<p>Complex sheet metal parts are manufactured through multi-stage plastic deformation processes. The geometric configuration design, feature recognition, and segmentation of these forming stages constitute a core challenge in intelligent die design. This demands systematic planning integrating material behavior, process constraints, and tool structure, thus necessitating precise geometric understanding of the part. To address this, we propose a spectral-graph-theory-based approach for analyzing forming-feature attributes. A topological model of geometric features is established to support optimal forming-stage geometry design. Beginning with the 3D surface model, elementary geometric features are extracted, and their mean curvatures are adopted as primary graph signals. A weighted Laplacian matrix is constructed using Gaussian kernel weights, with its eigen decomposition yielding the spectral graph representation of the part. Spectral clustering then partitions this graph into distinct subgraphs, each corresponding to an individual geometric forming feature. This facilitates the establishment of a topological structure graph for forming features, enabling robust feature recognition. Validation via recognition and design case studies demonstrate that the method efficiently translates geometric features into frequency-domain signals through mathematical modeling. This achieves effective feature extraction and classification, confirming its significant potential for advancing process design and manufacturing in sheet-metal forming.</p>