In the ultrasonic imaging detection of surface parts, the structure of determined data is three dimensional, thus the rotation, movement and translation operation are commonly needed to observe the defects. However, due to the large amount of determined data, the 3D reconstruction and display speed are very slow. So, the determined data should be in 2D form for defect inspection. And the plane image is usually required to describe the results in the actual detection. Based on the above request, a surface flattening method based fast Isomap algorithm is proposed. First, n points are selected as landmark points. The geodesic distances between the points are computed by the Dijkstra method with Fibonacci heap. Then, MDS method is used to map a set of points into a flat domain. The rest of points coordinates are determined by sample interval. Finally, the weighted optimization method is performed on the flattening meshes. The experimental results indicate that the proposed method can flattening the surface with less distortion. Compared with Isomap method, the proposed algorithm has lower error value and less time, which is about 0.1, 2~4 s, respectively.