Hyperspectral Image Dimension Reduction via Iterative Basis Vector Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current hyperspectral image dimension reduction methods, such as PCA and pixel purity techniques, suffer from limited performance when pure pixels are not found in the scene, lack control over error levels, and are computationally intensive, making them inadequate for detecting spectrally unique objects and managing large data volumes effectively.
Innovation Solution
A method and system for hyperspectral image dimension reduction that establishes a basis vector set by selecting initial pixels and iteratively adding spectral vectors with significant residual magnitudes, allowing for controlled error levels and efficient computation, while enabling lossless or lossy data reduction suitable for real-time processing and transmission.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If traditional dimension reduction methods (PCA, pixel purity) are used, then data volume is reduced, but information fidelity and detection precision deteriorate when pure pixels are not present
Solution Approach 1:
The patent transforms the dimension reduction problem from selecting entire spectral bands to selecting individual spectral coefficients within bands. By changing the parameter of selection from band-level to coefficient-level, the method achieves finer control over information retention, allowing precise preservation of spectrally unique objects while reducing overall data volume.
Solution Approach 2:
The patent segments each spectral band into multiple coefficients through mathematical transformation. Instead of treating each band as an indivisible unit, the segmentation of bands into coefficients enables selective retention of only those coefficients that contain relevant information about spectrally unique objects, thereby improving detection precision while reducing data volume.
2Measurement precision
If more spectral bands are captured to improve detection precision, then data volume increases, making transmission and processing slower
Solution Approach 1:
The patent extracts only the essential coefficients from the full spectral data that are necessary for detecting spectrally unique objects. By taking out and retaining only these critical coefficients while discarding redundant information, the method maintains high detection precision with a significantly reduced data set, thereby improving processing speed without sacrificing accuracy.
3Speed
If data is transmitted in real-time for disaster response or combat zones, then transmission speed must increase, but data volume is too large for limited bandwidth
Solution Approach 1:
The patent applies partial action by transmitting only a subset of the most important spectral coefficients rather than the complete hyperspectral data set. This partial transmission approach provides sufficient information for real-time decision-making in disaster response and combat zones, achieving acceptable detection precision with dramatically reduced data volume that can be transmitted over limited bandwidth.
Data Source
AI summary
Provided is a method of hyperspectral image dimension reduction. The method includes receiving a hyperspectral image having a plurality of pixels. A set of basis vectors is established at least in part with respect to the spectral vectors of the initial hyperspectral image. For each pixel of the hyperspectral image, the spectral vector is read and decomposed, i.e. unmixed, with the basis vector set to provide at least a reduced dimension vector for each pixel. Collectively the reduced dimension vectors for each pixel represent the dimensionally reduced image. A system operable to perform the method is also provided.