Abstrak

Microplastic morphology influences particle behavior, environmental fate, and ecological risk, yet commonly used two-dimensional (2D) shape descriptors often struggle to represent complex and irregular geometries. This study introduces the Shape Descriptor Index (SDI), a composite metric integrating area, length, and circularity, designed as an alternative and machine-compatible proxy for microplastic morphology. Using deep feature embeddings extracted from scanning electron microscopy (SEM) images with Inception V3, we evaluated the predictability of SDI relative to classical descriptors across multiple machine learning models. SDI demonstrated the strongest performance, particularly with the AdaBoost model, achieving an R² of 0.919 along with reduced root mean square error (RMSE) and mean absolute percentage error (MAPE) compared to the other descriptors. These findings indicate that SDI aligns well with deep visual representations and offers a robust, scalable metric for rapid morphology assessment. The approach supports high-throughput and objective analysis, making SDI particularly suitable for large-scale environmental monitoring and automated microplastic characterization.