The increasing demand for precise metrology and accurate annotations in semiconductor manufacturing, particularly in the analysis of Scanning Electron Microscopy (SEM) and Transmission Electron Microscopy (TEM) images, requires advanced segmentation techniques. These images are critical for evaluating the physical and dimensional characteristics of devices at the nanoscale. However, the inherent low contrast and noise in SEM/TEM images present substantial challenges for traditional image segmentation methods, often leading to misinterpretation of object boundaries and diminished accuracy.