A method for generating plate reinforcement layout process based on image recognition

Abstract

The present invention relates to the field of intelligent construction technology, and in particular to a method for generating a plate reinforcement layout process based on image recognition. By acquiring historical images of plate reinforcement during construction, the coverage area and layout type of the plate reinforcement are extracted, and the reinforcement spacing and binding spacing are obtained based on these image information. The first neural network model is trained using the plate reinforcement coverage area and layout type as input, and the reinforcement spacing and binding spacing as output. The second neural network model is used to learn the characteristics of different plate reinforcement layout types at different construction stages. By acquiring the current construction image, the reinforcement layout type is determined according to the current construction area, the construction stage is judged by the second neural network model, and the first neural network model is used to perform defect verification to generate a process to be processed. Real-time automatic identification and positioning of defects in construction are achieved, auxiliary guidance and monitoring of on-site construction personnel are provided, and the workload of manual evaluation and real-time judgment and correction is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent construction, and in particular to a method for generating a plate reinforcement arrangement process based on image recognition. Background Art

[0002] In construction projects, plate reinforcement is an important part of concrete structure, mainly used to enhance the tensile and compressive resistance of concrete slabs and ensure the strength and stability of building structures. The layout of plate reinforcement is directly related to the overall performance and safety of the building, so its layout rules are of great technical significance. Plate reinforcement is usually divided into bottom layer reinforcement and surface layer reinforcement, which are arranged at the lower and upper parts of the concrete slab respectively to resist stress in different directions. According to the design requirements, plate reinforcement needs to be arranged according to certain spacing and rules to ensure the uniform distribution and overall stability of the concrete slab when subjected to stress. The layout rules of plate reinforcement are closely related to the area of ​​the concrete slab. A larger slab area requires more reinforcement to support it to ensure stability and safety under various loads. Specifically, the layout type of plate reinforcement (such as single layer, double layer, grid type and reinforced type) and the parameters such as the spacing and binding spacing of the reinforcement need to be adjusted and optimized according to the coverage area of ​​the slab. This area-based layout method can ensure the rationality of the distribution of reinforcement and the uniformity of the structure, reduce stress concentration and potential structural defects.

[0003] At present, steel bar layout mainly relies on manual design and on-site construction, which is not only time-consuming and labor-intensive, but also prone to human errors, affecting construction quality and efficiency. As the scale and complexity of construction projects continue to increase, traditional manual operation methods are difficult to meet the needs of efficient and precise construction. Therefore, there is an urgent need for an automated and intelligent steel bar layout method to assist on-site workers in guiding and monitoring, so as to improve construction efficiency and quality and reduce errors and rework rates. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a plate reinforcement arrangement process generation method based on image recognition.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for generating a plate reinforcement arrangement process based on image recognition comprises the following steps:

[0007] S1. Obtain the slab reinforcement coverage area and slab reinforcement arrangement type in the slab reinforcement historical construction image of the building, and obtain the reinforcement spacing and reinforcement binding spacing based on the slab reinforcement historical construction image;

[0008] S2, using the plate reinforcement coverage area and the plate reinforcement arrangement type as input, and the reinforcement spacing and the binding spacing as output, training the first neural network model;

[0009] S3, based on the historical construction images of slab reinforcement, learning the construction stages of different slab reinforcement arrangement types through the second neural network model;

[0010] S4. Obtain the current construction image, determine the plate reinforcement layout type according to the current construction area, judge the current construction stage according to the plate reinforcement layout type and the current construction image through the second neural network model, and perform defect verification through the first neural network model. Generate a pending process based on the defect verification result. If the current pending processes have been completed, re-execute S4; if the current construction stage is the final stage and the pending processes have been completed, end the execution.

[0011] Furthermore, the plate reinforcement arrangement types include: a single-layer reinforcement arrangement type, a double-layer reinforcement arrangement type, a grid reinforcement arrangement type and a reinforced reinforcement arrangement type, wherein the reinforced reinforcement arrangement type is reinforced by adding additional reinforcements in the center of the grid reinforcement.

