CN-121996928-A - Brillouin optical fiber sensing noise reduction method based on double-stage DnCNN

CN121996928ACN 121996928 ACN121996928 ACN 121996928ACN-121996928-A

Abstract

The invention discloses a Brillouin optical fiber sensing noise reduction method based on a double-stage DnCNN, which relates to the technical field of distributed optical fiber sensing, and the method comprises the steps of firstly dividing a Brillouin gain curve into a near-end section and a far-end section according to noise level, then normalizing data of the two ends to a 0-1 section, and further normalizing the noise level of the data after normalization to be higher due to small data average value of the far-end section, the effective noise reduction threshold of the convolutional noise reduction neural network DnCNN can be achieved, and then the pre-training DnCNN model matched with the noise levels of the near-end section and the far-end section is used for denoising the two sections of data respectively, namely, double-stage DnCNN noise reduction is adopted, and compared with the traditional single-stage DnCNN noise reduction, the signal to noise ratio of the far-end data is greatly improved.

Inventors

LI ZONGLEI
PENG YULAN
CHEN HAO

Assignees

西南交通大学

Dates

Publication Date: 20260508
Application Date: 20260123

Claims (4)

1. The Brillouin optical fiber sensing noise reduction method based on the double-stage DnCNN is characterized by comprising the following steps of: Acquiring signal data of the scattered light containing an original noise signal in the transmission process of the Brillouin optical fiber sensing equipment, and acquiring a gain curve of Brillouin gain along with the change of distance based on the signal data; Amplifying the gain curve of the far-end section part to ensure that the Brillouin gain of the gain curve is in a normalized gain range so as to improve the noise level represented by the original noise signal of the far-end section part to be above the effective noise reduction threshold of the convolutional noise reduction neural network DnCNN; Training DnCNN a model by using an original noise signal of a near-end section to obtain a near-end pre-training DnCNN model, training DnCNN a model by using Gaussian noise matched with amplified noise level to obtain a far-end pre-training DnCNN model, and denoising gain curves of the near-end section and the far-end section by using the near-end pre-training DnCNN model and the far-end pre-training DnCNN model respectively to obtain gain curves after denoising of the near-end section and the far-end section; And merging the gain curves after denoising of the near-end section and the far-end section into a Brillouin gain denoising curve, and analyzing the temperature and stress distribution along the optical fiber based on the Brillouin gain denoising curve.
2. The dual-stage DnCNN-based brillouin optical fiber sensing noise reduction method according to claim 1, wherein when the gain curve is divided into the near-end stage and the far-end stage, the selection of the division range includes: the segmentation range is determined according to the original noise level and the amplitude of the Brillouin gain attenuated along with the distance; When the noise level of the gain curve of the measured Brillouin gain changing along with the distance is 1/255 after normalization, the noise level is increased to be more than 5/255, and then the gain curve is divided at the position that the gain value of the gain curve is attenuated to 0.2.
3. The dual-stage DnCNN-based brillouin optical fiber sensing noise reduction method according to claim 1, wherein the acquisition of the near-end pre-training DnCNN model and the far-end pre-training DnCNN model includes: Acquiring the relation between the signal-to-noise ratio improvement of the normalized data and the noise level when carrying out DnCNN model training of different noise levels, and measuring that the signal-to-noise ratio improvement brought by image denoising is 0 when the noise level represented by an original noise signal is lower than 1/255 according to the relation between the signal-to-noise ratio improvement and the noise level; When the noise level is increased from 1/255 to 5/255, the signal to noise ratio is increased from 0dB to 12dB, and the convolutional noise reduction neural network DnCNN is trained by using gain curve data with the noise level of 5/255, so that a far-end pre-training DnCNN model matched with the noise level of a far-end section is obtained; the near-end pre-training DnCNN model is obtained by training the convolutional noise reduction neural network DnCNN through gain curve data of a near-end segment.
4. The dual-stage DnCNN-based brillouin optical fiber sensing noise reduction method according to claim 1, wherein the gain curve of the far-end section is amplified, and the amplification factor is dynamically adjusted according to the target noise level, so that the brillouin gain is within the normalized gain range of [0,1], so as to raise the noise level of the far-end section to the effective noise reduction threshold of the convolutional noise reduction neural network DnCNN by between 5/255 and 10/255.

