Automated Classification of Carbon Nanomaterial Structures based on Computer Vision Model

Carbon nanomaterial structures hold significant promise across various industries, necessitating accurate and automated classification methods. Conventional approaches rely on handcrafted feature extraction techniques, often failing to capture complex spatial patterns inherent in nanostructures. Traditional Machine Learning (ML) and basic Deep Learning (DL) models suffer from low generalization and require manual feature engineering, making them inefficient for handling diverse and noisy microscopy images of nanostructures. The objective is to achieve a highly accurate and automated classification of carbon nanomaterial structures through an advanced framework. A novel approach Modified Water Wave-inspired Convolutional Autoencoder with Swin Transformer (MWW-CAE-ST), integrates optimization, and classification techniques to address existing challenges. A collection of microscopy images of carbon nanomaterials, including diamond particles, and nanotubes was used to evaluate the framework. Techniques, such as median filtering and histogram equalization (HE) were applied to enhance image quality by reducing noise and normalizing intensity levels. Local Binary Patterns (LBP) were employed to extract texture-based features that capture fine-grained details of the nanomaterial structures. Features generated by LBP were processed through the CAE for dimensionality reduction and refined by the Swin Transformer, which utilizes hierarchical self-attention to classify structures effectively.