[0012] Furthermore, the method of obtaining the steel bar spacing and steel bar binding spacing based on the historical construction image of the plate steel bar comprises the following steps:

[0013] Preprocess the historical construction images of the plate reinforcement to generate a planar grayscale image of the plate reinforcement;

[0014] The image recognition algorithm is used to mark the steel bar positions and steel bar intersections in the planar grayscale image;

[0015] The rebar spacing is determined by calculating the distance between adjacent rebars using rebar locations and rebar intersections.

[0016] Furthermore, the formula of the first neural network model is as follows:

[0017] Y=g(V·f(W·X+b)+c)

[0018] Among them, Y is the output vector, which represents the predicted value of the reinforcement spacing and the binding spacing; X is the input vector, including the characteristics of the plate reinforcement coverage area and the plate reinforcement arrangement type; W is the weight matrix of the hidden layer; b is the bias vector of the hidden layer; f is the activation function of the hidden layer; V is the weight matrix of the output layer; c is the bias vector of the output layer; g is the activation function of the output layer.

[0019] Furthermore, the second neural network is a combined neural network, including a convolutional neural network and a BP neural network.

[0020] Furthermore, S3 comprises the following steps:

[0021] Pre-classify different construction stages for different slab reinforcement arrangement types, and label the slab reinforcement arrangement types and corresponding construction stages;

[0022] The stage image features in the historical construction images of plate reinforcement are extracted through convolutional neural networks;

[0023] The BP neural network is trained with the slab reinforcement arrangement type label and stage image features as input and the construction stage label as output;

[0024] Optimize BP neural network parameters through back propagation.

[0025] Further, determining the slab reinforcement arrangement type according to the current construction area includes the following steps:

[0026] Obtain the current construction area, where the construction area is the required steel bar coverage area in the construction site;

[0027] The current construction area is compared with the preset slab reinforcement arrangement type coverage area standard to determine several slab reinforcement arrangement types.

[0028] Further, judging the current construction stage according to the plate reinforcement arrangement type and the current construction image by the second neural network model includes:

[0029] Taking several slab reinforcement arrangement types and the current construction image as input, the second neural network model outputs several construction stage labels;

[0030] Get the number of times a tag appears in several construction stage tags and sort them in descending order, and determine the current construction stage based on the descending order results.

[0031] Furthermore, the defect verification is performed by the first neural network model, and the process to be processed is generated according to the defect verification result, including the following steps:

[0032] Input the construction type corresponding to the current construction stage and the current construction image into the trained first neural network model to identify and detect defects in the steel bar arrangement, including steel bar spacing and binding spacing that do not meet design requirements;

[0033] Classify and mark the identified defects and locate the defects in the construction images;

[0034] Generate corresponding processing suggestions and correction steps based on the defect type, and generate the process to be processed based on the defect location and processing suggestions.

[0035] The beneficial effects of the present invention are as follows: the present invention obtains the historical images of plate reinforcement during the construction process, extracts the coverage area and layout type of the plate reinforcement, and obtains the reinforcement spacing and binding spacing based on these image information. With the plate reinforcement coverage area and layout type as input, and the reinforcement spacing and binding spacing as output, the first neural network model is trained to predict the reinforcement layout parameters under different conditions. In this way, the system can quickly and accurately generate a reinforcement layout plan that meets the design specifications, reducing the error and time cost of manual design. Based on the historical construction images, the second neural network model is used to learn the characteristics and laws of different plate reinforcement layout types at different construction stages. In this way, it can ensure that the system provides accurate layout suggestions at different construction stages and improves the construction quality. In the actual construction process, the current construction image is obtained, the reinforcement layout type is determined according to the current construction area, the construction stage is judged by the second neural network model, and the defect check is performed by the first neural network model based on the current construction stage and the plate reinforcement layout type, and the process to be processed is generated to ensure the construction quality. The system can automatically identify and locate defects in construction, and provide corresponding processing suggestions and correction steps in real time. It can generate corresponding processes at each stage of plate reinforcement layout, provide auxiliary guidance and monitoring to on-site staff, and greatly reduce the workload from manual evaluation of construction plans to real-time judgment and correction by professionals. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flowchart of the steps of a method for generating a plate reinforcement arrangement process based on image recognition in the present invention.

