AI intelligent body measurement analysis system

Abstract

The invention discloses an AI intelligent body measurement analysis system, and relates to the technical field of AI intelligent body measurement analysis. The system comprises a self-adaptive standardization module, a self-adaptive layout solving module, an edge and ground color unification module, a unified export and protocol module, a step-level trace reserving module, a layered fault-tolerant and quality gating module and a performance and resource optimization module. And the self-adaptive normalization module is used for executing safety margin setting and alpha channel unification processing on the multi-source heterogeneous body measurement image materials. According to the invention, through the cooperation of adaptive standardization, adaptive layout solving and edge and ground color unification modules, safe margin control, precise scaling and regular layout can be carried out on the multi-source heterogeneous body measurement image, the problems of image truncation, black edges and alignment dislocation are effectively avoided, the output consistency and definition of the body measurement image are ensured, and the quality of the multi-source heterogeneous body measurement image is improved. A high-quality image basis is provided for the AI model to accurately identify the body measurement mark, the body measurement data error is greatly reduced, and the high-precision body measurement scene requirement is met.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of AI intelligent body analysis, and particularly relates to an AI intelligent body analysis system. BACKGROUND

[0002] The mainstream AI intelligent body analysis system terminal data processing scheme in the prior art is a back-end system, which is a back-end processing system acting on a front-end interactive page. The back-end processing system mainly provides a stable, observable and maintainable back-end service framework for the front end or business. The back-end processing system standardizes and synthesizes “multi-source and inconsistent image materials” according to rules in the back end, provides consistent and available interfaces and results upward, and connects general storage and log systems downward to ensure availability and traceability. The general Web back-end architecture is generally used to realize the basic data processing function. The core logic of the scheme is as follows: receiving multi-source body images transmitted by the front end, analyzing the request through the routing layer and distributing it to the corresponding controller, completing parameter verification and permission checking by the controller, calling the general graphics library / Canvas interface by the service layer to decode, size adjust, simply splice (such as side-by-side or top-and-bottom stacking), and transcode and export the body images, finally returning the processing results to the front end in the URL, BASE64 or Buffer format, and storing the key data to the database and file storage system. Some schemes retain a small amount of intermediate result pictures in the local or storage for the convenience of quality traceability.

[0003] Firstly, the request is distributed to the corresponding controller by analyzing the request path and method through the routing. The controller completes the basic parameter verification and permission checking, and then standardizes the format, converts the type, checks the size and integrity of the input body data / image file. Then, the service layer arranges the body data processing capability according to the scene, including image loading, canvas creation, body marker drawing / image splicing, data encoding export, state recording, unified result packaging, data storage, and generation of debugging snapshots. Finally, the processing results are returned in a unified response format. The general graphics library / Canvas interface is used to realize body image decoding, size calculation, white background / transparency background filling, margin blanking, up-and-down / left-and-right splicing layout, transcoding to PNG / JPEG format and output buffer. In order to realize quality traceability, key intermediate result pictures are retained in the local / storage.

[0004] The existing system process is not observable, the problem is difficult to reproduce, there is no step-level debugging snapshot and structured log, when the image processing link is abnormal, the specific problem link cannot be located, the image download has no layered timeout and retry mechanism, network fluctuation is easy to cause timeout, there is no size gate, and the CPU / memory occupation is easy to be too high when processing a super large image, and even the service crashes. There is no minimum resolution and aspect ratio check threshold for the body image, and low-quality images enter the synthesis process, resulting in body marking recognition deviation and data error exceeding the allowed range. Multi-image synthesis adopts synchronous series and full resolution loading, and there is no resource reuse, so CPU / memory jitter is obvious, response delay increases, and it is difficult to support large-scale user demand.

[0005] To this end, we provide an AI intelligent body analysis system to solve the problems in the above. SUMMARY

[0006] The purpose of the present application is to provide an AI intelligent body analysis system, which solves the problems of unobservable system terminal data processing process, difficult problem reproduction, insufficient network and resource robustness, easy timeout and crash, large body data error and poor performance difficult to support high concurrency in the prior art.

