AI-based metal roof construction quality detection system

Abstract

The application relates to the technical field of roof image recognition, in particular to a metal roof construction quality detection system based on AI, which comprises an included angle mutation recognition module, a tangential abnormality extraction module, a curvature stability recognition module, a path fitting offset analysis module and a defect type judgment module. In the application, by constructing a boundary node included angle change rate sequence, a mutation point of a metal roof splicing part is accurately recognized, and a local structure change trend is effectively revealed; in combination with a tangential displacement sequence and an included angle mutation feature between adjacent boundary nodes, a direction continuity imbalance section is captured, abnormal direction variation presents a dynamic evolution track; on the basis, a stability discrimination is further conducted on a boundary node curvature distribution state, a curvature fluctuation concentrated area and an extreme point are recognized through differential analysis, and the extraction of geometric performance of abnormal paving precision is strengthened; in time sequence evolution, residual extraction capability of an abnormal trend can be obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of roof image recognition, and in particular to an AI-based metal roof construction quality detection system. BACKGROUND

[0002] The technical field of metal roof construction quality detection systems includes applications related to roof image recognition. This technical field is mainly based on computer vision and image processing, and through the collection and analysis of images of roof structures, defects, cracks, splicing errors, and quality problems during the construction process are identified.

[0003] Among them, the metal roof construction quality detection system refers to a system designed for quality detection during the construction process of metal roofs. It mainly uses image acquisition devices to collect images of roof structures at multiple angles and time periods, and uses image recognition methods to detect the splicing gaps of roof panels, panel misplacement, connection boundary node abnormalities, surface cracks, and sealing tape laying conditions.

[0004] In the prior art, static detection is performed based on image acquisition and recognition methods, which excessively relies on the boundary clarity and material characteristics of single-frame images. There is a lack of continuous analysis methods for structural defects, especially boundary misplacement and node deformation, making it difficult to identify abnormal areas in dynamic evolution. Only surface manifestations such as gaps and cracks are detected, ignoring changes in structural geometry and path continuity, which makes it difficult to discover deep-level laying deviations such as angle fractures and curvature disturbances in actual applications. The image recognition process lacks time series modeling capability and does not have a systematic extraction strategy for the outlying state of node distribution density changes, which can easily misjudge or miss key defect segments in dense structures. For example, in the connection transition area, if the local instability of the node does not manifest as a clear crack, only the tangential continuity is lost, traditional methods cannot detect such potential risks, which severely restricts the accuracy and coverage of the whole cycle monitoring of laying quality, causing an increase in maintenance costs and an increase in safety hazards. SUMMARY

[0005] The purpose of the present application is to solve the shortcomings in the prior art and to provide an AI-based metal roof construction quality detection system.

[0006] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme: the AI-based metal roof construction quality detection system comprises:

[0007] The included angle mutation recognition module obtains all boundary node sequences of the target position of the metal roof, extracts the included angle change rate of adjacent boundary nodes in the node sequence, screens the included angle change mutation points, and constitutes a first boundary node abnormality preliminary screening set;

[0008] The tangential anomaly extraction module obtains a tangential displacement sequence between adjacent boundary nodes of a target position of the metal roof through the first boundary node anomaly preliminary screening set, and performs continuity analysis on a tangential gradient to select boundary nodes with continuous anomalies to form a first boundary node set of continuous directional anomalies;

[0009] The curvature stability identification module obtains a sequence of all boundary nodes of the target position of the metal roof, judges the stability of the curvature variation state of the node sequence, and obtains a second boundary node set of curvature distribution anomalies;

[0010] The path fitting offset analysis module inputs the first boundary node set of continuous directional anomalies and the second boundary node set of curvature distribution anomalies into a bidirectional long short-term memory network to perform path fitting offset analysis, and constructs a path fitting offset sequence;

[0011] The defect type determination module performs anomaly type detection on the laying path of the target position of the metal roof based on the path fitting offset sequence, and obtains a metal roof quality detection result.

[0012] As a further scheme of the present application, the first boundary node anomaly preliminary screening set includes an angle change sudden increase point, a normal vector mutation boundary node, and a boundary direction fracture point, the first boundary node set of continuous directional anomalies specifically includes a multi-segment continuous mutation boundary node sequence, a tangential angle discontinuous region, and a local path disturbance zone, the second boundary node set of curvature distribution anomalies includes a curvature extreme point, a curvature fluctuation concentrated area, and a boundary curvature instability section, the path fitting offset sequence specifically refers to a fitting residual value sequence, a fitting trend mutation area, and a fitting error time sequence distribution, and the metal roof quality detection result includes a path tearing zone, a corner twisting zone, and a curvature instability zone.

[0013] As a further scheme of the present application, the angle mutation identification module includes:

[0014] The normal vector extraction submodule collects image data of the target position of the metal roof, obtains an arrangement order of the boundary nodes in a three-dimensional coordinate space, calculates a normal vector direction constructed by adjacent boundary nodes, and establishes a boundary node normal direction sequence based on spatial coordinate data between adjacent normal vectors;

[0015] The angle increment calculation submodule selects two normal vectors formed by every group of three adjacent boundary nodes in the boundary node normal direction sequence, calls a directional angle value between each group of vectors and constructs an angle difference value sequence, performs comparative analysis on the angle variation amplitude of the boundary nodes, and obtains a boundary node angle change rate sequence;

[0016] The mutation point screening submodule calls the angle change rate amplitude of each boundary node in the boundary node angle change rate sequence and sequentially compares the boundary node change trend, identifies the boundary node with inconsistent continuous change direction and amplitude mutation as a feature point, and generates a first boundary node abnormal preliminary screening set.

