Due to the complex surface profiles of blades, the current measurement point planning does not adequately capture the distinct zones, leading to low measurement efficiency and challenges in meeting online detection requirements. This study proposes a novel approach to blade zoning measurement point planning. First, it utilizes an improved centroid Voronoi extreme model and smoothness indicators to select seed points during the zoning process. Following that, feature tensor-driven regional clustering is carried out based on these seed points. Finally, measurement points are determined through adaptive sampling of profile deviations along the feature stacking axis within each zone. Simulation and real experiments demonstrate that the measurement points generated by this method effectively represent the blade surface's distinctive features, reducing the number of measurement points by 91.7% and significantly enhancing efficiency.