Machine vision-based laser screed concrete surface flatness anomaly recognition system

Abstract

This invention relates to the field of image recognition technology, specifically disclosing a machine vision-based system for identifying anomalies in the smoothness of concrete surfaces using a laser screed. The system includes a multispectral acquisition and fusion module, a speckle dynamic compensation module, a spectral noise suppression module, an energy entropy anomaly identification module, and a parameter correlation and tracing module. It acquires reflected light intensity distribution maps and depth point cloud data to generate a fused spatial grid; extracts dynamic gradient change features and combines them with real-time elevation values ​​to compensate for the spatial offset of the point cloud in the fused spatial grid, generating a deformation-compensated speckle map; performs multi-band spectral decomposition on the deformation-compensated speckle map and suppresses water film reflection noise to obtain a substrate texture feature set; constructs an energy entropy matrix to identify abnormal regions where the frequency offset exceeds the offset threshold; and correlates the spatiotemporal location of the abnormal regions with the time-series data of the feed valve opening to generate a smoothness anomaly tracing report. This invention can improve the accuracy and intelligence level of concrete surface smoothness anomaly identification.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a machine vision-based system for identifying abnormal flatness of concrete surfaces on laser leveling machines. Background Technology

[0002] Currently, during the paving and vibration operations of laser screed machines on concrete surfaces, online detection of surface smoothness primarily relies on machine vision and lidar technology. Laser screed machines emit laser beams to form a reference surface, and image sensors collect data on the reflected light intensity and depth of the concrete surface to assess surface quality in real time. In this process, multispectral imaging and point cloud data acquisition are fundamental to building a three-dimensional spatial model, with the key being the accurate capture of minute surface texture undulations to identify potential smoothness defects.

[0003] Existing technologies lack a deep spatiotemporal fusion mechanism for multispectral and point cloud data. This makes it difficult for single-sensor data to accurately compensate for spatial point cloud shifts when encountering dynamic laser speckle shifts, leading to geometric distortions in 3D reconstruction. Furthermore, due to the strong light noise generated by water film reflection on concrete surfaces, existing methods struggle to specifically suppress this spectral interference during multi-band decomposition, resulting in impure extraction of substrate texture features. Moreover, existing recognition logic often analyzes image features in isolation, lacking the ability to combine texture energy entropy with the fundamental frequency characteristic band of the screed's own vibration. This makes it difficult to distinguish frequency shifts caused by equipment vibration from actual concrete surface defects, resulting in low accuracy in identifying anomalous areas. Further, existing systems cannot correlate the spatiotemporal location of identified anomalous areas with construction parameters such as the screed's feed valve opening, making it impossible to trace the root cause of flatness defects. Therefore, there is an urgent need to develop an anomaly identification system with multispectral dynamic compensation, strong light noise suppression, and parameter correlation tracing capabilities to solve the problems of low data fusion, poor anti-interference ability, and inability to trace defects in existing technologies, thereby improving the accuracy and intelligence level of concrete surface flatness anomaly identification. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides a machine vision-based concrete surface smoothness anomaly identification system for laser screed machines, characterized in that the system includes a multispectral acquisition and fusion module, a speckle dynamic compensation module, a spectral noise suppression module, an energy entropy anomaly identification module, and a parameter correlation and tracing module, wherein:

[0005] S1. The multispectral acquisition and fusion module is used to simultaneously acquire the reflected light intensity distribution map and depth point cloud data of the leveled area through the multispectral imaging module, and generate a fused spatial grid.

[0006] S2. The speckle dynamic compensation module is used to extract the dynamic gradient change characteristics of laser speckle in the fused spatial grid, and combine it with the real-time elevation value of the leveling machine to compensate for the point cloud spatial offset of the fused spatial grid, and generate a deformation-compensated speckle map.

[0007] S3. The spectral noise suppression module is used to perform multi-band spectral decomposition on the deformation compensation speckle map, and based on the preset strong light interference range, suppress the water film reflection noise of the deformation compensation speckle map and extract the substrate texture feature set.

[0008] S4. The energy entropy anomaly identification module is used to construct the energy entropy matrix of the base texture feature set and match it with a preset vibration fundamental frequency feature band to identify abnormal regions where the frequency offset exceeds the offset threshold.

[0009] S5. The parameter association and traceability module is used to associate the spatiotemporal location of the abnormal area with the timing data of the feed valve opening of the leveling machine, and generate a flatness anomaly traceability report.

[0010] In a preferred embodiment, when the multispectral acquisition and fusion module performs the synchronous acquisition of reflected light intensity distribution map and depth point cloud data of the leveled area through the multispectral imaging module to generate a fused spatial mesh, it is specifically used for:

[0011] Multiple camera sensors of the multispectral imaging module are controlled to simultaneously acquire the original reflected light intensity distribution map of the leveled area at a preset exposure time;

[0012] Radiometric correction is performed on the original reflected light intensity distribution map to remove ambient light interference, resulting in a radiometrically corrected light distribution map.

[0013] Simultaneously acquire the original depth point cloud data corresponding to the radiation-corrected illumination distribution map, and project the original depth point cloud data onto the three-dimensional coordinate system;

[0014] The radiation-corrected illumination distribution map and the projected original depth point cloud data are spatiotemporally registered to generate registered two-dimensional light intensity and three-dimensional point cloud combined data.

[0015] Based on the registered two-dimensional light intensity and three-dimensional point cloud combined data, a grid topology structure is constructed to obtain the fused spatial grid.

[0016] In a preferred embodiment, when the speckle dynamic compensation module extracts the dynamic gradient change features of laser speckle in the fused spatial grid and combines the real-time elevation value of the leveling machine to compensate for the point cloud spatial offset of the fused spatial grid to generate a deformation-compensated speckle map, it is specifically used for:

[0017] Laser speckle cloud data is segmented from the fused spatial grid, and the spatial displacement vector of the laser speckle in the time series is calculated;

[0018] Based on the spatial displacement vector, a differential gradient function is derived to extract the dynamic gradient change characteristics of the laser speckle.

[0019] The real-time elevation value of the leveling machine is obtained, and the real-time elevation value is fused with the dynamic gradient change characteristics to generate an elevation gradient matrix;

[0020] The spatial offset of the point cloud is compensated by geometric transformation, and the speckle pattern image is reconstructed based on the compensated point cloud data to obtain the deformation-compensated speckle map.

[0021] In a preferred embodiment, when the spectral noise suppression module performs multi-band spectral decomposition on the deformation-compensated speckle map and suppresses water film reflection noise in the deformation-compensated speckle map based on a preset strong light interference range to extract the substrate texture feature set, it is specifically used for:

[0022] The deformation-compensated speckle pattern is imported into a spectral spectrometer for multi-band decomposition to obtain red band data, green band data, and blue band data.

[0023] Within the preset strong light interference range, the water film reflection noise signal in the red band data should be attenuated;

[0024] Morphological closing operations are performed on the attenuated red band data, and the processed red band data is combined with green and blue band data to form a noise-suppressed image.

[0025] Pixel-level texture features, including gray-level difference statistics and co-occurrence matrix properties, are acquired from the noise-suppressed image.

[0026] Pixel-level texture features are normalized and dimensionality reduced to generate a set of feature vectors;

[0027] The set of feature vectors is encapsulated into the base texture feature set.

[0028] In a preferred embodiment, when the energy entropy anomaly identification module constructs the energy entropy matrix of the base texture feature set and matches it with a preset vibration fundamental frequency feature band to identify abnormal regions where the frequency offset exceeds an offset threshold, it is specifically used for:

[0029] Perform a Fast Fourier Transform on the base texture feature set to obtain the frequency domain distribution of the texture features;

[0030] An energy entropy matrix is ​​constructed based on the frequency domain distribution of texture features, and the entropy value information in the energy entropy matrix is ​​obtained.

