half of the contour width, used to limit the lateral offset of the point-puncture point relative to the principal axis direction of the detection area within a unified normalized scale. The perspective position deviation of the i-th point-puncture point is calculated by normalizing the initial coordinate set of the point-position, the scale coordinate set, and the shooting distance data from the acquisition device to the forearm detection area using the perspective projection distance. The projection offset distance of the i-th point-puncture point relative to the scale coordinate set is used as the perspective offset basis, and scale normalization is performed in conjunction with the shooting distance data from the acquisition device to the forearm detection area. The perspective position deviation is... The ratio of the projection offset distance to the pixelated distance of the shooting distance data from the acquisition device to the forearm detection area on the image plane; the pixelated distance is calculated from the shooting distance data from the acquisition device to the forearm detection area according to the initial global pixel conversion ratio, used to eliminate the perspective offset scale difference caused by different shooting distances; the scale neighbor support of the i-th point is obtained by counting the number of nearest neighbor scales and calculating the reciprocal normalization of distances to the initial coordinate set of the point and the scale scale coordinate set, and the available support within the neighborhood of the i-th point is counted. The number of valid scale lines is matched, and the pixel distance from the i-th dot point to the adjacent valid scale line is normalized by taking the reciprocal. The ruler's neighbor support is calculated as the ratio of the number of valid scale lines in the preset neighborhood to the upper limit of the preset number of valid scale lines, and the distance support is calculated as the ratio of the reciprocal of the pixel distance from the i-th dot point to the upper limit of the preset distance reciprocal. The number support and the distance support are multiplied and then normalized to ensure that the ruler's neighbor support is within a uniform value range.

[0032] The pitch angle data of the acquisition device is calculated, and the absolute value is incremented by one to obtain the pitch angle distortion term. The effect of pitch changes on longitudinal perspective distortion is amplified through tangent mapping. Pitch direction differences are eliminated by taking the absolute value, and incrementing by one ensures that this term does not suffer from zero-value suppression when used as a product factor. The roll angle data of the acquisition device is also calculated, and the absolute value is incremented by one to obtain the roll angle distortion term. The effect of roll tilt on lateral scale stretching is characterized through tangent mapping, and incrementing by one ensures that the roll angle distortion term maintains its baseline contribution in the absence of roll. The surface position deviation of the i-th point is then considered. Add one to the perspective position deviation of the i-th point puncture point to obtain the surface deviation enhancement term and the perspective deviation enhancement term. This addition preserves the relative gain of the surface position deviation and the perspective position deviation, allowing the two types of deviations to amplify the contribution of local scale distortion through product. Add one to the scale neighbor support of the i-th point puncture point to obtain the scale support suppression term. Adding one to the denominator avoids division by zero and ensures that the scale neighbor support creates a reverse constraint on the point scale distortion value. Calculate the product of the pitch distortion term, roll distortion term, surface deviation enhancement term, and perspective deviation enhancement term, and then divide... Using a scale to support the suppression term and taking the arctangent of the result, the point scale distortion value of the i-th point is obtained. The product structure reflects the joint magnification relationship between shooting posture, surface position, and perspective offset. The arctangent function compresses the calculation result to a finite interval, ensuring that the point scale distortion value has a stable threshold comparison range. Compared to directly linearly superimposing the pitch angle data, roll angle data, surface position offset, and perspective position offset of the acquisition device, this calculation method can form a joint magnification effect when multiple scale offset factors occur simultaneously, and it also achieves this effect in adjacent scale locations. Sufficient support near the point of contact creates a suppression effect. Compared to proportional calculations without arctangent compression, this calculation method can limit numerical jumps caused by abnormal shooting postures or local offsets, making the point-scale distortion value more suitable for stable comparison with the point-scale distortion threshold, thereby improving the stability of the division of the directly measurable point set and the point set to be corrected. Since the pitch angle data and roll angle data of the acquisition equipment are angular quantities, and the surface position deviation, perspective position deviation, and scale adjacent support are all normalized, the point-scale distortion value is a dimensionless value. The specific calculation formula is as follows: ; In the formula, Indicates the first The point scale distortion value of each puncture point is used to characterize the degree of local scale shift of the corresponding puncture point under the combined effects of acquisition posture, surface position and perspective position. This represents the pitch angle data of the acquisition device, used to characterize the longitudinal tilt state of the acquisition device relative to the forearm detection area; This represents the roll angle data of the acquisition device, used to characterize the lateral tilt of the acquisition device relative to the forearm detection area; Indicates the first The deviation of the curved surface position of each puncture point is used to characterize the difference in the lateral curved surface position of the corresponding puncture point relative to the main axis direction of the forearm detection area. Indicates the first The perspective position deviation of each dot is used to characterize the degree of projection offset of the corresponding dot relative to the set of scale coordinates. Indicates the first The scale support near each dotted point is used to characterize the degree of scale support near the corresponding dotted point that can be used for scale conversion.

