Toward an Automatic Classification of SEM Images of Nanomaterials via a Deep Learning Approach

Abstract

Nanofibrous materials produced by electrospinning process may exhibit characteristic localized defects and anomalies (i.e., beads, speck of dust) that make the nanostructure a network of nonhomogeneous nanofibers, unsuitable for industrial production at large scale of the nanoproducts. Therefore, monitoring and controlling the quality of nanomaterials production has become increasingly important and intelligent anomalies detection systems have been emerging. In this study, we propose an innovative framework based on machine (deep) learning for automatic anomaly detection. Specifically, a deep convolutional neural network (CNN) is proposed to automatically classify scanning electron microscope (SEM) images of homogeneous (HNF) and nonhomogeneous nanofibers (NHNF), interpreted as two different categories. The proposed approach has been validated on experimental SEM images acquired through SEM images analyzer on polyvinylacetate (PVAc) nanofibers produced by electrospinning process. Experimental results showed that the designed deep CNN achieved an accuracy rate up to 80% and average precision, recall, F_score of, 78.5, 79, and 78.5%, respectively. These promising results encourage the use of this effective technique in industrial production.