Optimized deep networks for the classification of nanoparticles in scanning electron microscopy imaging

The major role of many research studies in nanotechnology is to identify, count, and measure nanoparticles. Images with particles are often handled by hand, employing a computer program ruler. The lack of available algorithms and specialized software tools for analyzing microscope images are limited in this research area. So, this study aims to encourage the development of a comprehensive model to predict the nanoparticles type in scanning electron microscopy (SEM). We present a dataset of 750 ordered nanoscale materials. Palladium nanoparticles (Pd nanoparticles) are placed on a carbon surface using a scanning electron microscope. This paper proposes an intelligent optimization model to classify Nanoparticle types in SEM images (e.g., lines, intersections, networks, ellipses, and circles). The utilized dataset is unlabeled with the imbalanced distribution of classes. Moreover, it contains a large amount of redundant information. These kinds of problems affect the performance of the nanoparticle's classifier. The proposed model can be divided into four phases: preprocessing, feature extraction, feature selection, and classification. The first phase of the Nanoparticle classification model is concerned with image labeling, where some morphological operations and image filtration are proposed to help experts in labeling the SEM dataset. VGG-19 deep networks in combination with Grey Wolf Optimization (GWO) are used for feature extraction and selection phases. The imbalanced input data classification phase proposes a class weight balancing support vector machine (SVM) with different kernel functions. Detailed results of all experiments show that the proposed intelligent optimization model is promising in providing a high-performance classifier for nanoparticle types in SEM images. Overall performance measures are 97% and 98% for overall accuracy and F1-score, respectively.