[0033] In this implementation scheme, this step achieves independent quantification of the risk of local scale inaccuracy at each puncture point. It uniformly maps the acquisition device's attitude, forearm surface position, perspective offset, and adjacent scale support status to point-level scale distortion values, eliminating the need for uniform scaling of puncture points at different spatial locations based on the assumption of a unified conversion across the entire image. By generating these point-level scale distortion values, the degree of scale offset caused by differences in the forearm surface, shooting tilt, and scale support can be distinguished. This establishes a criterion for directly measurable points and points to be corrected, providing a stable preliminary judgment basis for subsequent surface-perspective joint scale correction, point-level pixel conversion ratio generation, and reliable output of wind cluster features.

[0034] Specifically, the process of dividing the set of directly measurable points and generating the set of points to be corrected based on the results of the point scale distortion analysis is as follows: Real-time comparison of point scale distortion values ​​and point scale distortion thresholds: When the point scale distortion value is less than the point scale distortion threshold, the corresponding point is determined to meet the global scale acceptance condition. The point is then included in the set of directly measurable points. The initial coordinates and scale neighbor support of the corresponding point are retained in the set of directly measurable points as the index basis for subsequent consistency confirmation. The initial global pixel conversion ratio is used as the basic conversion ratio of the point, and the surface-perspective joint scale correction analysis is performed to confirm the scale consistency.

[0035] When the point scale distortion value is greater than or equal to the point scale distortion threshold, it is determined that the corresponding point has a risk of point-level scale offset. A point scale correction mark is generated, and the point scale correction mark is bound to the initial coordinates of the corresponding point to form a point-level correction index. The point is included in the set of points to be corrected. At the same time, the local window capture range of the point in the standardized detection image is recorded as the processing object for subsequent surface unfolding and perspective correction. The point scale distortion value, surface position deviation, perspective position deviation, and scale neighbor support are sent to the surface-perspective joint scale correction analysis.

[0036] This implementation scheme achieves point-based flow control based on point-scale distortion values, enabling different point-based punctures to enter corresponding processing paths according to their local scale offset. For point-based punctures that meet the global scale acceptance conditions, their initial coordinates and adjacent support on the scale are retained, and the initial global pixel conversion ratio is used as the basic conversion basis to ensure that scale consistency confirmation has a traceable index. For point-based punctures with point-level scale offset risks, point-based scale correction marks are generated and bound to point-level correction indexes, enabling these point-based punctures to subsequently enter the surface unfolding and perspective correction processes as local windows. Through the above flow control, all point-based punctures are prevented from being uniformly converted to global scale, improving the targeting of subsequent surface-perspective joint scale correction and the reliability of wind cluster feature extraction results.

[0037] Specifically, the process of performing surface-perspective joint dimensional correction analysis on the set of points to be corrected is as follows: the surface development residual of the i-th point is obtained by using the thin plate spline approximate development residual of the forearm contour coordinate set, the principal axis direction of the detection area, and the initial coordinate set of the points. The principal axis direction of the detection area is used as the development reference axis, and the two boundary points in the forearm contour coordinate set are used as the thin plate spline control points to establish a local approximate development mapping of the points, and the normalized residual of the local coordinate deviation before and after the mapping of the i-th point is calculated.

[0038] Calculate the ratio of the pixel-level scaling factor at the i-th dot spur point to the initial global pixel scaling factor. Take the natural logarithm and absolute value of the ratio, then apply a negative exponent to obtain the scaling factor deviation suppression term. This ratio represents the degree of deviation of the pixel-level scaling factor from the initial global pixel scaling factor. The natural logarithm compresses the scaling factor change, the absolute value eliminates the difference in the direction of deviation, and the negative exponent strengthens the suppression effect when the scaling factor deviation is more pronounced. Calculate the dot-scale distortion value of the i-th dot spur point and the ratio of the pixel-level scaling factor of the i-th dot spur point to the initial global pixel scaling factor. The sum of the surface unfolding residuals is taken as an inverse hyperbolic sine, then one is added, and the reciprocal is taken to obtain the distortion residual suppression term. This summation reflects the combined effect of point scale distortion and surface unfolding error. The inverse hyperbolic sine is used to compress the abnormal amplitude, and adding one avoids the denominator of the reciprocal calculation being zero. Furthermore, the distortion residual suppression term decreases as distortion and residuals increase. The product of the scale deviation suppression term and the distortion residual suppression term is calculated to obtain the scale correction bearing value for the i-th point. This product structure ensures that neither the scale deviation nor the unfolding residual satisfies the bearing value. When accepting conditions, the scale correction acceptance value is reduced; specifically, the product of the initial global pixel conversion ratio and the scale neighbor support of the i-th point is calculated, and then divided by the sum of the point scale distortion value, the surface position deviation of the i-th point, the perspective position deviation of the i-th point, and one, to obtain the point-level pixel conversion ratio of the i-th point. The support of the local scale availability for the point-level pixel conversion ratio is enhanced by the scale neighbor support. The point scale distortion value, surface position deviation, and one are used to further enhance the support of the local scale availability for the point-level pixel conversion ratio. The deviation and perspective position offset jointly constrain the correction range of the point-level pixel conversion ratio. Since the scale proximity support, point scale distortion, surface position offset, perspective position offset, and surface unfolding residual of the i-th point are all dimensionless values, the point-level pixel conversion ratio of the i-th point has the same dimension as the initial global pixel conversion ratio. The scale correction inherits a dimensionless value, and the specific calculation formula is as follows: ; ; In the formula, Indicates the first The scale correction acceptance value of each point is used to characterize whether the point-level pixel conversion ratio has the acceptance condition to enter the point-level corrected image block generation under the constraints of scale deviation and surface unfolding residual. Indicates the first The point-level pixel conversion ratio of each puncture point is used to convert the pixel length in the local image of the corresponding puncture point into the actual length. This indicates the initial global pixel conversion ratio, used to provide a baseline for the overall length conversion of the image. Indicates the first The scale neighbor support of each dot-punch point is used to characterize the degree of support near the corresponding dot-punch point in terms of local scale conversion using the scale. Indicates the first The point scale distortion value of each puncture point is used to characterize the degree of local scale shift of the corresponding puncture point. Indicates the first The deviation of the curved surface position of each puncture point is used to characterize the difference in the lateral curved surface position of the corresponding puncture point relative to the main axis direction of the forearm detection area. Indicates the first The perspective position deviation of each dot is used to characterize the degree of projection offset of the corresponding dot relative to the set of scale coordinates. Indicates the first The surface development residual of each puncture point is used to characterize the local coordinate residual deviation of the corresponding puncture point after approximate development by a thin plate spline.

