This study proposes an adaptive filtering non-local mean image (NLM) denoising algorithm based on the gradient of two-dimensional discrete functions and the optical intensity loss along the optical fiber, and applies it to the distributed optical fiber sensing system based on Brillouin scattering. On the basis of non-local mean filtering, in order to improve the signal-to-noise ratio of the sensing data of the Brillouin optical time-domain reflectometer (BOTDR) and solve the problem that the large difference in the noise standard deviation between the sensing rear end and the front end caused by the loss along the optical fiber affects the measurement accuracy and spatial resolution, this algorithm establishes relationships between the filtering coefficient and three parameters [the vertical gradient operator of Brillouin gain spectrum (BGS), the sensing distance, and the loss along the optical fiber] through a mathematical model. The denoising intensity is concentrated near the peak of the BGS and increases with the distance, so as to design an image noise reduction algorithm suitable for the distributed optical fiber sensing system. The research results show that compared with the original NLM algorithm, the proposed algorithm achieves an 8.56 dB improvement in the signal-to-noise ratio, and the standard deviation of the Brillouin frequency shift (BFS) is reduced by 0.91 MHz. In the face of the common multi-location and remote temperature change situations in engineering applications, the influence of the denoising of the proposed algorithm on the temperature accuracy is reduced by 15.29 percentage points, and the spatial resolution loss is controlled within 10%. In the field of distributed optical fiber sensing applications, the proposed algorithm improves the performance and cost-effectiveness of Brillouin detection instruments, and enhances its engineering practicability.