Description

Brillouin optical fiber sensing noise reduction method based on double-stage DnCNN Technical Field The invention relates to the technical field of distributed optical fiber sensing, in particular to a Brillouin optical fiber sensing noise reduction method based on a double-stage DnCNN. Background The Brillouin Optical Time Domain Analysis (BOTDA) is used as a distributed temperature and strain sensing technology with high precision and high stability in many practical applications, and the measurement precision of the BOTDA optical fiber sensor finally depends on the signal-to-noise ratio (SNR) of the measured Brillouin gain, but in the processes of acquisition, transmission, amplification, filtering and the like, the long-distance BOTDA system introduces some unavoidable random noise due to factors such as external environment, and the like, so that the measurement precision of the system is deteriorated, and the sensing effect is affected. There are many methods that have been demonstrated to improve SNR, such as pulse coding, distributed raman amplification, centralized amplification, and image denoising, where image denoising has been demonstrated to be very effective in enhancing the signal-to-noise ratio of BOTDA fiber optic sensors, and the more effective image denoising technique is a convolutional denoising neural network (DnCNN) based on residual learning and batch normalization that can provide signal-to-noise ratio improvement exceeding 10dB without introducing delay and edge blurring, and with GPU-accelerated denoising networks, the data processing time is negligible. However, the noise reduction effectiveness of DnCNN is highly dependent on the noise level of the raw data, but is poor when DnCNN is used for noise reduction due to the long brillouin fiber sensing distance. Disclosure of Invention The embodiment of the invention provides a Brillouin optical fiber sensing noise reduction method based on a double-stage DnCNN, which can solve the problems existing in the prior art. The embodiment of the invention provides a Brillouin optical fiber sensing noise reduction method based on a double-stage DnCNN, which comprises the following steps of: Acquiring signal data of the scattered light containing an original noise signal in the transmission process of the Brillouin optical fiber sensing equipment, and acquiring a gain curve of Brillouin gain along with the change of distance based on the signal data; Amplifying the gain curve of the far-end section part to ensure that the Brillouin gain of the gain curve is in a normalized gain range so as to improve the noise level represented by the original noise signal of the far-end section part to be above the effective noise reduction threshold of the convolutional noise reduction neural network DnCNN; Training DnCNN a model by using an original noise signal of a near-end section to obtain a near-end pre-training DnCNN model, training DnCNN a model by using Gaussian noise matched with amplified noise level to obtain a far-end pre-training DnCNN model, and denoising gain curves of the near-end section and the far-end section by using the near-end pre-training DnCNN model and the far-end pre-training DnCNN model respectively to obtain gain curves after denoising of the near-end section and the far-end section; And merging the gain curves after denoising of the near-end section and the far-end section into a Brillouin gain denoising curve, and analyzing the temperature and stress distribution along the optical fiber based on the Brillouin gain denoising curve. Preferably, when the gain curve is divided into the near-end segment and the far-end segment, the selecting of the division range includes: the segmentation range is determined according to the original noise level and the amplitude of the Brillouin gain attenuated along with the distance; When the noise level of the gain curve of the measured Brillouin gain changing along with the distance is 1/255 after normalization, the noise level is increased to be more than 5/255, and then the gain curve is divided at the position that the gain value of the gain curve is attenuated to 0.2. Preferably, the obtaining of the proximal pre-training DnCNN model and the distal pre-training DnCNN model includes: Acquiring the relation between the signal-to-noise ratio improvement of the normalized data and the noise level when carrying out DnCNN model training of different noise levels, and measuring that the signal-to-noise ratio improvement brought by image denoising is 0 when the noise level represented by an original noise signal is lower than 1/255 according to the relation between the signal-to-noise ratio improvement and the noise level; When the noise level is increased from 1/255 to 5/255, the signal to noise ratio is increased from 0dB to 12dB, and the convolutional noise reduction neural network DnCNN is trained by using gain curve data with the noise level of 5/255, so that a far-end pre-training DnCNN mo