A child sense of balance training AI intelligent evaluation and training coordination system

Abstract

The application discloses a kind of children sense training AI intelligent evaluation and training coordination system, based on AI intelligent algorithm construction, including perception acquisition module, quantification evaluation and re-evaluation module, individualized training scheme generation module, training execution and control module, data storage analysis and general control module, each module linkage constructs whole-process closed-loop coordination system, and is equipped with standardization intelligent training evaluation scene;It can also add multi-scene adaptation home-school residual link interactive module, adapt to rehabilitation training institutions, school, residual link and other multi-scene reuse.The application realizes the accurate quantification evaluation of children's sense ability, individualized training scheme automatic generation and dynamic regulation through multi-module AI coordination, solves the problem that existing children's sense training evaluation is not accurate, scheme individualization is insufficient, multi-scene adaptability is poor, and each link coordination is weak, improves training efficiency and evaluation accuracy, guarantees data security, adapts to different use scenarios and children individual needs, has strong practicality, and has wide protection range.

Description

Technical Field

[0001] This invention relates to the field of sensory integration training technology for children, specifically to an AI-powered intelligent assessment and training collaboration system for children's sensory integration training. It is applicable to various scenarios such as rehabilitation training institutions, families, schools, and disabled persons' federations, enabling accurate quantitative assessment of children's sensory integration abilities, personalized training, and full-process collaborative management, accurately supporting the scope of protection defined by the claims. Background Technology

[0002] Sensory integration dysfunction in children can severely impact their physical and mental health and growth and development. Existing technologies related to sensory integration training for children have significant shortcomings and cannot meet practical application needs: assessment methods are crude and lack precision, lacking multi-dimensional data support and reliability optimization mechanisms; training programs have low personalization, making it difficult to adapt to individual children's developmental patterns and different sensory integration ability levels; assessment and training are disconnected, with weak coordination, failing to achieve a dynamic cycle of assessment-training-optimization; scenario adaptability is poor, making it difficult to meet the reuse needs of multiple scenarios such as rehabilitation institutions, schools, and disability federations; system control precision is insufficient, data security is inadequate, and the training process lacks precise control and comprehensive safety protection. To address the shortcomings of the existing technologies, there is an urgent need for an AI-powered intelligent assessment and training collaboration system for children's sensory integration training that can solve the above problems, be stably implemented, and has a reasonable scope of protection, corresponding to the technical solutions defined in claims 1-19, filling the gap in existing technology. Summary of the Invention

[0003] This invention aims to overcome all the shortcomings of existing technologies and provide an AI-powered intelligent assessment and training collaboration system for children's sensory integration training. Based on AI intelligent algorithms, it constructs a multi-module closed-loop collaborative system with standardized scenarios and optional addition of multi-scenario interactive modules. This enables accurate assessment of children's sensory integration abilities, automatic generation of personalized training plans, dynamic control of the training process, and reuse across multiple scenarios. It solves the core problems of existing technologies, such as inaccurate assessments, insufficient personalization of plans, weak collaboration, and poor scenario adaptability, ensuring the stable implementation of the technical solution. Simultaneously, it provides comprehensive support for claims 1-19, achieving maximum protection.

[0004] To achieve the above-mentioned objectives, this invention provides the following technical solutions, which fully correspond to the technical features of claims 1-19, clearly defining the implementation methods of each feature, and ensuring that those skilled in the art can implement this system based on this specification: This system is built upon AI intelligent algorithms, and its core components include a perception and acquisition module, a quantitative evaluation and review module, a personalized training plan generation module, a training execution and control module, and a data storage, analysis, and overall control module. The specific architecture of these five core modules is as follows: Figure 1As shown; each core module is linked through wired or wireless communication protocols to construct a closed-loop collaborative system covering the entire process of "sensory integration assessment - solution development - training implementation - phase review - solution optimization"; simultaneously, standardized intelligent training and assessment scenarios (which can be physical spaces or virtual simulation spaces) are built to provide a standardized hardware environment and scenario support for the overall system operation; optional multi-scenario adaptation modules for home-school-disabled persons' federation interaction can be added to achieve multi-scenario reuse without affecting the operation of core modules; the operational logic of the above-mentioned closed-loop collaborative system corresponds to the numbered sequence. Figure 8 As shown.

[0005] Corresponding to claims 1, 2, 8, and 14, the perception and acquisition module is configured with multi-dimensional data acquisition components and a preprocessing process. A distributed deployment forms a full-domain perception network within a standardized intelligent training and assessment scenario. The collected raw data related to children's sensory integration training, after preprocessing, is transmitted in real-time to the quantitative assessment and review module and subsequent related modules, providing high-quality data support for assessment, training, and regulation. Its specific structure corresponds to the following in numerical order: Figure 2 As shown.

[0006] The multi-dimensional data acquisition components are used to collect multi-dimensional data on children's movements, physiology, and environment. Preferably, they include motion capture devices, visual acquisition devices, and physiological feature detection devices. Alternatively, one or more of these devices can be selected based on the scenario requirements. All acquisition devices work collaboratively to achieve comprehensive data acquisition: the motion capture devices include, but are not limited to, inertial sensors, pressure sensors, and laser devices. Inertial sensors are suitable for dynamic training scenarios, pressure sensors are suitable for contact training scenarios, and laser devices are suitable for unobstructed, long-distance scenarios. They can capture data such as children's limb movement posture and contact pressure, achieving the acquisition of minute limb displacements at the 0.05-0.1mm level. The visual acquisition devices can accurately capture children's minute limb displacements and movement angles, working in conjunction with the motion capture devices to achieve comprehensive, high-precision acquisition of children's limb movements. The physiological feature detection devices are wearable structures used to collect children's heart rate, blood oxygen, skin conductance, body temperature, and other physiological feature data. The heart rate acquisition accuracy is ±1 beat / minute, and the blood oxygen acquisition accuracy is ±1%, meeting the monitoring needs of children's sensory integration training. The environmental perception sensors are used to collect environmental parameters such as temperature, humidity, noise, and light within the scene, providing data for environmental adjustment.

[0007] The preprocessing process involves targeted processing of the raw data acquired by the acquisition components, including but not limited to noise reduction, time synchronization, data standardization, drift calibration of motion capture data (using an adaptive calibration algorithm assisted by a lightweight large model), background noise reduction and contour extraction of visual acquisition data, and baseline calibration of physiological feature data. This ensures data consistency and accuracy, and avoids invalid data from affecting subsequent evaluation and training control effects. The lightweight large model has been specially fine-tuned with children's sensory integration acquisition data, which can quickly identify data drift deviations and has a calibration accuracy superior to traditional algorithms, adapting to the low latency requirements of real-time acquisition scenarios.

[0008] Corresponding to claims 1, 3, 9, and 15, the quantitative assessment and reassessment module incorporates a trained feature fusion intelligent analysis model and an assessment credibility optimization module. It receives preprocessed data transmitted from the sensory acquisition module, completes the initial quantitative assessment of children's sensory integration abilities and the phased reassessment during the training process, and outputs quantitative results including the type of children's sensory integration abilities (vestibular dysfunction, proprioceptive dysfunction, tactile dysfunction), the degree of dysfunction (mild, moderate, severe), scores for each dimension, and assessment credibility. The quantitative results can be directly synchronized to the personalized training plan generation module and support comparative analysis of quantitative results at different training stages. Simultaneously, it is synchronized to the data storage analysis and overall control module, with its workflow corresponding to the numbered sequence. Figure 3 As shown.

