Many methods employ unsigned distance field (UDF) to reconstruct open surfaces and complex internal structures. However, due to the non-differentiability of UDF at the surface, neural networks suffer from inaccuracies in learning UDF values and gradients near the zero level set, leading to UDF-based methods struggle to generate high-quality surfaces. Therefore, we propose an UDF learning method that constructs geometric prior constraints via cross product of raw point cloud. We achieve this by aligning gradients and cross product results, combining with chamfer distance optimization to learn more continuous unsigned distance field. Extensive experiments demonstrate that our method outperforms the state-of-the-art methods in both accuracy and the quality of reconstructed surfaces.