[0031] A preset vibration fundamental frequency characteristic band is loaded, and based on the entropy information, the peak value of the energy entropy matrix is ​​detected within the vibration fundamental frequency characteristic band.

[0032] Calculate the frequency offset of the peak entropy value. When the frequency offset exceeds a predefined offset threshold, mark the corresponding spatial location of the frequency offset as an abnormal region.

[0033] In a preferred embodiment, when the parameter association and tracing module associates the spatiotemporal location of the abnormal region with the timing data of the feed valve opening of the leveling machine to generate a flatness anomaly tracing report, it is specifically used for:

[0034] Extract timestamp data and location coordinate data from the abnormal region;

[0035] Collect the timing opening data of the feed valve of the leveling machine, and smooth the high-frequency fluctuations of the timing opening data;

[0036] Calculate the rate of change of opening state based on the smoothed time-series opening data;

[0037] Spatial correlation analysis was performed between the opening state change rate and location coordinate data to detect the abnormal flow index of concrete slurry.

[0038] A causal map is generated by combining the aforementioned flow anomaly index and environmental humidity parameters;

[0039] The causal map is encoded into structured source tracing analysis results;

[0040] The output of the source analysis results is a flatness anomaly source tracing report.

[0041] In a preferred embodiment, when the parameter correlation tracing module outputs the tracing analysis result as a flatness anomaly tracing report, it is further specifically used for:

[0042] Acquire real-time temperature and humidity data, and correct the error tolerance of the flatness anomaly tracing report based on the temperature and humidity data;

[0043] The revised report is stored in the local cache of the leveling machine.

[0044] In a preferred embodiment, the energy entropy anomaly identification module, when performing the matching of a preset vibration fundamental frequency characteristic band, is specifically used for:

[0045] Detect the real-time vibration spectrum of the leveling machine;

[0046] When a spectral drift is detected in the real-time vibration spectrum, the fundamental frequency range of the vibration fundamental frequency characteristic band is dynamically updated;

[0047] The anomalous region was re-identified using the updated vibration fundamental frequency characteristic band.

[0048] In a preferred embodiment, when the energy entropy anomaly identification module calculates the frequency offset of the entropy peak value and marks the corresponding spatial location of the frequency offset as an abnormal region when the frequency offset exceeds a predefined offset threshold, it is specifically used for:

[0049] The frequency offset of the entropy peak value is evaluated based on the dynamic flatness deviation index:

[0050] The formula for calculating the dynamic flatness deviation index is as follows:

[0051]

[0052] in, This is the dynamic flatness deviation index. This is the upper limit frequency of the fundamental frequency characteristic band of the vibration. This is the lower limit frequency of the fundamental frequency characteristic band of the vibration. For frequency, The rate of change of entropy with frequency. It is an exponential function. This is a density correction factor based on the concrete grade. This represents the change in surface texture roughness. To preset the reference roughness, To prevent the protection constant from being reduced to zero, For the natural constant Logarithmic function with base 0. The vibration sensitivity coefficient of the leveling machine. The frequency domain texture gradient magnitude;

[0053] When the dynamic flatness deviation index exceeds the offset threshold, the corresponding spatial location of the frequency offset is marked as an abnormal region.

[0054] In a preferred embodiment, when the multispectral acquisition and fusion module performs the synchronous acquisition of the reflected light intensity distribution map and depth point cloud data of the leveled area through the multispectral imaging module to generate a fused spatial mesh, it is also specifically used for:

[0055] Monitor the working speed of the leveling machine and dynamically adjust the exposure time of the multispectral imaging module based on the working speed of the leveling machine;

[0056] The adjusted collected data is input into the spatial grid of the leveling machine to generate a fused spatial grid.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] 1. This invention significantly improves the accuracy and robustness of concrete surface data acquisition in complex construction environments through the collaborative design of a multispectral acquisition and fusion module and a speckle dynamic compensation module. The multispectral acquisition and fusion module simultaneously acquires reflected light intensity distribution maps and depth point cloud data, and generates a high-precision fused spatial grid through radiometric correction and spatiotemporal registration, overcoming the deficiency in existing technologies where single-sensor data cannot accurately represent three-dimensional spatial information. Based on this, the speckle dynamic compensation module extracts the dynamic gradient change characteristics of laser speckle and combines this with the real-time elevation value of the leveling machine to perform geometric transformation compensation for the spatial offset of the point cloud, effectively overcoming the geometric distortion problem of three-dimensional reconstruction caused by equipment vibration or laser speckle drift, generating a more accurate deformation-compensated speckle map. Furthermore, the spectral noise suppression module performs multi-band spectral decomposition on the deformation-compensated speckle map and specifically suppresses water film reflection noise based on a preset strong light interference range, extracting a pure substrate texture feature set. This avoids the negative impact of ambient light interference on subsequent anomaly identification, providing a reliable data foundation for flatness analysis.

[0059] 2. This invention achieves an intelligent closed loop from precise identification of abnormal areas to in-depth tracing of construction parameters by organically combining an energy entropy anomaly identification module and a parameter correlation tracing module. The energy entropy anomaly identification module constructs an energy entropy matrix from the substrate texture feature set and matches it with a preset vibration fundamental frequency feature band. By calculating the frequency offset of the entropy peak value, it accurately marks abnormal areas, solving the problem of existing technologies' difficulty in distinguishing between equipment vibration noise and actual surface defects, and significantly improving the accuracy of anomaly identification. Furthermore, the parameter correlation tracing module performs spatial correlation analysis between the spatiotemporal location of the abnormal area and the time-series data of the leveling machine's feed valve opening, and generates a causal spectrum by combining environmental humidity parameters. Finally, it outputs a structured flatness anomaly tracing report, breaking through the limitation of traditional methods that cannot trace the root cause of defects. This module also has the ability to correct the report error tolerance based on real-time temperature and humidity data, ensuring the reliability of the tracing results under different working conditions. This provides a quantitative basis for real-time optimization of construction parameters and process improvement, greatly improving the intelligent level of concrete surface flatness control. Attached Figure Description

[0060] Figure 1 A system architecture diagram of a machine vision-based laser screed machine concrete surface flatness anomaly recognition system provided in an embodiment of the present invention;

[0061] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments belong to some, but not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “said” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0064] Depending on the context, the word "if" or "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0065] Furthermore, the timing of the steps in the following method embodiments is merely an example and not a strict limitation.

[0066] In practice, the server-side equipment deployed in the machine vision-based laser screed machine concrete surface flatness anomaly recognition system may consist of one or more devices. This machine vision-based laser screed machine concrete surface flatness anomaly recognition system can be implemented as: a business instance, a virtual machine, or hardware equipment. For example, this machine vision-based laser screed machine concrete surface flatness anomaly recognition system can be implemented as a business instance deployed on one or more devices in a cloud node. Simply put, this machine vision-based laser screed machine concrete surface flatness anomaly recognition system can be understood as software deployed on a cloud node, used to provide machine vision-based laser screed machine concrete surface flatness anomaly recognition system to various user terminals. Alternatively, this machine vision-based laser screed machine concrete surface flatness anomaly recognition system can also be implemented as a virtual machine deployed on one or more devices in a cloud node. This virtual machine contains application software for managing various user terminals. Alternatively, the machine vision-based laser screed machine concrete surface flatness anomaly recognition system can also be implemented as a server consisting of numerous identical or different types of hardware devices, with one or more hardware devices set up to provide machine vision-based laser screed machine concrete surface flatness anomaly recognition system to each user terminal.

[0067] In terms of implementation, the machine vision-based laser screed machine concrete surface flatness anomaly recognition system and the user terminal are mutually compatible. Specifically, if the machine vision-based laser screed machine concrete surface flatness anomaly recognition system is implemented as an application installed on a cloud service platform, then the user terminal acts as a client establishing a communication connection with that application; or if the machine vision-based laser screed machine concrete surface flatness anomaly recognition system is implemented as a website, then the user terminal acts as a webpage; or if the machine vision-based laser screed machine concrete surface flatness anomaly recognition system is implemented as a cloud service platform, then the user terminal acts as a mini-program within an instant messaging application.