[0037] Figure 2 It is a flowchart of the steps of obtaining the steel bar spacing and steel bar binding spacing based on the historical construction image of the plate steel bars in the present invention. DETAILED DESCRIPTION

[0038] See also Figure 1-2 As shown, the present invention relates to a method for generating a plate reinforcement arrangement process based on image recognition, comprising the following steps:

[0039] S1. Obtain the slab reinforcement coverage area and slab reinforcement arrangement type in the slab reinforcement historical construction image of the building, and obtain the reinforcement spacing and reinforcement binding spacing based on the slab reinforcement historical construction image;

[0040] S2, using the plate reinforcement coverage area and the plate reinforcement arrangement type as input, and the reinforcement spacing and the binding spacing as output, training the first neural network model;

[0041] S3, based on the historical construction images of slab reinforcement, learning the construction stages of different slab reinforcement arrangement types through the second neural network model;

[0042] S4. Obtain the current construction image, determine the plate reinforcement layout type according to the current construction area, judge the current construction stage according to the plate reinforcement layout type and the current construction image through the second neural network model, and perform defect verification through the first neural network model. Generate a pending process based on the defect verification result. If the current pending processes have been completed, re-execute S4; if the current construction stage is the final stage and the pending processes have been completed, end the execution.

[0043] In some embodiments, first, historical construction images of plate reinforcement during construction are obtained by high-resolution camera equipment. The historical construction images should cover different construction stages and multiple types of reinforcement layouts. The acquired images are preprocessed, including grayscale transformation, noise removal, and image enhancement steps to improve the clarity and recognizability of the images. In the preprocessed images, the image recognition algorithm is used to accurately extract the coverage area and layout type of the plate reinforcement. These layout types include single-layer reinforcement layout, double-layer reinforcement layout, grid layout, and reinforced layout. Further, based on these image information, by detecting the position and intersection of the reinforcement in the image, the spacing and binding spacing of the reinforcement are calculated and obtained to form a data set. Then, using the above data set, a first neural network model is constructed and trained. The coverage area and layout type of the plate reinforcement are used as the input variables of the model, and the spacing and binding spacing of the reinforcement are used as the output variables. Using deep learning technology, the model parameters are optimized through multiple iterations of training and verification to ensure that the model can accurately predict the reinforcement layout parameters under different conditions. In the training process, the gradient descent method is used for back propagation to minimize the loss function between the predicted value and the actual value, thereby improving the prediction accuracy of the model. Subsequently, the second neural network model is used to learn the different construction stages in the historical construction images of the plate reinforcement. The specific operations include: first, feature extraction of the historical images, using the convolutional neural network (CNN) to extract the spatial features and structural features in the images; then, these features are input into a trained BP (back propagation) neural network to learn the characteristics and laws of different plate reinforcement layout types in each construction stage. Through the learning of a large amount of historical data, the model can accurately identify and classify the characteristics of different construction stages. In the actual construction process, the real-time image of the current construction site is obtained and the same preprocessing steps are performed. The current reinforcement layout type is determined by calculating the reinforcement coverage area in the current construction image and combining the second neural network model. Then, based on the determined layout type and the current construction image, the second neural network model is used to determine the current construction stage. The current construction stage, reinforcement layout type and current construction image are input into the trained first neural network model for defect verification. During the defect verification process, the first neural network model automatically detects potential defects in the reinforcement layout, including reinforcement spacing that does not meet the design requirements, loose or missing tying, etc. The identified defects are classified and marked, the defect locations are accurately located in the construction image, and corresponding processing suggestions and correction steps are generated. These suggestions and steps form a pending process and are sent to the construction management system or on-site construction personnel to guide them to make corrections and adjustments. If the current pending processes have been completed, S4 is re-executed to obtain new construction images and conduct the next round of inspection and verification to ensure continuous monitoring and optimization of construction quality. If the current construction stage is the final stage and the pending processes have been completed, the execution ends.

[0044] Furthermore, the plate reinforcement arrangement types include: a single-layer reinforcement arrangement type, a double-layer reinforcement arrangement type, a grid reinforcement arrangement type and a reinforced reinforcement arrangement type, wherein the reinforced reinforcement arrangement type is reinforced by adding additional reinforcements in the center of the grid reinforcement.

