快猫短视频

With help from artificial intelligence, we can now detect patches of floating marine plastic from satellite data. The technique may eventually help environmental researchers better monitor and manage plastic waste in the ocean.

Lauren Biermann at the Plymouth Marine Laboratory in the UK, and her colleagues have devised an AI capable of identifying sea plastic in imagery taken聽at varying wavelengths of light. They trained it with images from the European Space Agency鈥檚 Sentinel-2 satellites, which record visual information at multiple wavelengths along the electromagnetic spectrum, ranging from visible light to shortwave infrared.

The algorithm can distinguish floating macroplastics 鈥� those greater than 5 millimetres in size 鈥� from other materials such as sea foam and seaweed with 86 per cent accuracy.

Biermann had previously found that floating plastic reflects light in a particular way, giving it a unique spectral signature. For example, in contrast with seawater, which absorbs most near-infrared light, plastic reflects light at this wavelength.

The researchers trained an AI using of floating plastic taken over Durban in South Africa in April last year, when severe flooding washed vast amounts of plastic waste out to sea.

They also trained the algorithm on existing spectral signatures for other natural materials, including seaweed, woody debris and volcanic rock.

The AI learned to analyse individual pixels of satellite imagery, each corresponding to a 100-square-metre area. It then determined which type of material, such as clear water or seaweed, the pixel was most likely to contain.

Tested on four sites off the coasts of Canada, Ghana, the UK and Vietnam, the AI was able to correctly identify floating plastic 86 per cent of the time.

To date, most marine plastic monitoring is done manually on boats, says Biermann. 鈥淵ou鈥檙e always sampling small areas and hoping to get an idea of what that means on a larger scale,鈥� she says.

The is a step towards wider monitoring. Sentinel-2 only captures data over land and coastal waters, which is a current limitation. 鈥淚f I鈥檓 looking at a Sentinel-2 image, today, it means it was acquired yesterday or even the day before, and in the coastal zone things change so quickly,鈥� says Biermann.

In future, the team plan to expand the monitoring over open oceans, where currents move less rapidly. 鈥淭hat鈥檚 where we might be able to remove tonnes of aggregated materials at a time, because we can direct clean-up operations to places where we saw a big aggregation,鈥� says Biermann.