[0068] like Figure 1 The figure shown is a system architecture diagram of a machine vision-based laser leveling machine concrete surface flatness anomaly recognition system provided in an embodiment of the present invention.

[0069] The machine vision-based laser screed machine concrete surface flatness anomaly identification system 100 described in this invention can be set up in a cloud server. In terms of implementation, it can be used as one or more service devices, or as an application installed in the cloud (e.g., a mobile service operator's server, server cluster, etc.), or it can be developed into a website. Depending on the functions implemented, the machine vision-based laser screed machine concrete surface flatness anomaly identification system 100 may include a multispectral acquisition and fusion module 101, a speckle dynamic compensation module 102, a spectral noise suppression module 103, an energy entropy anomaly identification module 104, and a parameter correlation and tracing module 105. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device's processor and can perform a fixed function, stored in the electronic device's memory.

[0070] In this embodiment of the invention, in the machine vision-based laser screed machine concrete surface flatness anomaly recognition system, each of the above modules can be implemented independently and can be called by other modules. Here, "calling" can be understood as one module connecting to multiple modules of another type and providing corresponding services to those connected modules. In the machine vision-based laser screed machine concrete surface flatness anomaly recognition system provided by this embodiment of the invention, the applicable scope of the system architecture can be adjusted by adding modules and directly calling them without modifying the program code, achieving cluster-style horizontal expansion to quickly and flexibly expand the machine vision-based laser screed machine concrete surface flatness anomaly recognition system. In practical applications, the above modules can be set in the same device or different devices, or they can be set in a virtual device, such as a service instance in a cloud server.

[0071] The following describes, with reference to specific embodiments, each component and its specific workflow of the machine vision-based laser screed machine concrete surface flatness anomaly identification system:

[0072] S1. The multispectral acquisition and fusion module is used to simultaneously acquire the reflected light intensity distribution map and depth point cloud data of the leveled area through the multispectral imaging module, and generate a fused spatial grid.

[0073] In this embodiment of the invention, when the multispectral acquisition and fusion module performs the synchronous acquisition of the reflected light intensity distribution map and depth point cloud data of the leveled area through the multispectral imaging module to generate a fused spatial grid, it is specifically used for:

[0074] Multiple camera sensors of the multispectral imaging module are controlled to simultaneously acquire the original reflected light intensity distribution map of the leveled area at a preset exposure time;

[0075] Radiometric correction is performed on the original reflected light intensity distribution map to remove ambient light interference, resulting in a radiometrically corrected light distribution map.

[0076] Simultaneously acquire the original depth point cloud data corresponding to the radiation-corrected illumination distribution map, and project the original depth point cloud data onto the three-dimensional coordinate system;

[0077] The radiation-corrected illumination distribution map and the projected original depth point cloud data are spatiotemporally registered to generate registered two-dimensional light intensity and three-dimensional point cloud combined data.

[0078] Based on the registered two-dimensional light intensity and three-dimensional point cloud combined data, a grid topology structure is constructed to obtain the fused spatial grid.

[0079] Multiple camera sensors in the multispectral imaging module synchronously acquire raw reflected light intensity distribution maps of the leveled area at preset exposure times. The multispectral imaging module integrates multiple camera sensors, all uniformly set to the same preset exposure time, which is pre-calibrated based on the ambient light intensity of the leveled area and the reflective characteristics of the concrete surface. When the preset exposure time is reached, all camera sensors trigger simultaneously, capturing images of the leveled area from different spectral bands or angles, recording the reflected light intensity information of the area at each band. During acquisition, each camera sensor generates an independent raw reflected light intensity distribution map; these images collectively constitute the raw multispectral reflected light intensity data of the leveled area, recording the raw light intensity value at each pixel location.

[0080] Radiometric correction is performed on the original reflected light intensity distribution map to remove ambient light interference, resulting in a radiometrically corrected light distribution map. The radiometric correction process involves analyzing the grayscale variations in the original reflected light intensity distribution map caused by uneven ambient lighting or shadows. Using a known radiometric correction reference board and a reference image acquired under the same lighting conditions, or by statistically analyzing the uniformity of light distribution within the leveled area, the radiometric correction coefficient for each pixel is calculated. Each pixel value in the original reflected light intensity distribution map is multiplied by its corresponding radiometric correction coefficient, thereby eliminating the influence of ambient light variations. This ensures that the corrected image accurately reflects the reflective properties of the concrete surface, generating the radiometrically corrected light distribution map.

[0081] Simultaneously, the raw depth point cloud data corresponding to the radiometrically corrected illumination distribution map is acquired and projected onto a three-dimensional coordinate system. While acquiring the reflected light intensity distribution map, raw depth point cloud data of the leveled area is obtained using a lidar or depth camera at the same time reference. This data records the distance information from each sampling point to the sensor. The acquired raw depth point cloud data is then input into the coordinate transformation process. Based on the sensor's installation position and attitude parameters, the distance and angle of each point are converted into coordinate values ​​in three-dimensional space using the sensor's own geometric model. This projects the raw depth point cloud data from the sensor coordinate system to a preset three-dimensional world coordinate system or the leveling machine's own coordinate system, forming point cloud data with clearly defined three-dimensional coordinates.

[0082] Spatiotemporal registration is performed between the radiometrically corrected illumination distribution map and the projected original depth point cloud data to generate registered combined 2D illumination intensity and 3D point cloud data. Spatiotemporal registration first confirms that the radiometrically corrected illumination distribution map and the projected original depth point cloud data were acquired at the same time, ensuring temporal consistency. In the spatial dimension, using a pre-calibrated transformation matrix between the camera and the lidar, each 3D point in the projected original depth point cloud data is mapped to the pixel coordinate system of the radiometrically corrected illumination distribution map, allowing each 3D point to obtain a corresponding illumination intensity value. Simultaneously, each pixel in the illumination distribution map can be reverse-mapped to find its corresponding 3D point. This spatial alignment operation generates registered combined 2D illumination intensity and 3D point cloud data, where each record contains both precise 3D coordinates and illumination intensity information at that location.

[0083] A fused spatial mesh is obtained by constructing a mesh topology based on the registered 2D light intensity and 3D point cloud combined data. According to the horizontal coordinate range of the 3D points in the registered 2D light intensity and 3D point cloud combined data, regularly arranged mesh cells are divided on the horizontal plane, each with a fixed size. For each mesh cell, all 3D points falling within the cell and their corresponding light intensity values ​​are collected. The elevation and light intensity values ​​of the mesh cell are calculated using interpolation or weighted averaging methods, thus forming a regular mesh node. All mesh nodes are connected according to their row and column positions to form a fused spatial mesh with a topological structure. Each node in this mesh stores the 3D coordinates and light intensity information of its location, providing a unified spatial data foundation for subsequent speckle dynamic compensation and texture feature extraction.

[0084] The beneficial effects are as follows: By synchronously acquiring the original reflected light intensity distribution map through multiple camera sensors of the multispectral imaging module at a preset exposure time, the temporal consistency of multi-band image data is ensured, providing a synchronous foundation for subsequent fusion. Radiometric correction of the original reflected light intensity distribution map effectively removes interference from uneven ambient lighting and shadows, ensuring that the radiometrically corrected lighting distribution map accurately reflects the reflective characteristics of the concrete surface. Synchronously acquiring the original depth point cloud data and projecting it onto a three-dimensional coordinate system achieves spatial unification of depth and light intensity information. Spatiotemporal registration is performed between the radiometrically corrected lighting distribution map and the projected original depth point cloud data. Through precise spatial mapping, light intensity values ​​are mapped one-to-one with three-dimensional coordinates, generating registered combined two-dimensional light intensity and three-dimensional point cloud data, solving the problem of heterogeneous multi-source data. Based on this combined data, a mesh topology structure is constructed to obtain a fused spatial mesh. This mesh simultaneously contains accurate three-dimensional geometric information and high-fidelity light intensity information, providing high-quality and highly consistent data support for subsequent speckle dynamic compensation, texture feature extraction, and flatness anomaly identification, significantly improving the overall recognition accuracy and robustness of the system.

