This paper presents two experiences in the field of high-performance SAR processing. Getting high quality SAR images in reasonably short times is getting more and more strategic for remote sensing applications, still requiring the use of data-parallel processing. Fast formation of images from radar data requires dedicated machines suited to fine-grained, DSP-based parallelism, while improving the image quality through despeckling is suited to large-grained parallelism based on general-purpose machines. These two aspects are analyzed and evaluated by presenting: an industrial system for real-time SAR processing on a custom, DSP-based parallel system, called ACIP-XSAR; and the results of the Esprit PULSAR project, providing a portable and scalable parallel package for image despeckling and segmentation.

Anneal provides a means by which users can obtain significant image enhancement over other general purpose techniques for SAR noise despeckling. Its principle disadvantage lies in its rather steep computation requirements, particularly on standard sized images (2048 by 2048 and 4096 by 4096). However, PULSAR has been able to produce a portable parallel version of Anneal that has excellent parallel performance on both workstation clusters and dedicated distributed memory parallel platforms. Speed-ups of 90 have been observed using 96 processors on the GC PowerPlus with the relatively small 1024 by 1024 image. Similar performance is expected on other platforms with a reasonable interconnect. Thus, Anneal is suitable for both providers of satellite SAR images (who likely employ MPP's to perform the initial low-level processing) as well as for interpreters and direct users of SAR data, whose source of computation comes from networked workstations.

Synthetic aperture radar (SAR) offers the possibility of imaging the earth's surface independent of time of day or weather conditions. SAR images differ from optical images in that they contain speckle, multiplicative noise with a poor signal-to-noise ratio of about 1 to 1. Several routines for noise despeckling and segmentation were ported to distributed memory platforms within the EUROPORT-2 PULSAR project. Excellent parallel speedup was ob served for ANNEAL, the most promising despeckling rou tine. Reasonable but limited parallel performance was obtained on RWSEG, a segmentation routine, because relatively large overlap regions were involved. The nature of these operations means that there is no one correct result to which output images can be compared. However, differences between images obtained from sequential and parallel implementations were encouragingly small.

The PULSAR project, aiming for the parallelization and adaptation of SAR segmentation and filtering codes, is carried out under the EUROPORT 2 activities within the European funded ESPRIT 3 program and running over 2 years. An international consortium consisting of companies working in the informatics and earth observation domain was created to perform the project as an interdisciplinary study. Different actions carried out in the past year and a half were mainly the testing and parallelization of the different codes, their adaptation to selected applications and the benchmarking in terms of speed and reliability of results. A number of codes were improved and speeded-up even before the parallelization and better benchmarking results could thus be achieved. For the detailed testing of the codes three key applications were identified where radar remote sensing will play a key role in the next few years. These applications are the oil spill detection, the surface management of temperate (European) agriculture and the tropical agriculture (rice surface detection and monitoring). For all three application fields different parameters available in the software were tested and results show great improvements of the interpretation capabilities in relation to the initially speckled ERS SAR data. The segmentation specifically allowed us to obtain field boundaries of agriculture fields over the test site Bourges (France) and areas of similar growth conditions for rice areas in Indonesia. The speed of the processing of one 512 by 512 pixel image was at 15 minutes using the serial version and 2 minutes using the parallelized version with 4 processors. The comparison between the serial and parallel results allowed us to investigate the stability of the parallel process which was below 2% inaccuracy. Thus, it can be underlined that the PULSAR project allowed us to develop a new and very useful tool for the interpretation of radar images. For specific applications it is intended to further adapt the tool in order to be able to deliver turn-key solutions to the user.

The EMMA2(1) parallel architecture is a network of nodes, named 'Regions': each region is based on two shared-memory hierarchical bus levels, on which a large series of monocpu, multicpu and specialized I/O boards can be inserted, grouped in 'Families'.Its typical applications are image processing, pattern recognition and artificial intelligence, characterized by intensive computation requirements, real-time constraints and good price-performance ratio for building large industrial systems. This paper presents the main characteristics of the EMMA2 parallel operating system, highlighting the internal organization and mechanisms offering the parallel co-operation model for user programs.Section 1 briefly introduces the overall hardware architecture. It allows application partitioning into parallel independent subsystems, on two levels of computation: the lower level (family) is dedicated to massive number crunching, while the upper (region) permits the interaction of these lower machines.Section 2 deals with the system software environment, based on a host-target configuration: the host is mainly devoted to machine control, system monitoring, program development and debugging in a multi-user environment.Sections 3, 4 and 5 describe respectively the fundamental concepts of the operating system internal organization and the basic mechanisms with respect to the machine architecture; the basic functions that constitute the internal layers and services; the resulting parallel programming and co-operation model and the typical mapping rules of algorithms to this architecture, completed with some examples of their functionalities.The most important aspect of the EMMA2 operating system is the fact that it offers a uniform interface to user tasks, independently of their allocation on the target machine. Objects managed by system services can be referred to via logical names. Code for system services is distributed among processors and resides on a private read-only memory. The operating system's data reside in shared memory only for system functions that need to be distributed, while private functions are based on different copies of data in private memories.Section 6 gives more details on the most relevant EMMA2 industrial applications, also showing their mapping to the parallel architecture and some performance-related information. Section 7 contains a brief outline of the novel architecture currently in development and of the subsequent OS improvements.In conclusion, we report some considerations about the present status of the EMMA2 system, compared with some other well-known architectures, and about research efforts for its future evolution.

The development of a parallel system providing high computational power at a reasonable cost using inexpensive processors is described. The original motivation for the work, which led to the EMMA (Elaborate Multi-Mini Associativo) multicomputer, was the mechanization of mail sorting. The need for a system that could be the basis for a wider range of applications spurred the study and development of a new system, the EMMA2. The discussion covers EMMA2's characteristics, architecture, buses, standard modules, I/O, approach to coprocessors, and operating system.< >

The paper presents the experience of porting and using the PVM messagepassing environment on EMMA2E, an industrial, shared-memory multiprocessor architecture. The work is still ongoing in the framework of the ESPRIT “EUROPORT” project, aiming to demonstrate practical availability of portable parallel applications with related industrial and commercial benefits. The project framework, the EMMA2E architecture and the related system software are briefly introduced, followed by the main features of PVM porting solution. This last especially aims to join PVM general and portable programming model with the run-time efficiency achievable in a shared-memory environment. Finally, performance figures of PVM in the multiprocessor will be given and the quality of results evaluated.