[0039] In this embodiment, Table 1 is a data table of point-level scale correction parameters, which records in detail the point-level scale distortion value, surface unfolding residual, scale neighbor support, initial global pixel conversion ratio, point-level pixel conversion ratio and scale correction acceptance value of different point-stab points during the scale correction process. This is used to quantify the degree of local scale offset of each point affected by the forearm surface and shooting tilt and the reliability of its correction. Specifically: Point 1 has a scale distortion value of 0.10, a surface unfolding residual of 0.05, a scale neighbor support of 0.90, an initial global pixel conversion ratio of 0.100, a point-level pixel conversion ratio of 0.112, and a scale correction tolerance of 0.085; Point 2 has a scale distortion value of 0.40, a surface unfolding residual of 0.05, a scale neighbor support of 0.90, an initial global pixel conversion ratio of 0.100, a point-level pixel conversion ratio of 0.089, and a scale correction tolerance of 0.410; Point 3 has a scale distortion value of 0.40, a surface unfolding residual of 0.25, and a scale neighbor support of 0.90. The initial global pixel conversion ratio is 0.100, the point-level pixel conversion ratio is 0.089, and the scale correction tolerance value is 0.281; the point-level scale distortion value corresponding to point 4 is 0.10, the surface unfolding residual is 0.05, the scale neighbor support is 0.60, the initial global pixel conversion ratio is 0.100, the point-level pixel conversion ratio is 0.122, and the scale correction tolerance value is 0.042; the point-level scale distortion value corresponding to point 5 is 0.75, the surface unfolding residual is 0.10, the scale neighbor support is 0.10, the initial global pixel conversion ratio is 0.100, the point-level pixel conversion ratio is 0.050, and the scale correction tolerance value is 0.208.

[0040] Table 1. Point-level scale correction parameter data table like Figure 3 The bar chart shows the comparison between point-scale distortion values ​​and point-level pixel conversion ratios. Table 1 shows a negative correlation between the degree of distortion at different points and their local scale conversion results. Point 5 has the highest point-scale distortion value of 0.75, with a point-level pixel conversion ratio of only 0.050, far lower than the initial global pixel conversion ratio of 0.100, indicating that severe distortion led to a significant correction of the local scale. Points 2 and 3 both have point-scale distortion values ​​of 0.40 and point-level pixel conversion ratios of 0.089, indicating moderate correction. Points 1 and 4 have lower distortion values ​​(both 0.10), with point-level pixel conversion ratios of 0.112 and 0.122 respectively, closer to the global baseline. However, point 4's scale correction reliability is the weakest because the scale support value drops to 0.60, resulting in a scale correction acceptance value of 0.042. Overall, the figure visually reveals the driving effect of distortion degree on local pixel conversion ratio, while also indicating that the magnitude of distortion value is not the only factor determining the reliability of correction.

[0041] like Figure 4 The figure shows the relationship between the ternary components of distortion, residual, and support and the scale correction acceptance value. Combined with Table 1, it can be seen that the scale correction acceptance value is affected by the synergistic influence of the proportions of the point's scale distortion value, surface development residual, and scale neighbor support. The scale neighbor support proportions at points 2 and 3 are similar, with acceptance values ​​reaching 0.410 and 0.281 respectively. The support component has a positive gain effect on the acceptance value, but at point 3, the acceptance value is significantly suppressed due to the increased surface development residual to 0.25. At point 5, the distortion component far exceeds the other two, while the scale neighbor support accounts for only a very small proportion, resulting in an acceptance value of only 0.208, which is at a moderately low level. At point 4, although the proportions of distortion and residual are not large, the proportion of support decreases, causing the acceptance value to drop to a minimum of 0.042. This figure demonstrates the influence of multi-factor joint constraints on the correction acceptance conditions from the perspective of component balance, providing a multi-dimensional basis for scale correction verification and point selection.