[0085] S2. The speckle dynamic compensation module is used to extract the dynamic gradient change characteristics of laser speckle in the fused spatial grid, and combine it with the real-time elevation value of the leveling machine to compensate for the point cloud spatial offset of the fused spatial grid, and generate a deformation-compensated speckle map.

[0086] In this embodiment of the invention, when the speckle dynamic compensation module extracts the dynamic gradient change features of laser speckle in the fused spatial grid and combines the real-time elevation value of the leveling machine to compensate for the spatial offset of the point cloud of the fused spatial grid, and generates a deformation-compensated speckle map, it is specifically used for:

[0087] Laser speckle cloud data is segmented from the fused spatial grid, and the spatial displacement vector of the laser speckle in the time series is calculated;

[0088] Based on the spatial displacement vector, a differential gradient function is derived to extract the dynamic gradient change characteristics of the laser speckle.

[0089] The real-time elevation value of the leveling machine is obtained, and the real-time elevation value is fused with the dynamic gradient change characteristics to generate an elevation gradient matrix;

[0090] The spatial offset of the point cloud is compensated by geometric transformation, and the speckle pattern image is reconstructed based on the compensated point cloud data to obtain the deformation-compensated speckle map.

[0091] When the multispectral acquisition and fusion module performs the synchronous acquisition of reflected light intensity distribution map and depth point cloud data of the leveled area through the multispectral imaging module to generate a fused spatial mesh, it is also specifically used for:

[0092] Monitor the working speed of the leveling machine and dynamically adjust the exposure time of the multispectral imaging module based on the working speed of the leveling machine;

[0093] The adjusted collected data is input into the spatial grid of the leveling machine to generate a fused spatial grid.

[0094] Laser speckle cloud data is segmented from the fused spatial grid, and the spatial displacement vector of the laser speckle over time is calculated. The speckle dynamic compensation module first extracts point cloud data related to the laser speckle from the constructed fused spatial grid. This point cloud data records the three-dimensional coordinates of the laser speckle on the concrete surface. By comparing the fused spatial grid at multiple consecutive time points, the positional changes of the same speckle in the grid at adjacent time points are identified. Then, the direction and distance of movement of each speckle over time are calculated. The combination of these directions and distances constitutes the spatial displacement vector of the laser speckle.

[0095] A differential gradient function is derived based on the spatial displacement vector to extract the dynamic gradient variation characteristics of laser speckle. Using the calculated spatial displacement vector, the rate of change of the displacement vector in its spatial distribution is analyzed. By examining the difference in displacement vectors between adjacent speckle points, a differential gradient function that describes the drastic local changes in the displacement field is derived. This function reflects the dynamic characteristics of laser speckle on the concrete surface as a function of time and space, thereby extracting the dynamic gradient variation characteristics of the laser speckle. This characteristic quantifies the spatial non-uniformity of the speckle displacement.

[0096] The real-time elevation value of the screed is acquired and fused with the dynamic gradient change features to generate an elevation gradient matrix. The speckle dynamic compensation module acquires the elevation value of the screed in real-time from its control system or elevation sensor, representing the reference height of the screed plate or laser emitter. The acquired real-time elevation value is fused with the previously extracted laser speckle dynamic gradient change features, and the elevation information is superimposed onto the dynamic gradient change features according to spatial location, forming an elevation gradient matrix containing both elevation and speckle gradient information. This matrix reflects the correlation between the screed height change and the speckle displacement gradient.

[0097] Geometric transformations were performed to compensate for the spatial offset of the point cloud, and the speckle pattern image was reconstructed based on the compensated point cloud data to obtain a deformation-compensated speckle map. Based on the correspondence between the leveling machine height change and speckle displacement reflected in the elevation gradient matrix, the spatial offset of the original point cloud data in the fused spatial grid caused by the leveling machine's movement or vibration was analyzed. Geometric transformation methods were used to reverse these spatial offsets, correcting the coordinates of each point cloud to their theoretical positions, resulting in compensated point cloud data. Subsequently, the compensated point cloud data was remapped onto a two-dimensional image plane according to its spatial position and light intensity information. Interpolation and reconstruction operations were then performed to form a speckle pattern image that accurately reflects the morphology of the concrete surface, i.e., the deformation-compensated speckle map.

[0098] The system monitors the leveling machine's operating speed and dynamically adjusts the exposure time of the multispectral imaging module based on this speed. During acquisition, the multispectral acquisition and fusion module monitors the leveling machine's speed in real time. When the leveling machine's speed changes, it dynamically adjusts the exposure time of the camera sensor in the multispectral imaging module to ensure shorter exposure times during high-speed movement to avoid image blurring, and appropriately longer exposure times during low-speed or stationary movement to increase light intake and guarantee the quality of the acquired images.

[0099] The adjusted acquired data is input into the spatial grid of the leveling machine to generate a fused spatial grid. The original reflected light intensity distribution map and depth point cloud data acquired after dynamic adjustment of exposure time are processed according to the aforementioned radiometric correction, projection, spatiotemporal registration and grid topology construction process. The processed data is finally integrated into the spatial grid of the leveling machine to generate a more accurate fused spatial grid that adapts to changes in working conditions.

[0100] The beneficial effects are as follows: by segmenting laser speckle cloud data from the fused spatial grid and calculating the spatial displacement vector, the motion law of speckle can be accurately captured. Based on the spatial displacement vector, a differential gradient function is derived and dynamic gradient change features are extracted, which quantifies the spatial inhomogeneity of speckle displacement and provides a basis for subsequent compensation. The real-time elevation value of the screed machine is fused with the dynamic gradient change features to generate an elevation gradient matrix, enabling the correlation analysis between equipment status and surface deformation. Geometric transformation compensation is performed on the spatial offset of the point cloud, and the speckle pattern image is reconstructed based on the compensated point cloud data to obtain a deformation-compensated speckle map, effectively eliminating the geometric distortion of the point cloud caused by the movement or vibration of the screed machine and restoring the true morphology of the concrete surface. Furthermore, by monitoring the working speed of the screed machine and dynamically adjusting the exposure time of the multispectral imaging module, the clarity and consistency of the acquired data under different working conditions are ensured. Inputting the adjusted acquired data into the fused spatial grid generated by the spatial grid has higher environmental adaptability. These steps collectively improve the accuracy and anti-interference capability of data acquisition, providing a more reliable data foundation for subsequent texture feature extraction and flatness anomaly identification.

[0101] S3. The spectral noise suppression module is used to perform multi-band spectral decomposition on the deformation compensation speckle map, and based on the preset strong light interference range, suppress the water film reflection noise of the deformation compensation speckle map and extract the substrate texture feature set.

[0102] In this embodiment of the invention, when the spectral noise suppression module performs multi-band spectral decomposition on the deformation-compensated speckle map and suppresses water film reflection noise in the deformation-compensated speckle map based on a preset strong light interference range, and extracts the substrate texture feature set, it is specifically used for:

[0103] The deformation-compensated speckle pattern is imported into a spectral spectrometer for multi-band decomposition to obtain red band data, green band data, and blue band data.

[0104] Within the preset strong light interference range, the water film reflection noise signal in the red band data should be attenuated;

[0105] Morphological closing operations are performed on the attenuated red band data, and the processed red band data is combined with green and blue band data to form a noise-suppressed image.

[0106] Pixel-level texture features, including gray-level difference statistics and co-occurrence matrix properties, are acquired from the noise-suppressed image.

