Morphological classification of reinforcing nanoparticle aggregates: comparison between visual expert decision and machine learning techniques

This work compares several automatic classification methods with visual expert decision classification. The classification is based on carbon black aggregates into one of the four morphological classes that were proposed by Herd. The automatic methodologies were performed in several steps: transmission electron microscopy image processing to compute a set of twenty-one morphological characteristics, multivariate analysis of the dataset to avoid the curse of dimensionality in classification problems, and built, test and validate machine-learning techniques to classify the aggregates. In the other hand six expert researchers made a visual classification of the aggregates and later, a statistical analysis was conducted to classify the aggregates according to some defined weights. And finally, a comparison of the methods decided which method obtained the highest accuracy and was the most reliable between the compared ones.