The small-scale carbonate fracture-cavity reservoirs developed in the deep Tarim Basin exhibit a typical heterogeneous "beaded" seismic response signature, but the accuracy of their reservoir identification and spatial delineation is susceptible to interference from background seismic reflections of surrounding strata. Therefore, building upon conventional SVD for extracting fracture-cavity reservoir responses, this study addresses uncertainties in iterative selection of truncated singular values. We develop a filtering operator through DTW-based similarity constraint of right singular vectors. With energy-ratio constraints, enabling simultaneous seismic response separation and denoising. This establishes a novel singular vector similarity-constrained SVD method. Based on testing and validation with both synthetic and actual seismic data, it is found that the truncated singular value corresponding to the "inflection point" in the DTW similarity distance between right singular vectors, which can effectively distinguish background formation reflections from seismic responses of fractured-cavity bodies. This method further highlights seismic responses of fracture-cavity bodies while significantly suppressing background reflections and random noise. The enhanced beaded weak signals provide critical support for subsequent seismic facies identification and reservoir prediction.