[0009] The feature fusion intelligent analysis model adopts a lightweight large model architecture based on the attention mechanism (preferably a multimodal large model with fine-tuned features for children's sensory integration), replacing the traditional deep learning + temporal algorithm fusion architecture. It achieves accurate alignment and deep fusion of multi-dimensional features through cross-modal interaction, adapts to the heterogeneous feature fusion and deep analysis needs of children's multi-dimensional related samples, and has the advantage of lightweight deployment, meeting the real-time evaluation needs of the terminal. The model training data includes multi-dimensional samples and professionally labeled assessment data of children of different ages and with different sensory integration abilities. The samples are precisely divided into critical periods of sensory integration development (3-12 years old) and correspond to specific sensory integration ability grading standards (normal, mild dysfunction, moderate dysfunction, severe dysfunction). The data covers samples related to children's movements, physiology, and emotions. Emotion-related samples are collected through the visual acquisition device and physiological feature detection device of the perception acquisition module, and are input into the model after feature alignment processing. The model dynamically optimizes the training samples through adaptive data augmentation processing. Adaptive data augmentation processing adopts a multi-strategy fusion approach, including but not limited to geometric transformation, pixel adjustment, and sample interpolation enhancement. The enhancement strategy can be dynamically adjusted according to the differences in sample distribution to improve the data diversity of multi-dimensional samples, avoid model overfitting, and thus improve its generalization ability and assessment robustness. At the same time, relying on the transfer learning capability of the large model, it can quickly adapt to the assessment needs of different scenarios and reduce sample labeling costs.

[0010] The assessment credibility optimization module uses a multi-dimensional feature weighted fusion algorithm to fuse action, physiological, and emotional features acquired by the perception acquisition module with the results of a feature fusion intelligent analysis model, thereby reducing assessment errors and improving the credibility and consistency of assessment results. A learnable attention-weighted fusion method is preferably adopted, where the weight coefficients of each fused feature are iteratively optimized using a gradient descent algorithm and meet normalization constraints. Simultaneously, behavioral response features from the child's training process (collected collaboratively by the perception acquisition module) are incorporated for adaptive dynamic fusion. Feature weights are adjusted in real-time through backpropagation of the loss function to achieve dynamic optimization of assessment credibility. When the assessment credibility falls below a preset threshold (preferably 85%), a reassessment process is automatically triggered.

[0011] Corresponding to claims 1, 4, 10, and 16, the personalized training program generation module and the quantitative assessment and review module are equipped with a built-in professional knowledge base for children's sensory integration training and an AI intelligent program matching algorithm. These modules achieve bidirectional data linkage with the data storage, analysis, and control module. The module obtains the quantitative assessment results of children's sensory integration abilities from the quantitative assessment and review module, and retrieves children's basic information (age, gender, sensory integration development foundation) and historical training data from the data storage, analysis, and control module. Based on the quantitative assessment results and children's basic information, a personalized sensory integration training program is automatically generated, and manual adjustment of parameters (training intensity, duration, progression rules, etc.) of the personalized training program is supported. The workflow corresponds to the module in numerical order. Figure 4 As shown.

[0012] The professional knowledge base for children's sensory integration training covers standardized training equipment, training movements, training intensity, and progression rules corresponding to different age groups and different sensory integration ability characteristics. The standardized training equipment includes, but is not limited to, intelligent balance beams, intelligent trampolines, intelligent tactile mats, intelligent rotating discs, intelligent swings, intelligent large dragon balls, and other common intelligent equipment for children's sensory integration training in this field. The standardized training equipment corresponds one-to-one with the standardized training movements, and also includes standardized training equipment combinations, training movement sequences, single movement training parameters, training intensity levels, progression nodes, and prohibition movement prompts, providing scientific support for the generation of personalized training programs.

[0013] The AI ​​intelligent solution matching algorithm's response time meets practical usage requirements. It can call upon historical training-related data (training completion rate, training effect, physiological response) stored in the data storage and analysis and overall control module. Combining the semantic understanding and logical reasoning capabilities of the large model, it optimizes the accuracy of personalized training plan generation through algorithm iteration. Based on the professional knowledge base, it can divide the training intensity into multiple gradients suitable for children with different sensory integration ability levels, achieving individual adaptive matching. Furthermore, it can call upon the advanced rules in the professional knowledge base based on children's training feedback, and dynamically optimize the personalized training plan through algorithm iteration, ensuring that the plan can adapt to the individual developmental patterns of children and improve the targeting and effectiveness of training.

[0014] Corresponding to claims 1, 5, 11, and 17, the training execution and control module integrates a dynamic control sub-module, a motion correction unit, a training load early warning unit, and a training result recording unit. These units work collaboratively, deployed within a standardized intelligent training and assessment scenario. They achieve low-latency linkage (linkage latency ≤ 100ms, preferably ≤ 80ms) with the personalized training plan generation module, the sensing and acquisition module, and the quantitative assessment and review module via a standardized communication link. This integrates training execution, real-time monitoring, dynamic control, and motion correction functions, enabling precise implementation of personalized training plans. Furthermore, it adjusts training implementation strategies based on the dynamic training data transmitted by the sensing and acquisition module, and performs real-time correction of non-standard training movements. The logic corresponds to… Figure 5 As shown.

[0015] The dynamic control submodule can perform real-time data analysis based on the children's training dynamic data (action completion rate, physiological load data) transmitted by the sensing and acquisition module. The real-time data analysis frequency is ≥10 times / second to meet the timeliness requirements of training control. The analysis results are synchronously fed back to the training result recording unit, and the control instructions are synchronously fed back to the quantitative assessment and reassessment module for real-time adjustment of assessment parameters, which meets the collaborative optimization requirements of training and assessment. It can automatically adjust one or more of the following based on the analysis results: training intensity, training duration, and training action sequence. The control logic can be linked to the training load early warning unit to ensure the synergy between early warning and control.

[0016] The motion correction unit uses multimodal prompts (including one or more combinations of voice, light, and vibration) to correct non-standard training movements in real time, and the relevant prompts for motion correction are simultaneously stored in the training result recording unit.

[0017] The training load early warning unit is used to monitor and provide tiered early warnings of children's physiological load in real time during training. The tiered early warning mechanism is specifically divided into multiple early warning levels based on children's physiological load thresholds. Each early warning level corresponds to a different physiological load threshold and a different emergency response strategy. The early warning events and corresponding handling are transmitted synchronously to the training result recording unit.

[0018] The training result recording unit is used to record in real time the child's movement data, physiological load data, error correction records, early warning events and corresponding handling, training completion status and other relevant result data throughout the entire training process. All recorded data is synchronously transmitted to the data storage analysis and overall control module, and simultaneously fed back to the quantitative assessment and review module, providing accurate data support for subsequent review and training program optimization, and assisting in the collaborative optimization of training and assessment.

[0019] Corresponding to claims 1, 6, 12, and 18, the data storage analysis and overall control module incorporates data analysis algorithms and AI system overall control functions. This module serves as the core management and data processing center of the system, storing data throughout the entire system operation process, supporting multi-dimensional data retrieval and historical traceability (data retention time ≥ 3 years, meeting industry standards). Standardized communication links are established between modules to achieve collaborative work and centralized system management. Training data is fed back to the evaluation module and solution generation module in real time, realizing a dynamic cycle of evaluation-training-optimization, improving system operating efficiency and the accuracy of evaluation and training.

[0020] The data analysis algorithm can perform multi-dimensional classification and mining of data throughout the entire system operation process (collected data, evaluation data, training data, and module operation data), accurately identify evaluation data deviations, training data anomalies, and module operation data fluctuations, and output evaluation accuracy optimization suggestions and training scheme adjustment suggestions. The optimization suggestions can clearly define the specific optimization direction and quantitative reference values. At the same time, with the multi-dimensional data retrieval and historical traceability functions, the optimization suggestions can be accurately correlated with the original data. Combined with historical data trend analysis, it provides a practical data basis for the long-term iteration of the system.

