Deep-MAP: A user-friendly platform for deep learning-based microplastics classification

In recent years, microplastic (MPs) pollution in rivers and oceans has become a serious environmental problem, raising concerns about its impact on ecosystems and human health. While the identification and quantification of MPs using microscope images are essential for understanding the extent of this pollution, manual analysis is time-consuming, labor-intensive, and prone to analyst-dependent bias. To address this challenge, this study aimed to establish a high-precision automated analysis method for MPs images using deep learning. Specifically, we systematically compared and evaluated three different models: 1) an end-to-end segmentation model (YOLOv8m-seg), 2) a two-stage model for detection and classification, and 3) a model combined with classical image pre-processing. The results showed that the end-to-end segmentation model without pre-processing achieved the highest performance, with a Mask [email protected] of 0.555 and a Mask [email protected] of 0.873. It was also found that aggressive background removal during pre-processing degraded performance due to the loss of boundary information essential for model recognition. Based on these findings, we developed Deep-MAP, a user-friendly analysis tool built on the best-performing model and implemented on Google Colaboratory. With this tool, users can upload microscope images and automatically obtain aggregated outputs on MPs count, types, areas, and colors. Deep-MAP helps bridge the gap between developers and end-users by eliminating the need for specialized knowledge.