Spectroscopic Raw Material Identification With Hierarchical SVM

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing chemometric classification techniques for raw material identification (RMID) suffer from poor transferability and insufficient granularity, particularly in large-scale applications.

Innovation Solution

A hierarchical support vector machine (SVM) classifier is employed to generate a global classification model using a training set, followed by a local classification model for improved accuracy in raw material identification, utilizing a multi-stage classification technique.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If a global classification model is used for raw material identification, then the coverage of material classes is comprehensive, but the classification accuracy is insufficient

Engineering Contradiction:
Improvecoverage of material classesVSAvoidclassification accuracy
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent divides the classification process into two segments: a global classification model that handles broad material categories and a local classification model that provides detailed specific identification. This segmentation allows the system to first filter samples into major classes using the global model, then apply the more specialized local model only to relevant subsets, thereby improving overall accuracy while maintaining comprehensive coverage.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces a hierarchical dimension to the classification system, organizing classes into multiple levels (global categories and local subcategories). This dimensional structure transforms the flat classification approach into a tree-like hierarchy, enabling the system to leverage both broad pattern recognition and fine-grained discrimination capabilities.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Device complexity

If traditional chemometric classification techniques are used, then the implementation is simple, but the transferability across different spectrometers is poor

Engineering Contradiction:
Improveimplementation simplicityVSAvoidtransferability across spectrometers
Core Design Contradiction:
Device complexityVSReliability

Solution Approach 1:

The patent develops classification models that are designed to be universal across different spectrometer types and measurement conditions. By training the global and local classification models on diverse datasets from multiple spectrometers and environmental conditions, the system achieves broad applicability and robust transferability while maintaining a relatively simple hierarchical implementation structure.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Loss of time

If a single-stage classification is performed, then the processing time is short, but the classification accuracy is insufficient

Engineering Contradiction:
Improveprocessing timeVSAvoidclassification accuracy
Core Design Contradiction:
Loss of timeVSMeasurement precision

Solution Approach 1:

The patent implements preliminary classification using the global model to quickly categorize samples into major material classes before applying the more computationally intensive local classification model. This preliminary action filters the search space, allowing the system to achieve high accuracy without processing all samples through the complete hierarchical pipeline, thereby balancing speed and precision.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3822977B1Identification using spectroscopy
Publication Date: 2025.11.12 VIAVI SOLUTIONS INC(US)
  • EP3822977B1 patent drawingFigure 1A
  • EP3822977B1 patent drawingFigure 1B
  • EP3822977B1 patent drawingFigure 2

AI summary

A device may receive information identifying results of a spectroscopic measurement of an unknown sample. The device may perform a first classification of the unknown sample based on the results of the spectroscopic measurement and a global classification model. The device may generate a local classification model based on the first classification. The device may perform a second classification of the unknown sample based on the results of the spectroscopic measurement and the local classification model. The device may provide information identifying a class associated with the unknown sample based on performing the second classification.