[0021] The AI ​​system's overall control function can achieve real-time monitoring of the operating status of each module, intelligent scheduling of communication links, and unified issuance of operation commands through intelligent algorithms. It can automatically adjust the communication link scheduling strategy and command issuance priority according to the operating load of each module, capture module operation anomalies in real time and push early warning prompts, and achieve rapid handling of anomalies in conjunction with the visual operation platform. The visual operation platform optimizes the interaction logic, supports manual one-click start of the entire process of evaluation, training, and re-evaluation, reduces the system operation threshold, supports multi-dimensional data retrieval and historical traceability, and adapts to the operation needs of different users.

[0022] Corresponding to claims 1, 7, and 13, the standardized intelligent training and assessment scenario serves as the physical carrier of the system, providing operational support. It adopts a functional zoning design, with three mandatory zones: an assessment zone, a training zone, and a central control zone. Auxiliary functional zones can be added as needed. Its core purpose is to achieve the orderly separation and coordinated linkage of children's sensory integration assessment, training execution, and system control, taking into account the practicality and adaptability of the scenario, and providing reliable support for the efficient implementation of children's sensory integration assessment and training.

[0023] The assessment area serves as the core area for children's sensory integration assessment and training program generation. It is adapted to deploy assessment-related modules and is primarily responsible for collecting children's sensory integration ability assessment data, performing assessment calculations and analysis, and generating personalized training programs, providing accurate data support and program basis for subsequent training implementation.

[0024] The training area serves as the core area for the execution and dynamic control of children's sensory integration training. It is adapted to deploy training-related modules and is designed to enable the execution of training tasks, dynamic control of the training process, and real-time feedback of training data, ensuring that the training process aligns with the assessment results and the child's own abilities.

[0025] As the core hub for overall system control and data management, the central control area is adapted to deploy data storage and system control-related modules. Its core functions include overall system data management, collaborative scheduling of various partition modules, system parameter configuration, and monitoring of operational status, ensuring the efficient and coordinated operation of all system modules.

[0026] The auxiliary functional area (added as needed) is flexibly deployed with child-friendly auxiliary facilities, such as rest facilities, guidance facilities, and protective auxiliary facilities, according to actual usage needs. This is used to improve the comfort and safety of children during the assessment and training process and to help the core functional areas operate efficiently.

[0027] At the hardware deployment level, the hardware devices for the system's perception and data acquisition module adopt a distributed deployment mode within the scene, and are integrated and deployed in the assessment and training areas to ensure comprehensive and accurate collection of various perception data (such as movement data, physiological feedback data, etc.) of children during the assessment and training process, providing reliable data support for assessment analysis and training regulation. Each functional area deploys system module hardware devices adapted to its own core functions. Each module can be directly deployed in the corresponding functional area, or it can be linked and adapted with the corresponding functional area according to linkage requirements, ensuring that the module functions and the area positioning are highly consistent, and improving the overall operating efficiency of the system. Meanwhile, to ensure scene safety and system stability, the scene configuration includes safety protection and emergency support equipment adapted to each core functional area. The equipment is configured differently based on the functional characteristics of each area to ensure it meets the core needs of each area and is reasonably and practically configured: Safety equipment in the testing area meets the safety requirements of the entire testing module process, balancing data collection accuracy and child safety; safety equipment in the training area meets the operational safety requirements of the training module, focusing on preventing the risk of bumps and knocks during children's physical movements; safety equipment in the central control area meets the safety requirements of the data storage and management module, ensuring equipment stability and data security. Safety equipment in the auxiliary functional areas is configured as needed, tailored to the usage requirements of auxiliary facilities. The safety protection equipment includes, but is not limited to, various protective components (such as protective mats, protective covers, protective fences, etc.) and safe power supply devices (such as leakage current protectors, stable power supplies, etc.); the emergency support equipment includes, but is not limited to, emergency call devices (such as emergency call buttons, voice call devices, etc.) and emergency response related equipment (such as emergency lighting, first aid kits, etc.). All the aforementioned devices, along with the core functions, module deployments, and hardware configurations of each functional area, form a complete adaptation loop. This further refines and supplements the requirements for scenario safety and system stability, jointly ensuring the safety and continuity of the children's assessment and training process, as well as the stability of the operation of each system module, fully implementing the core requirements of the system scenario design. Its logical correspondence... Figure 6 As shown.

[0028] Corresponding to claim 19, the multi-scenario adaptation home-school-disabled persons' federation interaction module is an additional functional module of the system, which can be selectively configured and does not interfere with the normal use of the standardized intelligent training and assessment scenario. This module prioritizes adaptation to the sensory integration rehabilitation assessment scenario for children in rehabilitation training institutions, while also being compatible with home-school collaborative training scenarios and sensory integration rehabilitation scenarios for children in disabled persons' federations, achieving multi-scenario reuse and adapting to the needs of different usage scenarios. The multi-scenario adaptation home-school-disabled persons' federation interaction module can achieve bidirectional data linkage with the quantitative assessment and reassessment module and the data storage analysis and overall control module, building a multi-terminal interaction platform adapted to rehabilitation training institution terminals, teacher terminals, parent terminals, and disabled persons' federation terminals (terminal architecture corresponds in numerical order). Figure 7As shown in the diagram, the module covers various users, including medical staff, teachers, parents, and rehabilitation workers from the Disabled Persons' Federation in rehabilitation training institutions. Each terminal's permissions are adapted to the multi-level access control requirements of the data security protection module. The module enables the synchronous sharing of children's sensory integration assessment results and training process records, supporting rehabilitation training institution diagnosis and assessment, efficient home-school communication, rehabilitation guidance from the Disabled Persons' Federation, and extended family training guidance. Before data synchronization and sharing, it undergoes anonymization processing by the data security protection module (hiding key privacy fields such as name, ID number, and home address). It employs conventional encryption algorithms such as AES to ensure data transmission and storage security, and optimizes data display and operation permissions for different usage scenarios to ensure the security of children's privacy data. Beneficial effects

[0029] Compared with the prior art, this invention fully corresponds to the technical features of claims 1-19, and has the following significant beneficial effects, further supporting the protection scope of the claims and highlighting the innovation and practicality of the technical solution: High accuracy and reliability of assessment: It adopts multi-dimensional complementary data collection and targeted preprocessing (lightweight large model assisted calibration), combined with a multimodal large model based on attention mechanism and a reliability optimization module, which improves the accuracy of assessment by ≥30%, can accurately reflect the true level of children's sensory integration, solves the problem of rough assessment and large error in existing technologies, and fully supports the technical features of claims 3, 9 and 15. The training program is personalized and highly adaptable: It combines children's individual characteristics and historical training data, and generates and dynamically optimizes the training program through a professional knowledge base and AI intelligent program matching algorithm (integrating large model reasoning ability). It supports manual adjustment, improves training efficiency by ≥50%, and adapts to the individual needs of children of different ages and different sensory integration ability levels, supporting the technical features of claims 4, 10 and 16. Closed-loop collaboration and precise control throughout the entire process: The five core modules work together with low latency to build a closed-loop system of "assessment-solution-training-reassessment-optimization". Dynamic regulation and assessment are coordinated and optimized, and the training process is monitored, corrected and warned in real time, which improves the system's operating efficiency and rehabilitation training effect, and supports the technical features of claims 1, 5, 11 and 17. Data security and ease of use: Encryption and desensitization combined with multi-level access permissions effectively prevent the leakage of children's privacy data; The visual operation platform supports one-click full-process operation, reducing the operation threshold and supporting the technical features of claims 6, 12, and 18. Strong adaptability and high reusability across multiple scenarios: Optional interactive modules enable reuse in multiple scenarios such as rehabilitation institutions, families, schools, and disabled persons' federations, breaking down data barriers, adapting to the usage needs of different scenarios, and not affecting the operation of core functions, supporting the technical features of claim 19. Highly practical and safe: Standardized scenarios are equipped with comprehensive environmental adjustments, safety protection, and emergency response equipment. All five core modules are optimized based on existing mature hardware and large-scale model-related algorithms. Those skilled in the art can quickly implement the system according to the instructions, ensuring comprehensive safety for children's training and supporting the technical features of claims 7 and 13. Highly scalable and with broad protection: The modular design allows for flexible configuration and independent optimization of the five core modules, supporting large-scale technical iteration and upgrades to adapt to future needs, while covering all technical features of claims 1-19. Detailed Implementation

