CN-121982434-A - Foundation pit slope deformation monitoring method based on image recognition

CN121982434ACN 121982434 ACN121982434 ACN 121982434ACN-121982434-A

Abstract

The invention relates to the technical field of foundation pit engineering safety monitoring, in particular to a foundation pit side slope deformation monitoring method based on image recognition. And inputting the descriptors into a multi-level feature fusion network to perform cross-region association analysis and consistency verification to obtain a relative displacement vector set. And starting a multi-hypothesis deformation deduction flow to generate a potential deformation mode, performing geometric rationality evaluation and stability influence labeling by using a historical deformation case library, and iteratively optimizing and selecting a slope integral deformation field and generating a monitoring report. According to the method, the spatial consistency of the displacement field is improved through the fusion network, and the reliable conversion of the displacement data to the integral deformation mode with definite engineering meaning is realized by using case-driven deduction optimization.

Inventors

WANG SHENNI
LI HONGRU
Qiang Chunpeng
LI JIAQI

Assignees

西北有色勘测工程有限责任公司
西安理工大学

Dates

Publication Date: 20260505
Application Date: 20260409

Claims (10)

1. The foundation pit slope deformation monitoring method based on image recognition is characterized by comprising the following steps of: constructing an initial slope characteristic map through optical image acquisition equipment deployed at a fixed measuring station; Performing region segmentation and key point calibration on the initial slope characteristic map according to the design parameters of slope engineering to form a plurality of monitoring sub-region maps corresponding to the slope structure parts; Performing space-time reference unification and feature enhancement processing on each monitoring sub-region map to generate a region feature descriptor with high identification degree; inputting all the regional feature descriptors into a multi-level feature fusion network to perform cross-regional association analysis and feature consistency verification, and calculating to obtain a relative displacement vector set of each monitoring subarea based on the output of the multi-level feature fusion network; aiming at the relative displacement vector set, starting a multi-hypothesis deformation deduction flow to generate a plurality of potential deformation modes conforming to a mechanical rule; Performing geometric rationality evaluation and stability influence labeling on each potential deformation mode by utilizing a historical deformation case library, fusing the results of the geometric rationality evaluation and the stability influence labeling, and executing iterative optimization in a deformation mode decision space to select a slope integral deformation field; And quantifying the whole deformation field of the side slope into specific deformation parameters, and generating a deformation monitoring report containing positions and variation according to the deformation parameters.
2. The method for monitoring deformation of a foundation pit slope based on image recognition according to claim 1, wherein the constructing an initial slope characteristic map comprises: Collecting sequence monitoring images containing a target foundation pit slope according to a preset time interval; Preprocessing the acquired sequence monitoring images, wherein the preprocessing comprises distortion correction, brightness equalization and image registration, and a standardized monitoring image sequence is generated; extracting texture features and structural features of the slope surface from the standardized monitoring image sequence, and constructing an initial slope feature map; Preprocessing the acquired sequence monitoring images, wherein the preprocessing comprises distortion correction, brightness equalization and image registration, and generating a standardized monitoring image sequence comprises the following steps: performing geometric distortion correction on the sequence monitoring image by using an internal reference matrix and a distortion coefficient of the optical image acquisition equipment, and eliminating the influence of lens deformation; performing brightness and color equalization processing on the corrected image by adopting an algorithm based on histogram matching, and eliminating the influence of illumination change of different periods; Selecting a first frame image in the sequence monitoring images as a reference image, and performing spatial alignment on all subsequent images and the reference image by adopting a characteristic point matching and perspective transformation matrix calculation method; outputting the images subjected to the distortion correction, brightness equalization and image registration processing to form the standardized monitoring image sequence with consistent comparability in time and space.
3. The method for monitoring deformation of a foundation pit slope based on image recognition according to claim 2, wherein the step of extracting texture features and structural features of the slope surface from the standardized monitoring image sequence to construct an initial slope feature map comprises the steps of: applying a multi-scale texture filter group to carry out convolution operation on each frame image of the standardized monitoring image sequence, and extracting a texture response graph representing the slope surface material; Synchronously applying an edge detection operator and a corner detection operator to process each frame of image, and extracting a structural feature point set representing the geometrical outline of the side slope and key connection points; Carrying out space superposition and association coding on the texture response graph and the structural feature point set at the same time point to form a slope feature graph of a single frame; integrating all the slope feature graphs of the single frames according to the time sequence, and establishing a tracking relation of the feature points in the time dimension, so as to construct the initial slope feature map containing space-time information.
4. The method for monitoring deformation of a foundation pit slope based on image recognition according to claim 3, wherein the performing region segmentation and key point calibration on the initial slope characteristic map according to the design parameters of the slope engineering to form a plurality of monitored sub-region maps corresponding to the slope structure part comprises: Reading design parameters of the slope engineering, wherein the design parameters at least comprise the grading platform position of the slope, the slope foot line, the slope top line and the boundary of the important structure; mapping the design parameters into a space coordinate system of the initial slope characteristic map to serve as a reference boundary for region segmentation; Cutting the initial slope characteristic map into a plurality of independent map units according to the reference boundary, wherein each map unit corresponds to a structural part; automatically calibrating representative key points for subsequent deformation calculation in each map unit according to texture uniformity and structural feature point density; and storing each map unit containing the calibrated key points as an independent monitoring subarea map respectively.
