Wall surface flatness detection method based on laser area array radar

Abstract

The invention relates to the technical field of wall flatness detection, discloses a wall flatness detection method based on a laser area array radar, and aims to improve the efficiency and accuracy of wall flatness detection. According to the method, an area array laser radar system is used for scanning a wall surface, and reflected laser signals are collected and preprocessed. And constructing a three-dimensional point cloud model of the wall surface, and accurately matching the preprocessed signal with the wall surface model by using the spatial resolution capability of the laser radar to form a three-dimensional laser reflection image of the wall surface. The image is analyzed, features are extracted and classified through a machine learning algorithm, and the uneven position, type and degree of the wall surface are automatically recognized and marked. And constructing an evaluation model to evaluate the wall flatness, outputting a flatness score, and combining the flatness score with the detection report to provide a comprehensive wall flatness detection report. According to the method, the efficiency and accuracy of wall flatness detection are improved, the limitation of a traditional method is overcome, and the method has a wide application prospect.

Description

Technical Field

[0001] The present invention relates to the technical field of wall surface flatness detection, and in particular to a wall surface flatness detection method based on a laser array radar. Background Art

[0002] In the fields of construction engineering, decoration maintenance, and quality inspection, wall flatness is a key indicator for measuring wall quality. Traditional wall flatness detection methods mostly rely on manual measurement, which is not only inefficient, but also easily affected by subjective factors, making it difficult to ensure the accuracy and consistency of the measurement results. With the advancement of science and technology, optical measurement technology has gradually been introduced into the detection of wall flatness in order to improve measurement efficiency and accuracy. Early optical measurement methods mostly used single-point or line scanning methods. Compared with manual measurement, these methods have improved measurement efficiency to a certain extent, but there are still obvious limitations. The measurement range of single-point or line scanning methods is limited, and it is impossible to cover the entire wall at one time, resulting in insufficient data density and difficulty in fully and accurately reflecting the flatness of the wall.

[0003] In recent years, laser radar technology has been widely used in many fields due to its high precision, high efficiency and non-contact measurement characteristics, such as 3D modeling, topographic mapping, object recognition, etc. In particular, the array laser radar system can simultaneously obtain spatial information in a large range, has a higher data acquisition speed and density, and provides new possibilities for accurate detection of wall flatness. It is not easy to apply laser array radar to wall flatness detection, and it still faces many technical challenges: how to effectively collect and process the laser signal reflected by the wall is a key issue. Factors such as wall material, color, and texture may affect the reflection characteristics of the laser signal; how to construct an accurate 3D point cloud model is also a technical problem. The accuracy of the point cloud model directly affects the reliability of subsequent wall flatness analysis; it is also necessary to accurately map the laser signal to the wall model to form a 3D laser reflection image. In this process, the problem of spatial coordinate matching between the signal and the model needs to be solved. How to extract and classify the features in the 3D laser reflection image to automatically identify the uneven position, type and degree of the wall is also a key step in achieving automated detection. Summary of the invention

[0004] The purpose of the present invention is to provide a wall flatness detection method based on laser array radar to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a wall flatness detection method based on a laser array radar, the method comprising:

[0006] Use the area array laser radar system to scan the wall and collect the reflected laser signal from the wall;

[0007] Preprocessing the collected reflected laser signal;

[0008] Construct a 3D point cloud model of the wall, and use the spatial resolution capability of the LiDAR to accurately match the pre-processed reflected laser signal with the spatial coordinates of the wall model. Use the point cloud imaging algorithm to map the laser signal to the wall model to form a 3D laser reflection image of the wall.

[0009] Analyze the three-dimensional laser reflection image, use machine learning algorithms to extract and classify the features in the image, automatically identify and mark the uneven position, type and degree of the wall surface, and generate a flatness detection report;

[0010] Construct an evaluation model to evaluate the flatness of the wall and output a flatness score;

[0011] The evaluation model is used to score the flatness of the wall surface, and the scoring results are combined with the flatness test report to provide a comprehensive wall flatness test report.