[0017] As a further scheme of the present application, the tangential anomaly extraction module comprises:

[0018] The tangential vector construction submodule obtains the spatial coordinate information of the adjacent boundary nodes of the metal roof target position based on the first boundary node abnormal preliminary screening set, sequentially connects the adjacent boundary nodes to form a boundary line segment direction vector sequence, uniformly normalizes each group of boundary node direction vectors, and establishes a boundary tangential direction sequence;

[0019] The gradient sequence extraction submodule calls the adjacent vector pairs in the boundary tangential direction sequence, obtains the direction mutation degree between the continuous boundary nodes by comparing and operating the direction included angle, constructs a continuous sequence reflecting the overall path change trend, and obtains a tangential gradient sequence;

[0020] The abnormal section identification submodule sequentially extracts the boundary node direction change amplitude according to the tangential gradient sequence, judges the continuity of the change interval, and when the boundary node direction mutation continuously occurs and the gradient change trend is unstable, the corresponding boundary node section is delimited, and a first boundary node set with continuous direction anomaly is obtained.

[0021] As a further scheme of the present application, the curvature stability identification module comprises:

[0022] The coordinate sequence acquisition submodule acquires the boundary image of the metal roof target position, extracts all boundary node positions on the boundary line segment, and forms a spatial point sequence in the order of the boundary node numbers, and constructs a boundary node combined coordinate sequence from each group of adjacent three boundary node coordinates in the sequence;

[0023] The curvature trend extraction submodule constructs the geometric relationship among the continuous three points according to the boundary node combined coordinate sequence, extracts the curvature values of the geometric relationship arcs, and statistically obtains the continuous trend of the center difference values between the adjacent curvatures, and obtains a curvature change trend sequence;

[0024] The instability point identification submodule compares the amplitudes of the changes of each boundary node in the curvature change trend sequence, judges the fluctuation state of the continuous change segment formed by the adjacent multiple boundary nodes, and screens a second boundary node set with abnormal curvature distribution.

[0025] As a further scheme of the present application, the path fitting offset analysis module comprises:

[0026] The sequence construction submodule extracts the normal vector included angle change rate of the curvature abnormal boundary node in the corresponding image frame according to the first boundary node set of the continuous direction anomaly, extracts the curvature change gradient between adjacent coordinate boundary nodes according to the second boundary node set of the curvature distribution anomaly, integrates the two types of boundary node sets into an independent time sequence structure, and generates an input feature sequence set;

[0027] The feature combination submodule calls the time window segment of each type of sequence in the input feature sequence set, performs time sequence feature splicing in the order of boundary node number, and generates a path sequence feature group;

[0028] The fitting training submodule inputs the path sequence feature group into a bidirectional long short-term memory network, learns and extracts boundary node output residual values based on the time dimension expansion of the boundary node space variation trend, and obtains a path fitting offset sequence.

[0029] As a further scheme of the application, the defect type determination module comprises:

[0030] The abnormal segment extraction submodule extracts a time sequence segment with a continuous offset amplitude in the path fitting offset sequence based on the boundary node output residual values, matches the corresponding boundary node number and sequence position, and obtains a continuous abnormal offset segment set;

[0031] The outlier area identification submodule calls the boundary node offset data in the continuous abnormal offset segment set, performs density ratio judgment through a local anomaly factor algorithm, identifies the time sequence segment with sparse local distribution and lower density than the surrounding boundary nodes in the offset sequence, and obtains an outlier boundary node section.

[0032] The defect type determination submodule performs classification and judgment on the boundary node shape according to the spatial structure position and included angle change trend of the boundary nodes in the outlier boundary node section, combines the boundary node boundary trend and curvature fluctuation characteristics, and generates a metal roof defect type result.

[0033] Compared with the prior art, the application has the advantages and positive effects that:

[0034] In this invention, by constructing a sequence of boundary node angle change rates, abrupt change points at the splicing parts of the metal roof are accurately identified, effectively revealing the local structural change trend. Combining the tangential displacement sequence and angle change characteristics between adjacent boundary nodes, sections with unbalanced directional continuity are captured, allowing abnormal directional variations to exhibit a dynamic evolution trajectory. Furthermore, the stability of the curvature distribution state of the boundary nodes is determined, and differential analysis identifies concentrated curvature fluctuation areas and extreme points, enhancing the extraction of geometric representations of abnormal laying accuracy. The path fitting process embeds the structural space change sequence into a time-domain learning model, combining normal angle changes and curvature gradients to construct a feature sequence, obtaining residual extraction capabilities for abnormal trends during temporal evolution, making abnormal trends more three-dimensional and traceable. Further, the abnormal offset segments and density factor algorithms are integrated to identify outliers in the laying path, and classification is completed based on the geometric position and orientation of the boundary nodes, effectively delineating the structural range of various defects such as tearing, torsion, and instability. Attached Figure Description

[0035] Figure 1 This is a system flowchart of the present invention;

[0036] Figure 2 This is a flowchart of the angle mutation recognition module of the present invention;

[0037] Figure 3 This is a flowchart of the tangential anomaly extraction module of the present invention;

[0038] Figure 4 This is a flowchart of the curvature stability recognition module of the present invention;

[0039] Figure 5 This is a flowchart of the path fitting offset analysis module of the present invention;

[0040] Figure 6 This is a flowchart of the defect type determination module of the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0042] In the description of the present application, it should be understood that the terms "length", "width", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, in the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise explicitly and specifically limited.