[0030] The present invention will be further described in detail below with reference to specific embodiments. The present invention is not limited to the described embodiments; any modifications, equivalent substitutions, or improvements within the scope of claims 1-19 are all within the protection scope of the present invention. Devices and algorithms not explicitly described in this embodiment are all existing mature technologies and can be implemented using conventional products and methods in the field, ensuring the feasibility of the technical solution. Example

[0031] This embodiment provides an AI-powered intelligent assessment and training collaboration system for children's sensory integration training, applicable to rehabilitation training institutions, corresponding to claims 1-18, as detailed below: The system is built upon AI intelligent algorithms and comprises five core modules: a perception and acquisition module, a quantitative evaluation and review module, a personalized training plan generation module, a training execution and control module, and a data storage, analysis, and overall control module. The specific architecture of these five core modules is described in [reference needed]. Figure 1 Each module is linked through the WiFi wireless communication protocol, and the data storage, analysis and central control module uniformly schedules the entire process to build a closed-loop collaborative system; a physical standardized intelligent training and assessment scenario is built to meet the needs of rehabilitation institutions for large-scale and standardized assessment and training.

[0032] (I) Sensing and Acquisition Module: Equipped with motion capture equipment (inertial sensor + pressure sensor), high-definition visual acquisition equipment (1080P camera), children's wearable physiological monitoring equipment (wearable bracelet), and environmental perception sensors; the inertial sensors are worn on the child's limbs, and the pressure sensor is laid on the surface of the training mat to simultaneously collect the child's limb movement posture and action contact pressure data, which can capture minute limb displacements of 0.1mm; the high-definition camera is deployed at the top or side of the scene to capture the child's movement details and facial expressions, assisting the motion capture equipment to improve the acquisition accuracy; the optional wearable bracelet collects the child's heart rate (collection range 60-150 beats / minute, accuracy ±1 beat / minute), blood oxygen (collection range 90%-100%, accuracy ±1%), and skin conductance data in real time; the environmental perception sensor collects temperature, humidity, noise, and light data in the scene. After preprocessing (motion data is calibrated using an adaptive calibration algorithm assisted by a lightweight large model, visual data is denoised using Gaussian filtering, and data is synchronized in time and outliers are removed), the raw data is transmitted in real time to the subsequent correlation modules. This lightweight large model uses a MiniLLM architecture specifically fine-tuned for children's sensory integration data, with a calibration response time ≤20ms and a drift calibration error ≤0.01mm, adapting to the requirements of real-time acquisition scenarios. Its specific structure is described in [reference needed]. Figure 2 .

[0033] (II) Quantitative Assessment and Reassessment Module: This module includes a built-in multimodal large-scale model based on attention mechanisms (a finely tuned version specifically for children's sensory integration dysfunction) and an assessment credibility optimization module. The assessment credibility threshold is set at 85%. The large-scale model training data includes over 10,000 sets of action, physiological, and facial expression samples from children aged 3-12 with different ages and degrees of sensory integration dysfunction, as well as assessment data annotated by professional medical personnel. Data enhancement processing, such as rotation, translation, and brightness adjustment, optimizes the model's generalization ability. Simultaneously, leveraging transfer learning capabilities, it reuses the feature extraction capabilities of a general large-scale model, reducing the need for sample annotation. Cost; After receiving preprocessed data, this module completes the initial quantitative assessment (assessment duration 10-15 minutes, adapted to children's attention span) and periodic reassessments (reassessment cycle 1-2 weeks, dynamically set according to children's training effects), outputting quantitative results including sensory integration dysfunction type, dysfunction degree, scores for each dimension, and assessment reliability, synchronized to subsequent modules, supporting comparison of quantitative results at different stages; Testing shows that this large model's assessment response time is ≤500ms, and the assessment accuracy is more than 15% higher than the traditional deep learning + temporal algorithm fusion architecture. Its workflow is as follows... Figure 3 .

[0034] (III) Personalized Training Program Generation Module: Links the quantitative assessment and reassessment module and the data storage, analysis, and overall control module to obtain information such as children's assessment results, age, gender, sensory integration development foundation, and historical training data; it has a built-in professional knowledge base for children's sensory integration training, covering intelligent training equipment, training actions, intensity, and progression rules corresponding to different dysfunction types for children aged 3-12; it adopts an AI intelligent program matching algorithm that combines collaborative filtering and decision tree algorithms, integrating the logical reasoning and semantic understanding capabilities of large models to automatically generate personalized training programs. It supports medical staff to manually adjust parameters such as training intensity and duration, and can dynamically optimize programs based on training feedback. Its workflow is as follows: Figure 4 .

[0035] (iv) Training Execution and Control Module: This module integrates with related modules with low latency (interconnection latency ≤ 80ms). The dynamic control sub-module performs real-time data analysis at a frequency ≥ 10 times / second. Based on the action completion and heart rate data transmitted by the sensing and acquisition module, it automatically adjusts the training intensity. Control commands are synchronously fed back to the quantitative assessment and review module. The action correction unit provides real-time prompts for children to adjust their actions via voice prompts and flashing lights. Correction records are stored in the training result recording unit. The training load warning unit categorizes children's heart rates into three levels: safe (60-100 beats / minute), warning (101-120 beats / minute), and dangerous (above 121 beats / minute), corresponding to different emergency response strategies. Warning events are synchronously transmitted to the training result recording unit. The training result recording unit stores all training data and synchronously transmits it to the data storage analysis and overall control module and the quantitative assessment and review module. Its internal structure is similar to... Figure 5 .

[0036] (V) Data Storage, Analysis and Control Module: The system uses an industrial-grade server to store all data throughout the process, with a data retention period of ≥3 years, supporting multi-dimensional retrieval and historical traceability; it has a built-in deep learning data analysis algorithm to identify deviations in evaluation data and anomalies in training data, and output optimization suggestions and quantitative reference values; it is equipped with a touch-screen visual operation platform, allowing medical staff to view evaluation results and training data, schedule the operation of each module, and support one-click start of the entire process operation; through AI intelligent scheduling algorithms, the system achieves collaboration among modules to ensure efficient system operation. According to tests, the evaluation accuracy of this embodiment is improved by ≥35%, and the training efficiency is improved by ≥55%.

[0037] (VI) Standardized Intelligent Training and Assessment Scenario: Providing operational support for the system, the scenario is divided into an assessment area, a training area, and a control area (see Figure 6). The assessment area is equipped with high-definition cameras and optional wearable wristbands for data collection. The training area features intelligent training equipment such as intelligent balance beams, intelligent trampolines, and intelligent tactile mats, along with safety features like anti-collision railings and anti-slip mats. Equipment in the assessment and training areas is reusable and shared, including both data collection devices and various training equipment, allowing for simultaneous assessment and training tasks. The control area houses a data storage, analysis, and central control module, equipped with a visual operation platform for unified management of the scenario. This scenario also includes an environmental parameter adjustment system that can automatically adjust environmental parameters to a comfortable range by linking air conditioning, humidifiers, desk lamps, and noise reduction devices. Emergency support equipment such as first-aid kits and emergency call devices are also provided to ensure children's safety.

