● Detection of marine plastic pollution still relies heavily on laborious site surveys and visual identification of images, both of which require significant time and human resources.
● Since the late 2010s, academic research has tested artificial intelligence and computer vision approaches that utilize deep learning algorithms to address this problem.
● Increased automation may pave the way for wider and more frequent detection, but new solutions still need to address field limitations and lack of algorithm training data.
Current estimates put about 8 million tons of plastic waste in the world’s oceans. Efforts to contain this phenomenon will inevitably require monitoring the myriad chaotic trajectories this debris follows before reaching the high seas. They are subject to fluctuating river levels, the presence of dams, and a variety of unpredictable geographic conditions, making surveillance missions even more challenging. Given this situation, researchers have developed techniques to automate waste detection using machine learning, especially deep learning. . Satellites, planes, balloons, drones, and boats are thus equipped with computer vision systems trained to detect plastic waste, and in theory, these innovations could enable them to be detected in larger areas. It paves the way for more regular detection. However, it still suffers from many limitations.
Wider range of detection
“Convolutional neural networks integrate the ability to describe various visual characteristics of an image (color, texture, shape, etc.) and use this information to identify objects of interest (such as dust) in specific image regions. determine whether it isexplains Ricardo da Silva Torres, professor of data science and artificial intelligence at Wageningen University and Research School in the Netherlands and professor of visual computing at the Norwegian University of Science and Technology.
Convolutional neural networks integrate the ability to describe various visual properties of images (color, texture, shape, etc.).
In 2021, some of the responses to the European Space Agency’s (ESA) call for bids for a project to detect marine plastic litter adopted this approach. That same year, the nongovernmental organization Ocean Cleanup announced that it had developed its own artificial intelligence surveillance and mapping tool.
The role of material reflectance
But how effective have they proven? Here is an August 2022 article published in a scientific journal water research Researchers at the Federal Institute of Hydrology in Koblenz, Germany set out to evaluate a variety of different methods involved. Some of these are based on the reflectance of plastics (the level of light reflected by the material), which differs from that of natural materials (algae, wood, spray). In 2020, a team led by Lauren Biermann at the Plymouth Oceanographic Institute (UK) discovered that it was possible to find specks of marine plastic using optical satellite data, in this case data from ESA’s Sentinel 2 satellite. shown for the first time. . Other methods are aimed at recognizing the shape of objects (bottles, cans, bags, etc.), but they work only at short distances or in small areas.
The comparative studies of the Koblenz researchers also paid particular attention to short-range detection, i.e. detection at distances less than 120 meters, performed by aerial drones and cameras permanently mounted on bridges and other high places. increase. Contrary to their conclusion: “So far, none of the automated approaches discussed are alternatives to traditional macroplastic monitoring approaches such as visual counting or field surveys.Too many disturbing patterns caused by lighting, ambiguous shaped objects, local environmental factors, etc.
Serious shortage of databases
Another problem is the severe shortage of databases, which play an important role in machine learning. Those that exist are limited in scale and lack diversity. Building a new one is itself a large project. “Labeling images is a tedious and time-consuming task, and for some applications it relies on the expertise of trained annotators.”Ricardo da Silva Torres points out that in January 2022 he also published a comparative study of deep learning tools for detecting plastic debris. He realized that there was only one publicly available dataset containing 1,500 images of plastic waste, so he and his team decided that the new one, he called PlastOPol, would include 1,000 additional images. It is
Despite everything water research The article argues that automated detection of marine debris still has great potential. This mainly depends on improving the algorithms that classify objects. In the future, analysis of spectral wavelengths and better segmentation will allow us to: “Distinguish between a large number of objects that may float on the surface of the water or wash up on shore.”.