[0043] Please refer to Figure 1 The AI-based metal roof construction quality detection system comprises:

[0044] The included angle mutation identification module obtains the entire boundary node sequence of the target position of the metal roof, extracts the included angle change rate of adjacent boundary nodes in the node sequence, screens the included angle change mutation points to form a first boundary node abnormal preliminary screening set;

[0045] The tangential anomaly extraction module obtains the tangential displacement sequence between adjacent boundary nodes of the target position of the metal roof through the first boundary node abnormal preliminary screening set, and analyzes the continuity of the tangential gradient, and selects the boundary nodes with continuous anomalies to form a first boundary node set with continuous direction anomalies;

[0046] The curvature stability identification module obtains the entire boundary node sequence of the target position of the metal roof, judges the stability of the curvature change state of the node sequence, and obtains a second boundary node set with curvature distribution anomalies;

[0047] The path fitting offset analysis module inputs the first boundary node set with continuous direction anomalies and the second boundary node set with curvature distribution anomalies into a bidirectional long-short term memory network for laying path fitting offset analysis, and constructs a path fitting offset sequence;

[0048] The defect type determination module performs anomaly type detection on the laying path of the target position of the metal roof based on the path fitting offset sequence, and obtains a metal roof quality detection result;

[0049] The first boundary node abnormal preliminary screening set includes included angle change sudden increase points, normal vector mutation boundary nodes and boundary direction fracture points, the first boundary node set with continuous direction anomalies specifically includes a plurality of continuous mutation boundary node sequences, tangential included angle discontinuous regions and local path disturbance zones, the second boundary node set with curvature distribution anomalies includes curvature extreme points, curvature fluctuation concentrated areas and boundary curvature instability sections, the path fitting offset sequence specifically refers to a fitting residual value sequence, a fitting trend mutation area and a fitting error time sequence distribution, and the metal roof quality detection result includes a path tearing zone, a corner twisting zone and a curvature instability zone.

[0050] Referring to Figure 2 , the corner abruptness recognition module comprises:

[0051] The normal vector extraction submodule collects image data of the target position of the metal roof, obtains the arrangement order of the boundary nodes in the three-dimensional coordinate space, calculates the direction of the normal vector constructed by adjacent boundary nodes, and establishes a normal direction sequence of the boundary nodes based on the spatial coordinate data between adjacent normal vectors;

[0052] The image acquisition device is positioned directly above the target area of the metal roof to ensure that the captured image has a vertical viewing angle and eliminates perspective distortion interference. The target area is scanned and photographed at a frame rate of 0.5 fps through a continuous shutter mode, at least 5 image frames covering the complete boundary are obtained, the edge node coordinates in each image frame are mapped from two-dimensional pixel coordinates to a three-dimensional rectangular coordinate system using an image coordinate system conversion algorithm, the corresponding space conversion coefficients are set according to the material parameters and roof configuration characteristics of the metal roof (such as reflectivity of 0.85 and thickness of 5 mm), and the endpoint information of each boundary line segment in the image is called to construct a boundary node array , wherein represents the i-th boundary node, , and are the transverse, longitudinal and elevation coordinate components of the i-th boundary node in the three-dimensional space, respectively, is the index of the node in the boundary sequence, and the value range is , and the difference vector between any two adjacent nodes such as and is calculated, wherein represents the vector from the i-th node to the j-th node, and is constructed according to the vector cross product operation combined with the three points of the local patch , and the patch normal is calculated, represents the patch normal vector composed of the boundary adjacent vectors, and the unit normal vector is obtained by normalization , and is the unit normal vector of the i-th normal direction after normalization, and finally the normal vector sequence of the boundary nodes is constructed in the node order , and the arrangement order is determined by the clockwise or counterclockwise direction of the boundary nodes in the image, for example, selecting any three points , , , , , the cross product is , and the repetition of the process completes the spatial arrangement of the normal vectors of all adjacent boundary nodes.

[0053] The included angle increment calculation sub-module selects two normal vectors formed by each group of three adjacent boundary nodes in the normal direction sequence, calls the included angle value between each group of vectors, and constructs an included angle difference value sequence, compares the change amplitude of the included angle of the boundary nodes, and obtains the boundary node included angle change rate sequence;

[0054] In the established normal direction sequence, each group of three consecutive normal vectors is selected in turn, for example , wherein represents the unit normal vector at the th node, , the inverse cosine function is called between the vector pairs and to calculate the included angle values and , wherein is the included angle value of the unit vector between nodes and , and is the included angle value of the unit vector between nodes and , and the difference value represents the amplitude of the normal included angle change of the node , and forms an included angle difference value sequence , wherein represents the included angle change value of the th three-point group, and the dot product value of the adjacent vectors needs to be determined in the interval , otherwise the abnormal vector group should be removed, for example , then , , , and the normal included angle change value between each node is obtained in turn, recorded in the form of a sequence, and each difference value needs to be stored in an array structure in the form of a measurement unit (unit: °).

[0055] The mutation point screening sub-module calls the included angle change rate amplitude of each boundary node in the boundary node included angle change rate sequence and compares the boundary node change trend in sequence, identifies the boundary nodes with inconsistent continuous change direction and amplitude mutation as feature points, and generates a first boundary node abnormal preliminary screening set;

[0056] For each boundary node , read its corresponding included angle change rate , represents the included angle change value at the node , and the absolute value comparison is performed with the difference amplitude of the previous node , wherein ​denotes the angle change amount of the previous node, if the value changes in sign and the amplitude is greater than the set threshold value for two consecutive times , wherein is the threshold value for the angle mutation judgment, the unit is angle, and the experience setting range is , it is determined that the node has a mutation feature, and is preliminarily selected as a boundary feature point, such as , , , the node is recorded, and it is further judged whether the change is changed in direction, if the previous difference value is increased , the latter difference value is decreased , that is, there is a direction inconsistency condition, the node is recorded as an abnormal mutation point, and an abnormal preliminary screening set is constructed , wherein the set represents a set of normal mutation boundary feature points screened out preliminarily, and outputs the node index sequence for subsequent feature extraction operation.

