Hyperspectral Data Noise Reduction via Wavelet Transform

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for multi-species mapping using hyperspectral data, such as LIBS spectroscopy, face challenges with low signal-to-noise ratios and significant spectral interferences, making it difficult to accurately identify and map chemical elements due to high noise levels and interference.

Innovation Solution

A method combining discrete Wavelet transform with noise reduction thresholding and principal component analysis, where the Wavelet transform is applied to hyperspectral data to produce parsimonious representations, followed by a filtered covariance matrix analysis to enhance the signal-to-noise ratio and identify spectral signatures.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If laser beam size is reduced to improve lateral resolution, then mapping resolution is improved, but signal-to-noise ratio decreases

Engineering Contradiction:
Improvelateral resolutionVSAvoidsignal-to-noise ratio
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent applies wavelet transform to segment the spectral data into different frequency components (approximation coefficients and detail coefficients). This segmentation allows separate processing of signal and noise components, enabling resolution improvement while managing noise through selective filtering of detail coefficients.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transforms the spectral data from original domain to wavelet coefficient domain, changing the representation parameters. This parameter transformation enables better separation of signal and noise, and the sparsification process进一步优化s the signal representation to improve measurement precision without sacrificing signal-to-noise ratio.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If number of wavelength channels is increased to improve spectral resolution, then species identification accuracy is improved, but data dimensionality and processing complexity increase

Engineering Contradiction:
Improvespectral resolutionVSAvoiddata dimensionality
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts the essential signal information from high-dimensional spectral data through wavelet transform, separating meaningful approximation coefficients from noisy detail coefficients. This extraction process reduces data dimensionality while preserving critical spectral features needed for species identification.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies different processing strategies to different parts of the spectral data: approximation coefficients are retained for signal representation while detail coefficients are filtered or processed differently. This local quality approach optimizes the balance between spectral resolution and data complexity by treating different frequency components differently.

Inventive Principle:
Principle #3Local quality

3Productivity

If spectral data is processed using traditional methods, then processing speed is maintained, but noise reduction and species identification accuracy are insufficient

Engineering Contradiction:
Improveprocessing speedVSAvoidspecies identification accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent performs wavelet transform and sparsification as preliminary processing steps before species identification. This preliminary action prepares the data in an optimized format that facilitates more accurate and efficient subsequent analysis, improving both noise reduction and identification accuracy while maintaining processing speed.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4166931B1Method for multi-species mapping of an area from spectral data
Publication Date: 2024.04.17 COMMISSARIAT A LENERGIE ATOMIQUE ET AUX ENERGIES ALTERNATIVES
  • EP4166931B1 patent drawingFigure 1a
  • EP4166931B1 patent drawingFigure 1b
  • EP4166931B1 patent drawingFigure 2

AI summary

A multi-species mapping method for an area comprising the steps of: - Acquiring (401) a set of spectral data associating for each point of the area a spectrum, - Applying (402) to the spectral data a Wavelet transform so as to produce a set of coefficients comprising characteristic noise detail coefficients and characteristic signal approximation coefficients, - Setting to zero (403) the detail coefficients whose spectral energy density value is less than a predetermined threshold, - Applying (404) an inverse Wavelet transform to the set of coefficients to obtain a new sparse spectral data set - Applying (405) a principal component analysis to the new sparse spectral data set so as to: i.Determine a projection basis L in the spectral domain, each component of the projection basis corresponding to a spectral signature of a species, ii. Project the spectral data into the projection basis L to obtain, for each component, a map of the area indicating, for each point in the area, the proportion of the species.