Endmember Spectrum Database Construction for Hyperspectral Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Hyperspectral images obtained from the sky have low spatial resolution, making it difficult to accurately identify substances as pixel spectra often represent a mixture of multiple substances, and existing methods struggle to effectively construct an endmember spectrum database for objects comprising multiple substances.
Innovation Solution
A method for constructing an endmember spectrum database that involves obtaining input endmember spectra from hyperspectral images, calculating similarities with spectra in a library and existing database, and registering unique spectra below a predetermined similarity threshold, allowing for accurate identification and registration of new endmember spectra.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If hyperspectral imaging is used to obtain images from the sky, then large-area coverage is achieved, but spatial resolution becomes low making substance identification difficult
Solution Approach 1:
The patent segments the mixed spectrum of each pixel into multiple endmember spectra representing different substances. By decomposing the composite spectrum into constituent spectral components, the system can identify individual substances even when spatial resolution is low and multiple substances are mixed within a single pixel.
Solution Approach 2:
The patent transitions from analyzing data in the spatial domain to analyzing it in the spectral domain. By examining the spectral characteristics across multiple wavelength bands, the system can distinguish between different substances based on their unique spectral signatures, compensating for the loss of spatial detail.
2Measurement precision
If endmember spectra are extracted from mixed pixel spectra, then substance identification becomes possible, but the complexity of spectrum processing increases
Solution Approach 1:
The patent performs preliminary actions by pre-storing reference spectra of known substances in a database. Before analyzing new hyperspectral images, the system has ready-made spectral templates to compare against, which simplifies the identification process and reduces computational complexity during actual analysis.
Solution Approach 2:
The system uses feedback mechanisms by comparing extracted endmember spectra against a reference database and iteratively refining the identification process. The comparison results feed back into the analysis, allowing the system to adjust and improve substance identification accuracy through repeated comparison and validation.
3Measurement precision
If a comprehensive endmember spectrum database is constructed, then substance identification accuracy improves, but the time and computational resources required increase
Solution Approach 1:
The patent performs preliminary action by constructing and storing a comprehensive endmember spectrum database in advance, before actual hyperspectral image analysis is needed. This pre-computed reference database contains spectral information for numerous substances, allowing rapid comparison and identification during operational phases without requiring time-consuming real-time computations.
Data Source
AI summary
According to one embodiment, an endmember spectrum extraction unit generates an input endmember spectrum group from a hyperspectral image. A collation unit outputs an endmember spectrum group including endmember spectra that are registered with neither a spectrum library nor an endmember spectrum database as a new endmember spectrum group. An input endmember spectrum distribution image generation unit generates an input endmember distribution image every input endmember spectrum. A new endmember spectrum distribution image selection unit selects an input endmember spectrum distribution image that is included in input endmember spectrum distribution images and that corresponds to new endmember spectrum as a new endmember distribution image. The user recognizes what kind of object corresponds to each new endmember spectrum on the basis of the new endmember spectrum distribution image, the hyperspectral image and a high resolution image, and determines and orders whether to register each new endmember spectrum with the endmember spectrum database.