Novel Estimation of Nanofiber Diameter from SEM Images Using Deep Feature Embeddings and Machine Learning Models

Abstract

Accurate nanofiber diameter estimation is crucial for optimizing their functionality in materials science. Traditional measurement methods from Scanning Electron Microscopy (SEM) images are often labor-intensive and subjective. This study proposes a machine learning-based approach using deep feature embeddings to predict average nanofiber diameters directly from SEM images. Eight machine learning models—Linear Regression (LR), k-Nearest Neighbors (kNN), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), Neural Network (NN), Gradient Boosting (GB), and AdaBoost—are evaluated using 5-fold and 10-fold cross-validation. Performance is assessed via Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and R2. The kNN model consistently outperformed others across three datasets: smooth nanofibers, beaded nanofibers, and a combined set. It achieves the lowest error metrics and the highest R2 (0.950) for smooth nanofiber images while demonstrating strong generalization across morphologies. This study is among the first to integrate deep feature embeddings with machine learning for direct nanofiber diameter prediction, providing a reliable alternative to traditional methods.