[0107] Pixel-level texture features are normalized and dimensionality reduced to generate a set of feature vectors;

[0108] The set of feature vectors is encapsulated into the base texture feature set.

[0109] The deformation-compensated speckle map is imported into a spectral spectrometer for multi-band decomposition, yielding red, green, and blue band data. The spectral noise suppression module first acquires the generated deformation-compensated speckle map, which contains speckle information from the dynamically compensated concrete surface. This image is then input into the spectral spectrometer, which uses specific spectroscopic elements such as gratings or prisms to separate the incident light according to wavelength, extracting the intensity information for the red, green, and blue bands respectively. This generates three independent band data sets: red, green, and blue band data, each reflecting the reflectivity of the concrete surface within that specific spectral range.

[0110] Within a pre-defined strong light interference range, the water film reflection noise signal in the red band data should be attenuated. Based on this range, determined through experimental calibration or theoretical analysis, the strong light interference range corresponds to the main spectral band occupied by water film reflection noise, which is typically strongest in the red band. Within this range, pixel values ​​in the red band data are attenuated. Specifically, an attenuation coefficient is set, multiplying the bright pixel values ​​in the red band data within this range by an attenuation factor less than one. This weakens the overly bright areas caused by water film reflection, suppressing its adverse effects on subsequent texture analysis, resulting in attenuated red band data.

[0111] Morphological closing is performed on the attenuated red band data, and the processed red band data is then combined with the green and blue band data to form a noise-suppressed image. Morphological closing is an image processing technique that involves first dilating the attenuated red band data to expand the bright areas outwards, filling small holes and cracks. Then, an erosion operation is performed to shrink the expanded area back to near its original size while maintaining connectivity, thereby eliminating dark details and broken gaps. This connects and smooths the discontinuous noise points left after the water film reflection area is weakened. The morphologically closed red band data is then superimposed with the unprocessed green and blue band data according to their corresponding pixel positions to form a new three-channel color image, i.e., the noise-suppressed image.

[0112] Pixel-level texture features, including gray-level difference statistics and co-occurrence matrix attributes, are extracted from noise-suppressed images. A pixel-by-pixel analysis is performed on the noise-suppressed images. First, the gray-level difference between each pixel and its neighboring pixels is calculated, and gray-level difference statistics such as the mean and variance are obtained. These statistics reflect the contrast and drastic changes in local texture. Simultaneously, a gray-level co-occurrence matrix is ​​constructed. This matrix characterizes the roughness, directionality, and periodicity of the texture by statistically analyzing the probability of two pixel gray-level values ​​occurring simultaneously at specific directions and distances in the image. Attribute parameters such as energy, contrast, correlation, and entropy are extracted from the co-occurrence matrix. These attributes collectively constitute a quantitative description of the texture characteristics of the noise-suppressed image.

[0113] Pixel-level texture features are normalized and dimensionality reduced to generate a set of feature vectors. Since the collected gray-level difference statistics and co-occurrence matrix attributes have different dimensions and numerical ranges, normalization maps these feature values ​​to a unified interval, eliminating the influence of dimensions and making different features comparable. Subsequently, the normalized features are dimensionality reduced by analyzing the correlation between features, removing redundant information, and retaining the most representative feature components. This simplifies the original high-dimensional texture features into a set of lower-dimensional but information-rich feature vectors, with each feature vector corresponding to a local region in the noise-suppressed image.

[0114] The feature vector set is encapsulated into a base texture feature set. The generated feature vector set is organized according to its spatial location in the image, forming a structured data set corresponding to the spatial distribution of the noise-suppressed image; this is the base texture feature set. This feature set contains the essential texture information of the concrete surface after removing water film reflection noise, providing clean and representative input data for the subsequent energy entropy anomaly detection module.

[0115] The beneficial effects are as follows: by decomposing the deformation-compensated speckle map into multi-band data to obtain red, green, and blue band data, the separation and processing of different spectral information is achieved. Within a preset strong light interference range, the water film reflection noise signal in the red band data is attenuated, specifically eliminating the strong interference caused by the water film in the red band. Morphological closing operations are performed on the attenuated red band data, effectively filling the holes and breaks left after noise removal and maintaining the continuity of the texture. The processed red band data is combined with the green and blue band data to form a noise-suppressed image, restoring the true color and detail information of the concrete surface. Gray-level difference statistics and co-occurrence matrix attributes are collected from this image, achieving a comprehensive quantitative description of texture features. Pixel-level texture features are normalized and dimensionality reduced to generate a feature vector set, eliminating both feature dimension differences and removing redundant information. The feature vector set is encapsulated into a base texture feature set, providing a clean, refined, and representative texture data foundation for subsequent smoothness anomaly identification, significantly improving the system's adaptability to noisy environments and the accuracy of anomaly identification.

[0116] S4. The energy entropy anomaly identification module is used to construct the energy entropy matrix of the base texture feature set and match it with a preset vibration fundamental frequency feature band to identify abnormal regions where the frequency offset exceeds the offset threshold.

[0117] In this embodiment of the invention, when the energy entropy anomaly identification module constructs the energy entropy matrix of the base texture feature set and matches it with a preset vibration fundamental frequency feature band to identify abnormal regions where the frequency offset exceeds the offset threshold, it is specifically used for:

[0118] Perform a Fast Fourier Transform on the base texture feature set to obtain the frequency domain distribution of the texture features;

[0119] An energy entropy matrix is ​​constructed based on the frequency domain distribution of texture features, and the entropy value information in the energy entropy matrix is ​​obtained.

[0120] A preset vibration fundamental frequency characteristic band is loaded, and based on the entropy information, the peak value of the energy entropy matrix is ​​detected within the vibration fundamental frequency characteristic band.

[0121] Calculate the frequency offset of the peak entropy value. When the frequency offset exceeds a predefined offset threshold, mark the corresponding spatial location of the frequency offset as an abnormal region.

[0122] When the energy entropy anomaly identification module performs matching of a preset vibration fundamental frequency characteristic band, it is specifically used for:

[0123] Detect the real-time vibration spectrum of the leveling machine;

[0124] When a spectral drift is detected in the real-time vibration spectrum, the fundamental frequency range of the vibration fundamental frequency characteristic band is dynamically updated;

[0125] The anomalous region was re-identified using the updated vibration fundamental frequency characteristic band.

[0126] When the energy entropy anomaly identification module calculates the frequency offset of the entropy peak value, and marks the corresponding spatial location of the frequency offset as an abnormal region when the frequency offset exceeds a predefined offset threshold, it is specifically used for:

[0127] The frequency offset of the entropy peak value is evaluated based on the dynamic flatness deviation index:

[0128] The formula for calculating the dynamic flatness deviation index is as follows:

[0129]

[0130] in, This is the dynamic flatness deviation index. This is the upper limit frequency of the fundamental frequency characteristic band of the vibration. This is the lower limit frequency of the fundamental frequency characteristic band of the vibration. For frequency, The rate of change of entropy with frequency. It is an exponential function. This is a density correction factor based on the concrete grade. This represents the change in surface texture roughness. To preset the reference roughness, To prevent the protection constant from being reduced to zero, For the natural constant Logarithmic function with base 0. The vibration sensitivity coefficient of the leveling machine. The frequency domain texture gradient magnitude;

[0131] When the dynamic flatness deviation index exceeds the offset threshold, the corresponding spatial location of the frequency offset is marked as an abnormal region.

[0132] A Fast Fourier Transform (FFT) is performed on the substrate texture feature set to obtain the frequency domain distribution of the texture features. The energy entropy anomaly identification module first acquires the substrate texture feature set generated by the spectral noise suppression module. This feature set represents the texture characteristics of various local regions on the concrete surface in the form of a set of feature vectors. These spatial domain feature vectors are arranged into a two-dimensional matrix according to their corresponding spatial positions. A FFT is then performed on this matrix to transform the texture information from the spatial domain to the frequency domain. The FFT decomposes the matrix into sinusoidal wave components of different frequencies, calculates the amplitude and phase of each frequency component, and finally obtains the frequency domain distribution of the texture features. This distribution reflects the energy intensity of the concrete surface texture at different spatial frequencies.

