Hyperspectral Target Detection via Mean Spectrum Subtraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current hyperspectral image processing techniques face challenges in detecting sparse and weak targets due to high levels of false alarms and computational complexity, particularly when dealing with dispersed targets like gaseous effluents, which are complicated by atmospheric effects and background clutter.
Innovation Solution
The approach involves pre-adjusting hyperspectral images by subtracting a mean spectrum from each pixel to reduce artifacts and combining the resulting data with matched filter scores and partial un-mixing coefficients in a simplified manner to enhance detection resolution and reliability, using techniques such as spectral matching and adaptive coherence/cosine estimators.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional detection techniques are used on hyperspectral images, then detection capability is maintained, but false alarms increase and reliability decreases for sparse and weak targets
Solution Approach 1:
The patent segments the hyperspectral image into multiple subsets and processes each subset separately through matched filtering. This segmentation approach reduces the impact of background clutter and atmospheric effects on any single processing unit, thereby reducing false alarms while maintaining detection reliability for sparse and weak targets
Solution Approach 2:
The patent applies preliminary atmospheric correction and background subtraction before matched filter processing. By removing atmospheric effects and background clutter in advance, the detection algorithm operates on pre-processed data with reduced interference, significantly lowering false alarm rates while preserving target detection capability
2Measurement precision
If high resolution detection is applied to sparse targets, then detection precision improves, but computational complexity increases
Solution Approach 1:
The patent divides the hyperspectral image into multiple smaller subsets for parallel processing. Each subset is processed through matched filtering independently, which reduces the computational burden per processing unit while maintaining high detection precision through the aggregation of results from all subsets
Solution Approach 2:
The patent applies matched filtering selectively to identified subsets rather than processing the entire hyperspectral image uniformly. This partial action approach focuses computational resources on regions containing potential targets, achieving high detection precision without the excessive computational complexity of processing all pixels at maximum resolution
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This method significantly reduces false alarms and improves the detection of sparse and weak targets by enhancing the signal-to-noise ratio, allowing for more accurate identification of target substances within hyperspectral images, even in complex backgrounds.
Implementation Method 1
As different materials reflect wavelengths of visible and invisible light selectively, analysis of the contiguous wavelength spectrum permits finer resolution and greater perception of information contained in the image
Implementation Method 2
The spectral reflectance curves of organic materials, such as healthy green plants also have a characteristic shape that is dictated by various plant attributes, such as the absorption effects from chlorophyll and other leaf pigments
Data Source
AI summary
Provided is a process and system for detection of sparse or otherwise weak targets in a hyperspectral image. A hyperspectral image is received having a multitude of pixels, with each pixel having a respective spectrum. In some embodiments, multiple mean spectra are selectively determined for respective sub-regions of the hyperspectral image. The subset mean spectra can be selectively removed from respective pixels, thereby improving image fidelity due to sensor artifacts. Additionally, target detection of such an adjusted image can be determined by one or more of matched filter techniques or by partial un-mixing. In at least some embodiments target detection is enhanced by combining a measure of target match with a residual spectrum determined as a measure of un-match. Target detection can be further improved by application of rules, for example, related to target detection threshold.