Wide-angle Synthetic Aperture Radar (WSAR) imaging accounts for multi-azimuthal scattering and is feasible for retrieving more comprehensive features of complex targets. This paper proposes a civilian vehicle target discrimination algorithm. The algorithm firstly images the vehicle target data separately for each azimuth, and then calculates the local representative point of each sub-image to generate a spiral three-dimensional point cube of the target attributed scattering centers. The point cube contains both the attributed scattering centers and azimuth information of the target, so it is easier to classify than the two-dimensional imaging without azimuth. Manifold learning methods can be used to reduce the dimension of spiral three-dimensional point cube to two dimensions. A simple and fast loose iterative Multi- Dimensional Scaling (MDS) algorithm for dimensionality reduction is presented in this paper. Finally, the Convolution Neural Network (CNN) is used to train the two-dimensional attributed scattering centers with azimuth information, and the discrimination results are obtained. The proposed method is demonstrated on Gotcha WSAR dataset.