Hyperspectral Covariance Estimation via Iterative Pixel Decontamination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional hyperspectral imaging systems face significant computational challenges in estimating background spectral content and detecting anomalies due to the high number of operations required for computing spectral covariance matrices, which can lead to performance degradation and inefficiencies.
Innovation Solution
The method involves adaptive, iterative sampling of a small fraction of scene pixels, computing covariance matrices for each pixel set, identifying and removing contaminating pixels, and iteratively increasing the number of sampled pixels until consistency is achieved, thereby reducing the computational burden and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all scene pixels are used to compute the spectral covariance matrix, then the accuracy of background spectral content estimation is improved, but the computational complexity and processing time increase significantly
Solution Approach 1:
The patent segments the scene pixels into multiple subsets and processes each subset independently to compute partial covariance matrices. This segmentation allows parallel processing and reduces the computational burden of handling all pixels simultaneously, while still achieving accurate background spectral content estimation through aggregation of subset results.
Solution Approach 2:
The patent employs partial action by using a small fraction (e.g., 1-10%) of scene pixels to estimate the spectral covariance matrix. Through iterative refinement and contamination removal, this partial sampling achieves sufficient accuracy without requiring processing of all pixels, thereby significantly reducing computational complexity while maintaining measurement precision.
2Loss of information
If all scene pixels are used for covariance computation, then the representation of all eigenvectors is improved, but the processing time and computational operations increase
Solution Approach 1:
The patent performs preliminary action by first identifying and removing contaminating pixels before computing the spectral covariance matrix. This preliminary decontamination step ensures that the subsequent covariance computation using a reduced pixel set accurately represents the background spectral content and its eigenvectors, preventing information loss while reducing processing time.
Solution Approach 2:
The patent uses partial action by computing the spectral covariance matrix from a selected subset of non-contaminating pixels rather than all scene pixels. This approach maintains sufficient representation of eigenvectors for accurate anomaly detection while significantly reducing the computational operations and processing time required.
3Measurement precision
If outlier removal is performed to eliminate contaminating pixels, then the anomaly detection accuracy is improved, but the computational operations increase due to additional processing steps
Solution Approach 1:
The patent applies the taking out principle by extracting and removing contaminating pixels from the scene pixel set before covariance computation. By identifying outliers based on initial covariance estimates and removing them, the method improves anomaly detection accuracy while the computational cost is offset by the subsequent reduction in pixels requiring full covariance processing.
Solution Approach 2:
The patent performs outlier removal on only a small fraction of pixels (e.g., 1-10%) rather than all scene pixels. This partial decontamination approach achieves sufficient improvement in anomaly detection accuracy while limiting the increase in computational operations to a manageable level.
4Productivity
If a small fraction of pixels is sampled for covariance computation, then the computational operations are reduced, but the covariance estimates become wildly varying and inaccurate
Solution Approach 1:
The patent performs preliminary action by first computing an initial spectral covariance matrix from all or most pixels, then using this initial estimate to identify contaminating pixels. This preliminary step provides a stable reference for outlier detection, enabling subsequent covariance computation from a smaller pixel fraction to produce accurate and consistent estimates rather than wildly varying results.
Solution Approach 2:
The patent employs feedback by iteratively refining the covariance estimate: computing initial covariance, identifying outliers based on this estimate, removing outliers, and recomputing covariance from the cleaned subset. This feedback loop ensures that the final covariance matrix from a small pixel sample is accurate and stable, resolving the issue of wildly varying estimates.
Data Source
AI summary
Methods and systems are provided for estimating background spectral content in a hyperspectral imaging (HSI) scene. A HSI processor computes a scene covariance matrix for each of a plurality of sparsely sampled pixel sets, identifies and removes the spectral content of contaminating pixels from the covariance matrices, and checks the consistency among the plurality of decontaminated covariance matrices, iteratively re-sampling and re-computing said matrices until an acceptable consistency is achieved, and then computes a final decontaminated covariance matrix representative of the background spectral content of the scene. Alternate approaches to pixel sampling, and/or using fewer spectral dimensions than are available for the pixels are presented.