Issue 6.1, April 2002
by Vinciane Lacroix and Marc Acheroy, Signal and Image Centre (SIC), Royal Military Academy (RMA), and Eleonore Wolff, Institut de Gestion et d’Aménagement du territoire, Université Libre de Bruxelles
Introduction and Overview
The aim of the PARADIS project is to improve the planning of humanitarian demining (HD) campaigns using remote sensing data and Geographic Information System (GIS) techniques. In this context, a method and two software packages were built based on the needs expressed by the Bomb Disposal Unit of the Belgian Armed Forces. The main software package consists of management tools integrated in a common GIS platform, working from the country scale to the field scale. The planning method involves two main actors: a demining organization and an image interpretation team. A third actor is an image processing team for which the other software package was built. Applying image processing tools on satellite images and on scanned maps should speed up the extraction of objects of interest in an HD context. This aspect of the project is beyond the scope of this article.
The Belgian Armed Forces Bomb Disposal Unit, also known as SEDEE-DOVO, is responsible for rendering safe and destroying all munitions found in Belgium. Apart from giving direct support to Belgian units when deployed abroad, the unit also contributes directly to HD all over the world. At the moment, the unit has teams of technical advisors deployed in Laos and Cambodia.
Some basic knowledge in remote sensing can be found in the TELSAT guide.4 Satellite sensors vary in spatial, spectral, temporal and radiometric resolution. In this document, we will often refer to the spatial resolution. Spatial resolution is related to the resolving power to distinguish image details. In remote sensing, it is common to specify the spatial resolution as the size each pixel represents in the real world. For example, the SPOT satellite has a spatial resolution in the panchromatic mode of 10 m, meaning that the image is composed of pixels with a ground diameter of 10 m. Also, a sensor may be active or passive. Optical sensors are passive, as they measure the reflected sun-radiance and radiance emitted by the observed objects. Radar sensors are active, as they emit a beam of electronic radiation and detect the wave that is reflected by objects.
GIS and Remote Sensing for HD
We made a review of existing GIS tools for HD in A Geographical Information System for Humanitarian Demining1 when proposing the design of PARADIS. The review included the project undertaken at James Madison University (JMU); the Information Management System for Mine Action (IMSMA), developed by the Geneva International Center for Humanitarian Demining (GICHD); Minedemon, developed at ITC; the Digital Mine Documentation System Prototype, developed by the Business for Industry Concerns (IABG); "FOCUS HD," a Mapping Information system designed by Landair International Ltd.; and the integrative approach proposed by the Defense Evaluation and Research Agency. These projects are described in more details in Demining Technologies, a publication from International Exhibition, Workshops and Training Courses.2 Also, some studies aimed at using remote sensing data in order to detect mine fields, as for example, the "airborne mine field detection5" project.3
From the latter review, we conclude the following:
Therefore, we decided to build a tool based on IMSMA and ArcView in order to plan the humanitarian campaign, following the tasks of deminers during a demining mission.
Design of PARADIS
Identification of Spatial Data Needs
In order to identify the spatial data needs of deminers during an HD campaign, a collective work was started. As a starting point, a table of the user needs established by Paddy Blagden, technical director of the GICHD, was used and expanded during meetings with the Belgian deminers.
A demining mission involves the following tasks:
Table 1 lists the required information at each step of the mission, its possible sources and the appropriate scale. A short version of this table has been presented at the JRC, in ISPRA, at the workshop "Towards Harmonized Information Systems for Mine Action in South Eastern Europe6," and published later in A Geographical Information System for Humanitarian Demining.1
We have identified the tasks to be performed in the interface according to the routines of a mission and the identified data needs previously described. A software package working at four scales embedded in an ArcView platform that is compatible with IMSMA and the Belgian EOD Champassak database has been produced.
The global scale or Country Scale (1:1,000,000) may contain the Digital Chart of the World (DCW), topographic maps, meteorological data and maps of refugees. At this scale, the user should be able to combine information such as administrative limits, roads, relief, hydrographic networks, inhabited areas and local names, climate zones, soil types, refugees, airports, mine clearance center locations and responsibility areas.
The Region Scale (1: 250,000 to 1: 50,000) may contain satellite images (SPOT, Landsat TM, ERS and RADARSAT), topographic maps and information from field surveys. Based on this data, the user should see practical roads and bridges, village extensions, hydrographic networks, hospitals, military buildings, accident localization, campaign schedules, mine field locations and land cover. This overlay information comes from image interpretation and from the IMSMA database. At this scale, the demining staff has tools to plan its campaigns and organize its teams according to priorities, regional constraints and logistic facilities.