[0012] Preferably, the step of mapping the laser signal onto the wall model by using a point cloud imaging algorithm to form a three-dimensional laser reflection image of the wall includes:

[0013] S101: performing point cloud processing on the preprocessed reflected laser signal, and constructing a point cloud matrix P using point cloud data acquired by multiple scan lines of the laser radar, wherein each row of P corresponds to point cloud data of a scan line, and each column corresponds to the three-dimensional coordinates of a point;

[0014] S102: According to the geometric layout of the laser radar and the registration principle of the point cloud data, the normal vector of the point cloud matrix P is calculated to obtain the normal vector matrix N, where the elements of N represent the normal directions of different points; wherein the normal vector is calculated using the following formula:

[0015]

[0016] Among them, P i , P i+1 , P i+2 are three adjacent point cloud data points, N i For point P i The normal vector of

[0017] S103: using the normal vector matrix N and the three-dimensional point cloud model of the wall, performing spatial coordinate matching, mapping the information in the normal vector to the corresponding spatial position on the wall model, and obtaining an initial three-dimensional laser reflection image I_0;

[0018] S104: performing image optimization processing on the initial reflection image I_0 to obtain an optimized reflection image I_e;

[0019] S105: According to the reflection characteristics and geometric features of the wall material, the intensity threshold T_h and shape recognition rule R of the reflection signal are set, and the optimized reflection image I_e' is analyzed point by point, and the points whose signal intensity exceeds the threshold T_h and meets the shape recognition rule R' are marked as potential uneven areas to obtain the marked reflection image I_m.

[0020] Preferably, the types of wall surface unevenness identified include bulges, depressions, cracks, peeling or deformation of the wall surface.

[0021] Preferably, the image optimization processing step S104 further includes:

[0022] Perform contrast enhancement on the initial reflection image I_0;

[0023] Use edge detection algorithm to enhance edge features in images;

[0024] Apply a denoising algorithm to smooth the image.

[0025] Preferably, the three-dimensional laser reflection image is analyzed, and features in the image are extracted and classified using a machine learning algorithm to automatically identify and mark the uneven position, type and degree of the wall surface. The specific steps include:

[0026] S201: construct a deep convolutional neural network model, which includes multiple convolutional layers, pooling layers, fully connected layers and output layers; wherein the convolutional layer is used to extract local features in the reflection image, the pooling layer is used to reduce the feature dimension, the fully connected layer is used to integrate features and perform high-level abstraction, and the output layer is used to classify and annotate the location, type and degree of unevenness;

[0027] S202: preparing a training data set, which includes a large number of labeled wall lidar reflection images, each of which is labeled with specific location, type and degree information of the unevenness, for training a deep convolutional neural network model;

[0028] S203: input the training data set into the deep convolutional neural network model, calculate the error between the model prediction result and the true annotation through the back propagation algorithm and the gradient descent method, and iteratively adjust the weight and bias parameters in the model until the error of the model on the training data set converges to below a preset threshold;

[0029] S204: Validate the trained deep convolutional neural network model and use an independent validation dataset to evaluate the recognition accuracy, recall rate, and F1 score of the model;

[0030] S205: Input the wall lidar reflection image to be detected into the trained and verified deep convolutional neural network model. The model automatically extracts features, classifies and annotates the image, and outputs the specific location, type and degree of unevenness.

[0031] Preferably, the evaluation model is constructed based on a support vector machine (SVM) algorithm.

[0032] Preferably, the construction process of the evaluation model includes:

[0033] S301: Setting the kernel function of the model to radial basis function RBF;

[0034] S302: Collect training data, which includes a wall surface flatness score and corresponding reflected laser signal features, including signal strength, point cloud density, and normal vector change rate;

[0035] S303: input the training data into the SVM model, and optimize the parameters of the model, including the penalty parameter C and the kernel function parameter γ, through the model training algorithm until the performance of the model on the validation data set reaches a preset standard;

[0036] S304: Test the trained SVM model and use an independent test data set to evaluate the scoring accuracy and stability of the model.

[0037] Preferably, the construction of the evaluation model also includes a cross-validation step, using a K-fold cross-validation method to divide the training data into K subsets, one of which is used as a validation set in turn, and the rest are used as training sets.

[0038] Preferably, the preprocessing of the collected reflected laser signal includes: signal denoising, intensity calibration, point cloud registration and normal estimation.