[0057] Please refer to Figure 3 , the tangential abnormality extraction module includes:

[0058] The tangential vector construction submodule obtains the spatial coordinate information of the adjacent boundary nodes of the target position of the metal roof based on the first boundary node abnormal preliminary screening set, sequentially connects the adjacent boundary nodes to form a boundary line segment direction vector sequence, and uniformly normalizes each group of boundary node direction vectors to establish a boundary tangential direction sequence.

[0059] After obtaining the first boundary node abnormal preliminary screening set , each abnormal boundary node in the set is traversed in sequence , wherein denotes the i-th node in the abnormal preliminary screening set, denotes the index value of the node in the original boundary node sequence, the spatial coordinate values of the adjacent boundary nodes before and after the node are called , and , wherein denotes the horizontal, vertical and vertical coordinate values of the i-th node in the three-dimensional space, respectively, and the boundary direction vector is constructed through the spatial coordinate difference value , , wherein denotes the tangential direction vector from the i-th node to the next node, denotes the node index in the boundary line segment sequence. Then, each is uniformly normalized, and the normalization method is , wherein denotes the normalized i-th node , , , a unit tangent vector, Euclidean norm (length) of vector is calculated as:

[0060] ;

[0061] where and are the coordinate components of the start and end points, respectively. Taking points , as an example, , the length is , and after normalization , the sequence of unit tangent vectors is established , where represents the th normalized unit tangent vector, which is used as the boundary tangent direction sequence for subsequent analysis. The order of the vector sequence is arranged in order of the node spatial coordinate index, ensuring path continuity.

[0062] The gradient sequence extraction submodule calls adjacent vector pairs in the boundary tangent direction sequence, compares the direction angles, obtains the degree of direction mutation between consecutive boundary nodes, and constructs a continuous sequence reflecting the overall path change trend, obtaining the tangent gradient sequence;

[0063] For the tangent unit vector sequence , where represents the normalized unit vector of the th boundary line segment direction, every two adjacent vector pairs are selected, and their angle values are calculated to determine the degree of change in the tangent direction. The angle calculation method is , where represents the angle value between the th and th unit vectors, with units of degrees. The dot product calculation needs to ensure that the vectors are normalized to maintain ; in actual sampling, if , , then the dot product is , and the corresponding angle is . In this way, all the angles between adjacent vectors are constructed into a sequence , where represents the direction angle of the th pair of adjacent vectors, reflecting the degree of direction turning of the path. This sequence serves as an important reference for characterizing the continuity and polyline density of the path trend.

[0064] The abnormal section identification submodule extracts the direction change amplitude of the boundary nodes sequentially according to the tangential gradient sequence, makes a continuous judgment on the change interval, and delineates the corresponding boundary node section when the abrupt change of the boundary node direction continues and the gradient change trend is unstable, and obtains the first boundary node set with continuous directional anomalies.

[0065] Obtaining the tangential gradient sequence Then, the variation range of each included angle value is extracted sequentially and continuity analysis is performed. The determination method is as follows: for any continuous subsequence ,in This indicates the first sub-segment. If there are at least three consecutive included angle values ​​that satisfy:

[0066] , ,in This represents the angle threshold difference used to identify abrupt changes in direction, in degrees, with an empirical range of [value missing]. If the sequence exhibits an abnormal directional trend, then the stability of the gradient sequence's change trend is assessed, defining the gradient change rate as... ,in Indicates the first and For the absolute magnitude of the change in the angle between vectors, if there exists in this segment... It exhibits oscillating changes and satisfies ,in The threshold value representing the magnitude of gradient mutation trends, in degrees, is recommended to be [value missing]. If the direction of this segment changes abruptly and becomes unstable, then a continuous anomalous direction segment is defined by combining the two conditions. All boundary nodes that meet the conditions are then... Add to continuous anomaly boundary node set ,in This represents the first set of boundary nodes for the continuous directional anomalies finally identified.

[0067] Please see Figure 4 The curvature stability recognition module includes:

[0068] The coordinate sequence acquisition submodule acquires the boundary image of the target location of the metal roof, extracts the positions of all boundary nodes on the boundary line segment, and forms a spatial point sequence according to the boundary node number order. The coordinates of each group of three adjacent boundary nodes in the sequence are used to construct a boundary node combined coordinate sequence.

[0069] In the boundary extraction process of the target position of the metal roof, the boundary image of the target area needs to be extracted through the coordinate sequence first. In practical applications, the metal roof usually has a specific geometric shape, so the image processing module first acquires an image containing the contour of the metal roof, which is a binary image, where the boundary line is the white part. Next, an edge detection algorithm (such as the Canny algorithm) is used to detect the connected regions of edges in the image, marking out the boundaries in the image. Through algorithm processing, the coordinate information of the boundary line segments can be obtained, and then the coordinates are extracted to obtain the node positions of each boundary line segment. The further step involves arranging these nodes in order of numbering and forming a spatial point sequence. For example, assume that the first boundary node extracted is A(10, 15), the second node is B(20, 25), the third node is C(30, 35), and so on. On this basis, every three consecutive adjacent nodes form a boundary node combination. For example, if A, B, and C are three consecutive nodes, the boundary node combination is . This combination method can further adjust the number of nodes according to different needs, for example, four nodes can be considered to form a combination, but in actual implementation, the use of three-point combination is more common. This step finally forms a continuous boundary node combination coordinate sequence, providing data support for the subsequent curvature trend extraction module.

