A manual for the multiblock PARTI runtime primitives revision 4
A manual for the multiblock PARTI runtime primitives revision 4
Compiler and runtime support for structured and block structured applications
Proceedings of the 1993 ACM/IEEE conference on Supercomputing
Efficient transposition algorithms for large matrices
Proceedings of the 1993 ACM/IEEE conference on Supercomputing
IEEE Spectrum
Processing large-scale multi-dimensional data in parallel and distributed environments
Parallel Computing - Parallel data-intensive algorithms and applications
Evaluating the DIPORSI Framework: Distributed Processing of Remotely Sensed Imagery
Proceedings of the 8th European PVM/MPI Users' Group Meeting on Recent Advances in Parallel Virtual Machine and Message Passing Interface
| Hi-index | 0.00 |
Remotely sensed imagery of the earth has been used to develop and validate various studies regarding land-cover dynamics. However, the large amounts of imagery collected by satellites are significantly contaminated by the effects of atmospheric particles. The objective of atmospheric correction is to retrieve the real surface reflectance values from remotely sensed imagery by removing the atmospheric effects. Atmospheric correction can significantly improve the accuracy of image classification and other forms of image analysis, but it is very computationally intensive.We introduce a number of computational techniques that lead to substantial speedup of an existing atmospheric correction algorithm that is based on using lookup tables. These techniques are of five main types: reordering and classifying operations as image-based, window-based, and pixel-based; performing interpolations on subcubes for each group; data-dependent control; changing nonlinear interpolations to linear interpolations; and removing unnecessary interpolations. The result is a reduction in runtime of between five and six orders of magnitude.Excluding I/O, the previous known implementation processed one pixel at a time and required about 2.63 seconds per pixel on a Sparc-10 workstation. This is unsuitable for handling real images, as it would take more than 15 years to correct all five spectral bands of a single standard 180 ( 180-km image from the Thematic Mapper instrument aboard the Landsat satellite. Our implementation is based on processing the whole image and takes about 4 to 20 microseconds per pixel on the same machine, or as little as 13 minutes for a 180 ( 180-km image. We also develop a parallel version of our algorithm that is scalable in both computation and I/O. Experiments show that a standard Thematic Mapper image (five bands, 36 Mbytes per band) can be corrected in less than 4.3 minutes on a 32-node Thinking Machines CM-5, including I/O time.