Sizing Microplastic Particles Using Acoustic Imaging and Deep Neural Network

This study addresses microplastic pollution in oceans by developing a real-time sensor system to size microplastic particles from aquatic environments. Using a deep neural network model, the study aims to segment microplastic particles from acoustic images. The methodology involves acoustic imaging with ultrasonic multi-element probes, utilizing imaging strategies like Total Focusing Method (slow but accurate) and Circular Wave Imaging (fast but includes artifacts). By generating a dataset of acoustic simulations, we trained multiple deep neural network models using various image reconstruction strategies to evaluate the feasibility of sizing and counting particles through rapid measurements, even with degraded reconstructed images. The findings suggest that deep-learning-based acoustic imaging can enhance the monitoring of oceanic microplastics by potentially increasing frame rates and simplifying probe complexity. Additionally, the model succeeds on experimental data, not included in the training set, which shows good generalization of the model.