[0039] Preferably, the method of using the evaluation model to score the flatness of the wall surface, combining the scoring result with the flatness test report, and providing a comprehensive wall surface flatness test report includes:

[0040] The flatness score output by the evaluation model is quantified to form specific scoring indicators, including flatness index, uneven area ratio and maximum unevenness degree;

[0041] The quantified scoring indicators are combined with the information on the location, type and degree of unevenness in the flatness test report to generate a comprehensive report containing detailed scores and unevenness details.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] By introducing laser radar technology, the present invention can simultaneously obtain spatial information in a large range of areas with a very high data acquisition speed. Compared with the traditional single-point or line scanning method, the present invention can cover the entire wall surface at one time, greatly improving the measurement efficiency. Laser radar technology has the characteristics of non-contact and high-precision measurement, and can accurately capture subtle changes in the wall surface. By constructing an accurate three-dimensional point cloud model and accurately mapping the laser signal to the wall model, the present invention can comprehensively and accurately reflect the flatness of the wall surface, effectively avoiding subjective errors in manual measurement.

[0044] The present invention can effectively deal with walls of different materials, colors and textures. By optimizing the acquisition and processing methods of laser signals, accurate measurement results can be obtained under various wall conditions, so that the present invention has strong adaptability and can be widely used in various construction projects, decoration maintenance and quality inspection scenarios. The present invention has a high degree of automation and reduces labor costs. The present invention uses a machine learning algorithm to extract and classify features in a three-dimensional laser reflection image, and can automatically identify and mark the uneven position, type and degree of the wall. This automated process greatly reduces manual intervention and reduces labor costs. By constructing an evaluation model to evaluate the flatness of the wall and outputting a flatness score, the present invention can generate a comprehensive wall flatness detection report. The report contains measurement results and intuitive scores, which is convenient for users to quickly understand the flatness of the wall. The proposal and implementation of the present invention solve the technical difficulties in wall flatness detection. With the wide application of the present invention, wall flatness detection will become more efficient, accurate and convenient, providing a strong guarantee for the quality control of construction projects and the refined management of decoration maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a working principle diagram of the wall flatness detection method based on laser array radar described in the present invention;

[0046] Figure 2 A diagram showing the steps of mapping the laser signal onto the wall model through the point cloud imaging algorithm to form a three-dimensional laser reflection image of the wall;

[0047] Figure 3 A step-by-step diagram for analyzing 3D laser reflection images and using machine learning algorithms to extract and classify features in the images, automatically identifying and annotating the location, type, and degree of unevenness on the wall;

[0048] Figure 4 This is a flowchart for building a wall flatness evaluation model based on the SVM algorithm. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] See also Figure 1-4 The present invention provides a technical solution: a wall flatness detection method based on a laser array radar, the method comprising:

[0051] Use an array laser radar system to scan the wall: Use an array laser radar system to perform an all-round scan of the target wall. The system can simultaneously emit and receive multiple laser beams to quickly collect laser signals reflected from the wall. During the scanning process, ensure that the relative position of the laser radar system and the wall is stable to ensure the accuracy of the collected data.

[0052] Preprocessing the collected reflected laser signals: Perform preprocessing operations such as denoising and correction on the collected reflected laser signals to eliminate interference components in the signals and improve the quality of the signals. The preprocessing process includes filtering out noise, correcting the offset and distortion of the laser beam, etc., to ensure the accuracy of subsequent processing.

[0053] Constructing a 3D point cloud model of the wall: Based on the pre-processed reflected laser signal, a point cloud generation algorithm is used to construct a 3D point cloud model of the wall. This model can accurately reflect the spatial form and structural characteristics of the wall, providing a basis for subsequent flatness analysis.

[0054] Accurately match the preprocessed reflected laser signal with the spatial coordinates of the wall model: Use the spatial resolution capability of the laser radar to accurately match the preprocessed reflected laser signal with the spatial coordinates of the three-dimensional point cloud model of the wall. Through the point cloud imaging algorithm, the laser signal is mapped to the wall model to form a three-dimensional laser reflection image of the wall.

[0055] Analyze the 3D laser reflection image: Use machine learning algorithms to analyze the 3D laser reflection image and extract feature information from the image, such as uneven features such as bulges and depressions on the wall. Use classification algorithms to automatically identify and mark the uneven location, type, and degree of the wall.

[0056] Construct an evaluation model to evaluate the flatness of the wall: According to the evaluation criteria for the flatness of the wall, construct an evaluation model. The model can comprehensively consider the location, type and degree of the unevenness of the wall and output the scoring result of the flatness of the wall.

