Although Scaffold-GS improves the point-based 3D Gaussian splatting method by clustering anchor points with neural Gaussians, effectively reducing redundant Gaussians and enhancing rendering quality, it heavily relies on Structure from Motion (SfM) results. This dependency, especially in few-view scenes, exposes significant limitations in its anchor growth strategy, leading to a considerable decrease in scene reconstruction accuracy. To overcome this issue, we present a Spatial-Aware Anchor Growth (SAAG) strategy. This strategy analyzes the spatial distribution relationship between neural Gaussians and anchor points to accurately locate under-reconstructed regions and generate candidate anchor points, thus reducing the dependency on high-quality initial point clouds. Subsequently, a dual filtering mechanism is employed to optimize the selection of candidate anchors, effectively preventing the generation of redundant anchors and ensuring efficient coverage of under-reconstructed areas with the fewest anchors, enabling robust anchor expansion. To further enhance reconstruction accuracy, we extract depth information and edge features from the scene and use regularization techniques to correct the localization of neural Gaussians, thereby accurately recovering the scene's geometric structure. Extensive experiments on multiple benchmark datasets demonstrate significant improvements in reconstruction accuracy with our method. For example, in the campus scene of the UrbanScene3D dataset, our method achieves a PSNR and SSIM improvement of 1.29 dB and 0.037, respectively.