[0007] To solve the above technical problems, the present application is realized by the following technical scheme.

[0008] The present application is an AI intelligent body analysis system, which comprises an adaptive normalization module, an adaptive layout solving module, an edge and background unification module, a unified export and protocol module, a step-level trace module, a hierarchical fault tolerance and quality gate module, and a performance and resource optimization module.

[0009] The adaptive normalization module is used for performing safety margin setting and alpha channel unification processing on multi-source heterogeneous body image materials, first estimating the scaling ratio based on the content bounding box of the body image, then drawing, and simultaneously unifying the background and blank strategy, setting the original size of the body image material as (w i ,h i ), the target height as H * , then the scaling ratio s = H * / h i , the image width w′ i after scaling = s·w i , the canvas height H c = H * +2p, the canvas width W c =∑(w′ i )+p·(n+1)+(n-1)·g, wherein p is the safety margin, n is the number of body images, and g is the image spacing.

[0010] The adaptive layout solving module is used for combining the equal height and equal width strategy with the content proportion of the body image, dynamically solving the canvas size and image placement coordinates, adopting the dynamic padding / gap and baseline alignment strategy, and defining the alignment error E align =∑|y i -y ref |,y i is the bottom coordinate of the body image, y ref is the bottom reference baseline, and the height h′ of all the body images after scaling under the equal height strategy i =H * , and E align →0 is realized, for the equal width strategy, the horizontal alignment error E align,x =∑|x i -x ref |,x i is the left coordinate of the body image, X ref is the horizontal reference baseline, and the width w′ of all the body images after scaling is unified i =W * , W * is the target width, and E align,x →0 is realized.

[0011] The edge and background unification module is used for unifying the alpha channel processing mode of the body image, adopting the edge anti-aliasing technology, determining the unified background strategy, and the background strategy including white background and transparent two types.

[0012] The unified export and protocol module is used for unifying the export encoding format and response protocol of the body image, the export encoding format including PNG and JPEG, and the error code and state information in the response protocol being structured.

[0013] The step-level trace module is used for generating the snapshot of each step of the body image processing in the time stamp-step-description.png format, recording the key metadata and the time consumption of each step, and the key metadata including the image size, the scaling ratio and the layout parameter.

[0014] The hierarchical fault tolerance and quality gating module is used for setting the hierarchical timeout, retry and fallback strategy in the body image downloading and decoding link, the fallback strategy including returning the placeholder image and displaying the prompt information, and setting the minimum resolution R min , the aspect ratio p and the file integrity as the front quality gating standards, R min ∈[64, 2048] pixels, and p ∈ [0.1, 10].

[0015] The performance and resource optimization module is used for adopting the on-demand downsampling processing for the body image, performing the drawing operation in steps and reusing the resources, and adopting the hybrid strategy of parallel downloading and serial drawing in the body image acquisition and processing process.

[0016] The application is further configured that, in the adaptive normalization module, the transparent processing mode includes at least one of premultiplication of alpha, non-premultiplication of alpha, edge feathering, threshold suppression, and anti-aliasing resampling, the normalization processing order includes any one of uniform alpha channel first and then scaling, scaling first and then background filling, and performing quality gating first and then uniform alpha channel, and the running position includes any one of a front-end link in a controller, a service layer independent module, and a preprocessing pipeline before storage into a repository.

[0017] The application is further configured that, in the adaptive layout solving module, the objective function includes any one of minimum alignment error, minimum canvas area, maximum uniformity of white space, and multi-objective weighting, the solving strategy includes any one of analytical solution, heuristic algorithm, constraint optimization algorithm, and dynamic programming, the heuristic algorithm includes greedy algorithm and iterative algorithm, the constraint optimization algorithm includes linear programming and quadratic programming, the layout type includes any one of horizontal side-by-side, vertical stacking, grid table, waterfall, golden section, and equal proportion segmentation, the alignment reference includes any one of top, bottom, center line, grid reference, and reference frame, the gap control mode includes any one of fixed gap, adaptive gap, and segmented function gap, and the adaptive gap is related to the size, quantity, and canvas size of the figure.

