遥感图像海陆分割的GPU并行加速
发布时间：2018-12-20 11:31
【摘要】：海陆分割是光学遥感图像海上目标识别过程中一个重要步骤,海陆分割的结果直接影响下一步目标识别的效率和正确率;而遥感数据的大量增长,使得图像处理速度变慢,因此分割算法的执行效率变得越来越重要。利用Nvidia开发的统一计算架构CUDA(compute unified device architecture),将海陆分割流程的一系列函数移植到GPU(graphic processing unit)上进行并行处理,能够有效提高算法执行速度。经实验验证最终完成2 000×2 000大小的图片在11 ms内的海陆分割处理。该方法能够满足对于图像数据的传输在25 ms内的处理,适合用于建立地面或者航空搭载的"实时"处理平台。
[Abstract]:Land and sea segmentation is an important step in the process of ocean target recognition in optical remote sensing image. The result of sea and land segmentation directly affects the efficiency and accuracy of target recognition in the next step. Because of the rapid growth of remote sensing data, the speed of image processing becomes slower, so the efficiency of segmentation algorithm becomes more and more important. The unified computing architecture CUDA (compute unified device architecture), developed by Nvidia can transplant a series of functions of sea and land segmentation process to GPU (graphic processing unit) for parallel processing, which can effectively improve the execution speed of the algorithm. The experimental results show that the image with the size of 2 000 脳 2 000 can be segmented within 11 ms. This method can satisfy the processing of image data transmission within 25 ms, and is suitable for the establishment of a "real-time" processing platform on the ground or in aviation.
【作者单位】： 中国科学院长春光学精密机械与物理研究所;中国科学院大学;
【分类号】：TP751
，
本文编号：2387936
[Abstract]:Land and sea segmentation is an important step in the process of ocean target recognition in optical remote sensing image. The result of sea and land segmentation directly affects the efficiency and accuracy of target recognition in the next step. Because of the rapid growth of remote sensing data, the speed of image processing becomes slower, so the efficiency of segmentation algorithm becomes more and more important. The unified computing architecture CUDA (compute unified device architecture), developed by Nvidia can transplant a series of functions of sea and land segmentation process to GPU (graphic processing unit) for parallel processing, which can effectively improve the execution speed of the algorithm. The experimental results show that the image with the size of 2 000 脳 2 000 can be segmented within 11 ms. This method can satisfy the processing of image data transmission within 25 ms, and is suitable for the establishment of a "real-time" processing platform on the ground or in aviation.
【作者单位】： 中国科学院长春光学精密机械与物理研究所;中国科学院大学;
【分类号】：TP751
，
本文编号：2387936
本文链接：https://www.wllwen.com/guanlilunwen/gongchengguanli/2387936.html