[0057] Use the evaluation model to score the flatness of the wall and generate a comprehensive report: Use the constructed evaluation model to score the flatness of the wall, combine the scoring results with the flatness test report, and form a comprehensive wall flatness test report. This report can comprehensively and accurately reflect the flatness of the wall, providing strong technical support for construction engineering, decoration maintenance, quality inspection and other fields.

[0058] The present invention will be further described below in conjunction with Examples 1 to 6:

[0059] Embodiment 1:

[0060] The step of mapping the laser signal onto the wall model through the point cloud imaging algorithm to form a three-dimensional laser reflection image of the wall specifically includes the following implementation methods:

[0061] S101. Point cloud processing and point cloud matrix construction: Perform point cloud processing on the pre-processed reflected laser signal. The laser radar system works simultaneously through multiple scan lines to obtain point cloud data of the wall. These point cloud data are organized into a point cloud matrix P, where each row of P corresponds to the point cloud data of a scan line, containing the three-dimensional coordinate information of all points on the scan line; each column corresponds to a specific point, and the elements in the column represent the coordinates of the point in three-dimensional space (such as X, Y, and Z coordinates). In this way, the point cloud matrix P fully reflects the spatial form of the wall under the laser radar scanning.

[0062] S102, normal vector calculation: according to the geometric layout of the laser radar and the registration principle of the point cloud data, the normal vector of the point cloud matrix P is calculated to obtain the normal vector matrix N, where the elements of N represent the normal directions of different points; wherein, the normal vector is calculated using the following formula:

[0063]

[0064] Among them, P i , P i+1 , P i+2 are three adjacent point cloud data points, N i For point P i The normal vector of .

[0065] S103, spatial coordinate matching and initial reflection image generation: spatial coordinate matching is performed using the normal vector matrix N and the three-dimensional point cloud model of the wall. The purpose of this step is to map the information in the normal vector to the corresponding spatial position on the wall model, thereby generating an initial three-dimensional laser reflection image I_0.

[0066] In specific implementation, for each normal vector in the normal vector matrix N, find the corresponding spatial position in the three-dimensional point cloud model of the wall according to the coordinates of the corresponding point in the point cloud matrix P. Then, map the information of the normal vector (such as direction, length, etc., which can be selected according to actual needs) to the spatial position to form the initial three-dimensional laser reflection image I_0. Each pixel point in I_0 corresponds to a specific position on the wall and the normal vector information of the position.

[0067] S104, image optimization processing: performing image optimization processing on the initial reflection image I_0 to obtain an optimized reflection image I_e. The purpose of image optimization processing is to improve the quality of the image and make the features in the image clearer and more accurate.

[0068] Specific optimization methods may include but are not limited to: denoising, edge enhancement, contrast adjustment, etc. Through these processes, noise and interference in the image can be eliminated, and feature information in the image can be enhanced, providing a better basis for subsequent feature extraction and classification.

[0069] S105, potential uneven area marking: according to the reflection characteristics and geometric features of the wall material, the intensity threshold T_h of the reflection signal and the shape recognition rule R' are set. Then, the optimized reflection image I_e is analyzed point by point.

[0070] For each pixel in the image, determine whether its signal strength exceeds the threshold T_h and whether it meets the shape recognition rule R'. If both conditions are met, the point is marked as a potential uneven area. Finally, the marked reflection image I_m is obtained. The marked area in I_m is the uneven area that may exist on the wall.

[0071] Embodiment 2:

[0072] This embodiment is used to describe the steps of preprocessing the collected reflected laser signal, and the specific implementation steps are as follows:

[0073] Signal denoising: Since the laser radar may be affected by various factors such as environmental noise, equipment noise, and wall reflection characteristics during the scanning process, the collected reflected laser signal contains noise components. In order to improve the accuracy of subsequent processing, the original signal needs to be denoised. In specific implementation, a filtering algorithm can be used to denoise the reflected laser signal. For example, common filtering methods such as Gaussian filtering, mean filtering, or median filtering can be used. According to the characteristics of the signal and the type of noise, appropriate filtering parameters can be selected to smooth the signal to remove noise components and retain useful signal information.