[0042] In this implementation scheme, this step determines the local scale correction acceptance capability of the puncture points to be processed. It integrates the point-level pixel conversion ratio, the initial global pixel conversion ratio, the point-level scale distortion value, and the surface unfolding residual into the scale correction acceptance value calculation, ensuring that the local scale correction amplitude and the forearm surface unfolding state are synchronously controlled. Through the formation of the scale correction acceptance value, it can be determined whether the local scale conversion result of the corresponding puncture point possesses the stable conditions for generating point-level corrected image blocks, preventing puncture points with abnormal scale deviations or abnormal surface unfolding residuals from directly entering the wind cluster feature extraction process. This step also incorporates the scale neighbor support amount, surface position deviation, and perspective position deviation into the point-level pixel conversion ratio correction, improving the adaptability of the local scale conversion to the spatial position of the point, the scale support state, and the shooting perspective state. This provides a reliable scale acceptance basis for subsequent perspective correction, surface unfolding correction, scale resampling processing, and reliable wind cluster feature output.

[0043] Specifically, the process of generating point-level pixel conversion ratios and point-level corrected image blocks based on the surface-perspective joint scale correction analysis results is as follows: (e.g.) Figure 5 The diagram shows the flowchart for determining and processing the scale correction acceptance value, which compares the scale correction acceptance value and the scale correction acceptance threshold in real time. When the scale correction tolerance value is greater than or equal to the scale correction tolerance threshold, perspective correction is performed on the corresponding dot-matrix image window using a point-level pixel conversion ratio. This correction is based on the shooting distance data from the acquisition device to the forearm detection area, the acquisition device's pitch angle data, the acquisition device's roll angle data, and the projection offset relationship between the initial coordinates of the dot-matrix and the scale coordinate set. The dot-matrix image window is a local image area centered on the initial coordinates of the dot-matrix. The window side length is determined based on the distance between adjacent dot-matrix points calculated from the initial coordinate set and the distance between scale points calculated from the scale coordinate set, with the value range being the distance between adjacent dot-matrix points. The distance between the puncture points is one-third to one-half, and does not exceed twenty times the pixel length corresponding to the distance between adjacent scale marks. Based on the forearm contour coordinate set and the main axis direction of the detection area, an approximate unfolded mesh of the local surface of the point is constructed. The pixels in the point image window are mapped to the unfolded plane according to their lateral offset relative to the main axis direction of the detection area to complete the surface unfolding correction. Scale resampling is performed according to the point-level pixel conversion ratio to make the pixel length in the generated point-level corrected image block consistent with the actual length. The point-level pixel conversion ratio, the surface unfolding residual, and the point-level corrected image block are sent to the wind cluster feature output credibility analysis.

[0044] When the scale correction acceptance value is less than the scale correction acceptance threshold, the window of the corresponding point image is expanded by n times the distance between adjacent point positions as the window side length increment. n is determined according to the arrangement density between point positions and the scale interval in the scale coordinate set, and the value ranges from 0.05 to 0.20. The adjacent scale coordinates in the scale coordinate set and the adjacent forearm contour coordinates in the forearm contour coordinate set are rematched. If the recalculated scale correction acceptance value is still less than the scale correction acceptance threshold, a scale correction verification mark is generated, and the corresponding point position is prohibited from outputting wind mass feature data using the initial global pixel conversion ratio. The standardized detection image, initial coordinates of the point position, scale distortion value of the point position, and scale correction record corresponding to the point position are created and archived in the scale correction verification database. Point positions with scale correction verification marks are not included in the output point set, and their corresponding wind mass feature data are only used as verification retention data and are not written into the final output result.

[0045] This implementation scheme achieves point-level correction admission control based on scale correction acceptance values. This allows point punctures that meet the scale acceptance conditions to undergo perspective correction, surface unfolding correction, and scale resampling processing, forming point-level corrected image blocks consistent with the actual local scale. This provides a corrected image foundation for subsequent reliable analysis of wind cluster feature output. For point punctures with insufficient scale correction acceptance, the image window is enlarged and adjacent scale coordinates and forearm contour coordinates are re-matched for confirmation, reducing the risk of misjudgment caused by insufficient local scale or contour support. Point punctures that still do not meet the conditions generate scale correction verification marks and are prohibited from directly using the initial global pixel conversion ratio to output wind cluster feature data. This prevents wind cluster features that have not undergone reliable local correction from entering the final result, thereby improving the scale consistency, verification controllability, and process traceability of the wind cluster feature extraction results.