[0133] An energy entropy matrix is ​​constructed based on the frequency domain distribution of texture features, and the entropy values ​​in the matrix are obtained. Based on the frequency domain distribution of texture features, for each frequency point, the proportion of its energy value in the total energy of the entire frequency domain is calculated, yielding the energy probability distribution for that frequency point. Using the definition of information entropy, the energy probability of each frequency point is multiplied by the logarithm of that probability, and then the sum is taken over all frequency points, with the result negative, to obtain the energy entropy value corresponding to that local region. This process is repeated for all spatial locations, and the calculated energy entropy values ​​for each location are arranged according to their spatial location to form an energy entropy matrix. The specific entropy value for each location is extracted from this matrix; this entropy information quantifies the degree of disorder or uncertainty of the texture features.

[0134] A preset vibration fundamental frequency feature band is loaded, and based on entropy information, the peak entropy value of the energy entropy matrix is ​​detected within the vibration fundamental frequency feature band. The preset vibration fundamental frequency feature band is read from the system's storage unit; this feature band is a frequency range representing the frequency band where vibration energy is concentrated during normal operation of the leveling machine. Based on the upper and lower limits of this feature band, corresponding frequency intervals are defined in the texture feature frequency domain distribution. Within this interval, the maximum value of the entropy information in the energy entropy matrix is ​​searched; the location of this maximum value is the entropy peak value, and the frequency value corresponding to this peak value reflects the frequency point where the current texture has the most concentrated energy within that frequency band.

[0135] The frequency offset of the entropy peak value is calculated. When the frequency offset exceeds a predefined offset threshold, the corresponding spatial location of the frequency offset is marked as an abnormal region. The frequency value corresponding to the detected entropy peak value is compared with the center frequency of the vibration fundamental frequency characteristic band, and the difference between the two is calculated. This difference is the frequency offset. An offset threshold is preset in the system. When the calculated frequency offset is greater than this threshold, it indicates that the frequency domain distribution of the texture features in this local area has deviated significantly from the normal vibration frequency of the leveling machine. At this time, the spatial location corresponding to the entropy peak value, that is, the row and column position of the value in the energy entropy matrix, is mapped back to the actual coordinates of the concrete surface, and this area is marked as an abnormal flatness region.

[0136] The module detects the real-time vibration spectrum of the screed. The energy entropy anomaly identification module continuously collects vibration signals from the screed during operation using vibration sensors installed on it. It then performs real-time spectrum analysis on these signals to obtain the current real-time vibration spectrum of the screed, which reflects the frequency components and energy distribution of the screed's vibration at different times.

[0137] When a spectral drift is detected in the real-time vibration spectrum, the fundamental frequency range of the vibration fundamental frequency characteristic band is dynamically updated. The system monitors changes in the real-time vibration spectrum of the leveling machine. When the peak frequency of the vibration spectrum shifts due to equipment wear, changes in operating conditions, or changes in concrete material properties, this spectral drift phenomenon is identified. Based on the new peak frequency after the drift, the upper and lower limits of the vibration fundamental frequency characteristic band are recalculated and set, thereby dynamically updating the characteristic band.

[0138] The updated vibration fundamental frequency feature band is used to re-identify abnormal areas. Using the updated vibration fundamental frequency feature band as a new benchmark, the above-mentioned entropy peak detection and frequency offset calculation process is performed again on the subsequently acquired substrate texture feature set, so as to accurately identify abnormal areas under new vibration conditions and ensure that the abnormal identification always matches the actual working state of the leveling machine.

[0139] The frequency offset of the entropy peak value is assessed based on the dynamic flatness deviation index. The dynamic flatness deviation index is used to comprehensively evaluate the significance of this offset when calculating the frequency offset. The calculation of this index involves the sources and interactions of multiple parameters.

[0140] The upper and lower limits of the fundamental frequency characteristic band of vibration are extracted from real-time monitoring or preset vibration spectra. The rate of change of entropy with frequency is obtained by differentiating the frequency domain distribution of texture features, reflecting the degree of drastic change in texture energy with frequency.

[0141] The density correction factor based on the concrete grade is predetermined according to the specific grade of the concrete used in construction. Different grades of concrete have different densities and mechanical properties. This factor is used to correct the influence of roughness variations on smoothness.

[0142] The change in surface texture roughness is obtained by analyzing the feature set of the base texture, which quantifies the degree of undulation of the texture in the current local area relative to a flat surface.

[0143] The preset reference roughness is a reference value set based on a standard concrete surface during system initialization. The zero protection constant is a very small positive value used to avoid zero denominators during calculations.

[0144] The vibration sensitivity coefficient of a screed is pre-calibrated based on the mechanical characteristics and vibration transmission characteristics of the screed, and it reflects the sensitivity of vibration to the formation of surface texture.

[0145] The frequency domain texture gradient magnitude is obtained by performing gradient operations on the frequency domain distribution of texture features, and it characterizes the rate of change of texture in the frequency domain.

[0146] The calculation process of the dynamic flatness deviation index organically combines these parameters. Through integral operation, the entropy change rate, roughness-related attenuation factor and texture gradient-related amplification factor are integrated into a comprehensive index. The larger the value of this index, the higher the probability of flatness anomalies in the area.

[0147] When the dynamic flatness deviation index exceeds the preset offset threshold, the spatial location corresponding to the frequency offset is marked as an abnormal area.

[0148] The beneficial effects are as follows: By performing a Fast Fourier Transform on the base texture feature set to obtain the frequency domain distribution of texture features, the transformation of texture information from the spatial domain to the frequency domain is realized, laying the foundation for subsequent frequency analysis. Based on the frequency domain distribution of texture features, an energy entropy matrix is ​​constructed and entropy value information is obtained, quantifying the degree of texture disorder and enabling abnormal areas to exhibit identifiable characteristics in terms of entropy values. By loading a preset vibration fundamental frequency feature band and detecting the peak entropy value, the texture features are correlated with the vibration characteristics of the leveling machine itself, effectively distinguishing texture changes caused by normal equipment vibration from genuine surface defects. The frequency offset of the entropy peak value is calculated and compared with a threshold, achieving objective judgment and accurate marking of abnormal areas. By detecting the real-time vibration spectrum of the leveling machine and dynamically updating the vibration fundamental frequency feature band, the system can adaptively adjust the identification benchmark, avoiding misjudgments caused by changes in equipment status. The frequency offset is evaluated based on a dynamic flatness deviation index, integrating multiple factors such as the rate of change of entropy value, surface roughness change, and texture gradient, making the identification of abnormal areas more comprehensive and accurate, significantly improving the system's robustness and identification accuracy under complex working conditions.

[0149] S5. The parameter association and traceability module is used to associate the spatiotemporal location of the abnormal area with the timing data of the feed valve opening of the leveling machine, and generate a flatness anomaly traceability report.

[0150] In this embodiment of the invention, when the parameter association and tracing module performs the correlation between the spatiotemporal location of the abnormal region and the timing data of the feed valve opening of the leveling machine to generate a flatness anomaly tracing report, it is specifically used for:

[0151] Extract timestamp data and location coordinate data from the abnormal region;

[0152] Collect the timing opening data of the feed valve of the leveling machine, and smooth the high-frequency fluctuations of the timing opening data;

[0153] Calculate the rate of change of opening state based on the smoothed time-series opening data;

[0154] Spatial correlation analysis was performed between the opening state change rate and location coordinate data to detect the abnormal flow index of concrete slurry.