[0074] Intensity calibration: The intensity of the reflected laser signal collected by the LiDAR may be affected by many factors, such as laser emission power, wall reflectivity, scanning distance, etc. In order to accurately reflect the reflection characteristics of the wall, the signal intensity needs to be calibrated. The intensity calibration process may include: obtaining the reflected laser signal of a standard wall or a standard reflector with known reflectivity as a calibration reference; establishing an intensity calibration model based on the reflectivity of the standard wall and the collected signal intensity; using the model to calibrate the intensity of the actually collected reflected laser signal to obtain an accurate signal intensity value.

[0075] Point cloud registration: When the laser radar scans the wall, due to factors such as scanning angle, scanning speed or device movement, there may be position offset or rotation between the point cloud data collected by different scanning lines or different scanning cycles. In order to accurately splice these point cloud data together to form a complete three-dimensional point cloud model of the wall, point cloud registration is required. The process of point cloud registration can include: extracting feature points or feature surfaces in the point cloud data; using feature matching algorithms to find the correspondence between different point cloud data; calculating the transformation matrix based on the correspondence, and rotating and translating the point cloud data to achieve accurate point cloud registration.

[0076] Normal estimation: For subsequent spatial coordinate matching and image generation, it is necessary to estimate the normal vector of the wall point cloud. The normal vector is an important feature that describes the surface direction of the point cloud and is of great significance for the flatness detection of the wall. The normal estimation process may include: for each point in the point cloud, select several points in its neighborhood as reference points; then, use the coordinate information of these reference points to estimate the normal vector of the point through geometric calculation or statistical methods; finally, normalize the estimated normal vector to obtain the unit normal vector.

[0077] Embodiment 3:

[0078] The image optimization processing step S104 further includes the following detailed implementation:

[0079] The initial reflection image I_0 is subjected to contrast enhancement processing to improve the visual effect of the image and make the details in the image more clearly discernible. Contrast enhancement can be achieved by adjusting the grayscale value or color value of the image, so that the bright area in the image is brighter and the dark area is darker, thereby enhancing the layering and three-dimensional sense of the image. In specific implementation, the image contrast can be enhanced by using methods such as histogram equalization, linear stretching or nonlinear transformation. For example, the histogram equalization method can enhance the contrast of the image by calculating the grayscale histogram of the image and then redistributing the grayscale values ​​so that the grayscale values ​​are evenly distributed throughout the range.

[0080] Edge detection algorithms are used to enhance edge features in images to highlight edge information in the image, facilitating subsequent feature extraction and classification. Edges are places where the grayscale value in an image changes dramatically, usually corresponding to the outline or boundary of an object. Specific edge detection algorithms can include Sobel operators, Canny operators, Prewitt operators, etc. These algorithms detect the position and direction of edges by calculating the gradient value or second-order derivative of each pixel in the image. Then, the detected edges can be refined or connected to obtain more complete and accurate edge features.

[0081] De-noising algorithms are applied to smooth images to remove noise and interference in the image and improve the quality and credibility of the image. Noise is unwanted random changes or fluctuations in the image, which may mask useful information in the image or produce misleading results. Specific denoising algorithms can include mean filtering, median filtering, Gaussian filtering, bilateral filtering, etc. These algorithms smooth the image and reduce noise by calculating the average, median or weighted average of each pixel in the image and the pixels in its neighborhood. When selecting a denoising algorithm, it is necessary to select the appropriate algorithm and parameters according to the characteristics of the image and the type of noise to achieve the best denoising effect.

[0082] Embodiment 4:

[0083] In the present invention, the specific steps of analyzing the three-dimensional laser reflection image, extracting and classifying the features in the image using a machine learning algorithm, and automatically identifying and marking the uneven position, type and degree of the wall surface include:

[0084] S201. Build a deep convolutional neural network model:

[0085] Construct a deep convolutional neural network model, which consists of multiple convolutional layers, pooling layers, fully connected layers, and output layers. The convolutional layer is responsible for extracting local features in the reflection image, such as edges and textures, and converting the input image into a feature map through convolution operations. The pooling layer is used to reduce the dimension of the feature map and reduce the amount of calculation while retaining important features. The fully connected layer integrates the features output by the pooling layer and performs high-level abstraction to form an overall understanding of the image. The output layer classifies and annotates the location, type, and degree of unevenness based on the output of the fully connected layer.