[0045] It should be noted that in the single-layer steel bar arrangement type, the steel bars are arranged in one direction of the concrete slab, which is usually used for slabs with smaller spans and low bearing capacity requirements. For example, in the construction of a residential floor, the single-layer steel bar arrangement can meet the design requirements. By acquiring historical construction images, the system extracts the coverage area and steel bar spacing of the single-layer steel bar and trains the neural network model. In actual construction, through real-time image recognition technology, the system can automatically verify the accuracy of the steel bar arrangement to ensure that the spacing of each steel bar meets the design specifications and reduce the error of manual operation. The double-layer steel bar arrangement type arranges steel bars in the upper and lower layers of the concrete slab to enhance the tensile and compressive resistance of the slab under bidirectional stress. For example, in the construction of the floor of a large shopping mall, the use of a double-layer steel bar arrangement can significantly improve the bearing capacity of the floor. Through the analysis of historical construction images, the system identifies the characteristics of the double-layer steel bar arrangement, including the staggered arrangement and specific spacing of the upper and lower layers of steel bars. In the actual construction process, real-time image recognition can monitor the position and spacing of each layer of steel bars to ensure the accuracy of the arrangement and provide correction suggestions when problems are found. The grid-type reinforcement arrangement type forms a grid-like reinforcement structure in the concrete slab, and the reinforcement is arranged crosswise in both horizontal and vertical directions. It is suitable for large-span, high-load structures. For example, in a large-area construction, the grid-type reinforcement arrangement can ensure the stability of the bridge under multi-directional stress. By analyzing the historical images of bridge construction, the system extracts the intersection and spacing information of the grid reinforcement and uses the neural network model for learning. In actual construction, the system uses image recognition technology to monitor the layout of the grid reinforcement in real time to ensure that the position and spacing of the intersection of the reinforcement meet the design specifications, and any deviations are discovered and corrected in time. The reinforced reinforcement arrangement type is based on the grid-type reinforcement arrangement, and is reinforced by adding additional reinforcement in the central area of ​​the grid, specifically the reinforcement structure of the central cross. This design is suitable for structures that require local reinforcement, such as the core tube area of ​​a high-rise building. By acquiring and analyzing the historical images of the core tube construction of a high-rise building, the system identifies the layout position and spacing of the additional reinforcement in the center of the grid and trains the neural network model.

[0046] Furthermore, the method of obtaining the steel bar spacing and steel bar binding spacing based on the historical construction image of the plate steel bar comprises the following steps:

[0047] Preprocess the historical construction images of the plate reinforcement to generate a planar grayscale image of the plate reinforcement;

[0048] The image recognition algorithm is used to mark the steel bar positions and steel bar intersections in the planar grayscale image;

[0049] The rebar spacing is determined by calculating the distance between adjacent rebars using rebar locations and rebar intersections.

[0050] In some embodiments, first, historical construction images of plate reinforcement during construction are obtained, which can be taken by high-definition camera equipment or drones. These images contain information on different construction stages and various types of reinforcement layout. Next, these historical construction images are preprocessed to generate a planar grayscale image of the plate reinforcement. The preprocessing steps include the following aspects: Grayscale transformation: converting the original color image into a grayscale image to simplify the complexity of image processing. The grayscale image only contains brightness information, eliminating the interference of color information, making the outline of the reinforcement clearer; using image processing techniques such as Gaussian filtering or median filtering, the grayscale image is denoised to remove random noise and interference in the image and improve the image quality; by using methods such as histogram equalization or contrast stretching, the contrast of the grayscale image is enhanced to make the position and outline of the reinforcement more prominent, which is convenient for subsequent image recognition. After completing the preprocessing, the obtained planar grayscale image is clearer and more stable, which helps to accurately identify the position and intersection of the reinforcement. Next, the planar grayscale image is processed using an image recognition algorithm to mark the position and intersection of the reinforcement in the image. The specific process is as follows: Edge detection: Use the Canny edge detection algorithm to extract edge information from the grayscale image. The Canny algorithm detects the edge of the steel bar by calculating the image gradient, thereby determining the position of the steel bar. Morphological processing: Apply morphological operations (such as corrosion and expansion) to process the edge detection results to eliminate noise and small fractures, making the steel bar contour more complete and continuous. Steel bar position marking: By analyzing the image after morphological processing, the position of the steel bar is identified, and the starting and ending points of the steel bar are marked. Steel bar intersection marking: Further analyze the steel bar position data to identify the intersections between the steel bars. The intersection is the location where two or more steel bars intersect. By calculating these intersections, the arrangement of the steel bars can be determined. Finally, the distance between adjacent steel bars is calculated through the marked steel bar positions and steel bar intersections to determine the steel bar spacing. Binding spacing calculation: By analyzing the location of the steel bar intersections, the spacing between the steel bar binding points is calculated. These binding points are usually located at the intersection of the steel bars. By calculating the distance between these points, the steel bar binding spacing is determined.

