Superpixel Segmentation Based on Feature Fusion and Boundary Constraint for Ferrograph Image Segmentation

Accurate segmentation of wear particles in ferrograph image is pivotal for ferrography analysis. Although morphological processing techniques have made noteworthy strides in wear particle segmentation, issues such as oversegmentation and boundary distortion remain evident. These challenges compromise the segmentation efficiency of prevailing techniques, especially in separating irregular wear particles and delineating wear particle contours. In this study, we introduce an advanced superpixel segmentation technique based on feature fusion and boundary constraint (FBS). Key characteristics of FBS include: 1) the development of an innovative feature fusion framework to cater to wear particles of varied contents in ferrograph images and 2) the implementation of a boundary constraint strategy to refine superpixel boundaries, ensuring alignment with wear particle contours. Experimental results indicate that the proposed method can adeptly segment wear particles, with its performance matching or even surpassing the state-of-the-art segmentation techniques. Furthermore, the FBS method achieves a segmentation accuracy of 97.8% on the ferrograph image dataset.