Multi-Task Deep Learning for Verified Foot Reflex Zone Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for identifying foot reflex zones in traditional Chinese medicine are subjective and lack accuracy, impacting the effectiveness of treatments due to unverified identification results.
Innovation Solution
A foot reflex zones identification method using a multi-task deep learning network model to verify the accuracy of identified reflex zones by ensuring foot organ reflex zones are sub-regions of foot part reflex zones, employing semantic segmentation and key point detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional practitioner experience methods are used to identify foot reflex zones, then the method is simple to operate, but the identification accuracy is low and subjective
Solution Approach 1:
The patent segments the reflex zone identification task into two distinct outputs from the deep learning model: foot part reflex zones (coarse-grained regions like head, neck, shoulder) and foot organ reflex zones (fine-grained specific organ locations). This segmentation allows the system to first identify broader regions and then pinpoint specific organ zones within them, improving overall identification accuracy while maintaining manageable system complexity through hierarchical processing
Solution Approach 2:
The patent implements a nested structure where foot organ reflex zones are verified to be sub-regions within foot part reflex zones. The validation mechanism checks whether identified organ-level zones are properly contained within their corresponding part-level zones, creating a nested verification system that ensures hierarchical consistency and improves measurement precision without proportionally increasing system complexity
2Measurement precision
If image recognition algorithms like YOLOv8 are used to identify foot reflex zones, then identification efficiency is improved, but the accuracy of identified reflex zones cannot be verified
Solution Approach 1:
The patent performs preliminary identification of foot part reflex zones first, then uses these identified regions as constraints for subsequent organ reflex zone identification. By pre-establishing the broader anatomical regions and their spatial relationships, the system narrows down the search space for organ-level identification, improving accuracy while avoiding the need for exhaustive validation of all possible organ locations
Solution Approach 2:
The patent implements a feedback mechanism where the identified foot part reflex zones are used to validate the foot organ reflex zones. The system checks whether organ-level identifications fall within the previously identified part-level regions, creating a self-validating loop that confirms identification accuracy without requiring external validation, thus improving precision without significant time loss
Data Source
AI summary
This invention provides a foot reflex zones identification method, device, rehabilitation robot, and storage medium. The method comprises the following steps: inputting a foot image into a multi-task deep learning network model, and obtaining foot part reflex zones and foot organ reflex zones; the foot part reflex zones include a head reflex zone, a neck reflex zone, a shoulder-back reflex zone, a chest reflex zone, an abdominal reflex zone, and a genital reflex zone; judging whether the foot organ reflex zones and the foot part reflex zones satisfy a theoretical position relationship, if the relationship is satisfied, determining the foot organ reflex zones as target foot organ reflex zones; wherein, the theoretical position relationship is that the foot organ reflex zones are sub-regions of each of the foot part reflex zones. This invention improves the accuracy of the identified target foot organ reflex zones.