[0051] Furthermore, the formula of the first neural network model is as follows:

[0052] Y=g(V·f(W.X+b)+c)

[0053] Among them, Y is the output vector, which represents the predicted value of the reinforcement spacing and the binding spacing; X is the input vector, including the characteristics of the plate reinforcement coverage area and the plate reinforcement arrangement type; W is the weight matrix of the hidden layer; b is the bias vector of the hidden layer; f is the activation function of the hidden layer; V is the weight matrix of the output layer; c is the bias vector of the output layer; g is the activation function of the output layer.

[0054] Specifically, first, a large amount of historical image data of plate reinforcement during construction is obtained. These data include the coverage area and layout type of plate reinforcement, and the corresponding reinforcement spacing and binding spacing are marked. These data are organized into a structured data set for training the neural network model. The organized data set is divided into a training set and a validation set. The training set is used to train the model, and the validation set is used to evaluate the model performance. The training data includes an input vector X and a corresponding output vector Y. Input layer: The input vector X includes the characteristics of the plate reinforcement coverage area and the plate reinforcement layout type. The coverage area represents the distribution range of the reinforcement in the concrete slab, and the layout types include single layer, double layer, grid type and reinforced type. The hidden layer of the neural network is calculated through multiple neurons. Each neuron receives the input signal and is linearly transformed by the weight matrix W and the bias vector b, and then nonlinearly transformed by the activation function f. The output of the hidden layer is linearly transformed by the weight matrix V and the bias vector c, and then the final output vector Y is obtained by the activation function g of the output layer. Model verification: After training, the model is evaluated using the validation set to verify the performance of the model on unseen data. By calculating the mean square error (MSE) and the coefficient of determination (R 2 ) and other indicators to evaluate the prediction accuracy and generalization ability of the model.

[0055] Furthermore, the second neural network is a combined neural network, including a convolutional neural network and a BP neural network.

[0056] In some embodiments, the combined structure of the second neural network uses a convolutional neural network (CNN) to extract features from the image, and performs feature classification and regression analysis through a BP neural network. This combined neural network structure can not only capture complex features in the image, but also efficiently perform classification and regression tasks. First, a large number of historical construction images of plate reinforcement are obtained, which cover information of different construction stages and various types of reinforcement layout. Through these images, a training data set is constructed for training the combined neural network model. The historical construction images of plate reinforcement are input into the convolutional neural network. The input image is usually standardized to improve the training effect. The convolution layer performs a convolution operation on the input image through multiple convolution kernels to extract local features in the image. The size and number of convolution kernels are adjusted according to the complexity of the image features. The pooling layer downsamples the output of the convolution layer to reduce the size of the feature map while retaining important features. Commonly used pooling methods include maximum pooling and average pooling. After each convolution layer and pooling layer, a nonlinear activation function (such as ReLU) is applied to perform a nonlinear transformation on the output to enhance the expression ability of the network. The outputs of the convolutional layer and the pooling layer are flattened into a one-dimensional vector and input into the fully connected layer to further extract high-level features. The feature vector extracted from the convolutional neural network is input into the BP neural network. The hidden layer of the BP neural network consists of multiple neurons, each of which is linearly transformed by weights and biases, and nonlinearly transformed by activation functions. The output of the hidden layer is linearly transformed by the fully connected layer to generate the final output vector. The output vector can represent labels or specific regression values ​​for different construction stages.

