Pressure-Sensor Activity Classification With Subject-Adaptive Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing activity classification systems struggle to accurately distinguish between similar physical activities and account for individual variations, particularly in transitions between activity types and personal attributes of the subject.
Innovation Solution
A deep learning neural network (NN) system that utilizes pressure data from sensors, combined with other data types, to classify activities through supervised training, including population and subject-specific training, using convolutional neural networks (CNN) and long short-term memory (LSTM) units to model spatial and temporal components, respectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional activity classification systems use predetermined pre-programmed characteristics from sensors, then the system structure remains simple, but the system cannot accurately distinguish between similar physical activities or account for individual variations
Solution Approach 1:
The patent replaces traditional mechanical rule-based classification systems with a neural network-based intelligent system. The neural network learns complex patterns from sensor data automatically, enabling accurate distinction between similar activities and adaptation to individual users without requiring manual programming of classification rules.
Solution Approach 2:
The system dynamically adjusts classification parameters through neural network training. By changing from fixed predetermined thresholds to adaptive learned parameters, the system can accurately classify activities while accounting for individual variations in movement patterns, pressure distributions, and temporal characteristics.
2Adaptability or versatility
If the system uses fixed classification rules, then the system is easy to operate, but it cannot classify transition stages between activity types or distinguish unique occurrences of similar activities
Solution Approach 1:
The neural network performs self-learning and self-adjustment through automated training processes. The system automatically adapts to different activity types and individual users without requiring manual configuration or programming, achieving high adaptability while maintaining ease of operation through plug-and-play functionality.
Solution Approach 2:
The classification system transitions from static fixed rules to dynamic adaptive models. The neural network continuously learns from new data and adjusts its parameters to accommodate transition stages between activities and unique activity occurrences, making the system both versatile and easy to operate.
3Measurement precision
If the neural network is trained with population data only, then training efficiency is high, but classification accuracy for individual subjects is reduced
Solution Approach 1:
The system performs preliminary population-level training to establish baseline classification capabilities quickly. This preliminary training provides a strong foundation that can be rapidly fine-tuned with individual subject data, reducing the total training time while achieving high individual accuracy through a two-stage training approach.
Solution Approach 2:
The training process is segmented into population-level training and individual-level fine-tuning. This segmentation allows the system to efficiently learn general patterns from large population datasets first, then quickly adapt to individual characteristics with smaller subject-specific datasets, optimizing both training efficiency and individual accuracy.
4Productivity
If real-time processing is implemented, then the system provides immediate feedback, but processing complexity and computational requirements increase
Solution Approach 1:
The neural network performs preliminary processing and feature extraction during the training phase, preparing optimized models for rapid inference. This preliminary preparation enables real-time classification with reduced computational complexity during actual operation, as the heavy lifting of pattern recognition has already been accomplished during training.
Data Source
AI summary
A method and system for activity classification. A pressure sensor receives input data resulting from physical activity of a subject performing an activity. The input data includes pressure data from at least one pressure sensor, and may include other data acquired through other types of sensors. A deep learning neural network is applied to the input data for identifying the activity. The neural network is trained with reference to training data from a training database. The training data may include empirical data from a database of previous data of corresponding activities, synthesized data prepared from the empirical data or simulated data. The training data may include data from physical activity of the subject being monitored by the system. Different aspects of the neural network may be trained with reference to the training data, and some aspects may be locked or opened depending on the application and the circumstances.