A particle’s material and size shape its ultrasound echo. In a study published on 3 March 2026, a neural network learned those frequency patterns to distinguish polyethylene and acrylic (PMMA) plastics from glass and steel.[1]
The researchers scanned microspheres immobilised in agarose gel with a 40 MHz transducer. Their 8,045 signal traces included multiple echoes from each particle. All traces belonging to one particle stayed within the same training, validation or test partition, keeping familiar particles out of the test.[1]
Across ten random splits, mean material-classification accuracy on held-out particles was 97.14%, with a standard deviation of 3.11 percentage points. The dataset retained isolated, usable echoes; overlapping particles could obscure smaller neighbours.[1]
Acquiring one scan took around five to six minutes. The experiment establishes material recognition in fixed laboratory microspheres. Recovering reliable signatures from irregular fragments moving through environmental water remains an engineering and validation task.[1]
Sources & further reading
- High-frequency ultrasound combined with deep learning enables identification and size estimation of microplastics npj Emerging Contaminants · 3 Mar 2026 · Accessed 20 Oct 2026