CN-122023752-A - Effective signal enhancement and interference separation method for ground penetrating radar underground disease data

Abstract

The invention discloses a ground penetrating radar underground disease signal enhancement and interference separation method which is used for improving the detection precision of holes, void, looseness and leakage. The method comprises the steps of denoising B-scan data, normalizing amplitude and processing a time window to construct clean input, generating curvature guiding features by using a curvature modeling network to enhance disease reflection, extracting and strengthening disease features by using a time domain-frequency domain dual-domain attention network, simultaneously inhibiting earth surface and multipath interference, and separating the features into background and disease abnormal components by using low-rank sparse decomposition to highlight disease signals. Finally, generating an enhancement chart by fusing all the features, and realizing automatic identification and distinguishing of diseases. The method does not need artificial characteristic engineering, is suitable for various underground disease detection scenes and has high precision, robustness and good generalization.

Inventors

• SHI WENXING
• YANG FENG
• Li Fanruo
• PENG SUPING

Assignees

• 中国矿业大学(北京)

Dates

Publication Date

20260512

Application Date

20260409

Claims (5)

1. 1. An effective signal enhancement and interference separation method for ground penetrating radar underground disease data, comprising the following steps: s1, acquiring B-scan data of an underground ground penetrating radar, and performing amplitude standardization, time window calibration, DC drift removal and noise suppression on the data; S2, constructing a curvature guide feature network, estimating the reflection curvature in the B-scan data, and generating a curvature guide feature; S3, constructing a time domain and frequency domain double-domain feature extraction network, respectively carrying out time domain feature extraction and frequency domain feature extraction on the preprocessed data, and weighting the double-domain features through an attention mechanism; S4, carrying out low-rank sparse decomposition on the dual-domain features to obtain a low-rank background component and a sparse abnormal component; and S5, fusing the curvature guide feature, the double-domain feature and the sparse abnormal component to obtain an enhanced feature.
2. 2. The method of claim 1, wherein the curvature guide feature is obtained by means of a second derivative, a gradient operator or a local curve fitting.
3. 3. The method of claim 1, wherein the dual domain feature extraction network achieves feature fusion by a time domain and frequency domain attention weighting mechanism.
4. 4. The method of claim 1, wherein the low rank sparse decomposition is implemented by matrix decomposition to separate background structures from anomaly characteristics.
5. 5. The method of claim 1, wherein the method is implemented based on a deep learning framework and trained with multi-source data.

Description

Effective signal enhancement and interference separation method for ground penetrating radar underground disease data Technical Field The invention belongs to the technical field of underground target detection and signal processing, and particularly relates to a data enhancement method applied to underground disease detection of a ground penetrating radar, in particular to a deep learning signal enhancement and interference separation method aiming at diseases such as holes, void, loose, leakage and the like. Background Diseases such as underground cavities, void, loose and leakage are ubiquitous in road, tunnel and pipe network structures, and the method has important significance for effectively identifying the diseases. The ground penetrating radar has nondestructive testing and rapid scanning capability, and is widely applied to underground disease detection, but B-scan echo data of the ground penetrating radar is easily affected by surface reflection, metal edges, multipath scattering, environmental noise and other interference, so that background and abnormal characteristics in signals are aliased, and weak disease reflection is difficult to effectively separate. The existing method is independent of single characteristics of time domain or frequency domain for analysis, and multi-domain information is difficult to fully utilize. The traditional feature extraction method is insufficient in utilization of the supercurve reflecting structure generated by the underground diseases, and lacks effective modeling for geometric features. The existing method generally cannot effectively distinguish stable stratum background from local abnormal reflection, so that detection results are easily affected by interference. Therefore, it is necessary to provide an underground disease signal processing method combining curvature feature modeling, time-frequency dual-domain feature enhancement and low-rank sparse decomposition, so as to realize effective signal enhancement and interference separation and improve the accuracy and stability of underground disease identification. Disclosure of Invention In view of the above, the invention provides an effective signal enhancement and interference separation method for underground disease data of a ground penetrating radar, which realizes the deep learning signal enhancement and interference separation of underground diseases such as holes, void, looseness, leakage and the like. In order to achieve the above purpose, the invention adopts the following technical scheme: the invention provides a ground penetrating radar multi-frequency characteristic-based underground disease detection method, which comprises the following steps: (1) B-scan data preprocessing (S1) is to acquire and preprocess ground penetrating radar B-scan data of the underground area, and perform amplitude standardization, time window calibration, DC drift removal and noise suppression processing on echo data. (2) And (2) constructing a curvature guide characteristic network based on reflection curvature modeling, carrying out curvature estimation on supercurve reflection generated by underground diseases (including hollowness, void, looseness and seepage) in the B-scan data, generating a reflection curvature guide graph, and enhancing curvature sensitivity characteristics related to the diseases. (3) The time domain-frequency domain double-domain attention enhancement network (S3) is constructed, a time domain-frequency domain double-domain attention feature extraction network is constructed, the preprocessed B-scan data are respectively input into a time domain convolution network and a frequency domain transformation network, and enhancement of disease reflection energy and suppression of noise such as earth surface reflection, metal edge interference, multipath scattering and the like are realized through a double-domain attention mechanism. (4) And (S4) introducing a low-rank sparse interference separation network, and decomposing the feature map after the dual-domain enhancement into a low-rank background component and a sparse disease component, wherein the low-rank component is used for representing a stable stratum background structure, and the sparse component is used for representing local abnormal reflection caused by holes, void, looseness and leakage. (5) And the cross-domain fusion module (S5) fuses the curvature guide characteristic, the double-domain attention output characteristic and the sparse disease characteristic to generate a disease enhancement map, so as to realize the obvious presentation and distinction of the underground disease. The invention has the following beneficial effects: 1. Through reflection curvature modeling and curvature guiding feature network, accurate curvature estimation can be carried out on supercurve reflection formed by diseases, disease features such as weak reflection and thin layer abnormality which are difficult to identify by a traditional method are