[0057] Furthermore, S3 comprises the following steps:

[0058] Pre-classify different construction stages for different slab reinforcement arrangement types, and label the slab reinforcement arrangement types and corresponding construction stages;

[0059] The stage image features in the historical construction images of plate reinforcement are extracted through convolutional neural networks;

[0060] The BP neural network is trained with the slab reinforcement arrangement type label and stage image features as input and the construction stage label as output;

[0061] Optimize BP neural network parameters through back propagation.

[0062] Pre-classify different construction stages for different slab reinforcement layout types, and label the slab reinforcement layout types and corresponding construction stages. First, collect a large amount of historical construction image data, which includes different types of slab reinforcement layouts (such as single-layer, double-layer, grid-type and reinforced type) and corresponding construction stage information. Through expert annotation or based on construction records, label the slab reinforcement layout type and construction stage corresponding to each image to form a structured data set. Input the preprocessed historical construction images into the convolutional neural network. The input images are standardized to improve the model training effect. The convolution layer applies multiple convolution kernels (filters) to convolve the input image to extract local features of the image. The size and number of convolution kernels are designed according to the image complexity and task requirements. After extracting the stage image features, the BP (back propagation) neural network is trained with the slab reinforcement layout type label and the stage image features as input and the construction stage label as output. Through the above steps, the system can automatically extract features from the historical construction images of slab reinforcement and accurately classify different construction stages.

[0063] Further, determining the slab reinforcement arrangement type according to the current construction area includes the following steps:

[0064] Obtain the current construction area, where the construction area is the required steel bar coverage area in the construction site;

[0065] The current construction area is compared with the preset slab reinforcement arrangement type coverage area standard to determine several slab reinforcement arrangement types.

[0066] Specifically, after obtaining the current construction area, it is compared with the preset plate reinforcement layout type coverage area standard. The system database pre-stores the coverage area standards of different types of plate reinforcement layouts, which are determined by engineering design specifications and historical data, including the coverage area ranges of single-layer reinforcement layout types, double-layer reinforcement layout types, grid layout types, and reinforced layout types. When comparing the current construction area with these preset standards, a matching algorithm is used to compare the current construction area with the standard areas in the database one by one to find the closest layout types. For example, for the current construction area of ​​1000 square meters, the matching algorithm will calculate the difference between the current area and each standard area and determine the layout type with the smallest difference. In addition, a tolerance range, such as 50 square meters, can be set to select all layout types within the tolerance range for the current construction area. If the tolerance range is 50 square meters, the plate reinforcement layout type with the current construction area in the range of 950-1050 square meters is selected.

[0067] Further, judging the current construction stage according to the plate reinforcement arrangement type and the current construction image by the second neural network model includes:

[0068] Taking several slab reinforcement arrangement types and the current construction image as input, the second neural network model outputs several construction stage labels;

[0069] Get the number of times a tag appears in several construction stage tags and sort them in descending order, and determine the current construction stage based on the descending order results.

[0070] In some embodiments, during the construction of a large-scale construction project, the system needs to determine the current construction stage. First, several plate reinforcement layout types and current construction images are obtained as input data. The current construction image is taken in real time by a high-resolution camera or drone, and the image is preprocessed, including denoising, grayscale conversion and enhancement processing, to ensure image clarity and recognition accuracy. Next, these input data are input into a pre-trained second neural network model. The model consists of a convolutional neural network (CNN) and a back propagation (BP) neural network, which can efficiently extract image features and classify them. Specifically, the convolutional neural network extracts spatial features in the image, such as the position, spacing and layout type of the reinforcement, through multiple convolutional layers and pooling layers. Then, these features are converted into one-dimensional vectors through a fully connected layer and input into the BP neural network for classification. The second neural network model outputs several construction stage labels, which represent different prediction results of the model for the current construction stage. In order to improve the accuracy of the judgment, the system will count each label and obtain the number of times the label appears. By calculating the frequency of occurrence of each label, a descending list of labels can be obtained, and the label with the highest frequency is the current construction stage judged by the system. For example, suppose that after the current construction image is processed by the model, the output construction stage labels include "rebar tying completed", "formwork installation in progress" and "concrete pouring preparation". By counting the number of occurrences of these labels, it is found that "rebar tying completed" appears the most, followed by "formwork installation in progress", and "concrete pouring preparation" appears the least. According to the descending results, the system determines that the current construction stage is "rebar tying completed". In this way, the system can accurately determine the current construction stage and provide real-time guidance for construction management. For example, in actual construction, if the system identifies that the current stage is "rebar tying completed", it can automatically generate the next construction plan and resource allocation plan to ensure that the construction progress is as planned.