The Field Scale (1: 10,000 to 1: 1,500) may contain aerial photos, very high-resolution satellite images (IKONOS), statistics and sketches. Highly accurate maps of the suspected areas and cleared mine fields could be available as overlay. An Advancement Scale (1:º500) is added in order to produce a detailed description of each mine field.
The system was fed with data for test sites in Laos and Mozambique (see section entitled Test Sites). The design of the prototype was presented at the ISPRS 2000 conference in Amsterdam.1 A double-face summary illustrating the system for the Laos and Mozambique test sites is available at the SIC website.6 Table 2 describes all the functions to be implemented in the interface and their working scale.
Organization of a Campaign
The general planning method proposed involves a demining organization and an image interpretation team. Additionally, an image processing team may ease the work of the interpretation team.
In the PARADIS project, the SEDEE-DOVO played the role of the team of deminers. During the designing phase of the project, the interpretation team was distributed between the scientific partners of the project: the IGEAT and the SIC. However, as the method should be performed in routine, an operational partner had to be found inside the Belgian Defense itself. The General Intelligence and Security Service, Section Imagery (SGR-IM) assumed this task. This role could be given to a well-chosen subcontractor in the case of non-Belgian missions.
Extracting information from satellite images and scanned maps could be tedious work if a lot of images must be analyzed or if images are large: semi-automatic feature extraction could be a precious help for image interpreters. This work is not mandatory, but it should speed up the interpreters’ work.
The procedure of data collection goes as follows; it is summarized in Figure 1. At the mission announcement, the demining team contacts the interpretation team (SGR/IM) in order to identify the regions and the best season for acquiring satellite data over the areas of interest, and to purchase them. This team is also responsible for collecting maps and performing the georeferencing of the satellite images. The images are then sent to the image processing team (SIC) for automatic analysis: extraction of the hydrographic and road networks, identification of water areas, and classification of the images. This information should be used to facilitate the image analysis which the aim is to produce the vectorial overlays made of roads, inhabited and cultivated areas, infrastructures, etc., again performed by the interpretation team. The latter team will also produce color composite images using all bands, and black and white images displaying the panchromatic data. Meanwhile, the deminer team (SEDEE-DOVO) fills the IMSMA database with the field survey. When all this data has been collected, it is introduced in the prototype described in the "PARADIS Interface."
The first test site chosen was in Mozambique because of the co-lateral data already available from the "airborne mine detection" project mentioned above. In order to show the adaptability of the method in a different context, another test site was chosen in Laos, where the Belgian deminers are active. All missions involved data collection, data interpretation and ground survey, as well as work with local deminers, specifically Norwegian People’s Aid (NPA) in Mozambique and UXO Lao in Laos.
The test sites were located in the province of Tete close to Songo and Mameme. A first mission on the sites was conducted in order to obtain all the missing information, establish contacts with NPA and elaborate a working method with them, and test the use of high-resolution images for demining activities and check their interpretation. A second mission there aimed at presenting final results to NPA and confirming the use of very high-resolution data.
For the first mission, LANDSAT TM (resolution 30 m), SPOT multispectral (resolution 20 m) and panchromatic (resolution 10 m), RADARSAT (resolution 13 m) and LANDSAT MSS images (resolution 80 m) were available and processed. For the second one, IKONOS panchromatic data (resolution 1 m) was acquired.
High-resolution satellite images (resolution between 10 and 30 m) were used to extract information about roads, railways, villages and crops. Thanks to a sequence of images, it was possible to observe an increase of inhabited areas, crops and even tracks in the Songo area. On the RADARSAT image, we could only see roads and rivers, while the LANDSAT MSS images were of too low resolution for our purpose. It should be emphasized that the advantage of radar images is to provide information despite the clouds; thus, they are useful in very cloudy areas where maps do not exist.
For the test sites in Laos, in the Champassak province, we purchased SPOT multispectral data (resolution 20 m) thanks to the "Secrétariat pour la Coopération au Développement." UXO LAO lent IKONOS panchromatic data (resolution 1 m) in order to assess the use of high-resolution images for HD activities.
The assessment concerned the method, the use of remote sensing data of various resolutions and the validity of its interpretation, the completeness and the appropriateness of the table of needs, the tools provided in the PARADIS interface and the use of image processing tools. We will not speak about the latter as its interest is more for image interpreters than for deminers.
In order to assess the method at the end of the project, we simulated a campaign in Laos, following the scheme of Figure 1. SGR-IM found SPOT and IKONOS images in the archives and purchased the SPOT data. Image Processing tools were applied to the images and sent back to the interpreters. SPOT and IKONOS images were interpreted according to the legends defined by the team. The time schedule was satisfied, and the products (space maps, overlays, documentation) were up to the end-user expectation. The interface was fed with information from Level One and Level Two Surveys for which IMSMA forms were used.