[0038] System operation process (refer to) Figure 9 The process is as follows: 1. The child enters the scene, puts on the wearable device, and the perception and acquisition module is activated to collect and preprocess multi-dimensional data (with large-scale model-assisted calibration); 2. The quantitative assessment and reassessment module receives the preprocessed data, completes the initial quantitative assessment through the multimodal large-scale model, and outputs the assessment results; 3. The personalized training plan generation module automatically generates a personalized training plan based on the assessment results and the child's basic information, integrating the reasoning ability of the large-scale model, and is activated after confirmation by medical staff; 4. The training execution and control module executes the training plan, monitors the child's training status in real time, performs dynamic control and action correction, and records training data; 5. The data storage, analysis, and overall control module stores the data of the entire process, mines the data through data analysis algorithms, and outputs optimization suggestions; 6. According to the set reassessment cycle, the quantitative assessment and reassessment module completes the phased reassessment, and the personalized training plan generation module optimizes the training plan based on the reassessment results and historical training data, forming a closed-loop iteration. Example

[0039] This embodiment is applicable to multiple scenarios involving home-school collaboration and disability rehabilitation, corresponding to claims 1-19. Based on embodiment 1, it adds a multi-scenario adaptation home-school-disabled person interaction module. The structure and functions of the remaining modules are basically the same, with only minor optimizations for multi-scenario adaptation requirements. This module is configured with a multi-terminal interaction platform (refer to...). Figure 7(Listed in numerical order), it supports both computer-based (for medical staff and disabled persons' federation workers) and mobile app-based (for parents and teachers). It allows for five access levels: administrator, medical staff, parents, teachers, and disabled persons' federation workers. Users with different permissions can view and operate data and functions within their respective scopes. The module is interconnected with the data storage, analysis and control module, the quantitative assessment and review module, and the personalized training plan generation module, enabling cross-scenario synchronous sharing of assessment and training data: medical staff can view children's home training data and adjust training plans; parents can view children's institutional assessment and training results and receive extended home training guidance; disabled persons' federation workers can view children's rehabilitation data within their jurisdiction and develop rehabilitation work plans; teachers can view students' in-school training data and cooperate in conducting on-campus rehabilitation training. Simultaneously, children's privacy data is encrypted and anonymized, using AES encryption to ensure data transmission and storage security, hiding key privacy fields such as ID numbers and home addresses to prevent privacy leaks.

[0040] This embodiment improves the system's evaluation accuracy by ≥30% and training efficiency by ≥50%, enabling multi-scenario collaboration among rehabilitation training institutions, families, schools, and disabled persons' federations. It breaks down data barriers, enhances the continuity and systematic nature of children's sensory integration rehabilitation training, and fully adapts to the technical features of claim 19, solving the problems of poor scenario adaptability and insufficient multi-scenario collaboration in existing technologies. Attached Figure Description

[0041] The following is a brief introduction to the drawings used in the description of the embodiments or prior art. Obviously, the following drawings are only some embodiments of the present invention, and those skilled in the art can obtain other drawings without creative effort.

[0042] Figure 1 This is a diagram of the five major modules of the present invention; it is used to clearly illustrate the architecture and relationships of the perception and acquisition module, the quantitative evaluation and review module, the personalized training scheme generation module, the training execution and control module, and the data storage, analysis and overall control module, corresponding to the composition features of the modules in claim 1. Figure 2 This is a schematic diagram of the sensing and acquisition module structure of the present invention; it is used to illustrate the multi-dimensional data acquisition components and data flow of the sensing and acquisition module, corresponding to the structural and functional features of the sensing and acquisition module in claims 2, 8, and 14. Figure 3 This is a flowchart of the quantitative assessment and reassessment module of the present invention; it is used to illustrate the working logic, data input and output, and reassessment mechanism of the quantitative assessment and reassessment module, corresponding to the functional features of the module in claims 3, 9, and 15. Figure 4This is a flowchart of the personalized training scheme generation module of the present invention; it is used to illustrate the data linkage logic, scheme generation and optimization process of the module, corresponding to the functional features of the module in claims 4, 10 and 16. Figure 5 This is a schematic diagram of the training execution and control module structure of the present invention; it is used to illustrate the four sub-units built into the module and the collaborative relationship between each unit, corresponding to the internal structure and functional features of the module in claims 5, 11, and 17. Figure 6 This is a schematic diagram of the functional partitions of the standardized intelligent training and evaluation scenario of the present invention; it is used to show the layout of the scenario functional partitions that provide support for the core module and the device deployment logic of each partition, corresponding to the functional partitions and security protection features of the scenario in claims 7 and 13. Figure 7 This is a terminal architecture diagram of the multi-scenario adaptation home-school-disabled persons' federation interaction module of the present invention; it is used to show the multi-terminal interaction platform composition of the additional module and the adaptation subject of each terminal, corresponding to the terminal architecture and functional features of the module in claim 19. Figure 8 This is a flowchart of the entire process of the system's modules; corresponding to claim 1, it illustrates the closed-loop operation logic of "collection-evaluation-solution-training-analysis-optimization" formed by the linkage of the five core modules; Figure 9 The system operation flowchart of this invention corresponds to the system operation flow in Embodiment 1.

[0043] Explanation of reference numerals in the attached figures 1-Sensing and Acquisition Module, 11-Motion Capture Device, 12-Visual Acquisition Device, 13-Physiological Feature Detection Device, 14-Environmental Perception Sensor, 2-Quantitative Evaluation and Review Module, 21-Feature Fusion Intelligent Analysis Model (Multimodal Large Model), 22-Evaluation Credibility Optimization Module, 3-Personalized Training Scheme Generation Module, 31-Professional Knowledge Base, 32-AI Intelligent Scheme Matching Algorithm, 4-Training Execution and Control Module, 41-Dynamic Control Sub-module, 42-Motion Error Correction Unit, 43-Training Load Early Warning Unit, 44-Training Result Recording Unit, 5-Data Storage Analysis and Overall Control Module, 51-Data Analysis Algorithm, 52-Visual Operation Platform, 6-Standardized Intelligent Training and Evaluation Scenarios, 61-Evaluation Area, 62-Training Area, 63-Control Area, 64-Environmental Parameter Adjustment System, 65-Safety Protection Equipment, 7-Multi-Scenario Adaptive Home-School-Disabled Persons' Federation Interaction Module, 71-Multi-Terminal Interaction Platform.

[0044] Supplementary Explanation of Detailed Implementation Methods Each module and scenario of this invention is innovatively combined and optimized based on existing mature hardware and software algorithms (not a simple stacking of existing technologies). The core innovations fully correspond to the technical features of claims 1-19, and the specific innovations are as follows: The perception and acquisition module optimizes the distributed deployment logic of multiple devices and the multi-dimensional data preprocessing algorithm (introducing a lightweight large model to assist calibration), improving the accuracy and consistency of data acquisition (corresponding to claims 2, 8, and 14); The quantitative assessment module adopts a multimodal large model with special fine-tuning for children's sensory integration, optimizes the multimodal feature fusion architecture and data augmentation strategy, and achieves accurate assessment and dynamic optimization of credibility by combining the developmental characteristics of children's sensory integration (corresponding to claims 3, 9, and 15); The personalized training program generation module optimizes the individual characteristics of children and the training program. The matching logic integrates the reasoning capabilities of large models to achieve dynamic iteration of solutions (corresponding to claims 4, 10, and 16); the training execution and control module optimizes the collaborative control logic of low-latency linkage and physiological load and action error correction, improving training safety and effectiveness (corresponding to claims 5, 11, and 17); the data storage analysis and overall control module optimizes the data association mining and closed-loop management logic, driving efficient system operation (corresponding to claims 6, 12, and 18); the standardized scenario optimizes the linkage and adaptation logic between environmental parameters and children's physiological states and training scenarios, strengthening safety protection (corresponding to claims 7 and 13); the multi-scenario adaptation home-school-disabled persons' association interaction module optimizes the multi-terminal permission adaptation and multi-scenario data collaborative sharing logic, realizing multi-scenario reuse (corresponding to claim 19). These innovations synergistically overcome existing technological limitations, solving core problems such as low assessment accuracy, poor collaboration, and insufficient scenario adaptability in existing technologies. Simultaneously, relying on the advantages of large model technology, they improve the system's intelligence level and feasibility for implementation, demonstrating significant substantive features and substantial progress.