[0071] Furthermore, the defect verification is performed by the first neural network model, and the process to be processed is generated according to the defect verification result, including the following steps:

[0072] Input the construction type corresponding to the current construction stage and the current construction image into the trained first neural network model to identify and detect defects in the steel bar arrangement, including steel bar spacing and binding spacing that do not meet design requirements;

[0073] Classify and mark the identified defects and locate the defects in the construction images;

[0074] Generate corresponding processing suggestions and correction steps based on the defect type, and generate the process to be processed based on the defect location and processing suggestions.

[0075] In some embodiments, it is assumed that during the construction of a high-rise building project, the system needs to perform defect verification of steel bar arrangement and generate a process to be processed. First, the construction type corresponding to the current construction stage and the current construction image are obtained. The current construction image is collected in real time by a high-resolution camera device and undergoes preprocessing steps, including denoising, grayscale conversion and enhancement processing, to ensure image clarity and recognition accuracy. The preprocessed construction image and construction type are input into the first neural network model that has been trained. The model consists of multiple hidden layers and is fully trained by the back propagation algorithm (Backpropagation) to identify and detect various defects in the steel bar arrangement. The input vector of the model includes the feature vector and construction type of the construction image, and the output vector is the defect detection result. In the defect detection stage of the model, the steel bar position and layout features in the image are extracted by a convolutional neural network (CNN), and the steel bar spacing and binding spacing that do not meet the design requirements are identified and detected. For example, if the design requires the steel bar spacing to be 20 cm, and the model detects that the steel bar spacing in a certain area is 25 cm, it is determined that there is a defect in the area. The detected defects are classified and marked, including but not limited to steel bar spacing that does not meet the design requirements, loose binding, and steel bar misalignment. During the classification process, the model will accurately locate the defect location in the construction image according to the defect type and generate a defect mark. For example, if the system detects that the steel bar is loosely tied, it will highlight the area in the construction image and mark it with the label "loose tie". Based on the identified defect type, the system generates corresponding processing suggestions and correction steps. The processing suggestions and correction steps are provided by the expert knowledge base in the system and contain detailed operation instructions. For example, for defects where the steel bar spacing does not meet the requirements, the system will suggest readjusting the steel bar spacing and provide specific operation steps; for defects where the tie is loose, the system will suggest re-binding and specify the specific binding method. Finally, based on the defect location and processing suggestions, the system generates a pending process and outputs it to the construction management system. The pending process includes detailed information such as the defect location, defect type, processing suggestions, and correction steps. For example, the process to be processed generated by the system may include the following: "Location: Area A, 3rd row of steel bars; Defect type: Spacing is too large; Treatment suggestion: Adjust the spacing of steel bars to 20 cm; Operation steps: Reposition the steel bars to ensure that the spacing is 20 cm, and use binding wire to fix them." In this way, construction management personnel can intuitively and accurately understand the defects in the current construction, and make corrections according to the treatment suggestions and steps provided by the system to improve construction quality and efficiency. For example, in a high-rise building project, the system identified that the spacing of steel bars on a certain floor did not meet the design requirements by collecting construction images in real time, and generated a detailed process to be processed. According to the system's treatment suggestions, the construction personnel adjusted and corrected the spacing of steel bars in a timely manner to ensure the compliance and standardization of construction quality.

