Wireless crack width measurement method and system based on image recognition

Abstract

The invention relates to the technical field of civil engineering detection equipment, in particular to a wireless crack width measurement method and system based on image recognition, and the system is composed of a detection end, a handheld terminal and a cloud server. The detection end can collect a crack image and transmit position information; the handheld terminal realizes image preprocessing, crack identification and width calculation through an AI processor; the cloud server constructs a distributed storage system and is internally provided with a crack prediction model; according to measurement, crack data are collected through a detection end, an AI processor of a handheld terminal executes Gaussian filtering, CNN recognition and Canny edge detection to calculate the width, and a cloud server predicts the crack trend through an LSTM + SVR model. Through multi-module integration, a depth AI algorithm and cloud data management, the problems of low measurement precision, insufficient intelligence, difficulty in data management and the like are solved, high-precision and intelligent crack detection and prediction are realized, and the data management efficiency and safety are improved.

Description

Technical Field

[0001] This invention relates to the field of civil engineering testing equipment technology, and in particular to a wireless crack width measurement method and system based on image recognition. Background Technology

[0002] During various construction activities, such as large-scale building construction, subway tunnel excavation, and road construction, it is inevitable that surrounding houses will be affected to some extent, potentially causing damage such as cracks. Such problems often lead to disputes between the construction company and the homeowner. To clarify liability, preserving evidence of damage to surrounding houses is crucial, and the accurate measurement and recording of cracks is particularly important as key evidence.

[0003] Currently, common methods for preserving evidence of cracks in buildings surrounding construction sites have many shortcomings. On one hand, traditional manual recording methods rely on workers using simple measuring tools, such as crack comparison cards and steel rulers, to measure and manually record each crack. This method is extremely inefficient, requiring significant manpower and time for large-scale inspections. For example, when a large commercial building is under construction near an old residential area, dozens or even hundreds of buildings may need to be inspected; manual recording could take days or even weeks to complete the initial inspection. Furthermore, the accuracy of manual measurements is limited; different workers measuring the same crack may yield different results, leading to poor data accuracy and reliability.

[0004] On the other hand, while some portable crack detection instruments have improved measurement accuracy to a certain extent, they still have significant shortcomings. Most of these instruments are standalone devices and lack wireless data transmission capabilities. After completing measurements, inspectors must manually input the data into a computer for storage and processing, a cumbersome and error-prone process. Especially in complex and variable construction site environments, data loss or incorrect recording may occur due to interference during data entry. Furthermore, these instruments lack intelligent crack identification and analysis functions. For some fine cracks or complex-shaped cracks, inspectors still need to rely on experience for judgment, making automated and intelligent evidence preservation difficult.

[0005] Meanwhile, existing evidence preservation methods lack effective data management and long-term storage mechanisms. When disputes arise and data needs to be retrieved, improper data storage or inconsistent formats may lead to inaccurate data presentation or loss, posing significant difficulties for liability determination and dispute resolution. In conclusion, existing methods for preserving evidence of cracks in buildings surrounding construction sites cannot meet the demands of modern construction activities for efficient, accurate, and intelligent evidence collection and management, urgently requiring a new technological approach to address these issues. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a wireless crack width measurement method and system based on image recognition, so as to solve the problems of low crack detection efficiency, poor accuracy, insufficient intelligence and difficult data management in the prior art, and to achieve efficient, accurate and intelligent crack detection and data management.

[0007] The technical solution adopted by this invention to solve its technical problem is: a wireless crack width measurement method based on image recognition, comprising the following steps:

[0008] S1. Detection preparation: Ensure that the detection terminal has sufficient power and turn on the corresponding positioning module according to the detection environment;

[0009] S2. On-site inspection: Align the inspection end with the suspected crack area, maintain the preset distance and vertical angle to ensure that the industrial camera completely covers the crack area, and collect crack images and location information;

[0010] S3. Data Processing: The handheld terminal receives data transmitted from the detection end and automatically calls the algorithm module to process the data through the AI ​​processor, including: using Gaussian filtering algorithm to reduce noise and enhance image quality; identifying cracks and outlining their contours based on convolutional neural network (CNN); detecting edges through Canny algorithm and calculating the actual width by combining camera parameters and calibration model; and using morphological algorithm to analyze crack morphology and determine the direction by pixel line fitting.