[0070] The curvature trend extraction submodule constructs the geometric relationship between consecutive three points according to the boundary node combination coordinate sequence, extracts the curvature value of the geometric relationship arc, and counts the continuous trend of the center difference value between adjacent curvatures to obtain the curvature change trend sequence.

[0071] By analyzing the boundary node combination coordinate sequence, the geometric relationship between consecutive three nodes is first constructed. The geometric relationship is calculated by calculating the curvature of the geometric figure formed by the three adjacent nodes. Assume that three consecutive boundary nodes , , form a three-point combination, and by calculating the degree of curvature between these nodes, the curvature value between the three points can be determined. When calculating, the distance between each pair of adjacent points needs to be calculated first, such as the distance between and is

[0072] .

[0073] Similarly, the distance between and is

[0074] .

[0075] Then, by these distances and the geometric relationship of the three points, the curvature values between the three points are calculated. Each curvature value represents the degree of bending of this segment of the boundary. Next, based on the curvature values calculated for all boundary node combinations, the differences between adjacent curvatures are extracted and a trend sequence is formed. For example, if the first two curvature values are and , then their central difference value is In this way, the trend of curvature change is extracted.

[0076] The instability point identification submodule compares the magnitude of the change in the curvature trend sequence for each boundary node, judges the fluctuation state of the continuous change segment formed by adjacent boundary nodes, and screens a second set of boundary nodes with abnormal curvature distribution;

[0077] The magnitude of the change in the curvature trend sequence is analyzed, and potential instability points are identified. In specific implementation, the curvature trend of each boundary node is compared with a set magnitude threshold. If the difference between adjacent curvature values exceeds the preset threshold, it indicates that there is a risk of instability. The set threshold is usually adjusted based on historical data or experimental results. Assuming a threshold of 0.1 is set, if the curvature change between two adjacent nodes is greater than 0.1, it is considered a potential instability point. For example, assuming that in a certain boundary, the curvature values of nodes to increase from 0.1 to 0.3, the magnitude difference is , which exceeds the set threshold of 0.1, so node is determined to be an instability point. In this way, the system can identify a set of nodes with abnormal curvature distribution, and further judge and screen out areas that may need to be repaired or monitored. This step forms a second set of boundary nodes through magnitude comparison, fluctuation state judgment, and abnormal screening.

[0078] Please refer to Figure 5 , the path fitting offset analysis module includes:

[0079] The sequence construction submodule extracts the normal vector angle change rate of the curvature abnormal boundary nodes in the corresponding image frame according to the continuously directionally abnormal first set of boundary nodes, and extracts the curvature change gradient between adjacent coordinate boundary nodes according to the second set of boundary nodes with abnormal curvature distribution. The two sets of boundary nodes are integrated into independent time sequence structures to generate an input feature sequence set.

[0080] The rate of change of the normal vector angle between the first set of boundary nodes with continuous directional anomalies is extracted from the corresponding image frame to find the boundary nodes with curvature anomalies. This process begins by extracting the boundary node set from the image frame. These nodes exhibit obvious directional anomalies due to boundary discontinuity interruptions, folding deformations, or local concave-convex structures. For these nodes, the rate of change of the normal vector angle between adjacent nodes needs to be calculated. Assume the current node to be analyzed is the first... Each boundary node (denoted as node i with an anomalous index direction) has a normal vector. Its preceding and following adjacent nodes are respectively the first and second nodes. The and the first A boundary node. Here These represent the normal vector of the node along the path in the three-dimensional coordinate system. , , The components of the axis, expressed as dimensionless direction ratios, represent unit vectors; while the... The normal vector of each boundary node (i.e., the next node in sequence, denoted as node i plus one) is denoted as... ,in Similarly, this is its directional component along the three-dimensional coordinate axes. At this point, the included angle... The calculation formula is:

[0081] ;

[0082] To further calculate the rate of change of the included angle, we can use unit time or the distance between nodes as the denominator. Assuming the unit node spacing is 1, the rate of change is approximately equal to the difference between adjacent included angles. Example: If nodes... Normal vector ,node Normal vector The dot product is The module lengths are respectively Similarly, the second modulus is also 1, therefore: If the included angle of the previous node is The rate of change of the included angle is This is used to construct time series of directional mutations at boundary nodes.

[0083] The feature combination submodule calls the time window segment of each type of sequence in the input feature sequence set, and splices the time-series features according to the boundary node number order to generate path sequence feature groups;

[0084] Extract the curvature change gradient between adjacent coordinate boundary nodes based on the second boundary node set with anomalies in curvature distribution. For each of the following paths... There are boundary nodes (where the subscript "curvature" indicates that the node belongs to the curvature anomaly detection subset), and their corresponding curvature values ​​are denoted as . , which is derived from the curvature measure of the curve determined in the three-point fitting method, in units of "degree / meter" , describes the magnitude of the change in the turning angle of the unit length path segment at the node. The adjacent node is the boundary node, denoted as node , whose curvature value is , indicating the degree of curvature of the subsequent boundary. To calculate the curvature change gradient between the two nodes, the spatial distance is obtained, and the Euclidean distance is set as , in units of "meters". The calculation formula of the curvature change gradient is:

[0085] For example, assuming that the curvature value at node is , the curvature at node is , and the actual distance in the image coordinate space is , then the curvature change gradient is: , which indicates that the curvature increases at a rate of per meter of path length between the 5th and 6th nodes. This result can be used to identify areas of sudden curvature changes in the boundary segment. This calculation is sequentially performed on all adjacent node pairs in the entire curvature anomaly second node set, gradually constructing a complete sequence of curvature change gradients.