[0046] Specifically, the process of performing reliable analysis of wind feature output on point-level corrected image blocks is as follows: For point-level corrected image blocks, the wind boundary closure support of the i-th point is obtained through connected component contour extraction, boundary closure ratio calculation, and ellipse fitting residual calculation. Candidate connected component contours are first extracted from the point-level corrected image blocks, then the proportion of closed boundary lengths of the candidate connected component contours is calculated, and the normalized distance residual between the candidate connected component contours and the ellipse fitting contours is used as a contour regularity constraint. For the reflective areas, hair-occluded areas, marked ink areas, and shadow areas in the point-level corrected image blocks, the non-target interference residue of the i-th point is obtained through brightness saturation region extraction, fine line structure detection, and color anomaly region statistics. The reflective areas are obtained from brightness saturation pixel connectivity statistics, the hair-occluded areas from fine line structure response statistics, and the marked ink areas and shadow areas from color channel deviation statistics.

[0047] Calculate the product of the wind cluster boundary closure support at the i-th point and the scale correction acceptance value at the i-th point. Take the negative exponent of the product and subtract the negative exponent from the result to obtain the boundary correction coordination term. The product structure reflects the coordination constraint between the wind cluster boundary closure support and the scale correction acceptance value. The negative exponent compresses the coordination gain and makes the boundary correction coordination term tend to stabilize as the coordination degree increases. Take the negative exponent of the surface unfolding residual at the i-th point to obtain the surface residual suppression term. The negative exponent reduces the corresponding output contribution as the surface unfolding residual increases. Take the inverse hyperbolic sine of the non-target interference residue at the i-th point and add one to obtain the interference residue suppression term. The inverse hyperbolic sine compresses the interference residue amplitude. By adding one to avoid the denominator being zero, and by making the residual amount of non-target interference a reverse constraint on the confidence value of the wind cluster feature output, the product of the boundary correction synergy term and the surface residual suppression term is calculated, and then divided by the interference residual suppression term to obtain the confidence value of the wind cluster feature output at the i-th point. Through multiplication and division combination, the integrity of the wind cluster boundary, the scale correction acceptance status, the surface unfolding residual, and the residual amount of non-target interference at the i-th point are unified to the same confidence discrimination scale. Since the wind cluster boundary closure support, the scale correction acceptance value, the surface unfolding residual, and the residual amount of non-target interference at the i-th point are all dimensionless values, the confidence value of the wind cluster feature output is also dimensionless. The specific calculation formula is as follows: ; In the formula, Indicates the first The confidence value of the wind feature output of each puncture point is used to characterize whether the wind feature data of the corresponding puncture point meets the confidence output condition. Indicates the first The closure support of the wind cluster boundary at each puncture point is used to characterize the boundary continuity and fitting stability of the corresponding wind cluster candidate profile. Indicates the first The scale correction acceptance value of each dot point is used to characterize whether the scale correction result of the corresponding dot point meets the conditions for subsequent feature extraction. Indicates the first The surface development residual of each puncture point is used to characterize the local coordinate residual deviation of the corresponding puncture point after approximate development by a thin plate spline. Indicates the first The residual amount of non-target interference at each point is used to characterize the degree of residual non-wind cluster image interference in the corresponding point-level corrected image block.

[0048] In this implementation scheme, this step achieves a unified determination of the availability and output reliability of wind cluster contours in point-level corrected image blocks. It aggregates the wind cluster boundary closure support, scale correction acceptance value, surface unfolding residual, and non-target interference residue into a reliable output value for wind cluster features. This ensures that the integrity of wind cluster boundaries, local scale correction status, surface unfolding stability, and image interference residue are all within the same reliable judgment framework. This step effectively identifies low-reliability points caused by incomplete boundary closure, insufficient contour regularity, unstable scale correction, large surface unfolding errors, and issues such as reflections, hair occlusion, marker ink marks, and shadow residue. It guarantees that only points meeting the reliable output conditions are included in subsequent wind cluster feature calculations and result output, improving the stability, authenticity, and traceability of the extracted results for the longest, shortest, average, area, and perimeter of the wind cluster.

[0049] Specifically, the process of outputting wind feature data, generating verification markers, and archiving traceable data based on the credibility analysis results of wind feature output is as follows: Real-time comparison of the credibility value and the credibility threshold of the wind feature output: When the confidence value of the wind cluster feature output is greater than or equal to the confidence threshold of the wind cluster feature output, the wind cluster contour is extracted based on the point-level corrected image patch. Boundary closure checks, connected component preservation, and contour smoothing constraints are then applied to the wind cluster contour to maintain a spatial correspondence with the point puncture locations. The longest diameter, shortest diameter, average diameter, area, and perimeter of the wind cluster are calculated according to the point-level pixel conversion ratio. Specifically, the longest diameter is the actual length obtained by converting the maximum distance between any two contour points on the wind cluster contour using the point-level pixel conversion ratio; the shortest diameter is the actual length obtained by converting the width of the contour projection perpendicular to the longest diameter direction using the point-level pixel conversion ratio; the average diameter is the arithmetic mean of the longest and shortest diameters; the area is the actual area obtained by converting the pixel area of ​​the closed region of the wind cluster contour using the square of the point-level pixel conversion ratio; and the perimeter is the actual length of the pixel length of the wind cluster contour boundary using the point-level pixel conversion ratio. Finally, wind cluster feature data corresponding to the point puncture locations is generated.

