An Analysis of Thermoplastic Composite Rods by Segmentation of SEM Images using Improved U-Net Model

The segmentation of thermoplastic composite rods in SEM images is crucial for analysing and understanding their microstructures and optimizing their mechanical properties.This study proposes an improved U-Net deep learning model for accurate segmentation of SEM images of thermoplastic composite rods.The proposed model incorporates attention mechanisms and residual blocks to enhance feature extraction and improve segmentation performance in complex microstructural regions.XAI techniques, specifically Grad-CAM, are also utilized to visualize the model's decision-making process.The model is trained and evaluated on a dataset of SEM images, achieving high segmentation accuracy and demonstrating superior performance compared to traditional methods.The results indicate that the improved U-Net model is effective in SEM images, offering a reliable tool for thermoplastic composite material analysis.