[0086] The fitting training sub-module inputs the path sequence feature group into a bidirectional long short-term memory network, learns the spatial variation trend of the boundary nodes based on the time dimension, extracts the output residual values of the boundary nodes, and obtains the path fitting offset sequence;

[0087] The path sequence feature group is input into a bidirectional long short-term memory network (BiLSTM), the spatial variation trend of the boundary nodes is learned based on the time dimension, and the output residual values of the boundary nodes are extracted to obtain the path fitting offset sequence. The path sequence feature group is composed of the normal vector angle change rate sequence of the direction anomaly first boundary node set and the curvature change gradient sequence of the curvature anomaly second boundary node set, forming the node input feature vector at time step , denoted as , where: : represents the normal vector angle change rate of the boundary node numbered in the direction anomaly set at time step , in units of "degrees / second" ; : represents the curvature change gradient of the boundary node numbered in the curvature anomaly set at time step , in units of "degrees / meter²" ). The bidirectional LSTM neural network structure learns from the combination of forward and backward states, preserving the state dependence of past and future nodes simultaneously in the context of time series. Its state update formula is as follows:

[0088] ;

[0089] wherein: : current time step : hidden state of the current time step, representing the boundary node : change trend of the spatial position; , : forward (previous time) and backward (next time) states; : forward and backward state adjustment factor (set to 0.8, based on the average fitting precision maximization result of the training set cross-validation); , : weight matrix of forward and backward states, respectively (set to random value in the initialization training stage, then adjusted by gradient descent, finally determined as 0.7 and 0.6); : weight matrix of feature input (set to 0.8, determined by training error convergence); : bias term (set to 0.1); : hyperbolic tangent activation function, mapping the linearly superimposed state output to interval.

[0090] Assuming the current time step is , the previous time state is , the next time state is , and the input feature is: , , the input feature of node 5 is: combined into a scalar input: , substitute the formula to calculate:

[0091] ;

[0092] The hidden state is used to generate the output prediction value of the network, and the result is used for residual error calculation with the actual node position value. Assuming the true observed spatial position value of node 5 is , and the prediction value is , the output residual error value is: Finally, all time step residual error values The path fitting offset sequence is combined in chronological order: this sequence is the offset error measure of each node on the predicted path relative to the true path, which is used as data input for subsequent structural monitoring or correction steps. The results show that there is a direct quantitative correspondence between residual calculation and path space difference, and the BiLSTM network output can effectively reflect the trend of boundary node trajectory change and the difference between the model prediction.

[0093] Please refer to Figure 6 , the defect type determination module comprises:

[0094] The abnormal segment extraction submodule extracts the time sequence segment with a continuous offset amplitude that is prominent in the boundary node output residual value in the path fitting offset sequence, matches the corresponding boundary node number and sequence position, and obtains a continuous abnormal offset segment set;

[0095] Based on the residual value of the boundary node in the path fitting offset sequence, the residual value sequence of each time step (for example, the residual value is 0.02 meters at the first time, 0.03 meters at the second time, and so on to the T time) is analyzed in sequence, wherein each residual value represents the absolute difference between the actual coordinate position of the boundary node at the time step and the predicted coordinate position of the BiLSTM, and the unit is meter. The value reflects the fitting accuracy error of the boundary node. First, linearly traverse the entire time sequence, and in the traversal process, compare the residual value of each time step with the offset reference threshold set by the system. The threshold is , the value is 0.05 meters, which comes from the maximum local node displacement offset limit value acceptable to the metal roof structure in the industry standard. If the residual values of any three consecutive time steps (for example, the 8th, 9th, and 10th time) are all greater than , mark this segment as an offset anomaly, continue to judge whether the subsequent time steps meet the same condition, if they do, merge them into the current segment until a time step residual value falls below the threshold. The time point is the end point of the current abnormal segment, for example, the 11th and 12th time still meet the offset condition, and the 13th time is less than 0.05 meters. Finally, the abnormal segment time is determined to be 8 to 12, and the boundary node number is J8 to J12 (J is the node number prefix, and the subscript is the sequence index). Then match the segment time index position with the spatial boundary node number one by one to form an abnormal offset segment set, such as segment {J8, J9, J10, J11, J12}. All time segments that meet the offset condition are recorded and numbered in the above manner.

[0096] The outlier zone identification submodule calls the boundary node offset data in the continuous abnormal offset segment set, judges the density ratio through the local anomaly factor algorithm, identifies the time sequence segment with sparse local distribution and lower density than the surrounding boundary nodes in the offset sequence, and obtains the outlier boundary node section.

