Phase-sensitive optical time-domain reflectometer (Φ-OTDR) utilizes a laser as the detection light source for detecting vibration signals along the optical fiber. However, the spontaneous emission of the laser can lead to phase fluctuations in the optical field, directly impacting the signal-to-noise ratio (SNR) of the phase-demodulated signal. To tackle this issue, a block adaptive denoising method in wavelet domain noise estimation is proposed. The characteristics of phase noise caused by laser spontaneous emission has been analyzed. The phase noise levels at different decomposition scales are extracted using continuous wavelet transform (CWT). By combining with unbiased risk estimation to adjust the block length and threshold of the wavelet coefficients, adaptive denoising for different input signals is achieved. Experimental results demonstrate that compared to untreated signals, at 4.5 km of optical fiber, the SNR of single-frequency signals improves from 40.01 dB to 54.60 dB, and the system’s strain resolution optimizes from 66.15 pε/√Hz to 11.69 pε/√Hz. The SNR of linear swept-frequency signals improves from 18.31 dB to 26.40 dB. In comparison with other denoising algorithms, for single-frequency signals, root mean square error (RMSE) reduces to 0.0096 with an SNR gain of 14.59 dB; for linear swept-frequency signals, RMSE decreases to 0.0809 with an SNR gain of 8.09 dB. The study demonstrates that this method suppresses phase noise while preserving effective signals, thereby improving the accuracy of phase recovery.