[0018] The application is further configured that, in the uniform derivation and protocol module, the drawing back end includes any one of CPU Canvas, Skia, GPU, and WASM rendering back end, the interface supported by the GPU includes OpenGL, Metal, Vulkan, and DirectX, the sampling kernel includes any one of bilinear, bicubic, Mitchell-Netravali, and Lanczos, the derivation encoding format further includes WEBP and AVIF, the color space supported by the color processing includes sRGB, Display-P3, and linear space, and the gamma correction function is supported.

[0019] The application is further configured that, in the step-level trace module, the snapshot storage medium includes any one of local file, object storage, memory ring buffer, and only metadata storage, the log structure includes any one of text, JSON, and combination of text and JSON, the uniform protocol supports any one of JSON, CBOR, and Protobuf, and the protocol field name is configurable and the semantics are consistent.

[0020] The layered fault tolerance and quality gating module is further provided with a retry strategy including any one of fixed retry, exponential backoff retry and jitter backoff retry, and a maximum retry number and a total time limit are set, and the fallback strategy further includes any one of resolution reduction processing, quality degradation and returning only layout metadata, and the resource gating standard includes at least one of maximum image side length, maximum pixel number, concurrent download number and memory water level triggering degradation.

[0021] The edge and background color unification module is further provided with an automatic detection on the alpha channel format of the multi-source heterogeneous volume image when performing the alpha channel unification processing, different types such as pre-multiplied alpha, non-pre-multiplied alpha and no alpha channel are identified, and format conversion is performed based on the system preset alpha channel standard format, for the volume image without the alpha channel, if the system background color strategy is set to transparent, an alpha channel is automatically added and the alpha value of the background area is set to 0, if the background color strategy is set to white background, the alpha channel does not need to be added and a white background color is directly filled.

[0022] The adaptive normalization module is further provided with a safety margin p and an image spacing g, and the value range of p and g is [0, 10 3 ] or [0%, 20%] of the canvas length.

[0023] The performance and resource optimization module is further provided with a downsampling processing sampling ratio determined according to the ratio of the target size of the volume image to the original size, a step-by-step drawing executed according to the image area division, and resource reuse including image cache reuse and calculation result reuse.

[0024] The system based on the Node.js / Express web backend architecture provides an HTTP interface, and is divided into an access and routing layer, a controller layer, a service layer, a data access layer and a file object storage layer from top to bottom according to responsibilities, and is equipped with a cross-cutting middleware layer and a configuration layer.

[0025] The present application has the following advantages.

[0026] 1、The present application can control the safety margin, accurately scale and regularly layout the multi-source heterogeneous volume image through the cooperation of the adaptive normalization, adaptive layout solving and edge and background color unification modules, effectively avoids the problems of image truncation, edge black border and alignment misplacement, guarantees the output consistency and clarity of the volume image, provides a high-quality image basis for the accurate identification of volume markers by the AI model, greatly reduces the volume data error, and meets the demand of high-precision volume scene.

[0027] 2、The system of the application realizes traceability of the whole process of body image processing by means of the step-level trace module, facilitates rapid positioning of abnormal links, and, in combination with the hierarchical fault tolerance and quality gating modules, can resist network fluctuation influence, filter low-quality image materials, and, through the on-demand downsampling and serial hybrid processing strategies of the performance and resource optimization module, reduce CPU and memory peak occupancy, improve system response speed and concurrent processing capacity, reduce operation and maintenance troubleshooting costs, ensure service stability in high-concurrency scenarios, and optimize user experience. BRIEF DESCRIPTION OF DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used for describing the embodiments.

[0029] Figure 1 It is a main flow chart of an AI intelligent body analysis system.

[0030] Figure 2 It is an adaptive standardized processing flow chart in an AI intelligent body analysis system.

[0031] Figure 3 It is an adaptive layout solving flow chart in an AI intelligent body analysis system.

[0032] Figure 4 It is a step-level trace and traceable flow in an AI intelligent body analysis system. DETAILED DESCRIPTION

[0033] The technical solutions in the embodiments of the application will be described below with reference to the drawings of the embodiments of the application. The described embodiments are only some of the embodiments of the application, not all.