[0097] The residual values of each boundary node in the continuous abnormal offset segment set are obtained, each boundary node is processed one by one, a local neighborhood residual set centered on the boundary node is constructed, density analysis is performed on the local neighborhood residual set, and the density measurement mode is to calculate the reachable density of the node in the residual neighborhood of the node, which is defined as the residual value distance relationship between the node and its k nearest neighbor nodes, wherein k is the number of neighborhoods, and k is set to 5 in the embodiment, which is derived from the minimum support number used to construct the local distribution stability judgment in statistical analysis, to prevent misjudgment caused by a single isolated value in the abnormal recognition process. For the boundary node J10 (numbered J10, wherein J represents a boundary node and 10 is a time or space sequence subscript), assuming that the residual value of J10 is 0.11 meters, the residual values of the five nodes adjacent to J10 in the time sequence are {0.06 meters, 0.08 meters, 0.09 meters, 0.07 meters, 0.05 meters}, and the local point set constructed with J10 as the center is used to evaluate the distribution density difference of J10 in the offset dimension. The specific abnormal judgment index is the local anomaly factor, and the calculation formula is:

[0098] ;

[0099] In the formula: represents the current evaluation node (for example, J10), and the subscript is the number of the node to be judged; represents the k-neighborhood set of node , that is, the k nodes closest to node in the residual dimension, for example, the five neighbors of node J10 are J6, J7, J8, J9 and J11; represents any one adjacent node selected from the set , which is used for density comparison with ; represents the number of elements in the set, that is, the value of k, which is 5 here; is the local reachable density of node , which is defined as the reciprocal of the average value of the reachable distance of node and all points in its k-neighborhood, and the unit is , which represents the density; is the local reachable density of the neighbor node ; is the density ratio, and the greater the average ratio, the more sparse node is relative to the neighborhood.

[0100] Suppose the reachable density of node J10 is = 33.33 (the calculation method is that the average value of the five reachable distances is 0.03 meters, and the reciprocal is 33.33), and the average neighbor density is 25, then the calculation result is: If the system sets the LOF threshold to 1.5 (this value is derived by comparing anomalous and normal nodes in a large number of test samples), then nodes with an LOF value less than 1.5 will not be considered outliers. If we change the node to J13, its local density is 10, and its average neighborhood density is 25, then... If the value is greater than the threshold, it is ultimately determined to be an outlier boundary node, and all nodes that meet the conditions are assigned to the outlier boundary node segment.

[0101] The defect type determination submodule classifies the boundary node morphology based on the spatial structural position and angle change trend of the boundary node in the outlier boundary node segment, combined with the boundary orientation and curvature fluctuation characteristics of the boundary node, and generates the defect type results of the metal roof.

[0102] Based on the spatial structural position of each node in the identified outlier boundary node segment (i.e., the two-dimensional coordinates of the node in the image coordinate system) ,in , The physical position of node Jj on the horizontal and vertical axes (in meters), and the trend of the angle between the normal vectors (expressed as...). The unit is degrees per second, representing the rate of angular fluctuation of the boundary node normal vector between consecutive time steps, and the curvature change gradient (expressed as...). The unit is degrees per meter², reflecting the intensity of the boundary curvature change. A comprehensive evaluation is then performed based on the distribution characteristics of the direction vectors in each segment. Each outlier boundary segment is constructed as a node sequence. The boundary direction vector is obtained by calculating the difference in coordinates between its first and last nodes. The angle (deflection angle) between this vector and the principal axis of the overall structure is used to determine its direction change characteristics. For example, suppose nodes J13 to J18 constitute an outlier segment, and its direction vector is... After normalization and comparison with the principal axis direction, the included angle is 22°. Combined with the average rate of change of the included angle at nodes of 6.1 degrees / second and the average gradient of curvature change of 0.07 degrees / m², the system classifies it as a "node-concentrated protrusion type defect" according to the classification rules. If, for example, in a certain segment such as J21 to J25, the average rate of change of the included angle reaches 8.4 degrees / second, the gradient of curvature change is 0.11 degrees / m², the fluctuation range of the direction vector turning back direction is greater than 30 degrees, and the abrupt change points of the included angle are concentrated in more than three nodes, then it is classified as a "path tear zone"; if the trend of the included angle change between J30 and J35 shows a sudden change on one side while the other side remains continuous (e.g., , If the direction vector exhibits bidirectional reversal characteristics and the curvature change is asymmetrical (e.g., the average curvature on the left is 0.08 degrees / m², and on the right is 0.03 degrees / m²), it is identified as a "bending and twisting zone"; if a boundary segment, such as J40 to J45, continuously exhibits violent oscillations in curvature change ( If the fluctuation of the direction vector is more than 0.05 degree / meter2on 5 consecutive nodes, but the direction vector remains smooth without obvious deflection, and the change trend of the included angle is lower than 3 degrees / second, it is determined as a "curvature instability zone".

[0103] The above merely describes the preferred embodiments of the present application, but does not limit the present application in other forms. Any person skilled in the art can modify or change the above disclosed technical contents into equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made to the above embodiments without departing from the technical scheme of the present application, according to the technical essence of the present application, still belong to the protection scope of the technical scheme of the present application.

