Real-Time State Change Detection Using MEMD and Logistic Regression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
State-of-the-art machine learning algorithms require extensive training and prediction times, making them unsuitable for real-time predictions, which limits their applications in real-life scenarios, especially in detecting random state changes in physical subjects like eye states or honeybee dances.
Innovation Solution
A method involving multivariate signal processing using high pass filtering, multivariate empirical mode decomposition, and logistic regression classification to detect state changes in real-time, allowing for efficient extraction and prediction of features from intrinsic mode functions, thereby reducing computational complexity and improving processing time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If state-of-the-art machine learning algorithms are used for prediction, then prediction accuracy is improved, but training and prediction time increases excessively
Solution Approach 1:
The patent segments the multivariate signal processing into distinct stages: high pass filtering to remove DC components, multivariate empirical mode decomposition (MEMD) to generate intrinsic mode functions, feature extraction from IMFs, and logistic regression classification. This segmentation allows each component to be optimized independently, achieving real-time performance while maintaining accuracy.
Solution Approach 2:
The patent changes the parameter of the machine learning algorithm from complex state-of-the-art methods to logistic regression, which provides a favorable balance between computational speed and classification accuracy. Additionally, the sampling rate is adjusted and critical time intervals are identified to optimize processing speed without sacrificing prediction accuracy.
2Measurement precision
If complex machine learning algorithms are applied to detect random state changes, then detection accuracy is improved, but real-time processing capability deteriorates
Solution Approach 1:
The patent applies preliminary high pass filtering to remove DC components from the multivariate signals before analysis. This preliminary action simplifies the subsequent detection process and enables real-time processing by eliminating unnecessary computational steps while preserving the ability to detect random state changes accurately.
Solution Approach 2:
The patent extracts only the critical components from the multivariate signals through MEMD decomposition, generating intrinsic mode functions that contain the essential information for state change detection. This extraction approach focuses computational resources on the most relevant signal features, achieving both high detection accuracy and real-time processing speed.
3Measurement precision
If extensive signal processing is performed to improve prediction accuracy, then classification precision is improved, but computational complexity increases
Solution Approach 1:
The patent employs dynamic multivariate empirical mode decomposition to adaptively generate intrinsic mode functions from the input signals. This dynamic approach allows the system to adjust to varying signal characteristics in real-time, maintaining high classification precision while managing computational complexity through efficient adaptive processing rather than static complex algorithms.
Data Source
AI summary
Methods and systems detect a random state change in a subject in real time. Eye state changes may be identified in encephalogram brain signals, or honeybee dance patterns may be classified. Multivariate signals including state change information are received via a plurality of channels. The signals are sampled and may be filtered to remove DC components. Statistical characteristics of the signals are monitored. When the statistical characteristics exceed a threshold during a critical time interval, a potential change of state is detected. The critical time segment of the signals may be filtered to generate respective state change artifact signals. The state change artifact signals are decomposed by MEMD, and intrinsic mode functions are generated. Features are extracted from the intrinsic mode functions. These steps may be repeated while the extracted features are provided to a logistic regression classifier that is used to predict a state of the subject.