[0086] In specific implementation, you can select appropriate convolution kernel size, step size, padding method and other parameters to construct the convolution layer, and select appropriate pooling window size and step size to construct the pooling layer. The number of neurons in the fully connected layer can be adjusted according to the specific task, and the number of neurons in the output layer corresponds to the number of categories classified and labeled.

[0087] S202, prepare training data set:

[0088] Prepare a training dataset that contains a large number of labeled wall LiDAR reflection images. Each image is labeled with the specific location, type, and degree of unevenness, such as cracks, bulges, depressions, and the severity of the unevenness. These labeled information will serve as supervisory signals for model training to guide the model to learn how to identify and label the location, type, and degree of unevenness on the wall.

[0089] The images in the training data set should cover various wall conditions, including different materials, different lighting conditions, different types and degrees of unevenness, etc., to ensure that the model has generalization capabilities and can adapt to wall flatness detection tasks in different scenarios.

[0090] S203, training deep convolutional neural network model:

[0091] The training data set is input into the deep convolutional neural network model, and the error between the model prediction result and the true annotation is calculated through the back propagation algorithm and gradient descent method. Then, the weight and bias parameters in the model are iteratively adjusted to minimize the error until the error of the model on the training data set converges to below the preset threshold.

[0092] During the training process, you can choose a suitable optimizer (such as Adam, SGD, etc.) and learning rate to accelerate convergence, and use regularization techniques (such as L2 regularization, Dropout, etc.) to prevent overfitting. In addition, you can also use data enhancement techniques (such as rotation, scaling, cropping, etc.) to expand the training data set and improve the robustness of the model.

[0093] S204. Verify the deep convolutional neural network model:

[0094] The trained deep convolutional neural network model is validated and the recognition accuracy, recall rate and F1 score of the model are evaluated using an independent validation dataset. The validation dataset should contain wall lidar reflection images that are different from the training dataset but have a similar distribution to ensure the objectivity and accuracy of the evaluation results.

[0095] By calculating various indicators of the model on the validation data set, it is possible to evaluate whether the performance of the model meets the requirements. If the model performance is poor, it is possible to return to step S201 to adjust the model structure or step S203 to optimize the training process until the model performance reaches the expected level.

[0096] S205. Apply deep convolutional neural network model to detect wall flatness:

[0097] The laser radar reflection image of the wall to be inspected is input into the trained and verified deep convolutional neural network model, which automatically extracts, classifies and annotates the image features and outputs the specific location, type and degree of unevenness. This information can be used to generate a wall flatness inspection report, providing a basis for subsequent maintenance and renovation.

[0098] In practical applications, the deep convolutional neural network model can be integrated into the wall flatness detection system to realize an automated and intelligent detection process. At the same time, new data can be continuously collected to update and optimize the model to adapt to the ever-changing wall conditions and detection needs.

[0099] Embodiment 5:

[0100] The evaluation model is constructed based on the support vector machine (SVM) algorithm. The detailed construction process of the model includes:

[0101] S301, setting the kernel function of the model: selecting the radial basis function (RBF) as the kernel function of the SVM model. The RBF kernel function is a commonly used nonlinear kernel function that can map data in the original space to a high-dimensional space, thereby processing nonlinear problems.

[0102] S302. Collect training data: The data includes the wall surface flatness score and the corresponding reflected laser signal characteristics. The wall surface flatness score can be a continuous or discrete value obtained based on expert evaluation or standard measurement. The reflected laser signal characteristics include signal strength, point cloud density, and normal vector change rate, which can reflect the geometric and reflective characteristics of the wall. Ensure that the training data is representative and diverse, covering different types of walls, different flatness conditions, and different environmental conditions.

[0103] S303, model training: input the training data into the SVM model, and optimize the model parameters through the model training algorithm. The parameters of the SVM model include the penalty parameter C and the kernel function parameter γ. The penalty parameter C controls the degree of penalty for classification errors. A larger C value will cause the model to pay more attention to the accuracy of the training data, but may overfit; a smaller C value will cause the model to be more generalized, but may underfit.

[0104] The training data set is used to train the model, and the optimal combination of C and γ parameters is selected through cross-validation or other methods. Specifically, a method such as grid search can be used to traverse within a preset parameter range and select the parameter combination that makes the model perform best on the validation data set.

