Profession: remote sensing researcher

Remote sensing gives rise to many professions. From the engineer to the technician, from the satellite manager to the marketing expert who distributes the space data. We spoke with Alexandre Carleer, a Belgian researcher who is currently testing a method for the automatic identification of objects from very high definition satellite images.

Alexander Carleer researcher at the Institute of environmental management and spatial planning (IGEAT) at the Université Libre de Bruxelles. Alexander Carleer researcher at the Institute of environmental management and spatial planning (IGEAT) at the Université Libre de Bruxelles.

The walls of his study, at the Université Libre de Bruxelles, are full of satellite images. They range from cities to mountains and from the countryside to deserts. “Beautiful images,” he remarks. “But you have to be able to decode, read and interpret them.”

When the regions observed are vast and the image definition – or spatial resolution to use the jargon – is low, it is easy to recognise major structures, such as a mountain range, lake, fields, or forests. But when you have to go into more detail, the task becomes almost impossible.

“Except since 1999,” explains this agricultural engineer who took a PhD in sciences and remote sensing. “Since that notable year, we have had very high resolution civilian satellite images. Before that, we had to make do with data from the Landsat satellites, for example, the early generations of which were unable to pick out surface details at anything under about 80 metres. The spatial resolution then increased to 30 metres. With the Spot family we started with pixels representing 20 metres on the ground before achieving 2.5 metres with Spot 5.”

The 1999 revolution bore a double name: Ikonos and Quickbird, two very high resolution US satellites. “With these, the image precision approached that of aerial photographs: between a few dozen centimetres and a metre. This resulted in an explosion of new data,” explains the researcher.

This mass of data then had to be identified, classified and interpreted. For aerial photographs the interpretation was done manually. But over larger areas that satellites revisit regularly this process becomes very long and costly. So automatic methods had to be found.

Methods of this kind had already been developed for satellites of lower resolution and were based on classification by pixel.

Image segmentation

“But for very high definition images, these methods did not suffice,” continues Alexandre Carleer. “We needed another method and this was classification by region. This segments the image according to the colour of each pixel, but it can also take account of criteria related to object shape and even texture. The programmes we are currently testing seek to recreate objects observed on the ground in a way that is homogenous and coherent.We can observe and identify many elements, including houses, industrial buildings, trees, cars, and roads. In short, objects that could not be seen previously using images with a less fine spatial resolution. The process also makes it possible to distinguish between objects of the same colour (or same spectral signature) depending on their shape. Take, for example, the spectral signature of tarmac that in itself was insufficient to say whether it was a section of road or a roof. Today, this latest method permits such a distinction.”

This digital processing of data of course has many applications. “One example is the verification of areas that European farmers receiving subsidies declare as being cultivated areas. The method makes it possible to check and calculate areas that are genuinely being farmed while identifying any inclusions in these parcels of a lake or copse, for example, and to check the types of crops grown. Another example where this method is also proving effective is for the regular updating of land use in a given province, region, or district,” concludes the researcher.