Accelerating plastic particle morphology analysis: Comparing manual, semi-automated, and machine learning-based image analysis methods

Determining plastic particle metrics, such as diameter and surface area, from soils and compost is crucial for assessing their behavior and potential impacts in terrestrial environments, yet it remains challenging due to the lack of standardized methods and the labor-intensive nature of manual image analysis. This study compares three image analysis approaches for particle characterization: manual segmentation in ImageJ, semi-automated analysis with the pliman package in R, and a machine-learning–based workflow using the trainable Weka segmentation plugin in ImageJ. A total of 926 plastic particles, extracted from organic fertilizers and ranging in Feret diameter from 0.23 to 13.9 cm, were analyzed. Manual processing required over 46 hours, whereas the Weka method took approximately 2-3 h, and the pliman method was fastest, completing batch processing of all images in approximately 20 minutes, with an additional 1 hour for manual verification. However, it correctly segmented only 42 % of the particles. The Weka approach performed better, achieving 64 % segmentation accuracy, though it required thorough model training. Both automated approaches correlated strongly with the manual results (R > 0.9, p < 0.001). On average, Weka slightly underestimated particle diameter (-2.1 %) and overestimated area (+2.3 %), while pliman overestimated area more substantially (+12.8 %). This comparative assessment highlights the potential of machine-learning approaches to accelerate plastic analysis workflows while maintaining accuracy. The study supports the development of reproducible, automated pipelines by offering practical recommendations for imaging plastic particles, an open-source Weka workflow, and a model trained to characterize particles commonly found in organic fertilizers and, thus, in soils.