Chunhua Zhang

Real-Time Implementation of Synthetic Aperture Sonar Imaging on High Performance Clusters

A flexible and scalable architecture that implements real-time synthetic aperture sonar (SAS) imaging on high performance clusters (HPCs) is introduced in this paper. Combing HPCs with application of a hybrid MPI and OpenMP parallel programming model, the lookup table algorithm for SAS image reconstruction based on the time-domain correlation algorithm is realized in parallel. Considering the huge computation load of the lookup table algorithm, the pulse-compressed data is partitioned into strips along the range direction which are distributed to the nodes of HPCs. The scalability allows the lookup table imaging cluster nodes to be added or removed without needing to recompile the existing software, and automatic burden balancing is considered. The result of lake trial verifies that this parallel algorithm on HPCs can provide good quality images and meantime have highly real time rate.

An automatic ellipse and line targets detection method from synthetic aperture sonar images

Detection of ellipse and line targets is important for the analysis of Synthetic Aperture Sonar (SAS) images. An automatic ellipse and line targets detection method from synthetic aperture sonar images is presented. The method mainly has three procedures: preprocessing of SAS images, Zernike Orthogonal Moment Edge Detection Algorithm (ZOMEDA), line and ellipse detection. The guidance is presented firstly on how to perform the preprocessing of SAS images. Then, ZOMEDA is utilized to produce edge points with both the direction and position information. Principles of ZOMEDA with the 7x7 template are analyzed and the coefficients to carry out the ZOMEDA are calculated and listed. The idea of Random Sample Consensus (RANSAC) is applied to the Line and ellipse detection procedure to improve the robustness and the computing efficiency. Detail procedures of RANSAC are analyzed in the article. Calculating of line and ellipse parameters is pivotal to carry out the idea of RANSAC. Principles are analyzed on how to calculate the parameters of the line and ellipse based on the direction and position information. Another important procedure, parameters refinement, is also discussed. At last, the line and ellipse detection method is applied to simulated datasets and lake-trial datasets for validation.