[0034] Embodiment one

[0035] Please refer to Figures 1-4 The application is an AI intelligent body analysis system. When processing multi-source heterogeneous body images, the existing scheme of rear-end image material synthesis splicing commonly used in the industry is taken as a baseline, and the problems of insufficient material standardization, lack of fault tolerance and quality gating are solved by focusing on the adaptive standardization module and the quality gating module.

[0036] Specifically, when the system receives body images of different sources, the adaptive standardization module first extracts the body core area of each image through the content recognition unit, determines the content bounding box, and then sets the preset target height H * of the body image as the reference, calculates the scaling ratio s = H * / image original height h i , ensures that the body core area is not compressed, sets a safety margin p, which is 5% of the canvas long side, and calculates the canvas height Hc = H * + 2p, canvas width W c = ∑(scaled width w' i i ) + p-(n+1) + (n-1)g, n=2 is the number of images, g=15 pixels is the image spacing, calculate the canvas size, finally convert the image alpha channel to the premultiplied format, fill the white bottom as the uniform background color, complete the normalization processing.

[0037] Verification method: select 10 groups of different sources and different sizes of quantity image as test set, respectively adopt the prior art and the embodiment scheme, through image analysis tool to count the cut-off rate R clip and blank consistency Δp, if the cut-off rate R clip of the embodiment scheme approaches to 0, the prior art is usually ≥15%, Δp fluctuation range ≤2 pixels, the prior art is usually ≥8 pixels, then it proves that the material normalization effect meets the standard, solves the material confusion problem of the prior art.

[0038] The effect of the embodiment is that: through the front-end quality gate to filter low-quality images, avoid entering the subsequent process to cause the quantity data deviation, the adaptive normalization solves the material scaling confusion and the blank non-uniformity in the prior art, makes the output format of the quantity images of different sources consistent, provides a stable basis for subsequent synthesis and analysis.

[0039] Embodiment two

[0040] The system aims at the single layout strategy and non-uniform protocol of the existing scheme of image material synthesis and splicing, optimizes the multi-image synthesis and result output process through the adaptive layout solving module and the uniform export protocol module.

[0041] Specifically: when three normalized quantity images, front view, back view and side view, the scaled height is 400 pixels, and the width is 300 pixels, 290 pixels and 270 pixels respectively, need to be synthesized, the adaptive layout solving module first judges the image quantity and size characteristics through the feature analysis unit, because the height difference of the three images is <5%, the equal height layout strategy is automatically selected, the module sets the bottom of the image as the uniform reference baseline y ref , calculates the deviation value of the bottom of each image and the reference baseline, the initial deviation is 5 pixels, 3 pixels and 7 pixels respectively, adjusts the Y axis position of the image in real time through the coordinate adjustment unit, so that E align = ∑|y i -y ref |, the initial deviation is 5 pixels, 3 pixels and 7 pixels respectively, adjusts the Y axis position of the image in real time through the coordinate adjustment unit, so that E alignApproaching to 0, while dynamically adjusting the image spacing g according to the number of images, taking the value of the safety margin p = 1.2 times of 20 pixels, that is, 24 pixels, finally generating a composite image with a canvas size of width = 300 + 290 + 270 + 20 * 4 + 2 * 24 = 988 pixels, height = 440 pixels, ensuring that the image is not crowded and misaligned.

[0042] Verification method: for the above three volume images, respectively, using the prior art, fixed left and right side by side, spacing 10 pixels, and the processing of the embodiment scheme, through the coordinate detection tool to calculate the alignment error E of the two groups of results align and the crowded rate C, the number of image pairs with edge spacing <5 pixels / total number of pairs, if the E align≤1 of the embodiment scheme is less than 1 pixel, the prior art is usually greater than or equal to 12 pixels, C = 0, and the prior art is usually greater than or equal to 30%, it is proved that the adaptive layout effect meets the standard, solving the layout misalignment and crowdedness problem of the prior art.