5. The method for monitoring deformation of foundation pit slope based on image recognition according to claim 4, wherein the performing space-time reference unification and feature enhancement processing on each monitored sub-region map to generate a region feature descriptor with high recognition degree comprises: Establishing a local coordinate system for each monitoring subarea map, and converting historical position data of all key points in the monitoring subarea map into a corresponding local coordinate system; Smoothing and filtering the position data of each key point on the time sequence to inhibit measurement jitter caused by image noise; according to the uniqueness of the surrounding textures of the key points, constructing appearance feature vectors of the key points by adopting a feature description algorithm; and combining and normalizing the smoothed position data of all key points in the same monitoring subarea map and the appearance characteristic vector thereof to generate the area characteristic descriptor representing the overall state of the monitoring subarea.
6. The method for monitoring deformation of a foundation pit slope based on image recognition according to claim 5, wherein the inputting all the regional feature descriptors into a multi-level feature fusion network for cross-regional correlation analysis and feature consistency verification comprises: the multi-level feature fusion network comprises independent input channels for receiving each regional feature descriptor and a shared feature interaction layer; Each independent input channel firstly carries out deep feature coding on the input regional feature descriptors; the coded features are sent to the shared feature interaction layer, and feature correlation weights between any two monitoring subareas are calculated in the feature interaction layer; According to the characteristic correlation weight, carrying out weighted fusion on the characteristics of the monitoring subareas with high correlation, and feeding back to each independent input channel to correct the initial coding characteristics; through repeated iterative coding, interaction and correction processes, a group of optimized region state characteristics with global consistency and without local mismatching are output.
7. The method for monitoring deformation of a foundation pit slope based on image recognition according to claim 6, wherein the calculating a set of relative displacement vectors of each monitored sub-region based on the output of the multi-level feature fusion network comprises: analyzing the optimized regional state characteristics output by the multi-level characteristic fusion network, and separating the key point set coordinates of each monitoring sub-region at the latest monitoring moment and the initial reference moment from the optimized regional state characteristics; For each monitoring subarea, calculating the integral rigid body transformation parameters of the coordinates of the key point set from the initial reference moment to the latest monitoring moment; Correcting the initial key point set coordinates by applying the integral rigid body transformation parameters, and then calculating the residual displacement of each key point; counting the average value and the direction distribution of residual displacement of all key points in each monitoring subarea, and defining an average value vector as the relative displacement vector of the monitoring subarea; and summarizing the relative displacement vectors of all the monitoring subareas to form the relative displacement vector set.
8. The method for monitoring deformation of a foundation pit slope based on image recognition according to claim 7, wherein the starting a multi-hypothesis deformation deduction process for the set of relative displacement vectors to generate a plurality of potential deformation modes conforming to a mechanical rule comprises: Based on the motion trend of each subarea reflected by the relative displacement vector set, presetting a plurality of typical slope deformation mechanical models as a hypothesis basis; substituting the relative displacement vector set into each slope deformation mechanical model respectively to perform forward calculation by taking the relative displacement vector set as a boundary condition, and estimating a continuous displacement field in the whole slope range; Each deduced displacement field forms a potential deformation mode, and the potential deformation mode comprises predicted displacement data of each point of the slope surface; and for each potential deformation mode, calculating fitting residual errors of the potential deformation modes and the measured relative displacement vector set at the position of the corresponding subarea, and taking the fitting residual errors as a preliminary measure of mode reliability.
9. The method for monitoring deformation of a foundation pit slope based on image recognition according to claim 8, wherein the performing geometric rationality evaluation and stability influence labeling on each potential deformation mode by using the historical deformation case library comprises: Retrieving the occurred cases similar to the geological conditions, the support forms and the current deformation stage of the current foundation pit from the historical deformation case library; extracting the final deformation morphological characteristics of each generated case, and taking the final deformation morphological characteristics as a rationality template; Carrying out geometrical morphology similarity calculation on each potential deformation mode and all matched rationality templates; Meanwhile, based on the displacement field corresponding to each potential deformation mode, evaluating the potential influence degree of the displacement field on the overall stability of the slope by adopting a simplified mechanical criterion, and classifying and marking; outputting the geometric similarity score and the stability influence class label of each potential deformation mode.
10. The method for monitoring deformation of a foundation pit slope based on image recognition according to claim 9, wherein the fusing the results of the geometric rationality evaluation and the stability influence labeling performs iterative optimization in a deformation mode decision space, and the selected overall deformation field of the slope comprises: Constructing a two-dimensional decision space taking geometric similarity scoring and stability influence category as coordinate axes, wherein each potential deformation mode occupies a point in the two-dimensional decision space; according to the conservation principle of engineering monitoring, higher decision weight is given to the stability influence category, and the points in the space are weighted and adjusted; In the decision space after the weighting adjustment, a dominant mode cluster is identified by adopting a clustering algorithm, and a potential deformation mode corresponding to a point with the highest geometric similarity score is selected from the dominant mode cluster; Based on the selected potential deformation mode, using the deduced displacement field as an initial value, performing reverse optimization fitting on the initial value and the measured relative displacement vector set again, and finely adjusting deformation field parameters; and determining the continuous displacement field which is obtained after the parameter fine adjustment and has the highest matching degree with the measured data as the whole deformation field of the side slope.

