In view of the fact that there are pseudo boundary points close to the real hole boundary points in the model, the large hole recognition algorithm cannot extract the complete boundary point set and is prone to misrecognition of boundary points. A fusion of first-order tensor and second-order tensor is proposed. Hole boundary feature point detection method based on voting algorithm. First, establish a new attenuation function to achieve the preliminary extraction of the boundary points of the first-order rod tensor voting in the neighborhood; then perform the eigenvalue and saliency of the semi-definite matrix obtained by the addition of the second-order rod, board, and ball tensors correspondence analysis, further extract the boundary points of the hole; finally, the boundary point set extracted twice is combined, and the noise is removed at the same time to achieve the purpose of detecting the boundary point of the hole. Experimental results show that this method can effectively eliminate the influence of false boundary points on the extraction of hole boundary features, can achieve a more ideal detection effect, is more robust to noise, and has a low algorithm complexity.