[0043] The technical effect of this embodiment is that it breaks the limitations of the fixed layout of the prior art, solves the misalignment and crowdedness problem that often occurs when synthesizing multiple volume images by dynamically selecting the layout strategy and correcting the alignment deviation in real time, makes the overall appearance of the synthesized image neat, and facilitates subsequent volume marking and feature recognition, improving the convenience and accuracy of volume analysis.

[0044] Embodiment three

[0045] In view of the problems of edge sawtooth and mixed background color of existing volume images, an edge and background unification module is set to optimize the visual effect and consistency of the image.

[0046] Specifically: after the volume image is completed and the layout is adjusted, the edge and background unification module first detects the image through the alpha channel, converts the alpha channel of the transparent PNG to the premultiplied format, identifies the edge noise and sawtooth area through the edge detection unit for JPEG images, starts the anti-aliasing processing unit, and uses the bilinear interpolation algorithm to smooth the edge pixels, so that the edge gray value gradually transitions, and at the same time selects white background, RGB value 255, 255, 255 as the unified background, fills the blank area of all images, and ensures that there is no transparent area mixed or background difference.

[0047] Verification method: select 5 groups of mixed volume images containing transparent PNG and JPEG, respectively, using the prior art without unified alpha processing, without anti-aliasing, and the processing of the embodiment scheme, through the image quality analysis tool to count the edge artifacts A edge of the two groups of results, if the single image A edgeIf the deviation of the RGB values ​​of the background color of all images is ≤10 (compared to ≥50 in existing technologies), and the deviation is ≤2 (compared to ≥15 in existing technologies), it proves that the effect of the edge and background color is consistent and solves the problems of edge defects and background color confusion in existing technologies.

[0048] The technical effect of this embodiment is that it effectively solves the visual confusion problem caused by black edges and inconsistent background colors in the existing technology of body image, making the image edges clearer and the style more consistent, reducing the cost of secondary beautification processing on the terminal side, and avoiding interference with the AI ​​model's recognition of body features due to jagged edges or differences in background color, thereby improving the extraction accuracy of body data.

[0049] Example 4

[0050] To address the issues of inconsistent export formats and protocols for existing solid-state images, a unified export and protocol module is used to standardize the output results.

[0051] Specifically: After the body image is synthesized and optimized, the unified export and protocol module first converts the image into a unified PNG format according to preset rules, generates an image data buffer, and then encapsulates the processing result according to a standardized response protocol. The protocol fields include status code, result type, image metadata, and error information. All interfaces return JSON format data with a consistent structure. Even if the front end requests a single or multiple images, the response structure is the same. If an export error occurs, the front end can quickly locate the problem type through the error code.

[0052] Verification method: Simulate 1000 front-end requests, process them using existing technology and the solution of this embodiment respectively, and calculate the integration error rate E of the two sets of results. api MTTR (Mean Time To Repair) api If the E of this embodiment is api≤ 0.5%, while existing technologies typically have ≥8%, MTTR api If the time is ≤10 minutes, while existing technologies typically take ≥30 minutes, it proves that the unified export and protocol achieves the desired effect and solves the problem of complex integration in existing technologies.

[0053] The technical effects of this embodiment are as follows: it solves the problem of complex front-end integration caused by chaotic export formats and inconsistent protocols in the prior art, reduces interaction failures caused by format differences, reduces the difficulty of system integration, and the standardized error response also makes it easier for the front-end to quickly locate problems and improve the efficiency of exception handling. At the same time, it provides stable interface support for the subsequent canary release and version rollback of the system, and enhances the maintainability of the system.

[0054] Example 5

[0055] To address the issues of unobservable and difficult-to-reproduce volumetric image processing processes, a step-level traceability module is set up to achieve full-process traceability.

[0056] Specifically: when the equivalent body image enters the processing flow, the step-level trace module establishes a trigger association with each link of the service layer. After the image download is completed, the snapshot generation unit automatically generates a snapshot in the format of timestamp-step-description.png, synchronously records the original size, download source and other metadata. After the image decoding link is completed, a decoded snapshot is generated and the decoding format and decoding time consumption are recorded. The layout and rendering links generate corresponding snapshots respectively, and record the canvas size, scaling ratio, image spacing and other key parameters. All snapshots and metadata are associated by a unique task ID. The key step snapshots are stored in the object storage system by the storage management unit, and the ordinary step snapshots are stored in the local file. At the same time, a 30-day storage validity period is set to automatically clean up expired data. If the subsequent composite image label misplacement problem occurs, technicians can retrieve each step snapshot through the task ID to determine whether the problem is caused by coordinate calculation error in the rendering link.

