Topologically associating domains (TADs) are fundamental three-dimensional structural units of mammalian genomes and play critical roles in regulating genome organization and gene expression. However, the extreme sparsity of single-cell high-throughput chromosome conformation capture (single-cell Hi-C) data substantially degrades the features of TAD-like domains, posing significant challenges for their accurate identification. In this study, we proposed RecTAD, a single-cell TAD-like domain identification method based on Hi-C contact matrix reconstruction. Specifically, RecTAD reconstructed single-cell Hi-C contact matrices by integrating graph embedding with spectral propagation, thereby enhancing the underlying community information. Subsequently, TAD-like domains were identified from the reconstructed contact matrices using multi-scale Haar features. Extensive experimental results on downsampled bulk Hi-C data, simulated single-cell Hi-C data, and experimental single-cell Hi-C datasets demonstrate that RecTAD achieves superior identification accuracy and exhibits strong biological relevance. These results indicate that RecTAD provides an effective technical approach for analyzing three-dimensional genome architecture and investigating its regulatory mechanism at the single-cell level. The source code of RecTAD is publicly available at: https://github.com/doubletigerzju/RecTAD.