Claims

1. An AI-based metal roof construction quality detection system, characterized in that, The system comprises: The included angle mutation recognition module acquires all boundary node sequences of the metal roof target position, extracts the included angle change rate of adjacent boundary nodes in the node sequence, screens the included angle change mutation points to form a first boundary node abnormal preliminary screening set; The tangential abnormality extraction module acquires the tangential displacement sequence between adjacent boundary nodes of the metal roof target position through the first boundary node abnormal preliminary screening set, and performs continuity analysis on the tangential gradient to select the boundary nodes with continuous abnormalities to form a first boundary node set with continuous direction abnormalities; The curvature stability recognition module acquires all boundary node sequences of the metal roof target position, judges the stability of the curvature change state of the node sequence, and obtains a second boundary node set with curvature distribution abnormalities; The path fitting offset analysis module inputs the first boundary node set with continuous direction abnormalities and the second boundary node set with curvature distribution abnormalities into a bidirectional long-short term memory network for laying path fitting offset analysis to construct a path fitting offset sequence; The defect type judgment module performs abnormal type detection on the laying path of the metal roof target position based on the path fitting offset sequence to obtain a metal roof quality detection result. 2.The AI-based metal roof construction quality detection system of claim 1, wherein The first boundary node abnormal preliminary screening set includes an included angle change sudden increase point, a normal vector mutation boundary node and a boundary direction fracture point. The first boundary node set with continuous direction abnormalities specifically refers to a multi-segment continuous mutation boundary node sequence, a tangential included angle discontinuous region and a local path disturbance zone. The second boundary node set with curvature distribution abnormalities includes a curvature extreme point, a curvature fluctuation concentrated area and a boundary curvature instability section. The path fitting offset sequence specifically refers to a fitting residual value sequence, a fitting trend mutation area and a fitting error time sequence distribution. The metal roof quality detection result includes a path tearing zone, a corner twisting zone and a curvature instability zone. 3.The AI-based metal roof construction quality detection system of claim 1, wherein The included angle mutation recognition module comprises: The normal vector extraction submodule collects image data of the metal roof target position, acquires the arrangement order of the boundary nodes in the three-dimensional coordinate space, calculates the normal vector direction constructed by adjacent boundary nodes, and establishes a boundary node normal direction sequence based on the spatial coordinate data between adjacent normal vectors; The included angle increment calculation submodule selects two normal vectors formed by every three adjacent boundary nodes in the boundary node normal direction sequence, calls the directional included angle value between each group of vectors and constructs an included angle difference sequence, performs comparative analysis on the included angle change amplitude of the boundary nodes, and acquires a boundary node included angle change rate sequence; The mutation point screening submodule calls the included angle change rate amplitude of each boundary node in the boundary node included angle change rate sequence and sequentially compares the boundary node change trend, identifies the boundary nodes with inconsistent continuous change directions and amplitude mutations as feature points, and generates a first boundary node abnormal preliminary screening set. 4.The AI-based metal roof construction quality detection system of claim 3, wherein The tangential abnormality extraction module comprises: The tangential vector construction submodule acquires the spatial coordinate information of adjacent boundary nodes of the metal roof target position based on the first boundary node abnormal preliminary screening set, sequentially connects the adjacent boundary nodes to form a boundary line segment direction vector sequence, uniformly normalizes each group of boundary node direction vectors, and establishes a boundary tangential direction sequence; The gradient sequence extraction submodule calls adjacent vector pairs in the boundary tangent direction sequence, obtains the degree of direction mutation between consecutive boundary nodes by comparing the direction included angle, and constructs a continuous sequence reflecting the overall path change trend to obtain a tangent gradient sequence; The abnormal section identification submodule sequentially extracts the boundary node direction change amplitude according to the tangent gradient sequence, judges the continuity of the change interval, and when the boundary node direction mutation continuously occurs and the gradient change trend is unstable, the corresponding boundary node section is delimited to obtain a first boundary node set with continuous direction abnormalities. 5.The AI-based metal roof construction quality detection system of claim 4, wherein The curvature stability identification module includes: The coordinate sequence acquisition submodule acquires the boundary image of the target position of the metal roof, extracts all boundary node positions on the boundary line segment, and forms a spatial point sequence in the order of boundary node numbers, and constructs a boundary node combined coordinate sequence from each group of adjacent three boundary node coordinates in the sequence; The curvature trend extraction submodule constructs the geometric relationship between consecutive three points according to the boundary node combined coordinate sequence, extracts the curvature value of the geometric relationship arc, and statistically obtains the continuous trend of the central difference value between adjacent curvatures to obtain a curvature change trend sequence; The instability point identification submodule compares the amplitude of the change of each boundary node in the curvature change trend sequence, judges the fluctuation state of the continuous change segment formed by adjacent boundary nodes, and screens a second boundary node set with abnormal curvature distribution. 6.The AI-based metal roof construction quality detection system of claim 5, wherein The path fitting offset analysis module includes: The sequence construction submodule extracts the normal vector included angle change rate of the curvature abnormal boundary node in the corresponding image frame according to the first boundary node set with continuous direction abnormalities, extracts the curvature change gradient between adjacent coordinate boundary nodes according to the second boundary node set with abnormal curvature distribution, integrates the two types of boundary node sets into an independent time sequence structure, and generates an input feature sequence set; The feature combination submodule calls the time window segment of each type of sequence in the input feature sequence set, performs time sequence feature splicing in the order of boundary node numbers, and generates a path sequence feature group; The fitting training submodule inputs the path sequence feature group into a bidirectional long short-term memory network, learns and extracts boundary node output residual values based on the time dimension expansion of the boundary node space change trend, and obtains a path fitting offset sequence. 7.The AI-based metal roof construction quality detection system of claim 6, wherein The defect type determination module includes: The abnormal segment extraction submodule extracts a time sequence segment with a continuous abnormal offset amplitude in the path fitting offset sequence based on the boundary node output residual values, matches the corresponding boundary node number and sequence position, and obtains a continuous abnormal offset segment set; The outlier region identification submodule calls the boundary node offset data in the continuous abnormal offset segment set, judges the density ratio by a local outlier factor algorithm, identifies the time sequence segment with a sparse local distribution and a lower density than the surrounding boundary nodes in the offset sequence, and obtains an outlier boundary node section; The defect type determination submodule performs classification judgment on the boundary node form according to the spatial structure position and the included angle change trend of the boundary nodes in the outlier boundary node section, combines the boundary node boundary trend and the curvature fluctuation feature, and generates a metal roof defect type result.

Images