[0050] When the confidence value of the wind feature output is less than the confidence threshold of the wind feature output, a wind feature verification mark is generated, and the verification reason type is recorded according to the calculation item corresponding to the confidence value of the wind feature output. The verification reason types include insufficient boundary closure, insufficient scale correction, abnormal surface unfolding residual, and abnormal non-target interference residue. The point-level corrected image block, wind contour, point-level pixel conversion ratio, non-target interference residue, and surface unfolding residual corresponding to the point puncture point are retained. The wind feature data corresponding to the point puncture point is not written into the final output result, and the corresponding point puncture point is removed from the output point set to avoid low-confidence wind feature data from participating in subsequent statistics or report generation.

[0051] The wind cluster feature data, point-level pixel conversion ratio, wind cluster feature output confidence value, scale correction verification mark, and wind cluster feature verification mark are created and archived into the wind cluster feature extraction result database. The initial coordinates of the corresponding point puncture points, point-level corrected image block index, and wind cluster contour index are archived simultaneously, forming a complete processing chain record from the original image and scale correction to the wind cluster feature output, which is used for subsequent result verification and image processing record traceability.

[0052] In this implementation scheme, this step achieves final output control based on the confidence value of wind cluster features. This enables point puncture locations that meet the confidence conditions to complete wind cluster contour extraction and calculate the longest, shortest, average, area, and perimeter of the wind cluster, forming wind cluster feature data that can be used for subsequent statistics and report generation. Point puncture locations that do not meet the confidence conditions generate wind cluster feature verification marks, retaining point-level corrected image blocks, wind cluster contours, point-level pixel conversion ratios, non-target interference residues, and surface unfolding residuals, preventing low-confidence wind cluster feature data from entering the final output result. The wind cluster feature data, point-level pixel conversion ratios, wind cluster feature output confidence values, scale correction verification marks, wind cluster feature verification marks, and corresponding index information are uniformly archived in the wind cluster feature extraction result database, forming a complete processing record from the original image and scale correction to the wind cluster feature output, improving the usability, verification convenience, and full-process traceability of the wind cluster feature extraction results.

[0053] like Figure 2 As shown, the second aspect of this invention provides a skin prick test image data processing and wheal feature extraction system, comprising: an acquisition and preprocessing module for acquiring prick test image acquisition data, performing image normalization and standardization processing on the prick test image acquisition data to form a unified data foundation required for subsequent point location, scale conversion, and forearm contour analysis; and a point scale distortion discrimination module for performing point scale distortion analysis based on the prick test image acquisition data, dividing the set of directly measurable points and generating the set of points to be corrected based on the point scale distortion analysis results, thereby realizing the scale risk of different prick test points. The module performs surface-perspective joint scale correction analysis on the set of points to be corrected. Based on the results of the surface-perspective joint scale correction analysis, it generates point-level pixel conversion ratios and point-level corrected image blocks, ensuring that the points to be corrected have a locally consistent image foundation. The module extracts and outputs wind features and performs reliable analysis on the point-level corrected image blocks. Based on the results of the reliable analysis, it outputs wind feature data, generates verification markers, and archives the data, achieving reliable output and process traceability of wind feature data.

[0054] In this implementation scheme, through progressive processing of acquisition and preprocessing, point scale distortion discrimination, surface-perspective joint scale correction, and wind cluster feature extraction and reliable output, the point puncture sites in the skin prick test images can be diverted and corrected according to local scale risk, avoiding measurement inaccuracies caused by using the same global scale conversion for point puncture sites at different locations. At the same time, through point-level corrected image blocks and reliable analysis of wind cluster feature output, the generation, verification, and archiving of wind cluster feature data are uniformly controlled, improving the scale consistency, output reliability, and process traceability of the extracted results of wind cluster longest diameter, wind cluster shortest diameter, wind cluster average diameter, wind cluster area, and wind cluster perimeter.