[0045] The scope of protection of this invention strictly corresponds to the technical solutions defined in claims 1-19. All modifications, equivalent substitutions, and improvements within the scope of these claims fall within the protection scope of this invention. The five core modules can be flexibly combined and optimized according to actual needs. All variations based on the core technical solutions of this invention are within the protection scope of this invention. Technical details not explicitly described in this invention can be implemented using existing mature technologies without affecting the implementation effect of this invention or the scope of protection of the claims.

Claims

1. A children's sensory integration training AI intelligent assessment and training collaboration system, characterized in that, Built upon AI intelligent algorithms, the system includes a perception and acquisition module, a quantitative evaluation and review module, a personalized training plan generation module, a training execution and control module, and a data storage, analysis, and overall control module. These modules are linked through wired or wireless communication protocols to construct a closed-loop collaborative system covering the entire process from sensory evaluation to plan development, training implementation, phase review, and plan optimization. A standardized intelligent training and evaluation scenario (which can be a physical space or a virtual simulation space) is also provided to support the system's overall operation with standardized hardware and scenario support. The perception and acquisition module is equipped with multi-dimensional data acquisition components, which are distributed and deployed in a standardized intelligent training and assessment scenario to form a full-domain perception network. The raw data related to children's sensory integration training collected is preprocessed and transmitted in real time to the quantitative assessment and reassessment module and subsequent related modules. The quantitative assessment and reassessment module incorporates a trained feature fusion intelligent analysis model and an assessment credibility optimization module. It adapts to the assessment function requirements of standardized intelligent training and assessment scenarios, and can be deployed in the scenario assessment area or linked with the assessment area. It receives preprocessed data transmitted from the perception acquisition module, completes the initial quantitative assessment of children's sensory integration abilities and the phased reassessment during the training process, and outputs quantitative results including the type of children's sensory integration abilities, degree of dysfunction, scores of each dimension, and assessment credibility. The quantitative results can be directly synchronized to the personalized training plan generation module, and supports comparative analysis of quantitative results at different training stages. The personalized training program generation module is adapted to the assessment function requirements of standardized intelligent training and assessment scenarios. It can be deployed in the scenario assessment area or linked with the assessment area, data storage analysis and overall control module. It can achieve bidirectional data linkage with the quantitative assessment and reassessment module and the data storage analysis and overall control module. It obtains the quantitative assessment results of children's sensory integration ability from the quantitative assessment and reassessment module and retrieves the children's basic information from the data storage analysis and overall control module. Based on the quantitative assessment results and the children's basic information, it automatically generates personalized sensory integration training programs and supports manual adjustment of parameters of the personalized training programs. The training execution and control module is adapted to the training function requirements of standardized intelligent training and assessment scenarios. It can be deployed in the scenario training area or linked with the training area. It can achieve low-latency linkage with the personalized training plan generation module, the perception and acquisition module, and the quantitative assessment and review module. It integrates training execution, real-time monitoring, dynamic control, and action correction functions to achieve accurate implementation of personalized training plans. It can also adjust the training implementation strategy based on the dynamic training data of children transmitted by the perception and acquisition module, and correct non-standard training actions in real time. The data storage, analysis, and control module stores all system operation data, adapts to the control function requirements of standardized intelligent training and evaluation scenarios, and can be deployed in the scenario control area or linked with the control area. It supports multi-dimensional data retrieval and historical tracing, and has built-in data analysis algorithms to provide data support for system optimization. Simultaneously, it is configured with control components and a visual operation platform, establishing standardized communication links between modules to achieve collaborative work and centralized system management. Collaborative work among modules is achieved through AI intelligent scheduling, enabling low-latency linkage between modules. Training data is fed back to the evaluation module and solution generation module in real time, realizing a dynamic cycle of evaluation-training-optimization, improving system operating efficiency and the accuracy of evaluation and training.

2. The AI-powered intelligent assessment and training collaboration system for children's sensory integration training according to claim 1, characterized in that, The multi-dimensional data acquisition component of the perception acquisition module is used to collect multi-dimensional data on children's movements, physiology, and environment. It includes one or more of motion capture devices, visual acquisition devices, environmental perception sensors, and physiological feature detection devices. The acquisition devices work together to achieve full-domain data acquisition. The preprocessing is a targeted processing of the raw data related to children's sensory integration training acquired by the acquisition components, including one or more of noise reduction, time synchronization, and data standardization.

3. The AI-powered intelligent assessment and training collaboration system for children's sensory integration training according to claim 1, characterized in that, The quantitative assessment and reassessment module incorporates a feature fusion intelligent analysis model and an assessment credibility optimization module. The feature fusion intelligent analysis model employs a deep learning-based algorithm architecture. Its training data includes multi-dimensional relevant samples of children of different ages and sensory integration abilities, as well as professionally labeled assessment data. Furthermore, the model dynamically optimizes the training samples through adaptive data augmentation to enhance its generalization ability and assessment robustness. The assessment credibility optimization module uses a multi-dimensional feature weighted fusion algorithm to optimize the action, physiological, and emotional features acquired by the sensory acquisition module. By combining the results of the feature fusion intelligent analysis model, the evaluation error can be reduced and the credibility and consistency of the evaluation results can be improved. The quantification results can be directly synchronized to the personalized training scheme generation module, providing accurate data support for the generation of personalized training schemes.

4. The AI-powered intelligent assessment and training collaboration system for children's sensory integration training according to claim 1, characterized in that, The personalized training program generation module incorporates a professional knowledge base on children's sensory integration training and an AI-powered intelligent program matching algorithm. The knowledge base covers standardized training equipment, exercises, intensity levels, and progression rules for different age groups and sensory integration abilities, providing scientific support for the generation of personalized training programs. The AI-powered intelligent program matching algorithm can access historical training data stored in the data storage and analysis module and the overall control module. Through algorithmic iteration, it optimizes the accuracy of personalized training program generation, ensuring that the program adapts to the individual developmental patterns of children and improves the targeting and effectiveness of training.

5. The AI-powered intelligent assessment and training collaboration system for children's sensory integration training according to claim 1, characterized in that, The training execution and control module is deployed within a standardized intelligent training and assessment scenario. It incorporates a dynamic control submodule, a motion correction unit, a training load early warning unit, and a training result recording unit. It achieves low-latency linkage (linkage latency ≤ 100ms) with the personalized training plan generation module, the sensing and acquisition module, and the quantitative assessment and review module via a standardized communication link. The dynamic control submodule performs real-time data analysis based on the children's training dynamic data transmitted by the sensing and acquisition module, and automatically adjusts one or more of the following based on the analysis results: training intensity, training duration, and training motion sequence. The motion correction unit provides real-time correction of non-standard training motions through multimodal prompts. The training load early warning unit monitors and provides tiered early warnings for the children's physiological load during training. The training result recording unit is used to record relevant result data such as children's movement data, physiological load data, error correction records, and training completion status in real time throughout the entire training process, and synchronously transmits the recorded data to the data storage analysis and overall control module to provide accurate data support for subsequent review and optimization of training programs.

