Parallel Markov Chains for SAR Image Boundary Delineation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for delineating targets in synthetic aperture radar images, such as using simulated annealing, are not rapid enough for real-time applications due to the presence of noise and complex target shapes, leading to difficulties in accurately identifying boundaries between targets and background clutter.
Innovation Solution
The use of parallel Markov chains with a Metropolis algorithm and adaptive cooling schedule, where multiple chains start from different states, exchange information, and adjust their temperature to optimize the fitness value of image contours, allowing for faster and more accurate target delineation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If simulated annealing is used to optimise contour placement for target delineation, then robustness to local maxima is improved, but processing speed deteriorates
Solution Approach 1:
The patent divides the single simulated annealing process into multiple parallel Markov chains that operate simultaneously. Each chain independently explores the solution space, and the best solution among all chains is selected. This segmentation allows the system to maintain robustness against local maxima while significantly reducing processing time through parallel computation.
Solution Approach 2:
The patent transitions from a sequential single-chain approach to a parallel multi-chain approach, adding the dimension of concurrency. Multiple Markov chains execute simultaneously across different processors or threads, transforming the problem from a time-sequential optimization into a spatially-parallel optimization that achieves both reliability and speed.
2Measurement precision
If conventional simulated annealing is used for target delineation, then accuracy in noisy conditions is improved, but real-time processing capability deteriorates
Solution Approach 1:
The patent segments the optimization task into multiple parallel Markov chains that simultaneously search for optimal contour placements. This segmentation enables the system to maintain high delineation accuracy through multiple independent searches while reducing processing time by executing these searches in parallel rather than sequentially.
Solution Approach 2:
The patent implements continuous parallel optimization where multiple Markov chains continuously explore the solution space simultaneously. This continuous parallel action ensures that accurate delineation is achieved through ongoing optimization while minimizing processing time by eliminating sequential waiting periods.
3Productivity
If multiple parallel Markov chains are used for simulated annealing, then processing speed is improved, but system complexity increases
Solution Approach 1:
The patent segments the optimization problem into multiple independent but parallel Markov chains. Each chain maintains its own state and exploration process, simplifying the implementation by allowing independent optimization of each chain while achieving overall speedup through parallel execution. The segmentation isolates complexity into manageable, identical units.
Solution Approach 2:
The patent changes the parameter of chain quantity from one to multiple, transforming the system's productivity. By adjusting this parameter and implementing parallel execution, the system achieves faster processing while managing complexity through standardized, repeatable chain structures that can be easily replicated and coordinated.
Data Source
Figure 1~4
AI summary
An image processing method using an algorithm which incorporates simulated annealing by parallel Markov chains, the calculation of fitness values of states of the Markov chains which have substantially the same simulated annealing temperature, the calculation of the standard deviation of these fitness values, and the use of this standard deviation in setting the simulated annealing cooling schedule. The method may be used to delineate an object of interest in an image against a background by estimating the boundary of the object and optimising the fit of the region within this boundary to the region occupied by the object.