[0105] In addition, the K-fold cross-validation method is used to further verify the performance of the model. The training data is divided into K subsets, one of which is used as the validation set and the rest as the training set in turn. Through K-fold cross-validation, the stability and generalization ability of the model on different data sets can be evaluated.

[0106] S304, model testing: Test the trained SVM model and use an independent test data set to evaluate the scoring accuracy and stability of the model. The test data set should contain wall flatness scores and reflected laser signal features that are different from the training data set but have similar distributions.

[0107] By calculating the scoring accuracy and stability of the model on the test data set, it is possible to evaluate whether the performance of the model meets the requirements. If the model performance is poor, it is possible to return to step S301 to adjust the kernel function or step S303 to optimize the model parameters until the model performance meets expectations.

[0108] Embodiment 6:

[0109] The present invention uses an evaluation model to score the flatness of a wall surface, and combines the scoring result with a flatness detection report to provide a method for providing a comprehensive wall surface flatness detection report, which specifically includes the following steps:

[0110] ① Quantify the flatness score output by the evaluation model to form specific scoring indicators. These scoring indicators include but are not limited to:

[0111] Flatness index: This is an indicator that comprehensively reflects the overall flatness of the wall surface. It can be obtained by normalizing or standardizing the original score output by the evaluation model. The higher the flatness index, the flatter the wall surface; conversely, it means that there are more uneven areas on the wall surface.

[0112] Ratio of uneven area: This refers to the ratio of the uneven area in the wall to the entire wall area. It can be obtained by calculating the area of ​​the uneven area identified by the evaluation model and comparing it with the area of ​​the entire wall. The larger the ratio of uneven area, the more uneven areas there are in the wall.

[0113] Maximum unevenness: This refers to the score or measurement value of the area with the most serious unevenness in the wall. The unevenness scores output by the evaluation model can be sorted and the maximum value can be selected as the maximum unevenness indicator.

[0114] ②Integrate the quantified scoring index with the unevenness location, type and degree information in the flatness test report. The specific integration method is as follows:

[0115] In the flatness test report, the scoring indicators such as flatness index, uneven area ratio and maximum unevenness degree are clearly listed to intuitively show the overall flatness condition of the wall. At the same time, the unevenness location, type and degree information identified by the evaluation model are recorded in detail in the report. This information can include the specific location of the uneven area (such as coordinates, area number, etc.), the type of unevenness (such as cracks, protrusions, depressions, etc.) and the degree of unevenness (such as slight, medium, severe, etc.). By combining the scoring indicators with the unevenness details, a comprehensive report containing detailed scores and unevenness details is generated. This report can not only comprehensively reflect the flatness condition of the wall, but also provide specific unevenness information, providing a strong basis for subsequent maintenance and renovation.

[0116] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0117] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A wall flatness detection method based on a laser array radar is characterized in that: The method comprises: Use the area array laser radar system to scan the wall and collect the reflected laser signal from the wall; Preprocessing the collected reflected laser signal; Construct a 3D point cloud model of the wall, and use the spatial resolution capability of the LiDAR to accurately match the pre-processed reflected laser signal with the spatial coordinates of the wall model. Use the point cloud imaging algorithm to map the laser signal to the wall model to form a 3D laser reflection image of the wall. Analyze the three-dimensional laser reflection image, use machine learning algorithms to extract and classify the features in the image, automatically identify and mark the uneven position, type and degree of the wall surface, and generate a flatness detection report; Construct an evaluation model to evaluate the flatness of the wall and output a flatness score; The evaluation model is used to score the flatness of the wall surface, and the scoring results are combined with the flatness test report to provide a comprehensive wall flatness test report.

2. The wall flatness detection method based on laser array radar according to claim 1 is characterized in that: The step of mapping the laser signal onto the wall model through the point cloud imaging algorithm to form a three-dimensional laser reflection image of the wall includes: S101: performing point cloud processing on the preprocessed reflected laser signal, and constructing a point cloud matrix P using point cloud data acquired by multiple scan lines of the laser radar, wherein each row of P corresponds to point cloud data of a scan line, and each column corresponds to the three-dimensional coordinates of a point; S102: According to the geometric layout of the laser radar and the registration principle of the point cloud data, the normal vector of the point cloud matrix P is calculated to obtain the normal vector matrix N, where the elements of N represent the normal directions of different points; wherein the normal vector is calculated using the following formula: Among them, P i , P i+1 , P i+2 are three adjacent point cloud data points, N i For point P i The normal vector of S103: using the normal vector matrix N and the three-dimensional point cloud model of the wall, performing spatial coordinate matching, mapping the information in the normal vector to the corresponding spatial position on the wall model, and obtaining an initial three-dimensional laser reflection image I_0; S104: performing image optimization processing on the initial reflection image I_0 to obtain an optimized reflection image I_e; S105: According to the reflection characteristics and geometric features of the wall material, the intensity threshold T_h and shape recognition rule R of the reflection signal are set, and the optimized reflection image I_e' is analyzed point by point, and the points whose signal intensity exceeds the threshold T_h and meets the shape recognition rule R' are marked as potential uneven areas to obtain the marked reflection image I_m.

