Hyperspectral Image Striping Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional hyperspectral imaging systems face challenges with gain and offset errors in focal plane pixels, leading to measurement biases and false alarms, which manifest as stripes in target and anomaly detection, limiting mission performance.
Innovation Solution
A method for on-platform, non-uniformity correction of pixels using an orthogonal subspace approach to estimate and remove striping noise from hyperspectral images, preserving the useful signal and avoiding artifacts, as demonstrated by the 'Subspace-Based Striping Noise Reduction' algorithm.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional hyperspectral imaging systems are used without correction, then the system structure remains simple, but gain and offset errors cause measurement biases and false alarms that manifest as stripes in target and anomaly detection
Solution Approach 1:
The patent extracts and removes the striping noise component from the hyperspectral image data. By decomposing the image into a low-rank component (representing the true scene) and a sparse component (representing the striping noise), the method isolates and eliminates the harmful striping artifacts caused by pixel gain and offset errors, thereby improving detection accuracy without requiring complex hardware modifications
Solution Approach 2:
The patent changes the parameter representation of the image data by transforming it into a different domain (e.g., wavelet domain or frequency domain) where the striping noise can be identified and separated from the useful signal. This parameter transformation enables the differentiation between the low-rank scene content and the sparse noise patterns, allowing for effective noise removal
2Measurement precision
If on-platform non-uniformity correction is implemented, then measurement precision improves, but computational complexity increases
Solution Approach 1:
The patent creates a computational model (a simplified representation) of the striping noise pattern based on the observed data. By constructing a sparse model that captures the essential characteristics of the noise without requiring full characterization of each pixel's error, the method achieves accurate correction with reduced computational burden compared to pixel-by-pixel calibration approaches
Solution Approach 2:
Instead of performing complete calibration on all pixels, the patent applies partial correction by focusing computational resources only on identifying and removing the dominant striping patterns. This selective approach achieves sufficient measurement precision for target detection without the excessive computational cost of full pixel-wise non-uniformity correction
3Reliability
If striping noise is not removed, then data processing is simpler, but false alarms increase and target detection reliability decreases
Solution Approach 1:
The patent converts the harmful striping noise into a useful diagnostic tool by first identifying and characterizing the noise patterns. The same sparse modeling technique used to remove the noise also provides insight into the pixel calibration errors, allowing the system to both eliminate the harmful effects of striping and potentially use the noise characterization for quality assessment or further correction refinements
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In accordance with various aspects of the disclosure, a system, a method, and computer readable medium having instructions for processing images is disclosed. For example, the method includes receiving an input datacube from which an input image is derived. The input datacube is transformed into a residual datacube by projecting out basis vectors from each spatial pixel in the input datacube, the residual datacube being used to derive a residual image. A statistical parameter value for samples of each focal plane pixel in the residual image is determined. Anomalous focal plane pixels are identified based upon a comparison of the determined statistical parameter value with the respective determined statistical parameter values of remaining focal plane pixels. Another comparison of residual values for each scanned sample of the identified anomalous focal plane pixels with values of corresponding scanned samples in the input datacube is performed.