Use of Remote Sensing Data
For the first mission in Mozambique, the visual interpretation performed by our scientist was useful for non-experts; however, some errors occurred, so we recommend a check on the field and feedback to the interpreters.
For the mission in Laos, the image interpretation of the IKONOS panchromatic image was good for punctual and linear elements such as houses, trees, roads, rivers, etc. The land cover interpretation seemed more difficult to use due to the complexity of the land use in that region (e.g., the mixture of coffee plants and bushes). In fact, the land cover interpretation seemed useless, since the local population—both deminers and villagers— could read and work with the raw image. Indeed, when asked to delineate an area selected on the image on the field (and conversely, to draw on the image a "cleared area" on the field) deminers had no hesitation. Local people also preferred the raw image because the interpreted layers hid the reality of the image.
Thus, with a limited budget, panchromatic images (resolution 1 m) should be preferred to multi-spectral data (resolution 4 m) because field details are probably more useful than the spectral information (i.e., the "color").
The survey and the roving teams might also use the IKONOS panchromatic images, for example, in order to locate the UXO on the image instead of drawing schematic maps or leading the team to the right place.
The assessment is different for the SPOT image. Due to its lower resolution and maybe to the chosen color composition, the raw image is much more difficult to read. Thus, the visual interpretation was very helpful for the deminers. They could use it as a map, associating each color or symbol with an object or an affectation. A villager had almost no difficulties reading the interpreted image; he could guide us along paths while showing his position on the document. Similar observations were obtained from the last mission in Mozambique.
Both types of satellite images need to be properly geo-referenced if they are used as reference maps to locate mine fields or UXO. If topographic maps exist, the user has to identify points in the image and give their correspondent on the map. However, the map may not be precise enough to geometrically correct the satellite images. Therefore, a set of ground control points collected on the field with a GPS should be used for the geometric correction.
In conclusion, very high-resolution satellite images (resolution 1 m) are useful for the work in mine fields since they do not require an interpretation. On the other hand, high-resolution images (resolution of 10 m to 30 m) are useful as regional maps for planning the teams’ work. However, they require an interpretation by an expert, and this interpretation could be sufficient for deminers.
The PARADIS interface was not finished in time, so only part of it could be tested. However, its philosophy has been explained to other end-users, and much appreciated. The presentations made to various people have sparked great interest and encouragement. Deminers showed their willingness to use it later, because (among other things) it simplifies the representation of the clearance areas, and it optimizes—both in easiness and speed—the encoding of data in IMSMA.
Furthermore, we noted that specific tools (such as clearing the grid, automatically integrating GPS measurements and shifting the scanned maps) might improve the deminers’ work both for the everyday job and for the planning process at the office.
Conclusions and Further Work
As a general conclusion, we may say that PARADIS was a successful project, thanks to a motivated multi-disciplinary team and the motivated and cooperative end-users.
The following is a list of conclusions from this project:
These conclusions should be validated by other field missions. Moreover, the table of spatial data needs could be enhanced by defining the precision of the location (implicit in the scale) and the relevant attributes (e.g., for a road, its width) as well as their translation in terms of graphical differences (i.e., defining a symbology dedicated to demining activities) for each object. Last but not least, the interface should be finalized and distributed as a complementary tool to the existing ones.
This project involved a multi-disciplinary team. At the SIC, Dirk Borghys, Mahamadou Idrissa and Vinciane Lacroix are Image Processing researchers; Youssef Ouaghli is the industrial engineer and Michal Shimomi is the geologist. At the ULB, Eleonore Wolff and Renzo Fricke are geographers. Finally, at the SEDEE-DOVO, Michel Lambrechts and Vincent Muylkens are deminers. In both missions, local NGOs were very helpfuI; we wish to thank Norwegian People’s Aid and UXO Lao. Also, the project used a lot of data collected in the previous project, "Airborne Mine field Detection," in which the SIC was involved. The PARADIS project was supported by the TELSAT program of the Scientific, Technical and Cultural Affairs of the Prime Minister’s Service (Belgian State).
1. V. Lacroix, M. Shimoni, E. Wolff and M. Acheroy, A Geographical Information System for Humanitarian Demining, ISPRS2000 GeoInformation for All, Amsterdam, July 2000, paper 738.
2. Demining Technologies; International Exhibition, Workshops and Training Courses, JRC (Joint Research Center European Commission), Ispra, 1998.
3. J. van Genderen and B. Maathuis, "Airborne Detection of Landmines: A Review of the Techniques and Some Practical Results," Disarmament and International Security, Regensburg Germany,1998.
5. http://www.itc.nl/ags/projects/mine field_detection/mine field_detection.htm