Modesty, warmth, artificial intelligence, forensic science and the monitoring of marine-litter pollution

Andrew R W Jackson, Hugh F D Jackson, Claire M B Gwinnett and Mohamed Sedky    *C.Gwinnett@staffs.ac.uk (Project leader and main contact)

We surround ourselves with textile fibres. Literally. These are the miniscule building blocks from which our clothes are made. Think of a world without them. Not very comfortable, is it? They keep us warm and dry, and we use them to signal to the world who we are and where we belong. They also preserve our modesty. To be without them is the stuff of nightmares.

In the modern world, plastics, including polyester, nylon, and polypropylene, are used to make vast amounts of textile fibres. These synthetic particles can be made to mimic the look and feel of natural fibres but outperform them in many ways. They are fashioned into a variety of high-quality products, including clothes, carpets and ropes.

Pile of Clothes.

Nothing is permanent. This includes the clothes we wear. From the moment we buy them, they are falling apart. Wherever we go and whatever we do, we each leave a trail of fibres from these clothes, and from the textile products that we have touched. This trail can be used by forensic scientists to infer our past whereabouts and actions. To do this we need knowledge of how readily fibres are moved from place to place. This is gained through what are known as transfer and persistence studies. During these, scientists, armed with strong magnifying glasses or low-power microscopes, count thousands of fibres by hand. This is time consuming, laborious and can lead to mistakes.

One of the principles of scientific studies is that they should be repeatable. In recent years, this has been developed into the concept of reproducible research, the goal of which “…is to tie specific instructions to data analysis and experimental data so that scholarship can be recreated, better understood and verified.” (https://cran.r-project.org/web/views/ReproducibleResearch.html, accessed on 9 Jan 2018). Accordingly, where ethically possible, scientists are encouraged to report not only their methods and findings but also their raw data and the computer code that they have used to process them. This has its practical limitations, however. In transfer and persistence studies, it is feasible to report the raw fibre counts. It would even be possible to publish images of the fibres that were counted. However, currently, it is unlikely that the fibres shown in any such images would be re-counted en masse, as to do so would be prohibitively time consuming.

One of the reasons that synthetic fibres are valued is their high resistance to damage. This means that they can be made into strong, hard-wearing textiles and ropes. However, even these fall apart. This means that their constituent fibres enter the natural environment where they are very slow to degrade.

Each time we wash our clothes, huge numbers of fibres fall from them into the water and are washed down the drain (just think of the fluff that collects in a tumble dryer’s filter). We now know that these water-borne fibres find their way through sewage works, down rivers and into the sea, where they are eaten by fish and other animals (https://www.nature.com/articles/srep33997). However, this knowledge has only been gleaned recently and, currently, there are no standard scientific procedures for the monitoring of this type of pollution.

We at Staffordshire University are developing artificial intelligence systems of fibre counting for use in both pollution monitoring and forensic research. Our motivation for this is threefold:

  • To reduce the labour required.
  • To extend the reach of reproducible research.
  • To produce standardised procedures.

Real Life Fibres (left) and Computer Generated Fibres (right)

We have started this work by automating the counting of fluorescent fibres. These were retrieved from a variety of surfaces during a forensic method validation study carried out for a police force here in the UK. These fibres had been counted by hand during that study and so we can be sure how many were present. The approach taken to automation involves computer vision and machine learning. Images were taken of the fibres and they were then counted by a model created by a convolutional neural network. Previously, this model had been trained, using computer-generated images, to recognise fibre ends. The number of such ends were then counted by computer in the images of the real fibres and divided by two to give the model’s count of fibres present. We have then compared the model’s count with the known number of fibres that can be seen in the images (Figure 1).

Figure 1. A graph illustrating the performance of the fibre counting convolutional neural network model used when tested against images containing known number of fibres that had been counted by hand. The blue line is not the line of best fit. Instead, it is there to show where all the points would have been found had the model produced exactly the same counts as found by a human.

Full technical details of the work to date is available online (https://hughfdjackson.com/machine-learning/automatic-fibre-counting-with-machine-learning/ and https://github.com/hughfdjackson/fluorescent-fibre-counting). The code written is given on the second of those sites, as are instructions explaining how to set it up so that you can run it yourself if you would like to do so. As can be seen from those sites, the work is still under development and the results are not yet as accurate as can be achieved by a human, but good progress has been made. We will be building on that work over the coming months, but you are welcome to take the code and develop it for yourself. If you do so, we would be very interested to know of the progress that you make with it. Our plans include the refinement of the approach taken to date to improve its accuracy and the application of that method to the counting of fibres and other microplastics from marine environments.

We also plan to draw on our previous research here at Staffordshire University in areas of computer vision and artificial intelligence (www.spectral360.com). Our research team has developed breakthrough computer vision technology (Spectral-360®) which emulates human vision to detect camouflaged objects under rapid illumination changes and in the presence of severe environmental hindering conditions. Field testing demonstrates the new technology’s ability to detect events missed by a vigilant operator. Four patents (UK, PCT and US) were successfully filed and granted filing numbers. Our US patent was granted in November 2015. This prior research has already created tools that allow computers to automatically detect, count and track objects from their images, and do so even when the lighting conditions change or when the camera used is swapped for another. Our intention is to use this to extend the current study to the counting of marine-litter from images captured by drones and of microfibres and other micro-litter extracted from aqueous environments.

We depend on textile fibres for our wellbeing and surround ourselves with them. They are used to help solve crime but this work is limited by it time-consuming nature. Also, those fibres enter the seas in unknown but vast numbers, as do other forms of marine-litter pollution. Our challenge is to develop automated systems to allow:

  • Better use to be made of the forensic value of fibres and
  • The standardised monitoring of marine-litter pollution to improve its  quantification, identification and interpretation.

Acknowledgements.                                                                                                Thanks are due to Zoe Jones for collecting and counting the fibres used in the work to date, Jolien Casteel for taking photographs of fibres and to Chelsie Maxwell for her technical work on the website.

Leave a Reply

Your e-mail address will not be published. Required fields are marked *