6. The AI-powered intelligent assessment and training collaboration system for children's sensory integration training according to claim 1, characterized in that, The built-in data analysis algorithm of the data storage analysis and overall control module can mine and analyze the data of the entire system operation process, and output suggestions for optimizing evaluation accuracy and adjusting training schemes, providing data support for the overall optimization of the system. The AI ​​system overall control function of the data storage analysis and overall control module can realize real-time monitoring of the operation status of each module, intelligent scheduling of communication links and unified issuance of operation instructions through intelligent algorithms. The equipped visual operation platform supports manual one-click start of the entire process of evaluation, training and re-evaluation, reducing the threshold of system operation and supporting multi-dimensional data retrieval and historical traceability.

7. The AI-powered intelligent assessment and training collaboration system for children's sensory integration training according to claim 1, characterized in that, The standardized intelligent training and assessment scenario adopts a functional zoning design; it must include an assessment area, a training area, and a central control area, with auxiliary functional areas that can be added as needed; the hardware devices of the sensing and acquisition modules are distributed throughout the scenario, and integrated in the assessment and training areas to achieve data collection. Each functional zone deploys system module hardware devices adapted to its core functions. Specifically, the assessment area is adapted to deploy assessment-related modules to support functions related to children's sensory integration assessment and training program generation; the training area is adapted to deploy training-related modules to implement training execution and dynamic adjustment functions; and the central control area is adapted to deploy data storage and system management modules to support data management and module collaborative scheduling functions. Each module can be deployed in its corresponding functional zone or linked and adapted with its corresponding functional zone. Auxiliary functional areas deploy child-adaptive auxiliary facilities as needed. The scenario is also equipped with safety protection equipment and emergency support equipment adapted to each functional zone to ensure the safety of children's assessment and training and the stable operation of the system.

8. The AI-powered intelligent assessment and training collaboration system for children's sensory integration training according to claim 2, characterized in that, The multi-dimensional data acquisition component includes a motion capture device, a visual acquisition device, and a physiological feature detection device. The motion capture device includes, but is not limited to, conventional motion capture devices in the field such as inertial sensors, pressure sensors, and laser devices. The visual acquisition device, as an independent component of the multi-dimensional data acquisition component, can work with the motion capture device to accurately capture children's limb movements. The physiological feature detection device is a wearable structure used to collect children's physiological feature data such as heart rate, blood oxygen, skin conductance, and body temperature, with acquisition accuracy meeting the needs of children's sensory integration training monitoring. The motion capture device and the visual acquisition device work together to achieve comprehensive and high-precision acquisition of children's limb movements, complementing the physiological feature data collected by the wearable physiological feature detection device and the environmental data collected by the environmental perception sensor. The preprocessing further refines the raw data from each acquisition device, including but not limited to drift calibration of motion capture data, background noise reduction and contour extraction of visual acquisition data, and baseline calibration of physiological feature data, ensuring data consistency and accuracy, and providing high-quality multi-dimensional data support for subsequent evaluation and control.

9. The AI-powered intelligent assessment and training collaboration system for children's sensory integration training according to claim 3, characterized in that, The feature fusion intelligent analysis model adopts a deep learning algorithm architecture based on an attention mechanism, adapting to the heterogeneous feature fusion and in-depth analysis needs of children's multi-dimensional related samples. The multi-dimensional related samples of children in the feature fusion intelligent analysis model include, but are not limited to, children's action, physiological, and emotion-related samples, and cover the entire critical period of children's sensory integration development and different sensory integration ability levels. Among them, emotion-related samples are collected collaboratively by the visual acquisition device and physiological feature detection device of the perception acquisition module, and are input into the model after feature alignment processing. The adaptive data augmentation processing adopted by the feature fusion intelligent analysis model includes, but is not limited to, geometric transformation, pixel adjustment, and sample interpolation enhancement. The enhancement strategy can be dynamically adjusted according to the differences in sample distribution to improve the data diversity of multi-dimensional samples. The evaluation credibility optimization module adopts a learnable attention-weighted fusion method. The weight coefficients of each fused feature are iteratively optimized through the gradient descent algorithm and meet the normalization constraints to ensure that the fusion results are reasonable and accurate, further improving the evaluation credibility.

10. The AI-powered intelligent assessment and training collaboration system for children's sensory integration training according to claim 4, characterized in that, The professional knowledge base includes standardized training equipment combinations, training movement sequences, single-movement training parameters, training intensity levels, advancement nodes, and prohibited movement prompts. The AI ​​intelligent solution matching algorithm has a response time that meets actual usage requirements. It can divide the training intensity in the professional knowledge base into multiple gradients based on the child's historical training data to achieve individual adaptive matching. It can also call upon the advancement rules in the professional knowledge base based on the child's training feedback, and dynamically optimize the personalized training plan through algorithm iteration, further improving the targeting and effectiveness of training.

11. The AI-powered intelligent assessment and training collaboration system for children's sensory integration training according to claim 5, characterized in that, The real-time data analysis frequency of the dynamic control submodule of the training execution and control module meets the timeliness requirements of training control. Its analysis results are synchronously fed back to the training result recording unit, and the control instructions of the dynamic control submodule are synchronously fed back to the quantitative assessment and review module for real-time adjustment of assessment parameters, which meets the collaborative optimization requirements of training and assessment. The real-time data analysis frequency is ≥10 times / second. The action error correction unit of the training execution and control module adopts a multimodal prompting method including one or more combinations of voice, light, and vibration, and the relevant prompts for action error correction are synchronously stored in the training result recording unit. The training load early warning unit of the training execution and control module adopts a tiered early warning mechanism, which is to divide multiple early warning levels according to the child's physiological load threshold. Each early warning level corresponds to a different emergency handling strategy, and the early warning event and the corresponding handling situation are synchronously transmitted to the training result recording unit. The training result recording unit of the training execution and control module records the child's action data, physiological load data, error correction records, early warning events and corresponding handling, training completion status and other relevant result data in real time throughout the entire training process. All recorded data is synchronously transmitted to the data storage analysis and overall control module, providing accurate data support for subsequent review and training program optimization. At the same time, it is synchronously fed back to the quantitative assessment and review module to help achieve collaborative optimization of training and assessment.

12. The AI-powered intelligent assessment and training collaboration system for children's sensory integration training according to claim 6, characterized in that, The built-in data analysis algorithm of the data storage analysis and overall control module further optimizes data processing capabilities, accurately mining evaluation data, training data, and module operation data throughout the entire system operation process. It optimizes the output accuracy of evaluation precision suggestions and training scheme adjustment suggestions. Simultaneously, in conjunction with multi-dimensional data retrieval and historical tracing functions, it achieves precise correlation between optimization suggestions and original data, enhancing the targeted nature of overall system optimization. The AI ​​system overall control function of the data storage analysis and overall control module further refines operational management capabilities. Based on real-time monitoring of the operating status of each module, intelligent scheduling of communication links, and unified issuance of operation commands, it optimizes the interactive logic of the visual operation platform. In addition to supporting one-click manual initiation of the entire process of evaluation, training, and re-evaluation, it further lowers the system's operational threshold and enhances the response efficiency of collaborative linkage between modules, ensuring precise control over the entire process. Working in conjunction with the built-in data analysis algorithm, it provides comprehensive support for overall system optimization.

