Search

CN-122017778-A - Method and system for restraining reinforcement signal and reconstructing reinforced concrete structural defects

CN122017778ACN 122017778 ACN122017778 ACN 122017778ACN-122017778-A

Abstract

The invention relates to the technical field of nondestructive testing and intelligent signal processing, in particular to a method and a system for inhibiting reinforced bar signals and reconstructing reinforced concrete structural defects. The method has the advantages that the characteristic modeling is carried out through a transducer architecture, a dynamic attention mask mechanism is introduced, the explicit modeling and suppression are carried out on the steel bar echo, meanwhile, the defect related characteristics are enhanced, the problem that defect information is easy to lose due to dependence on manual experience parameters or simple filtering in the traditional method is solved, meanwhile, on the basis of suppressing steel bar interference, the enhancement and spatial reconstruction are carried out on weak reflection defect signals such as cracks and holes in concrete, so that a reconstruction result has higher definition and higher interpretation, and the image quality of the ground penetrating mines in a complex steel bar environment and the reliability of defect identification are improved. In addition, the end-to-end trainable network structure is adopted, so that the method has good generalization capability and noise resistance, can be adapted to detection scenes under different structural forms and different construction environments, and has higher engineering application value.

Inventors

  • WANG HUILI
  • Gao Yuanbiao

Assignees

  • 大连理工大学

Dates

Publication Date
20260512
Application Date
20260122

Claims (10)

  1. 1. The method for inhibiting the defects of the reinforced concrete structure by using the reinforced bar signals is characterized by comprising the following steps of: Performing fixed-size two-dimensional patch division on an original B-scanned GPR image, performing linear projection processing on each patch, merging position codes, generating a sequence embedded vector, and completing format conversion from two-dimensional imaging data to serialization input; inputting the sequence embedded vector into a transducer encoder, carrying out global feature modeling by means of a multi-head self-attention module and a feedforward neural network, and extracting reinforcement response features; determining a reinforcement echo region and constructing a reinforcement mask by a self-adaptive threshold algorithm according to the reinforcement energy diagram, and applying controllable weight attenuation to the characteristics of the reinforcement echo region by using the dynamic mask attention, and simultaneously enhancing the characteristic response of the defect related region; Inputting the features subjected to dynamic mask attention processing into a transducer decoder, and reconstructing the GPR image subjected to defect signal enhancement through multi-layer feature fusion and spatial relationship recovery operation; and adopting a joint loss function to synchronously optimize the reinforcement inhibition effect and the defect reconstruction quality.
  2. 2. The method for suppressing defects of a reinforced concrete structure according to claim 1, wherein the sequence embedding vector is: Wherein: for the initial input sequence to be entered, Is an original B-swept GPR image; Embedding a function for the patch; is a position code.
  3. 3. The method for suppressing and reconstructing defects of a reinforced concrete structure according to claim 1, wherein the global feature modeling mode of the multi-head self-attention module and the feedforward neural network is as follows: Wherein, the Is that The characteristics of the layers are such that, Is that The characteristics of the layers are such that, Is a multi-head self-attention module; For feed-forward neural networks, encoder output characteristics are utilized ; 、 、 Respectively representing a query vector, a key vector and a value vector which are obtained by linear mapping of input features, The dimension size of the key vector is represented, A function for normalizing the attention weight is shown.
  4. 4. The method for suppressing defects of a reinforced concrete structure according to claim 1, wherein the method for generating the reinforced energy map is as follows: Wherein, the Is an output characteristic of the encoder and, Representing a matrix of weights that are to be used, The amount of offset is indicated and, Representing the activation function and, Is a reinforcing steel bar energy diagram.
  5. 5. The method for suppressing defects of a reinforced concrete structure according to claim 1, wherein the method for constructing a reinforcement mask is as follows: Wherein, the In order to adapt the threshold value to be used, Is an indication function; The dynamic mask attention is: Wherein, the Is a reinforcement inhibition strength factor.
  6. 6. The method for inhibiting and reconstructing defects of a reinforced concrete structure according to claim 1, wherein the GPR image after the defect signal enhancement is: Wherein, the The output is a reconstructed defect image, which is a transducer decoder.
  7. 7. The method for suppressing defects of a reinforced concrete structure according to claim 1, wherein the joint loss function is: Wherein, the Reconstructing an error for the defect; The precision of the recognition and mask for the steel bars; as a noise suppression term, Is a weight coefficient.
  8. 8. A system for reinforcement signal suppression and reinforced concrete structure defect reconstruction, comprising: the image sequence embedding module is used for dividing an original B-scanning GPR image into two-dimensional patches with fixed sizes, performing linear projection processing on each patch, merging the linear projection processing into position codes, generating a sequence embedding vector, and completing format conversion from two-dimensional imaging data to serialization input; The system comprises a reinforcement bar characteristic extraction and energy map generation module, a reinforcement bar response characteristic extraction and energy map generation module, a reinforcement bar energy map generation module and a reinforcement bar energy map generation module, wherein the reinforcement bar characteristic extraction and energy map generation module inputs a sequence embedded vector into a transducer encoder, and develops global characteristic modeling by means of a multi-head self-attention module and a feedforward neural network to extract reinforcement bar response characteristics; The method comprises the steps of constructing a reinforcement mask, constructing a dynamic attention suppression module, determining a reinforcement echo region and constructing the reinforcement mask through a self-adaptive threshold algorithm according to a reinforcement energy diagram, applying controllable weight attenuation to the characteristics of the reinforcement echo region by utilizing the dynamic mask attention, and enhancing the characteristic response of a defect related region; The defect image reconstruction module inputs the characteristics subjected to dynamic mask attention processing into a transducer decoder, and a GPR image with enhanced defect signals is obtained through reconstruction by multi-layer characteristic fusion and spatial relationship recovery operation; and the multi-objective joint optimization module adopts a joint loss function to synchronously optimize the reinforcement inhibition effect and the defect reconstruction quality.
  9. 9. An electronic device comprising a memory, a processor and a computer program stored for execution on the memory, wherein the processor, when executing the program, implements a method for reinforcement signal suppression and reinforced concrete structure defect reconstruction as claimed in any one of claims 1 to 7.
  10. 10. A computer readable storage medium having stored thereon a computer program which when executed by a processor implements a method of reinforcement signal suppression and reinforced concrete structure defect reconstruction as claimed in any one of claims 1 to 7.