[0011] S4. Data Upload and Storage: The handheld terminal uploads the processed data to the cloud server, which then categorizes, stores, and indexes the data.

[0012] S5. Crack Prediction: The cloud server uses a crack prediction model based on Long Short-Term Memory Network (LSTM) and Support Vector Regression (SVR) algorithm to train on historical detection data to predict the development trend of cracks. LSTM learns the long-term dependence of crack width on time, and SVR performs regression analysis on the features output by LSTM.

[0013] S6. Data Interaction: View detection results and prediction information via handheld terminal or cloud platform, and support historical data query and data export.

[0014] Furthermore, in step S3 of this invention, the calculation formula for the Gaussian filtering algorithm is as follows:

[0015]

[0016] Where I(x+m,y+n) is the pixel value at coordinates (x+m,y+n) in the original image, w(m,n) is the weight value of the Gaussian weight matrix at position (m,n), and k determines the size of the Gaussian filter kernel.

[0017] Furthermore, in step S3 of this invention, crack identification uses the convolutional layer of a convolutional neural network (CNN) to extract image features through convolution operations. The calculation formula is as follows:

[0018]

[0019] Where O(i,j) is the value of the convolutional output feature map at position (i,j), M and N are the height and width of the convolutional kernel, respectively, I is the input image, and K is the convolutional kernel; the pooling layer uses max pooling, and the calculation formula is:

[0020]

[0021] Where P(i,j) is the value of the pooling output feature map at position (i,j), s is the pooling window size, and F is the input feature map;

[0022] Fully connected layers use the softmax function to convert the network output into a probability distribution, calculated as follows:

[0023]

[0024] Where σ(z)j is the probability of category j, zj is the network output for category j, and K is the total number of categories.

[0025] Furthermore, in step S3 of this invention, when the Canny algorithm performs edge detection, it first calculates the image gradient magnitude G and direction θ, with the following formulas:

[0026]

[0027] Where Gx(x,y) and Gy(x,y) are the gradients of the image in the x and y directions, respectively; the formula for calculating the actual crack width W is:

[0028]

[0029] Where p is the pixel spacing of the crack edge measured in the image, d is the pixel size of the image sensor, and f is the camera focal length.

[0030] Furthermore, in step S3 of this invention, the crack morphology analysis is performed using a corrosion operation formula. For the binarized crack image B and the structural element S, the calculation formula for the corrosion operation is:

[0031]

[0032] in, It is the value of the eroded image at position (x,y).

[0033] Furthermore, in step S5 of this invention, the input gate i of the LSTM t Forgotten Gate t Output gate o t and memory unit C t The calculation formulas are as follows:

[0034] i t =σ(W ii x t +b ii +W hi h t-1 +b hi )

[0035] f t =σ(W if x t +b if +W hf h t-1 +b hf )

[0036] o t =σ(W io x t +b io +W ho h t-1 +b ho )

[0037]

[0038] h t =o t ×tanh(C t )

[0039] Where σ is the sigmoid function, tanh is the hyperbolic tangent function, W and b are the weight matrix and bias vector, respectively, and x... t It is the input at the current moment, h t-1 It is the hidden state from the previous moment, C t-1 It is the memory unit from the previous moment;

[0040] The feature vector output by the LSTM is used as the input to the SVR, which predicts the crack width by finding an optimal hyperplane.

[0041] Meanwhile, the present invention also provides a wireless crack width measurement system based on image recognition, including...

[0042] The detection end is used to acquire crack images and location information. The detection end includes an optical module, a positioning module, and a wireless communication module.

[0043] A handheld terminal, wirelessly connected to the detection terminal, is used to receive and process data transmitted by the detection terminal. The handheld terminal includes an AI processor and a touch screen.

[0044] A cloud server, which communicates with the handheld terminal, is used to store detection data and predict crack development trends.

[0045] Furthermore, the optical module of the present invention includes an industrial camera with a macro lens and a pixel count of ≥5 million and an LED ring light; the positioning module includes at least one of a GPS positioning module, a Bluetooth positioning module, and a WiFi positioning module, used to automatically or manually switch the positioning mode according to the detection environment; the wireless communication module supports WiFi, 4G, or 5G communication protocols, used to integrate the image data collected by the optical module with the location information obtained by the positioning module and transmit it to the handheld terminal in real time.