[0076] The above implementation modes are merely descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering and technical personnel in the field shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for generating a plate reinforcement arrangement process based on image recognition, characterized in that: The following steps are involved: S1. Obtain the slab reinforcement coverage area and slab reinforcement arrangement type in the slab reinforcement historical construction image of the building, and obtain the reinforcement spacing and reinforcement binding spacing based on the slab reinforcement historical construction image; S2, using the plate reinforcement coverage area and the plate reinforcement arrangement type as input, and the reinforcement spacing and the binding spacing as output, training the first neural network model; S3, based on the historical construction images of slab reinforcement, learning the construction stages of different slab reinforcement arrangement types through the second neural network model; S4, obtaining the current construction image, determining the plate reinforcement arrangement type according to the current construction area, judging the current construction stage according to the plate reinforcement arrangement type and the current construction image through the second neural network model, and performing defect verification through the first neural network model, generating a process to be processed according to the defect verification result, and if the current process to be processed has been completed, re-execute S4; if the current construction stage is the final stage and the processes to be processed have been completed, terminate the execution; The defect verification is performed by the first neural network model, and the process to be processed is generated according to the defect verification result, including the following steps: Input the construction type corresponding to the current construction stage and the current construction image into the trained first neural network model to identify and detect defects in the steel bar arrangement, including steel bar spacing and binding spacing that do not meet design requirements; Classify and mark the identified defects and locate the defects in the construction images; Generate corresponding processing suggestions and correction steps based on the defect type, and generate the process to be processed based on the defect location and processing suggestions.

2. The method for generating a plate reinforcement arrangement process based on image recognition according to claim 1, characterized in that: The plate reinforcement arrangement types include: a single-layer reinforcement arrangement type, a double-layer reinforcement arrangement type, a grid reinforcement arrangement type and a reinforced reinforcement arrangement type, wherein the reinforced reinforcement arrangement type is reinforced by adding additional reinforcements in the center of the grid reinforcement.

3. The method for generating a plate reinforcement arrangement process based on image recognition according to claim 1, characterized in that: The method of obtaining the steel bar spacing and steel bar binding spacing based on the historical construction image of the plate steel bar comprises the following steps: Preprocess the historical construction images of the plate reinforcement to generate a planar grayscale image of the plate reinforcement; The image recognition algorithm is used to mark the steel bar positions and steel bar intersections in the planar grayscale image; The rebar spacing is determined by calculating the distance between adjacent rebars using rebar locations and rebar intersections.

4. The method for generating a plate reinforcement arrangement process based on image recognition according to claim 1, characterized in that: The formula of the first neural network model is as follows: ， in, is the output vector, which represents the predicted value of the reinforcement spacing and the binding spacing; is the input vector, including the characteristics of the slab reinforcement coverage area and the slab reinforcement arrangement type; is the weight matrix of the hidden layer; is the bias vector of the hidden layer; is the activation function of the hidden layer; is the weight matrix of the output layer; is the bias vector of the output layer; is the activation function of the output layer.

5. The method for generating a plate reinforcement arrangement process based on image recognition according to claim 1, characterized in that: The second neural network is a combined neural network, including a convolutional neural network and a BP neural network.

6. The method for generating a plate reinforcement arrangement process based on image recognition according to claim 5, characterized in that: The S3 comprises the following steps: Pre-classify different construction stages for different slab reinforcement arrangement types, and label the slab reinforcement arrangement types and corresponding construction stages; The stage image features in the historical construction images of plate reinforcement are extracted through convolutional neural networks; The BP neural network is trained with the slab reinforcement arrangement type label and stage image features as input and the construction stage label as output; Optimize BP neural network parameters through back propagation.

7. The method for generating a plate reinforcement arrangement process based on image recognition according to claim 1, characterized in that: Determining the slab reinforcement arrangement type according to the current construction area comprises the following steps: Obtain the current construction area, where the construction area is the required steel bar coverage area in the construction site; The current construction area is compared with the preset slab reinforcement arrangement type coverage area standard to determine several slab reinforcement arrangement types.

8. The method for generating a plate reinforcement arrangement process based on image recognition according to claim 7, characterized in that: The determining of the current construction stage according to the plate reinforcement arrangement type and the current construction image by the second neural network model includes: Taking several slab reinforcement arrangement types and the current construction image as input, the second neural network model outputs several construction stage labels; Get the number of times a tag appears in several construction stage tags and sort them in descending order, and determine the current construction stage based on the descending order results.

Images