[0057] Verification method: 20 body measurement tasks containing abnormal processing records are selected, and the existing technology and the embodiment scheme are used for processing respectively. The average alarm response time MTTA and the average repair time MTTR of the two groups of tasks are counted. If the MTTA of the embodiment scheme is ≤5 minutes, the MTTA of the existing technology is usually ≥30 minutes, the MTTR is ≤15 minutes, the MTTR of the existing technology is usually ≥60 minutes, and 100% of the tasks can pass through the snapshot to backtrack the complete processing flow, it is proved that the step-level trace effect meets the standard, and the problem of untraceable process in the existing technology is solved.

[0058] The technical effect of this embodiment is to realize the observability and replayability of the whole process of body measurement image processing, greatly shorten the problem positioning and repair time, reduce the operation and maintenance troubleshooting cost, avoid the repeated occurrence of similar faults caused by the problem that cannot be reproduced, and ensure the stability and continuity of the body measurement service.

[0059] Embodiment six

[0060] In view of the problems of poor network adaptability and low-quality materials affecting the analysis results in the prior art, the system stability and data quality are improved through hierarchical fault tolerance and quality gating modules.

[0061] Specifically: before the image processing, the quality gating module first performs pre-checking, the resolution detection unit determines whether the minimum resolution of the image is within the range of 64-2048 pixels, the aspect ratio checking unit confirms whether the aspect ratio is within the interval of 0.1-10, the file integrity unit checks the damaged files through CRC checking, the images that do not meet the standard are directly returned with the prompt of “substandard material quality”, the images that pass the checking enter the download link, the hierarchical fault tolerance module sets an 8-second timeout threshold for the download link, triggers a 2-second fixed interval retry for the first timeout, switches to 4-second and 8-second exponential backoff retry for 2 consecutive retry failures, and automatically returns a preset white placeholder image if it still fails, the decoding link sets a 5-second timeout threshold, directly triggers a rollback after the timeout, records error logs, and in the network fluctuation scenario, the system can still ensure that more than 80% of the tasks normally output results through the retry and rollback mechanism.

[0062] Verification method: simulate a mixed scene test, initiate 1000 times of volume requests, process the requests using the prior art and the embodiment respectively, and count the system success rate SR, timeout rate TR, and quality unqualified rejection rate RR of the two groups of requests. If the SR of the embodiment is ≥92%, the SR of the prior art is usually ≥60%, the TR of the embodiment is ≤5%, the TR of the prior art is usually ≥25%, the RR of the embodiment is 50%, the RR of the prior art is 10% and contains a large number of low-quality image omissions, and there is no case of volume data error exceeding the standard caused by low-quality materials, and the error exceeding the standard accounts for 20% of the prior art, it is proved that the hierarchical fault tolerance and quality gating effect meet the standard, and the problems of poor stability and low data quality of the prior art are solved.

[0063] The technical effect of the embodiment is that low-quality materials are effectively filtered to avoid entering the subsequent process and causing volume data deviation, the anti-risk ability of the system in a complex network environment is improved, the timeout and crash probability is reduced, and the accuracy of the volume analysis result and the stability of the system operation are ensured.

[0064] Embodiment seven

[0065] In view of the problems of low multi-image processing performance and high resource occupation of the prior art, the performance and resource optimization module is used to improve the system throughput and resource utilization.

[0066] Specifically: when processing a 4K resolution multi-image body task, the performance and resource optimization module first determines a 1 / 2 downsampling ratio according to the target size to reduce the image resolution to 1920x1080 pixels through the downsampling processing unit, and divides the synthesis process into left region-middle region-right region, loads and draws the left two images first, releases the memory to load the right two images, avoids memory overflow caused by loading four full-resolution images at once, the resource reuse unit establishes a user image cache library to store user historical body images and processing results, and directly calls the cached data when the user initiates the same image processing request again, and adopts the parallel download and serial drawing strategy, that is, four images are downloaded in parallel and drawn in series after the download is completed.