[0055] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0056] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for processing image data and extracting wheal features from skin prick test images, characterized in that, Includes the following steps: S1, acquire point puncture image data, and perform image normalization and standardization processing on the point puncture image data; S2, perform point scale distortion analysis based on point puncture image acquisition data, and divide the set of directly measurable points and generate the set of points to be corrected based on the point scale distortion analysis results; The specific process of performing point-scale distortion analysis based on point-puncture image acquisition data is as follows: The surface position deviation of the i-th puncture point is calculated by normalizing the distance from the point to the principal axis using the initial coordinate set of the point, the forearm contour coordinate set, and the principal axis direction of the detection area. The perspective position deviation of the i-th puncture point is calculated by normalizing the perspective projection distance using the initial coordinate set of the point, the scale coordinate set, and the shooting distance data from the acquisition device to the forearm detection area. The scale neighbor support of the i-th puncture point is calculated by normalizing the nearest neighbor scale number and the reciprocal of the distance using the initial coordinate set of the point and the scale coordinate set. Calculate the tangent of the pitch angle data of the acquisition device, take the absolute value and add one to obtain the pitch angle distortion term; calculate the tangent of the roll angle data of the acquisition device, take the absolute value and add one to obtain the roll angle distortion term; add one to the surface position deviation of the i-th point and the perspective position deviation of the i-th point respectively to obtain the surface deviation enhancement term and the perspective deviation enhancement term. Add one to the scale neighbor support of the i-th point to obtain the scale support suppression term; Calculate the product of pitch distortion, roll distortion, surface deviation enhancement, and perspective deviation enhancement, then divide by the scale support suppression term, and take the arctangent of the result to obtain the point scale distortion value of the i-th point. S3, Perform surface-perspective joint scale correction analysis on the set of points to be corrected, and generate point-level pixel conversion ratio and point-level corrected image block based on the results of the surface-perspective joint scale correction analysis. The specific process of performing surface-perspective combined scalar correction analysis on the set of points to be corrected is as follows: The surface development residual of the i-th point puncture point is obtained by approximating the forearm contour coordinate set, the principal axis direction of the detection area, and the initial coordinate set of the point position through thin plate spline. Calculate the ratio of the pixel-level conversion ratio of the i-th puncture point to the initial global pixel conversion ratio. Take the natural logarithm and absolute value of the comparison value, and then take the negative exponent to obtain the ratio deviation suppression term. Calculate the sum of the point scale distortion value of the i-th point puncture point and the surface unfolding residual of the i-th point puncture point, take the inverse hyperbolic sine of the sum, add one, and then take the reciprocal to obtain the distortion residual suppression term. The product of the scaling deviation suppression term and the distortion residual suppression term is calculated to obtain the scale correction value of the i-th dot puncture point. Specifically, the product of the initial global pixel conversion ratio and the scale neighbor support of the i-th dot puncture point plus one is calculated, and then divided by the sum of the dot scale distortion value of the i-th dot puncture point, the surface position deviation of the i-th dot puncture point, the perspective position deviation of the i-th dot puncture point, and one, to obtain the dot-level pixel conversion ratio of the i-th dot puncture point. S4 performs a reliable analysis of wind cluster feature output on the point-level corrected image block, and outputs wind cluster feature data, generates verification marks, and archives them based on the results of the reliable analysis of wind cluster feature output.

2. The method for processing skin prick test image data and extracting wheal features according to claim 1, characterized in that: The specific process for acquiring the dot-puncture image data is as follows: During the skin prick test image acquisition stage, the forearm detection area is placed within the imaging range of the acquisition device to obtain prick image acquisition data. The prick image acquisition data includes: original skin prick detection image, millimeter-level calibration scale image, point mark image, forearm outer contour image, shooting distance data from the acquisition device to the forearm detection area, acquisition device pitch angle data, acquisition device roll angle data, and image acquisition timestamp data.

3. The method for processing skin prick test image data and extracting wheal features according to claim 1, characterized in that: The specific process of image normalization and standardization processing of the collected dot-matrix image data is as follows: The pixel position of each scale line is identified based on the millimeter-level calibration scale image, and a set of scale coordinates and an initial global pixel conversion ratio are established according to the actual distance between adjacent scales. Using the outline of the marker in the point marker image as the recognition object, the center coordinates corresponding to each point are extracted to form an initial set of point coordinates; The forearm outer contour image is processed by contour boundary extraction and detection region principal axis localization to obtain the forearm contour coordinate set and detection region principal axis direction; the original skin puncture detection image is processed by exposure equalization, noise smoothing, reflection suppression and hair weakening to obtain a standardized detection image; The numerical format of the shooting distance data from the acquisition device to the forearm detection area, the pitch angle data of the acquisition device, the roll angle data of the acquisition device, and the image acquisition timestamp data are unified and the timestamps are bound together. The data collected for dot puncture images were standardized using a zero-mean, unit-variance standardization algorithm. The standardized detection images are normalized using a maximum and minimum value normalization algorithm.

4. The method for processing skin prick test image data and extracting wheal features according to claim 1, characterized in that: The specific process of dividing the set of directly measurable points and generating the set of points to be corrected based on the results of point scale distortion analysis is as follows: Real-time comparison of point scale distortion values ​​and point scale distortion thresholds: When the point scale distortion value is less than the point scale distortion threshold, the point puncture point is included in the set of directly measurable points, and the initial global pixel conversion ratio is used as the basic conversion ratio of the point puncture point, and then the surface-perspective joint scale correction analysis is entered. When the point scale distortion value is greater than or equal to the point scale distortion threshold, a point scale correction mark is generated, the point puncture point is included in the set of points to be corrected, and the point scale distortion value, surface position deviation, perspective position deviation, and scale adjacent support are sent to the surface-perspective joint scale correction analysis.