Description

Method and system for restraining reinforcement signal and reconstructing reinforced concrete structural defects Technical Field The invention relates to the technical field of nondestructive testing and intelligent signal processing, in particular to a method and a system for inhibiting reinforced bar signals and reconstructing reinforced concrete structural defects. Background Ground Penetrating Radar (GPR) is used as a core technical means in the field of nondestructive testing of concrete structures, is widely applied to internal disease identification of bridges, tunnels and various building structures by virtue of the advantages of non-destructiveness and high efficiency, and provides an important technical support for infrastructure safety evaluation. However, in an actual engineering detection scene, steel bars commonly distributed in a concrete structure can strongly scatter radar waves to form high-amplitude strong reflection echoes, the echoes show obvious hyperbolic characteristics in a GPR image, the energy of the echoes is far higher than weak reflection signals generated by deep defects such as cracks and holes, the signals are very easy to cover or interfere with the defect signals, the defect recognition difficulty is greatly increased, the imaging quality is obviously reduced, and the accuracy of detection results is seriously affected. In the prior art, related researches on road and underground structure disease identification of GPR images are carried out, but the related researches still have obvious limitations. For example, patent CN120783024A proposes a method for detecting urban road hidden diseases based on YOLO11n-GPR, realizes automatic identification of cracks and holes through a lightweight target detection model, but does not deal with the problem of reinforcement interference, the defect identification accuracy is greatly affected by reinforcement reflection, patent CN119206354A proposes a method for discriminating underground structure images based on KAN, only can realize image level identification of defect types, can not output accurate position information of defects, and is difficult to meet engineering actual demands, patent CN111965711B proposes a method for simulating depth development of road reflection cracks based on GPR image forward modeling, only analyzes the evolution rule of cracks from a theoretical level, does not consider the problem of shielding defect signals by reinforcement strong reflection, and has limited practicability. In summary, in the prior art, the characteristic differences of the strong reflection and the weak reflection of the reinforced steel bar are not effectively distinguished from the signal modeling layer, so that the defect information in the reinforced concrete structure is easily shielded and difficult to reconstruct accurately, the overall recognition precision and engineering applicability are limited, and the high-precision detection requirement of complex structures such as bridges, urban roads and the like is difficult to meet Disclosure of Invention The invention aims to provide a method and a system for inhibiting a reinforced bar signal and reconstructing a reinforced concrete structure defect, which solve the problem that the defect identification and the defect signal are difficult to effectively separate and reconstruct in the detection of a reinforced bar strong reflection interference in the detection of a ground penetrating radar of the reinforced concrete structure. According to a first aspect of the embodiments of the present disclosure, there is provided a method for suppressing a reinforced signal and reconstructing a defect of a reinforced concrete structure, including the steps of: Performing fixed-size two-dimensional patch division on an original B-scanned GPR image, performing linear projection processing on each patch, merging position codes, generating a sequence embedded vector, and completing format conversion from two-dimensional imaging data to serialization input; inputting the sequence embedded vector into a transducer encoder, carrying out global feature modeling by means of a multi-head self-attention module and a feedforward neural network, and extracting reinforcement response features; determining a reinforcement echo region and constructing a reinforcement mask by a self-adaptive threshold algorithm according to the reinforcement energy diagram, and applying controllable weight attenuation to the characteristics of the reinforcement echo region by using the dynamic mask attention, and simultaneously enhancing the characteristic response of the defect related region; Inputting the features subjected to dynamic mask attention processing into a transducer decoder, and reconstructing the GPR image subjected to defect signal enhancement through multi-layer feature fusion and spatial relationship recovery operation; and adopting a joint loss function to synchronously optimize the reinforcement inhibi