3. The wall flatness detection method based on laser array radar according to claim 2 is characterized in that: The types of wall unevenness identified include bulges, depressions, cracks, peeling or deformation of the wall surface.

4. The wall flatness detection method based on laser array radar according to claim 2 is characterized in that: The image optimization processing step S104 further includes: Perform contrast enhancement on the initial reflection image I_0; Use edge detection algorithm to enhance edge features in images; Apply a denoising algorithm to smooth the image.

5. The wall flatness detection method based on laser array radar according to claim 1 is characterized in that: The three-dimensional laser reflection image is analyzed, and features in the image are extracted and classified using a machine learning algorithm to automatically identify and mark the uneven position, type and degree of the wall surface. The specific steps include: S201: construct a deep convolutional neural network model, which includes multiple convolutional layers, pooling layers, fully connected layers and output layers; wherein the convolutional layer is used to extract local features in the reflection image, the pooling layer is used to reduce the feature dimension, the fully connected layer is used to integrate features and perform high-level abstraction, and the output layer is used to classify and annotate the location, type and degree of unevenness; S202: preparing a training data set, which includes a large number of labeled wall lidar reflection images, each of which is labeled with specific location, type and degree information of the unevenness, for training a deep convolutional neural network model; S203: input the training data set into the deep convolutional neural network model, calculate the error between the model prediction result and the true annotation through the back propagation algorithm and the gradient descent method, and iteratively adjust the weight and bias parameters in the model until the error of the model on the training data set converges to below a preset threshold; S204: Validate the trained deep convolutional neural network model and use an independent validation dataset to evaluate the recognition accuracy, recall rate, and F1 score of the model; S205: Input the wall lidar reflection image to be detected into the trained and verified deep convolutional neural network model. The model automatically extracts features, classifies and annotates the image, and outputs the specific location, type and degree of unevenness.

6. The wall flatness detection method based on laser array radar according to claim 1 is characterized in that: The evaluation model is constructed based on the support vector machine (SVM) algorithm.

7. The wall flatness detection method based on laser array radar according to claim 6 is characterized in that: The construction process of the evaluation model includes: S301: Setting the kernel function of the model to radial basis function RBF; S302: Collect training data, which includes a wall surface flatness score and corresponding reflected laser signal features, including signal strength, point cloud density, and normal vector change rate; S303: input the training data into the SVM model, and optimize the parameters of the model, including the penalty parameter C and the kernel function parameter γ, through the model training algorithm until the performance of the model on the validation data set reaches a preset standard; S304: Test the trained SVM model and use an independent test data set to evaluate the scoring accuracy and stability of the model.

8. The wall flatness detection method based on laser array radar according to claim 7 is characterized in that: The construction of the evaluation model also includes a cross-validation step, which uses a K-fold cross-validation method to divide the training data into K subsets, and use one of the subsets as a validation set and the rest as training sets in turn.

9. The wall flatness detection method based on laser array radar according to claim 1 is characterized in that: The preprocessing of the collected reflected laser signal includes: signal denoising, intensity calibration, point cloud registration and normal estimation.

10. The wall flatness detection method based on laser array radar according to claim 1 is characterized in that: The method of using the evaluation model to score the flatness of the wall surface, combining the scoring result with the flatness test report, and providing a comprehensive wall surface flatness test report includes: The flatness score output by the evaluation model is quantified to form specific scoring indicators, including flatness index, uneven area ratio and maximum unevenness degree; The quantified scoring indicators are combined with the information on the location, type and degree of unevenness in the flatness test report to generate a comprehensive report containing detailed scores and unevenness details.

Images