The paper introduces a novel point-based representation called Gaussian surfels, which combines the advantages of flexible optimization in 3D Gaussian points and the surface alignment property of surfels. By setting the z-scale of 3D Gaussian points to zero, the method effectively flattens the ellipsoid into a 2D ellipse, improving optimization stability and surface alignment. The method includes a self-supervised normal-depth consistency loss to address the issue of zero derivatives with respect to the local z-axis. Monocular normal priors and foreground masks are incorporated to enhance reconstruction quality. A volumetric cutting method is proposed to remove erroneous points in depth maps generated by alpha blending. The screened Poisson reconstruction method is applied to extract the surface mesh from the fused depth maps. Experimental results show that the method achieves superior performance in surface reconstruction compared to state-of-the-art neural volume rendering and point-based rendering methods, demonstrating a good balance between reconstruction quality and training speed.The paper introduces a novel point-based representation called Gaussian surfels, which combines the advantages of flexible optimization in 3D Gaussian points and the surface alignment property of surfels. By setting the z-scale of 3D Gaussian points to zero, the method effectively flattens the ellipsoid into a 2D ellipse, improving optimization stability and surface alignment. The method includes a self-supervised normal-depth consistency loss to address the issue of zero derivatives with respect to the local z-axis. Monocular normal priors and foreground masks are incorporated to enhance reconstruction quality. A volumetric cutting method is proposed to remove erroneous points in depth maps generated by alpha blending. The screened Poisson reconstruction method is applied to extract the surface mesh from the fused depth maps. Experimental results show that the method achieves superior performance in surface reconstruction compared to state-of-the-art neural volume rendering and point-based rendering methods, demonstrating a good balance between reconstruction quality and training speed.