[0155] A causal map is generated by combining the aforementioned flow anomaly index and environmental humidity parameters;

[0156] The causal map is encoded into structured source tracing analysis results;

[0157] The output of the source analysis results is a flatness anomaly source tracing report.

[0158] When the parameter association tracing module outputs the tracing analysis result as a flatness anomaly tracing report, it is also specifically used for:

[0159] Acquire real-time temperature and humidity data, and correct the error tolerance of the flatness anomaly tracing report based on the temperature and humidity data;

[0160] The revised report is stored in the local cache of the leveling machine.

[0161] Timestamp data and location coordinate data are extracted from the abnormal areas. The parameter correlation and tracing module first acquires the flatness anomaly areas marked by the energy entropy anomaly identification module. These anomaly areas have a clear spatial range. For each anomaly area, the system reads the acquisition time corresponding to when the area was identified as an anomaly from its corresponding storage record and records this time as timestamp data. At the same time, the system extracts the specific coordinate range or center point coordinates of the anomaly area in the leveling machine coordinate system or global coordinate system from the coordinate system of the fused spatial grid, forming location coordinate data. The timestamp data and location coordinate data together constitute the spatiotemporal location information of the anomaly area, providing a spatiotemporal reference for subsequent correlation analysis with construction parameters.

[0162] The system collects time-series opening data of the feed valve of the screed and smooths out high-frequency fluctuations in this data. The parameter correlation and traceability module collects the opening value of the feed valve in real time from the screed's control system or sensor network. This opening value reflects the supply of concrete slurry. The data collection process is performed at fixed time intervals, forming a time-varying sequence of opening data, known as the time-series opening data. Because the feed valve may experience momentary minor vibrations or sensor noise during adjustment, the opening data may contain high-frequency fluctuations. A moving average method is used to calculate the average value of each data point in the time-series opening data and several neighboring data points. This average value replaces the original data points, thus filtering out high-frequency fluctuations and obtaining smoothed time-series opening data.

[0163] Based on smoothed time-series opening data, the rate of change of opening status is calculated. The smoothed time-series opening data is analyzed, and the difference between the opening values ​​at two adjacent moments is calculated. This difference is then divided by the time interval to obtain the rate of change of opening status within that time interval. This process is repeated for all adjacent moments to generate a series of rate of change values, forming a sequence of opening status change rates. This sequence reflects the drastic degree and direction of the change in the feed valve opening over time. An increase in the rate of change of opening status indicates a rapid increase or decrease in the feed rate, while a rate of change approaching zero indicates a stable feed rate.

[0164] Spatial correlation analysis is performed between the opening change rate and location coordinate data to detect the concrete slurry flow anomaly index. Location coordinate data extracted from the anomaly area is mapped onto the working path of the leveling machine to determine the construction time corresponding to the anomaly area. The opening change rate sequence near this time is correlated with the spatial location of the anomaly area to analyze whether the feed valve opening changed drastically within the time window of the anomaly area's formation. By calculating the correlation coefficient or matching degree between the opening change rate and the spatial location of the anomaly area, the degree of correlation between the two is quantified, yielding the concrete slurry flow anomaly index. The higher the index, the stronger the correlation between the smoothness defects in the anomaly area and the fluctuation in the feed rate.

[0165] A causal graph was generated by combining the flow anomaly index and environmental humidity parameters. Real-time environmental humidity parameters of the construction area were obtained from environmental monitoring sensors. The calculated flow anomaly index was fused with the environmental humidity parameters for analysis, considering the impact of humidity on concrete flowability and setting properties. This determined whether, under specific humidity conditions, the material supply fluctuations reflected by the flow anomaly index were sufficient to cause smoothness defects. The spatiotemporal location of the anomalous area, the flow anomaly index, the environmental humidity parameters, and the logical relationships among them were organized to form a causal graph describing the causes of the defects. This graph graphically illustrates how material supply fluctuations and environmental factors work together to lead to the appearance of anomalous areas.

[0166] The causal graph is encoded into structured source analysis results. The generated causal graph is encoded and converted according to a pre-defined data structure, transforming node information such as the location of abnormal areas, flow anomaly indices, humidity parameters, and causal relationships between nodes (e.g., causing / exacerbating) into computer-readable formatted data. This process converts graphical causal information into structured text or data files, facilitating subsequent storage, transmission, and display.

[0167] The output of the source tracing analysis results is a flatness anomaly source tracing report. The structured source tracing analysis results are formatted according to a preset report template, generating a complete document that includes the location of the anomaly area, the time of occurrence, related changes in the feed valve opening, the impact of environmental humidity, and an analysis of the cause of the defect. This report is provided to construction management personnel to guide the adjustment of construction parameters and the improvement of processes.

[0168] The system acquires real-time temperature and humidity data and adjusts the error tolerance of the flatness anomaly tracing report based on this data. The parameter correlation tracing module obtains real-time temperature and humidity values ​​of the construction environment from temperature and humidity sensors. Based on the influence of temperature and humidity on concrete setting time and surface forming quality, the system dynamically adjusts the judgment thresholds or error ranges used in the flatness anomaly tracing report. For example, in high-temperature, low-humidity environments, where concrete moisture evaporates quickly, the error tolerance can be appropriately relaxed; in low-temperature, high-humidity environments, the error tolerance can be appropriately tightened. The adjusted error tolerances are applied to the data analysis and conclusion determination in the report to ensure the report's adaptability and accuracy under different environmental conditions.

[0169] The corrected report is stored in the leveling machine's local buffer. The leveling anomaly traceability report, after error tolerance correction, is written to the leveling machine's local buffer, which uses non-volatile storage media to ensure the report is not lost after the leveling machine is shut down or powered off. The stored report can be used for subsequent historical data retrieval, quality traceability, or remote transmission.

[0170] The beneficial effects are as follows: By extracting timestamps and location coordinates from abnormal areas, precise spatiotemporal positioning is provided for the correlation analysis between abnormal events and construction parameters. High-frequency fluctuations in the timing data of the feed valve's opening position are collected and smoothed, eliminating sensor noise and valve vibration interference, and obtaining data reflecting the actual trend of material supply changes. Based on the smoothed data, the rate of change in the opening position is calculated, quantifying the dynamic adjustment process of the material supply. Spatial correlation analysis is performed between the rate of change in the opening position and location coordinates, and a flow anomaly index is detected, realizing a quantitative correlation between flatness defects and fluctuations in material supply parameters. A causal graph is generated by combining the flow anomaly index and environmental humidity parameters, intuitively displaying the multiple factors contributing to defect formation and their logical relationships. The causal graph is encoded into a structured source tracing analysis result and output as a flatness anomaly source tracing report, providing clear directions for construction process optimization. Furthermore, by acquiring real-time temperature and humidity data and adjusting the error tolerance of the report accordingly, the source tracing results can adapt to different environmental conditions, improving the reliability and practicality of the report. The revised report was stored in the leveling machine's local cache, ensuring data security and traceability, and accumulating valuable historical data for subsequent quality analysis and process improvement.

[0171] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0172] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0173] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A machine vision-based system for identifying abnormal surface flatness of concrete surfaces in laser screed machines, characterized in that, The system includes a multispectral acquisition and fusion module, a speckle dynamic compensation module, a spectral noise suppression module, an energy entropy anomaly identification module, and a parameter correlation and tracing module, wherein: S1. The multispectral acquisition and fusion module is used to simultaneously acquire the reflected light intensity distribution map and depth point cloud data of the leveled area through the multispectral imaging module, and generate a fused spatial grid. S2. The speckle dynamic compensation module is used to extract the dynamic gradient change characteristics of laser speckle in the fused spatial grid, and combine it with the real-time elevation value of the leveling machine to compensate for the point cloud spatial offset of the fused spatial grid, and generate a deformation-compensated speckle map. S3. The spectral noise suppression module is used to perform multi-band spectral decomposition on the deformation compensation speckle map, and based on the preset strong light interference range, suppress the water film reflection noise of the deformation compensation speckle map and extract the substrate texture feature set. S4. The energy entropy anomaly identification module is used to construct the energy entropy matrix of the base texture feature set and match it with a preset vibration fundamental frequency feature band to identify abnormal regions where the frequency offset exceeds the offset threshold. S5. The parameter association and traceability module is used to associate the spatiotemporal location of the abnormal area with the timing data of the feed valve opening of the leveling machine, and generate a flatness anomaly traceability report.

