Automated Shape and Two-dimensional Projected Area Characterization of Nanoparticles from Scanning Electron Microscopy Images Using Deep Learning and OpenCV

Automated analysis of microscopy images is increasingly important for engineering applications where manual measurements are time-consuming and inconsistent.In this work, we present a lightweight computer vision pipeline that integrates bounding-box annotation, a compact convolutional neural network (CNN), and OpenCV-based contour routines to classify nanoparticle shapes and estimate two-dimensional (2D) projected areas.Four classes were considered-spherical, cylindrical, hexagonal, and triangular-using labeled scanning electron microscopy (SEM) images.The CNN achieved stable multiclass performance with high accuracy and balanced F1-scores (harmonic mean of precision and recall), while the contour-based approach enabled direct estimation of 2D projected area from particle boundaries.Validation on synthetic SEM-like images indicated that the proxy closely tracked true mask areas with an error of about 3 % under controlled conditions.In addition to single-particle measurements, the system generates batch-level statistics such as total projected area, mean particle area, random sample particle areas, and overall particle coverage, providing insights into size distributions and synthesis consistency.The emphasis is on automation, reproducibility, and computational efficiency, making the framework a practical engineering solution rather than a replacement for gold-standard segmentation methods.This demonstrates how computer vision and lightweight deep learning can provide accessible tools for microscopy data analysis in engineering workflows.