Description

Foundation pit slope deformation monitoring method based on image recognition Technical Field The invention relates to the technical field of foundation pit engineering safety monitoring, in particular to a foundation pit slope deformation monitoring method based on image recognition. Background The foundation pit side slope deformation monitoring mainly relies on single-point measurement technologies such as total stations and GNSS, displacement information of discrete points is obtained, and continuous deformation fields of the side slope surface cannot be represented. The monitoring method based on image recognition calculates the multi-point displacement through feature matching, and although the data density is improved, each feature point or local area is still treated as an independent unit. The conventional image processing method lacks modeling capability of inherent deformation relevance of the slope as a continuous mechanics whole, so that contradiction can exist in the output displacement vector set, and the real deformation is difficult to distinguish from the abnormality caused by image noise and local shielding. There are two main drawbacks to the prior art solutions. The method is characterized in that the displacement information fragmentation is formed by a plurality of isolated and unverified local measurement values of relevance, the overall harmony and mechanical rationality of a displacement field on spatial distribution cannot be ensured, the reliability of subsequent analysis is affected, engineering semantic conversion capability is lacked, the existing method is terminated by outputting displacement values, discrete and noisy displacement data cannot be converted into an overall deformation mode with definite stability significance, and engineering risk assessment and decision are difficult to directly support. A monitoring method is needed, which can extract the whole displacement field with space and mechanical consistency from the image and establish a reliable reasoning link from the bottom displacement data to the deformation mode and stability research of the high-rise engineering. The present invention aims to solve the above-mentioned problems. Disclosure of Invention The invention aims to solve the defects in the prior art, and provides a foundation pit slope deformation monitoring method based on image recognition. In order to achieve the purpose, the invention adopts the following technical scheme that the foundation pit side slope deformation monitoring method based on image recognition comprises the following steps: constructing an initial slope characteristic map through optical image acquisition equipment deployed at a fixed measuring station; Performing region segmentation and key point calibration on the initial slope characteristic map according to the design parameters of slope engineering to form a plurality of monitoring sub-region maps corresponding to the slope structure parts; Performing space-time reference unification and feature enhancement processing on each monitoring sub-region map to generate a region feature descriptor with high identification degree; inputting all the regional feature descriptors into a multi-level feature fusion network to perform cross-regional association analysis and feature consistency verification, and calculating to obtain a relative displacement vector set of each monitoring subarea based on the output of the multi-level feature fusion network; aiming at the relative displacement vector set, starting a multi-hypothesis deformation deduction flow to generate a plurality of potential deformation modes conforming to a mechanical rule; Performing geometric rationality evaluation and stability influence labeling on each potential deformation mode by utilizing a historical deformation case library, fusing the results of the geometric rationality evaluation and the stability influence labeling, and executing iterative optimization in a deformation mode decision space to select a slope integral deformation field; And quantifying the whole deformation field of the side slope into specific deformation parameters, and generating a deformation monitoring report containing positions and variation according to the deformation parameters. As a further scheme of the invention, the construction of the initial slope characteristic map comprises the following steps: Collecting sequence monitoring images containing a target foundation pit slope according to a preset time interval; Preprocessing the acquired sequence monitoring images, wherein the preprocessing comprises distortion correction, brightness equalization and image registration, and a standardized monitoring image sequence is generated; extracting texture features and structural features of the slope surface from the standardized monitoring image sequence, and constructing an initial slope feature map; Preprocessing the acquired sequence monitoring images, wherein the preprocessing c