2. The machine vision-based concrete surface flatness anomaly identification system for laser screed machines as described in claim 1, characterized in that, When the multispectral acquisition and fusion module performs the synchronous acquisition of reflected light intensity distribution map and depth point cloud data of the leveled area through the multispectral imaging module to generate a fused spatial mesh, it is specifically used for: Multiple camera sensors of the multispectral imaging module are controlled to simultaneously acquire the original reflected light intensity distribution map of the leveled area at a preset exposure time; Radiometric correction is performed on the original reflected light intensity distribution map to remove ambient light interference, resulting in a radiometrically corrected light distribution map. Simultaneously acquire the original depth point cloud data corresponding to the radiation-corrected illumination distribution map, and project the original depth point cloud data onto the three-dimensional coordinate system; The radiation-corrected illumination distribution map and the projected original depth point cloud data are spatiotemporally registered to generate registered two-dimensional light intensity and three-dimensional point cloud combined data. Based on the registered two-dimensional light intensity and three-dimensional point cloud combined data, a grid topology structure is constructed to obtain the fused spatial grid.

3. The machine vision-based concrete surface flatness anomaly identification system for laser screed machines as described in claim 1, characterized in that, The speckle dynamic compensation module, when extracting the dynamic gradient change characteristics of laser speckle in the fused spatial grid and combining it with the real-time elevation value of the leveling machine to compensate for the spatial offset of the point cloud in the fused spatial grid, and generating a deformation-compensated speckle map, is specifically used for: Laser speckle cloud data is segmented from the fused spatial grid, and the spatial displacement vector of the laser speckle in the time series is calculated; Based on the spatial displacement vector, a differential gradient function is derived to extract the dynamic gradient change characteristics of the laser speckle. The real-time elevation value of the leveling machine is obtained, and the real-time elevation value is fused with the dynamic gradient change characteristics to generate an elevation gradient matrix; The spatial offset of the point cloud is compensated by geometric transformation, and the speckle pattern image is reconstructed based on the compensated point cloud data to obtain the deformation-compensated speckle map.

4. The machine vision-based concrete surface flatness anomaly identification system for laser screed machines as described in claim 1, characterized in that, The spectral noise suppression module, when performing multi-band spectral decomposition on the deformation-compensated speckle map and suppressing water film reflection noise in the deformation-compensated speckle map based on a preset strong light interference range, and extracting the substrate texture feature set, is specifically used for: The deformation-compensated speckle pattern is imported into a spectral spectrometer for multi-band decomposition to obtain red band data, green band data, and blue band data. Within the preset strong light interference range, the water film reflection noise signal in the red band data should be attenuated; Morphological closing operations are performed on the attenuated red band data, and the processed red band data is combined with green and blue band data to form a noise-suppressed image. Pixel-level texture features, including gray-level difference statistics and co-occurrence matrix properties, are acquired from the noise-suppressed image. Pixel-level texture features are normalized and dimensionality reduced to generate a set of feature vectors; The set of feature vectors is encapsulated into the base texture feature set.

5. The machine vision-based concrete surface flatness anomaly identification system for laser screed machines as described in claim 1, characterized in that, When the energy entropy anomaly identification module constructs the energy entropy matrix of the base texture feature set and matches it with a preset vibration fundamental frequency feature band to identify abnormal regions where the frequency offset exceeds the offset threshold, it is specifically used for: Perform a Fast Fourier Transform on the base texture feature set to obtain the frequency domain distribution of the texture features; An energy entropy matrix is ​​constructed based on the frequency domain distribution of texture features, and the entropy value information in the energy entropy matrix is ​​obtained. A preset vibration fundamental frequency characteristic band is loaded, and based on the entropy information, the peak value of the energy entropy matrix is ​​detected within the vibration fundamental frequency characteristic band. Calculate the frequency offset of the peak entropy value. When the frequency offset exceeds a predefined offset threshold, mark the corresponding spatial location of the frequency offset as an abnormal region.

6. The machine vision-based concrete surface flatness anomaly identification system for laser screed machines as described in claim 1, characterized in that, When the parameter association and tracing module associates the spatiotemporal location of the abnormal area with the timing data of the feed valve opening of the leveling machine to generate a flatness anomaly tracing report, it is specifically used for: Extract timestamp data and location coordinate data from the abnormal region; Collect the timing opening data of the feed valve of the leveling machine, and smooth the high-frequency fluctuations of the timing opening data; Calculate the rate of change of opening state based on the smoothed time-series opening data; Spatial correlation analysis was performed between the opening state change rate and location coordinate data to detect the abnormal flow index of concrete slurry. A causal map is generated by combining the aforementioned flow anomaly index and environmental humidity parameters; The causal map is encoded into structured source tracing analysis results; The output of the source analysis results is a flatness anomaly source tracing report.

7. The machine vision-based concrete surface flatness anomaly identification system for laser screed machines as described in claim 6, characterized in that, When the parameter association tracing module outputs the tracing analysis result as a flatness anomaly tracing report, it is also specifically used for: Acquire real-time temperature and humidity data, and correct the error tolerance of the flatness anomaly tracing report based on the temperature and humidity data; The revised report is stored in the local cache of the leveling machine.

8. The machine vision-based concrete surface flatness anomaly identification system for laser screed machines as described in claim 1, characterized in that, When the energy entropy anomaly identification module performs matching of a preset vibration fundamental frequency characteristic band, it is specifically used for: Detect the real-time vibration spectrum of the leveling machine; When a spectral drift is detected in the real-time vibration spectrum, the fundamental frequency range of the vibration fundamental frequency characteristic band is dynamically updated; The anomalous region was re-identified using the updated vibration fundamental frequency characteristic band.

9. The machine vision-based concrete surface flatness anomaly identification system for laser screed machines as described in claim 5, characterized in that, When the energy entropy anomaly identification module calculates the frequency offset of the entropy peak value, and marks the corresponding spatial location of the frequency offset as an abnormal region when the frequency offset exceeds a predefined offset threshold, it is specifically used for: The frequency offset of the entropy peak value is evaluated based on the dynamic flatness deviation index: The formula for calculating the dynamic flatness deviation index is as follows: ； in, This is the dynamic flatness deviation index. This is the upper limit frequency of the fundamental frequency characteristic band of the vibration. This is the lower limit frequency of the fundamental frequency characteristic band of the vibration. For frequency, The rate of change of entropy with frequency. It is an exponential function. This is a density correction factor based on the concrete grade. This represents the change in surface texture roughness. To preset the reference roughness, To prevent the protection constant from being zero, For the natural constant Logarithmic function with base 0. The vibration sensitivity coefficient of the leveling machine. The frequency domain texture gradient magnitude; When the dynamic flatness deviation index exceeds the offset threshold, the corresponding spatial location of the frequency offset is marked as an abnormal region.

10. The machine vision-based concrete surface flatness anomaly identification system for laser screed machines as described in claim 2, characterized in that, When the multispectral acquisition and fusion module performs the synchronous acquisition of reflected light intensity distribution map and depth point cloud data of the leveled area through the multispectral imaging module to generate a fused spatial mesh, it is also specifically used for: Monitor the working speed of the leveling machine and dynamically adjust the exposure time of the multispectral imaging module based on the working speed of the leveling machine; The adjusted collected data is input into the spatial grid of the leveling machine to generate a fused spatial grid.

Images