A Novel Low-Cost Approach For Detection, Classification, and Quantification of Microplastic Pollution in Freshwater Ecosystems using IoT devices and Instance Segmentation
Saketh Sundar
Abstract
Microplastic pollution in freshwater is an extremely important issue, affecting numerous ecosystems and species. The lack of infrastructure for wide-scale analysis of microplastics makes it difficult for scientists and governments to develop conservation efforts. New studies concerning this topic usually involve complex and expensive processes, making them hard to scale. The objective was to develop a novel approach for microplastic detection, classification and quantification using IoT devices and deep learning to make the process faster and affordable (total cost was 75 dollars). A prototype device was created using Raspberry Pis to capture hundreds of images of water samples in the Chesapeake Bay watershed for training. The final method demonstrated excellent scores for all three tasks. In the future, this novel approach will enable large-scale data collection for data-driven policies to minimize microplastic pollution.