[0067] Verification method: in the same hardware environment, initiate 500 multi-image body requests, process them by using the prior art and the embodiment respectively, and count the P95 response time, memory peak M peak , the number of successful processing per unit time TPS, if the P95 response time of the embodiment is ≤8 seconds, the prior art is usually ≥25 seconds, M peak≤ 4GB, the prior art is usually ≥10GB, TPS ≥60, and the prior art is usually ≥20, then it is proved that the performance and resource optimization effect meets the standard, and the problems of insufficient performance and resource waste of the prior art are solved.

[0068] The technical effect of the embodiment is that: under the premise of ensuring the quality of the body image, the CPU and memory peak occupation are greatly reduced, the system response speed and concurrent processing capacity are improved, the system can stably support high concurrent body demand, and the decline of user experience caused by performance bottleneck is avoided.

[0069] The preferred embodiments of the application disclosed above are only used to help explain the application, the preferred embodiments do not describe all the details, and the application is not limited to the specific embodiments described, the embodiments are selected and described in detail in the specification, in order to better explain the principles and practical applications of the application, so that the persons skilled in the art can well understand and utilize the application.

Claims

1. An AI intelligent somatometer analysis system, characterized in that: The adaptive normalization module, the adaptive layout solving module, the edge and background unification module, the unified export and protocol module, the step-level trace module, the hierarchical fault tolerance and quality gating module, and the performance and resource optimization module are comprised. The adaptive normalization module is used for performing security margin setting and alpha channel unification processing on multi-source heterogeneous volume image materials, first estimating a content-priority scaling ratio based on a volume image content bounding box, then performing rendering, and simultaneously unifying the background color and white space strategy, assuming that the original size of the volume image material is (w i ,h i ), the target height is H * , the scaling ratio s=H * / h i , the scaled image width w i ′=s·w i , the canvas height H c =H * +2p, and the canvas width W c =∑(w i ′)+p·(n+1)+(n-1)·g, wherein p is a security margin, n is the number of volume images, and g is an image spacing. The adaptive layout solving module is used for combining the equal high and wide strategy with the content proportion of the volume image, dynamically solving the canvas size and the image placement coordinate, adopting the dynamic padding / gap and baseline alignment strategy, and defining the alignment error E align =∑|y i -y ref |, y i is the bottom coordinate of the volume image, y ref is the bottom reference baseline, the height h i ′=H * of all volume images is scaled under the equal high strategy, E align →0 is realized, for the equal wide strategy, the horizontal alignment error E align,x =∑|x i -x ref |, x i is the left coordinate of the volume image, X ref is the horizontal reference baseline, and the width w i ′=W * of all volume images is uniformly scaled, W * is the target width, and E align,x →0 is realized. The edge and background unification module is configured to unify the alpha channel processing mode of the volume image, adopt an edge anti-aliasing technology, and determine a unified background strategy, which includes white background and transparent. The unified export and protocol module is configured to unify the export encoding format and response protocol of the volume image, the export encoding format includes PNG and JPEG, and the error code and state information in the response protocol are structured. The step-level trace module is configured to generate snapshots of each step of the volume image processing in the timestamp-step-description.png format, record key metadata and the time consumption of each step, and the key metadata includes image size, scaling ratio, and layout parameters. The hierarchical fault-tolerant and quality gating module is used to set hierarchical timeout, retry, fallback strategy in volume image downloading and decoding link, the fallback strategy includes returning a placeholder image, displaying prompt information, and setting minimum resolution R min , aspect ratio p and file integrity as pre-quality gating standards, R min ∈ [64, 2048] pixels, p ∈ [0.1, 10]; The performance and resource optimization module is configured to perform on-demand downsampling processing on the volume image, perform drawing operations in steps and reuse resources, and adopt a hybrid strategy of parallel downloading and serial drawing in the volume image acquisition and processing process.

