Nanoparticles, which exhibit characteristics different from those of larger particles in terms of physical, chemical, and function al attributes. Accurate segmentation of nanoparticles in SEM images is significant to materials science because it offers vital information concerning the size, shape, positioning, and surface properties of nanoparticles. However, traditional methods of image segmentation face challenges in segmenting particles with fine borders and adapting to variations in scale and contrast, thus limiting their widespread adoption. In this research paper, a new deeplearning approach for nanoparticle segmentation referred to as ResidualFormer is introduced. The method involves employing a Convolutional Neural Network (CNN) encoder with residual blocks to capture spatial features, while the SegFormer-based decoder architecture is utilized to merge multi-scale feature maps. In particular, the images employed in this experiment were taken from SEM datasets of TiO₂ nanoparticles, which have been significantly data augmented. According to the findings made through the experimental approach, ResidualFormer proves to be effective thanks to such metrics as the average Dice score being 0.9540, the average IoU being 0.9121, and the average AUC-ROC being 0.9655, with minimal variability across different folds. Furthermore, it should be noted that, while computing the Dice score, there was a standard deviation of 0.00126, and when computing the IoU, there was one of 0.002329. Based on the given information, it is possible to conclude that, owing to ResidualFormer, it is possible to achieve highly accurate results, without sacrificing robustness.