Abstract: By effectively isolating noise, deep borehole seismic observations offer clear advantages related to signal identification and seismic source research. However, quantitative assessments of noise reduction and signal-to-noise ratio (SNR) improvements in km-scale boreholes are limited. In this study, a deep borehole seismic observation platform in Changde, China, was used to investigate noise reduction and SNR improvement. Two Chinese TDE-120VB very broadband seismometers were deployed at a depth of 1972.7 m and synchronous observations were made by co-located surface instruments. The results indicate that the background noise power spectral density (PSD) of the borehole system was 20–40 dB lower than that of the surface instruments at 0.001–25 Hz, with decreases of up to 60 dB in certain sub-bands. This corresponds to SNR improvements of 2–4 orders of magnitude. For the first time in deep borehole seismology, spectral entropy was applied to downhole seismic data, and cumulative sum change-point detection within a mean-variance statistical framework was introduced into the entropy series, which enhanced automatic seismic signal identification and noise discrimination. Spectral entropy analysis of the 2025 Dingri earthquake indicates that the borehole station maintained a low-entropy state for ~4.5 h, which is more than twice the length of that observed at the surface (~2 h). The findings support the optimization of deep borehole observation systems and provide a valuable reference for underground observation network development and seismic source research.