质量引导和最小不连续相结合的相位解缠及其并行化方法研究

Phase unwrapping is a crucial step in the signal processing of interferometric synthetic aperture radar and interferometric synthetic aperture sonar, which affects the accuracy and efficiency of the generated digital elevation model directly. This subject investigates a combined phase unwrapping algorithm and its parallelization method based on quality guided and minimum discontinuity, which helps to solve the problem of phase unwrapping with both high accuracy and efficiency. The main contents studied include: (1) The preprocessing algorithm is investigated. It can get a higher efficiency by parallel computing the phase quality map and optimizing the guiding strategy. (2) The reliability partition method of the initial unwrapped result is investigated. The residual balance method, thresholding and morphological operation are adopted to partition the initial unwrapped result. (3) The minimum discontinuity algorithm performed on the initial unwrapped result is investigated. The design method of non-recursive algorithm is adopted, and the result of the reliability partition is introduced to restrict the search area in the low quality areas. It can accelerate the convergence of the minimum discontinuity algorithm. (4) The parallelization method of phase unwrapping algorithm is investigated, including the whole phase unwrapping parallelization and the blocked phase unwrapping parallelization. The whole phase unwrapping parallelization can be implemented by removing the improving loops in the low quality areas simultaneously. The blocked phase unwrapping parallelization can be implemented by unwrapping those blocked wrapped phase firstly, then merging those unwrapped blocks by minimum discontinuity algorithm. The balanced task distribution strategy is carried out according to the relationship between the optimization time and the size of optimization areas. It can improve the efficiency of the parallel phase unwrapping methods.

相位解缠是干涉合成孔径雷达和干涉合成孔径声纳信号处理中极其关键的步骤，直接影响所生成的数字高程模型精度和速度。本课题通过对质量引导和最小不连续相结合的相位解缠及其并行化进行研究，实现高精度、高效率相位解缠。主要研究内容包括：(1)研究预处理算法。通过并行计算相位质量图和优化引导策略，实现快速预处理；(2)研究初始解缠相位可信度区分方法。通过残差点平衡、阈值法和形态学方法实现初始解缠相位可信度区分；(3)研究初始解缠结果上的最小不连续算法。通过非递归优化算法设计，并引入可信度区分结果限制搜索区域，加速算法收敛；(4)研究相位解缠算法并行化方法，包括整体并行解缠和分块并行解缠。通过低质量区域并行增长圈消除实现整体并行解缠。通过先分块解缠，最后采用最小不连续思想融合分块解缠结果的方式实现分块并行解缠。两种方法都根据优化时间与优化区域大小的关系进行均衡任务分配，提升解缠效率。

相位解缠是干涉合成孔径雷达和干涉合成孔径声纳信号处理中极其关键的步骤，直接影响所生成的数字高程模型精度和速度。本课题针对质量引导相位解缠算法与最小不连续相位解缠算法相结合的具体方法进行了深入研究，实现了高精度、高效率相位解缠。在质量引导相位解缠算法方面，设计了一种高质量相位质量图，并实现了共享内存和异构环境下的两种高效率相位解缠方法。在最小不连续相位解缠算法方面，设计了非递归最小不连续相位解缠算法，并将其与质量引导相位解缠算法相结合极大提高了解缠效率。针对大块干涉相位图解缠问题，提出了共享内存环境下的整块干涉相位图解缠和分块相位解缠两种方法，在保持相位解缠结果精度的同时，极大提升了相位解缠效率。