Probabilistic Solid Material Identification in Hyperspectral Imagery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional hyperspectral target detection methods face challenges in accurately identifying materials, particularly solid materials, due to overlapping target and background distributions, leading to missed targets and false alarms, and are not well-suited for handling correlated confusers and within-class variability.
Innovation Solution
The system employs probabilistic identification of solid materials (PRISM) using model averaging and whitening transforms to generate a model search space, calculate probabilities, and integrate statistical and physical quantities, enabling accurate estimation of target material presence and visualization of sub-pixel targets by approximating spectral signatures and background suppression.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a decision threshold is applied to separate target scores from background scores, then target detection capability is improved, but false alarms increase due to overlapping distributions
Solution Approach 1:
The patent transforms the detection problem from using a single decision threshold to using multiple parameters including probability scores, model weights, and spectral fit metrics. By changing from a binary threshold parameter to a multi-parameter probabilistic framework, the system can distinguish targets more accurately without increasing false alarms.
Solution Approach 2:
The patent introduces an intermediary probabilistic model that mediates between the raw spectral scores and the final target detection decision. This intermediary layer computes likelihood ratios and model probabilities, allowing the system to account for distribution overlaps and reduce false alarms while maintaining target detection sensitivity.
2Reliability
If conventional target detection methods are used, then processing simplicity is maintained, but reliability decreases due to missed targets and false alarms
Solution Approach 1:
The patent segments the target detection process into distinct modules: spectral signature extraction, model generation, probability computation, and decision making. By dividing the complex detection algorithm into separable functional segments, the system achieves higher reliability through comprehensive analysis while managing complexity through modular architecture.
Solution Approach 2:
The patent implements a dynamic detection framework where model weights and probability thresholds are adjusted based on the specific characteristics of each detection scenario. This dynamic adaptation allows the system to optimize reliability for different target types and environmental conditions, rather than using fixed simple thresholds.
3Measurement precision
If spectral signatures are compared directly to reference libraries, then computational efficiency is improved, but measurement precision decreases due to within-class variability
Solution Approach 1:
The patent performs preliminary actions by pre-processing spectral signatures into standardized formats, pre-computing model probabilities for reference materials, and organizing the reference library into hierarchical structures before actual detection. This preliminary preparation reduces the computational burden during real-time detection, maintaining processing speed while improving identification precision through more sophisticated spectral analysis.
Data Source
AI summary
Systems, methods and computer program products, for identification of materials based on hyperspectral imagery, are disclosed. An example system comprises one or more processors, a memory, a library of spectral signatures, a receiver, a model generator, and a material identifier. The receiver module is configured to receive a first spectral signature corresponding to a region of interest contained in the hyperspectral image. The model generator is configured to create a model search space including one or more model signatures based on the spectral signatures in the library, wherein each of the one or more model signatures approximate the first spectral signature. The material identifier is a material identifier configured to calculate a probability associated with a presence or absence of a material within the first spectral signature, based on the first spectral signature and the model search space and determine the presence or absence of the material in the region of interest based on the probability.