5. The method for processing skin prick test image data and extracting wheal features according to claim 1, characterized in that: The specific process for generating point-level pixel conversion ratios and point-level corrected image blocks based on the surface-perspective joint scale correction analysis results is as follows: Real-time comparison of scale correction acceptance value and scale correction acceptance threshold: When the scale correction acceptance value is greater than or equal to the scale correction acceptance threshold, the point image window of the corresponding point is subjected to perspective correction based on the shooting distance data from the acquisition device to the forearm detection area, the pitch angle data of the acquisition device, the roll angle data of the acquisition device, and the projection offset relationship between the initial coordinates of the point and the scale coordinate set. Point-level correction image blocks are generated, and the point-level pixel conversion ratio, surface unfolding residual and point-level correction image blocks are sent to the wind cluster feature output credibility analysis. When the scale correction acceptance value is less than the scale correction acceptance threshold, the window of the corresponding point image is expanded by n times the distance between adjacent point punctures as the window side length increment. The adjacent scale coordinates in the scale scale coordinate set and the adjacent forearm contour coordinates in the forearm contour coordinate set are rematched. If the recalculated scale correction acceptance value is still less than the scale correction acceptance threshold, a scale correction verification mark is generated, and the corresponding point puncture is prohibited from outputting wind cluster feature data using the initial global pixel conversion ratio. The standardized detection image, initial coordinates of the point, scale distortion value of the point, and scale correction record corresponding to the point puncture are created and archived in the scale correction verification database. Point punctures with scale correction verification marks are not included in the output point set.

6. The method for processing skin prick test image data and extracting wheal features according to claim 1, characterized in that: The specific process of performing wind cluster feature output credibility analysis on point-level corrected image blocks is as follows: The point-level corrected image patch is obtained by extracting the contours of the connected components, calculating the boundary closure ratio, and calculating the ellipse fitting residual. The windblown boundary closure support of each point puncture site; the reflective areas, hair-occluded areas, marked ink areas, and shadow areas in the point-level corrected image block are obtained by extracting the brightness saturation area, detecting fine line structures, and statistically analyzing the color anomaly areas. Non-target interference residue at each puncture point; Calculate the product of the wind cluster boundary closure support at the i-th point and the scale correction bearing value at the i-th point. Take the negative exponent of the product and subtract the negative exponent from the result to obtain the boundary correction synergistic term. The surface residual of the i-th puncture point is negatively exponentially expanded to obtain the surface residual suppression term; the non-target interference residual of the i-th puncture point is inversely hyperbolic sine plus one to obtain the interference residual suppression term. Calculate the product of the boundary correction synergistic term and the surface residual suppression term, and then divide by the interference residual suppression term to obtain the confidence value of the wind cluster feature output at the i-th point.

7. The method for processing skin prick test image data and extracting wheal features according to claim 1, characterized in that: The specific process of outputting wind cluster feature data, generating verification markers, and archiving traceability based on the reliable analysis results of wind cluster features is as follows: Real-time comparison of the confidence value and confidence threshold of wind mass feature output: When the confidence value of the wind mass feature output is greater than or equal to the confidence threshold of the wind mass feature output, the wind mass contour is extracted based on the point-level corrected image block, and the longest diameter, shortest diameter, average diameter, area and perimeter of the wind mass are calculated according to the point-level pixel conversion ratio to generate the wind mass feature data corresponding to the point puncture point. When the confidence value of the wind cluster feature output is less than the confidence threshold of the wind cluster feature output, a wind cluster feature verification mark is generated. The point-level correction image block, wind cluster outline, point-level pixel conversion ratio, non-target interference residue and surface unfolding residual corresponding to the point puncture point are retained. The wind cluster feature data corresponding to the point puncture point is not written into the final output result, and the corresponding point puncture point is removed from the output point set. The wind cluster feature data, point-level pixel conversion ratio, wind cluster feature output confidence value, scale correction verification mark, and wind cluster feature verification mark are created and archived into the wind cluster feature extraction result database.

8. A skin prick test image data processing and wheal feature extraction system, applied to the skin prick test image data processing and wheal feature extraction method according to any one of claims 1-7, comprising: The acquisition and preprocessing module is used to acquire point puncture image acquisition data and perform image normalization and standardization processing on the point puncture image acquisition data; The point scale distortion discrimination module is used to perform point scale distortion analysis based on point puncture image acquisition data, and to divide the set of directly measurable points and generate the set of points to be corrected based on the point scale distortion analysis results. The surface-perspective joint scale correction module is used to perform surface-perspective joint scale correction analysis on the set of points to be corrected, and generate point-level pixel conversion ratios and point-level corrected image blocks based on the results of the surface-perspective joint scale correction analysis. The wind cluster feature extraction and reliable output module is used to perform reliable analysis of wind cluster feature output on point-level corrected image blocks, and to output wind cluster feature data, generate verification marks, and archive traceability based on the results of the reliable analysis of wind cluster feature output.

Images