Method and Apparatus for Automatic Pattern Recognition

Simple SummaryContent extracted from patent full text and abstract with AI.

This invention provides a method and apparatus for automatically recognizing patterns in sequences of electronic data using data processing systems. The core innovation is a hybrid approach that converts training data sequences (of varying lengths) into feature vectors of fixed length using a dynamic time warping (DTW) technique, preserving essential temporal information. Such encoded data can then be analyzed by classical vector-based classifiers instead of more complex or restrictive models like Hidden Markov Models (HMMs). The system can efficiently identify whether incoming data matches known patterns based on a calculated similarity measure.

Use CasesContent extracted from patent full text and abstract with AI.

• Knock detection in combustion engines (identifying abnormal combustion events based on sensor data).
• Speech recognition systems for identifying spoken commands.
• Medical diagnostics such as analyzing EKG signals for abnormal heart rhythms.
• Virus scanners that identify malicious code patterns despite variations in the code structure.
• Gene sequence analysis in bioinformatics for detecting genes or variations.
• Handwriting and signature recognition for user authentication (e.g., signing on a touchscreen).
• Machine and process monitoring for fault detection in industrial automation.
• Quality control using thermal imaging data to detect anomalies in manufactured parts.

BenefitsContent extracted from patent full text and abstract with AI.

• Allows more reliable and simpler pattern recognition by combining DTW-based sequence normalization with vector-based classifiers, avoiding the complexity of stochastic models like HMMs.
• Accommodates variations in input data lengths, making cross-comparisons between sequences possible.
• Reduces the amount of information lost during feature extraction by retaining temporal distortion details.
• Enables continuous, real-time analysis of streaming data with immediate detection capabilities.
• Offers wide applicability across various domains including automotive, medical, manufacturing, software security, and biometrics.
• Easier parameterization and model training compared to neural networks or HMMs, leading to faster deployment and more consistent results.

Technical Classifications (CPCs)

Inventors & Applicants

Patent Abstract

The invention relates to a method for automatic pattern recognition in a sequence of electronic data by means of electronic data processing in a data processing system, in which in an analysis the sequence of electronic data is compared with parametrized model data which represent at least one pattern sequence, and in which the at least one pattern sequence is recognized if it is ascertained during the analysis that model data, which the parametrized model data comprise and which are associated with the at least one pattern sequence, occur with a similarity measure which exceeds a similarity measure threshold, wherein during the formation of the parametrized model data, training data are processed by means of a dynamic time warping method to form a set of feature vectors of the same length and with the same information content as the training data from which the parametrized model data are derived. Furthermore, the invention relates to an apparatus for automatic pattern recognition in a sequence of electronic data by means of electronic data processing with a data processing system.