13. The AI-powered intelligent assessment and training collaboration system for children's sensory integration training according to claim 7, characterized in that, The safety protection equipment and emergency support equipment are configured differently according to the hardware deployment specifications, system module operation requirements and scenario usage characteristics of each functional area (evaluation area, training area, main control area and auxiliary functional areas added as needed). The configuration method is completely matched with the setting logic, core function positioning and corresponding system module deployment adaptation logic of each functional area, and is consistent with the module adaptation requirements of each functional area as described in claim 7. The safety protection and emergency support equipment in the assessment area are adapted to the assessment-related modules deployed in the assessment area as described in claim 7, and meet the safety requirements of such modules throughout the entire process of sensory integration assessment, data collection, assessment calculation, and training program generation. The safety protection and emergency support equipment in the training area are adapted to the training-related modules deployed in the training area as described in claim 7, and meet the safety requirements of such modules throughout the entire process of training execution, dynamic control, action guidance, and data feedback. The safety protection and emergency support equipment in the central control area are adapted to the data storage and system management-related modules deployed in the central control area as described in claim 7, and meet the equipment safety and data security requirements of such modules during data management, module collaborative scheduling, and overall system operation, and coordinate and operate synchronously with the system module hardware devices deployed in each zone. The auxiliary functional areas added as needed are equipped with safety protection equipment and emergency support equipment as required, adapting to the usage needs of the child-friendly auxiliary facilities deployed in the area; the safety protection equipment includes, but is not limited to, various protective components and safe power supply devices, and the emergency support equipment includes, but is not limited to, emergency call and emergency response related equipment, forming an adaptation closed loop with the core functions, module deployment and hardware configuration of each functional area, jointly ensuring the safety and continuity of the child's assessment and training process, as well as the stability of the operation of each module of the system, meeting the core requirements of scene safety and stable system operation as described in claim 7.

14. The AI-powered intelligent assessment and training collaboration system for children's sensory integration training according to claim 8, characterized in that, The motion capture device is adapted to different training scenarios. Specifically, the inertial sensor is adapted to dynamic training scenarios, collecting real-time data on children's limb movement postures; the pressure sensor is adapted to contact training scenarios, collecting data on contact pressure during training; the laser device is adapted to unobstructed and long-distance scenarios, enabling accurate collection of children's movement data; the visual acquisition device can accurately capture subtle limb displacements and movement angles, meeting the needs for capturing subtle movements and further enhancing the accuracy of limb movement acquisition in conjunction with the motion capture device. The coordinated adaptation of all devices further ensures comprehensive and high-precision acquisition of children's limb movements and physiological characteristics, providing accurate data support for subsequent evaluation and control.

15. The AI-powered intelligent assessment and training collaboration system for children's sensory integration training according to claim 9, characterized in that, The feature fusion intelligent analysis model is based on a deep learning algorithm architecture with an attention mechanism, and further adopts a multimodal feature fusion neural network structure. It achieves accurate alignment and deep fusion of multi-dimensional features through cross-modal interaction. The samples of the feature fusion intelligent analysis model are accurately divided into critical periods of sensory integration development according to age groups and correspond to exclusive sensory integration ability grading standards. Combined with the individual developmental differences of children, the data is dynamically adapted to ensure the representativeness and relevance of the samples. The adaptive data augmentation processing of the feature fusion intelligent analysis model adopts a multi-strategy fusion method. Through the collaborative optimization of geometric transformation and pixel adjustment, the model avoids overfitting. Combined with sample difficulty grading enhancement, the model's assessment accuracy and adaptability to complex scenarios are improved. The assessment credibility optimization module uses a learnable attention-weighted fusion method. Combined with the results of the feature fusion intelligent analysis model, it further incorporates the behavioral response features of children during training (collaboratively collected by the perception acquisition module) for adaptive dynamic fusion. Through backpropagation of the loss function, the feature weights are adjusted in real time to achieve dynamic optimization of assessment credibility.

16. The AI-powered intelligent assessment and training collaboration system for children's sensory integration training according to claim 10, characterized in that, The standardized training equipment in the professional knowledge base includes, but is not limited to, intelligent balance beams, intelligent trampolines, intelligent tactile mats, intelligent rotating discs, intelligent swings, and intelligent large exercise balls, which are common intelligent equipment for children's sensory integration training in this field. The standardized training equipment corresponds one-to-one with the standardized training movements. The AI ​​intelligent solution matching algorithm adapts its response time to actual usage needs. It can divide the training intensity into multiple gradients suitable for children with different sensory integration ability levels based on the professional knowledge base. It can also dynamically optimize personalized training programs by calling the advancement rules in the professional knowledge base in real time based on children's training feedback, adapting to the dynamic adjustment of single-movement training parameters and advancement nodes, further ensuring the pertinence and rationality of the training.

17. The AI-powered intelligent assessment and training collaboration system for children's sensory integration training according to claim 11, characterized in that, When the dynamic control submodule of the training execution and control module performs real-time training control, its control logic can be linked with the training load early warning unit to ensure the synergy between early warning and control, which meets the needs of collaborative optimization of training and evaluation. The multimodal prompting method used in the action error correction unit of the training execution and control module allows the error correction record and the warning event of the training load warning unit to be transmitted synchronously and stored together in the training result recording unit during the error correction operation. The training load warning unit of the training execution and control module adopts a tiered warning mechanism, which is specifically divided into multiple warning levels according to the child's physiological load threshold. Each warning level corresponds to a different physiological load threshold and a different emergency handling strategy. The training result recording unit of the training execution and control module records relevant result data such as children's movement data, physiological load data, and error correction records in real time throughout the entire training process. Simultaneously, it records the warning events and corresponding handling of the training load warning unit, ensuring the integrity of the recorded data and providing accurate support for subsequent review and optimization of training programs.

18. The AI-powered intelligent assessment and training collaboration system for children's sensory integration training according to claim 12, characterized in that, The built-in data analysis algorithm of the data storage analysis and overall control module can perform multi-dimensional classification and mining of data throughout the entire system operation process, accurately identify evaluation data deviations, training data anomalies, and module operation data fluctuations, and output evaluation accuracy optimization suggestions and training scheme adjustment suggestions that clearly correspond to specific optimization directions and quantitative reference values. At the same time, combined with historical data trend analysis, it provides a data basis that can be implemented for long-term system iteration. The AI ​​system overall control function of the data storage analysis and overall control module can automatically adjust the communication link scheduling strategy and command issuance priority according to the operating load of each module, capture module operation anomalies in real time and push early warning prompts, and achieve rapid handling of anomalies in conjunction with the visual operation platform, further improving the overall system operation stability, response efficiency and control accuracy, forming a collaborative closed loop with the built-in data analysis algorithm, and comprehensively supporting the overall system optimization.

19. The AI-powered intelligent assessment and training collaborative system for children's sensory integration training according to any one of claims 1-18, characterized in that, It also includes a multi-scenario adaptation module for home-school-disabled persons' federation interaction. This module is primarily adapted to the sensory integration rehabilitation assessment scenario for children in rehabilitation training institutions, while also being compatible with home-school collaborative training scenarios and sensory integration rehabilitation scenarios for children in disabled persons' federations, enabling multi-scenario reuse and adapting to the needs of different usage scenarios. The multi-scenario adaptation module for home-school-disabled persons' federation interaction achieves bidirectional data linkage with the quantitative assessment and reassessment module and the data storage, analysis, and overall control module, building a multi-terminal interactive platform adapted to rehabilitation training institution terminals, teacher terminals, parent terminals, and disabled persons' federation terminals, covering different users such as medical staff, teachers, parents, and disabled persons' federation rehabilitation workers. The permissions of each terminal are adapted to the multi-level access permission requirements of the data security protection module, realizing the synchronization of children's sensory integration assessment results and training process recording data. The system supports the sharing of information and services, including diagnosis and assessment by rehabilitation training institutions, efficient communication between home and school, rehabilitation guidance from the Disabled Persons' Federation, and extended training guidance for families. The multi-scenario adaptive home-school-disabled persons' federation interaction module adapts to the encryption, anonymization, and multi-level access permission requirements of the data security protection module, optimizing data display and operation permissions for different usage scenarios to ensure the security of children's privacy data. The synchronous sharing of children's sensory integration assessment results and training process records undergoes anonymization processing by the data security protection module to ensure the security of children's privacy data. The multi-scenario adapted home-school-disabled persons' federation interaction module is an additional functional module of the system, which can be selectively configured and does not affect the closed-loop operation of the core modules such as sensory acquisition, assessment and review, and training control, nor does it interfere with the normal use of standardized intelligent training and assessment scenarios.

Images