Infrared Focal Plane Array Radiation Tolerance via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Infrared Focal Plane Arrays (IRFPAs) used in high radiation environments suffer from permanent damage due to cosmic rays, leading to high costs and limited lifetimes, as existing mechanisms like redundant pixels and detector resets are inefficient and costly.
Innovation Solution
An imaging system that utilizes a subset of active pixels, with machine learning to identify and compensate for radiation damage by dynamically replacing damaged pixels and using inpainting methods to reconstruct images, reducing the need for redundant pixels and extending the detector's lifetime.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If redundant pixels and reference pixels are added to mitigate radiation damage, then radiation tolerance is improved, but device complexity and cost increase
Solution Approach 1:
The patent replaces the mechanical/hardware approach of adding redundant pixels with a computational/software approach using machine learning algorithms. The system uses inpainting algorithms to computationally reconstruct damaged pixel data from remaining functional pixels, eliminating the need for physical redundant pixels while maintaining radiation tolerance.
Solution Approach 2:
The patent creates computational copies of pixel data through machine learning reconstruction. Instead of physical duplicate pixels, the system generates virtual pixel values by learning from patterns in the remaining functional pixels and reconstructing the missing or damaged pixel information algorithmically.
2Reliability
If addressable pixels are added to identify and remove radiation damage, then radiation damage mitigation is improved, but device complexity and cost increase
Solution Approach 1:
The patent replaces the hardware-based addressable pixel identification system with a software-based machine learning identification system. The algorithm automatically identifies radiation-damaged pixels by analyzing signal patterns and statistical properties, eliminating the need for additional addressable control circuitry.
3Reliability
If machine learning and computational correction are implemented, then radiation tolerance is improved, but processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary computational work by pre-training machine learning models on complete, undamaged images before actual imaging. During radiation exposure, the pre-trained model can quickly reconstruct damaged images without requiring extensive real-time computation, thus reducing processing time during critical operations.
4Object-affected harmful factors
If a subset of pixels is used for imaging, then radiation damage effect is reduced, but image quality may deteriorate
Solution Approach 1:
The patent replaces the physical complete pixel array with a computationally reconstructed image from a subset of pixels. Machine learning algorithms infer the values of missing pixels by learning spatial correlations and patterns from the available pixels, maintaining image quality without requiring all physical pixels to be functional.
Data Source
AI summary
An imaging system includes a focal plane array, readout electronics, and a computing system in which the number of active pixels is either set at a low-fraction of the total pixels thereby reducing the effect of radiation damage, or radiation damage over time is detected and automatically compensated. Machine learning is used to identify radiation damaged pixels and damaged regions which are subsequently eliminated and replaced by the computational system. The machine learning is used to identify changes in the fixed pattern signal/noise and/or noise of the system, and is then computationally corrected.