Novel Framework for Enhanced Nanoparticle Detection in Scanning Electron Microscopy Using Synthetic Data

Nanoparticle detection in scanning electron microscopy (SEM) is crucial for various applications. Existing techniques for detecting nanoparticles in SEM images need help to handle dispersed particles and need more accuracy. This research uses a deep learning strategy to enhance recognition efficiency and precision. To overcome the challenges, we develop a robust Multi fused Spectral Deep Convolute Neural Net (MS-DCNN) based model for nanoparticle detection, utilizing synthetic data generation to facilitate practical neural network training and collecting the SEM image dataset for detecting the nanoparticle. Created an algorithm to generate synthetic data, combining random particle distributions to simulate SEM micrographs and allows the development of annotated datasets that are essential for neural network training. Compared to existing approaches; the results are reduced pixel (0.62), warp errors (0.0008), decreased computing time (398s) and greater accuracy (92.5%). The suggested MS-DCNN framework is practical and better than conventional techniques, exhibiting improved precision in the identification of dispersed nanoparticles. The generation of synthetic data helps in the development of a trained model that will deal with a variety of particle distributions. The model is trained using synthetic data, demonstrating the technique's potential to improve nanoparticle analysis in SEM imaging which got proven result over the existing method.