2. The AI intelligent body analysis system according to claim 1, characterized in that: In the adaptive normalization module, the transparent processing mode includes at least one of premultiplied alpha, non-premultiplied alpha, edge feathering, threshold suppression, and anti-aliasing resampling, the normalization processing order includes any one of first unifying the alpha channel and then scaling, first scaling and then background filling, and first performing quality gating and then unifying the alpha channel, and the running position includes any one of a front-end link in a controller, a service layer independent module, and a preprocessing pipeline before storage. 3.The AI intelligent body analysis system of claim 1, wherein: In the adaptive layout solving module, the objective function includes any one of minimum alignment error, minimum canvas area, maximum uniformity of white space, and multi-objective weighting, the solving strategy includes any one of analytical solution, heuristic algorithm, constraint optimization algorithm, and dynamic programming, the heuristic algorithm includes greedy algorithm and iterative algorithm, the constraint optimization algorithm includes linear programming and quadratic programming, the layout type includes any one of horizontal side-by-side, vertical stacking, grid table, waterfall flow, golden section, and equal proportion segmentation, the alignment reference includes any one of top, bottom, center line, grid reference, and reference box, the interval control mode includes any one of fixed gap, adaptive gap, and segmented function gap, and the adaptive gap is related to the size, number, and canvas size of the volume image.

4. The AI intelligent body analysis system of claim 1, wherein: In the unified export and protocol module, the drawing backend includes any one of CPU Canvas, Skia, GPU, and WASM rendering backend, the GPU supports interfaces including OpenGL, Metal, Vulkan, and DirectX, the sampling kernel includes any one of bilinear, bicubic, Mitchell-Netravali, and Lanczos, the export encoding format further includes WEBP and AVIF, the color space supported by the color processing includes sRGB, Display-P3, and linear space, and gamma correction function is supported.

5. The AI intelligent body analysis system according to claim 1, characterized in that: In the step-level trace module, the snapshot storage medium includes any one of local files, object storage, memory ring buffer, and metadata storage only, the log structure includes any one of text, JSON, and a combination of text and JSON, the uniform protocol supports any one of JSON, CBOR, and Protobuf, the protocol field name is configurable and the semantics are consistent.

6. The AI intelligent body analysis system according to claim 1, characterized in that: In the hierarchical fault tolerance and quality gating module, the retry strategy includes any one of fixed retry, exponential backoff retry, and jitter backoff retry, and the maximum number of retries and the total time limit are set, the fallback strategy further includes any one of resolution reduction processing, quality reduction derivation, and layout metadata return only, and the resource gating standard includes at least one of the maximum image side length, the maximum number of pixels, the number of concurrent downloads, and the memory water level triggering degradation.

7. The AI intelligent body analysis system according to claim 1, characterized in that: In the edge and background color unification module, when performing alpha channel unification processing, the alpha channel format of the multi-source heterogeneous volume image is first automatically detected, different types such as premultiplied alpha, non-premultiplied alpha, and no alpha channel are identified, then format conversion is performed based on the system preset alpha channel standard format, for the volume image without an alpha channel, if the system background color strategy is set to transparent, an alpha channel is automatically added and the alpha value of the background area is set to 0, if the background color strategy is set to white background, no alpha channel needs to be added and a white background color is directly filled. 8.The AI intelligent body analysis system of claim 1, wherein: In the adaptive normalization module, the safety margin p and the value range of the image spacing g are [0, 10 3 ] or [0%, 20%] of the canvas length. 9.The AI intelligent body analysis system of claim 1, wherein: In the performance and resource optimization module, the sampling ratio of the downsampling processing is determined according to the ratio of the target size of the volume image to the original size, the step-by-step rendering is performed according to the image area, and the resource reuse includes image cache reuse and calculation result reuse.

10. The AI intelligent volume analysis system according to any of claims 1-9, characterized in that: The above system is based on the Node.js / Express Web backend architecture, provides an HTTP interface, and is divided into an access and routing layer, a controller layer, a service layer, a data access layer, and a file object storage layer from top to bottom according to responsibilities, and is equipped with a cross-cutting middleware layer and a configuration layer.

Images