Image segmentation method for pure iron grain combining multiscale feature aggregation and iterative refinement

This paper proposes a segmentation method that combines convolutional neural network feature aggregation with recurrent neural network iterative refinement to address challenges in segmenting polycrystalline pure iron grains, such as handling surface noise on grain boundaries and repairing missing boundaries. The method effectively addresses issues related to surface noise and missing boundaries, thereby improving the accurate localization of grains. Experimental results demonstrate that the method can effectively remove noise, accurately identify grain boundaries of pure iron grains, and exhibit high computational efficiency.