[0046] Furthermore, the AI ​​processor of this invention incorporates an image preprocessing algorithm, a crack recognition algorithm, a width calculation algorithm, and a feature analysis algorithm. The image preprocessing algorithm uses a Gaussian filtering algorithm to reduce noise in the image. The crack recognition algorithm is implemented based on a convolutional neural network (CNN). The width calculation algorithm uses the Canny algorithm for edge detection and combines camera imaging parameters and a preset calibration model to calculate the actual width of the crack. The touch screen is used to display detection data, analysis results, and operation menus in real time, and supports setting detection parameters and exporting data in multiple formats.

[0047] Furthermore, the cloud server described in this invention constructs a distributed storage system for classifying and storing detection data and establishing indexes according to multiple dimensions such as house address, detection time, and crack type; the cloud server also has a built-in crack prediction model based on Long Short-Term Memory (LSTM) network and Support Vector Regression (SVR) algorithm, which is used to predict crack development trends based on historical detection data.

[0048] The beneficial effect of this invention is that it solves the defects existing in the prior art.

[0049] 1. High-precision measurement: Utilizing an industrial camera with a macro lens and ≥5 megapixels, coupled with advanced image recognition and processing algorithms, cracks can be accurately identified and their width measured. Through Gaussian filtering, Canny edge detection, and width calculation methods based on imaging parameter calibration, measurement errors can be controlled to an extremely small range. Compared to traditional methods, this significantly improves measurement accuracy, providing accurate data support for the crack situation in surrounding buildings and facilitating a precise assessment of the extent of damage.

[0050] 2. Intelligent Analysis: The AI ​​processor is equipped with multiple intelligent algorithms, which can not only automatically identify cracks, but also perform in-depth analysis of crack morphology, direction, and other characteristics. Convolutional neural networks are used for crack identification, morphological analysis algorithms determine crack shape, and line fitting determines crack direction. These intelligent methods can quickly and accurately extract key crack information without much manual intervention, greatly improving detection efficiency and data analysis capabilities, and enabling the timely detection of potential structural safety hazards in buildings.

[0051] 3. Wireless Data Transmission and Convenient Export: The wireless communication module enables real-time uploading of detection data to the cloud server, ensuring data immediacy and security, and facilitating multi-departmental collaboration and data sharing. Simultaneously, the handheld terminal supports one-click data export, allowing data to be exported in various common formats. This facilitates in-depth analysis with other professional software, or direct use for report writing, archiving, and one-click report export, significantly improving data usability and flow efficiency.

[0052] 4. Long-term Data Management and Prediction: The cloud server categorizes, stores, and manages large amounts of detection data, creating indexes for easy historical data retrieval. The built-in crack prediction model based on LSTM and SVR algorithms can predict crack development trends based on historical data and issue early warnings. This helps construction companies develop response strategies in advance and take targeted protective measures, such as reinforcing buildings and adjusting construction plans, effectively preventing more serious building damage caused by further crack development, protecting the lives and property of surrounding residents, reducing the risk of construction disputes, and maintaining social harmony and stability. Attached Figure Description

[0053] Figure 1 This is a system architecture block diagram of the present invention;

[0054] Figure 2 This is a flowchart of the method of the present invention;

[0055] Figure 3 This is a block diagram of the handheld terminal architecture of the present invention. Detailed Implementation

[0056] The present invention will now be described in further detail with reference to the accompanying drawings and preferred embodiments. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0057] like Figure 1 , Figure 3 The wireless crack width measurement system based on image recognition shown consists of three main parts: a detection end, a handheld terminal, and a cloud server. These parts work together to achieve efficient detection, data processing, and management of cracks in buildings around the construction site.

[0058] Measurement method and process as follows Figure 2 As shown, it includes the following steps:

[0059] 1. Testing Preparation: Ensure the testing device has sufficient power and all modules are functioning properly. If testing indoors, confirm that Bluetooth or WiFi positioning is enabled and connected correctly; for outdoor testing, ensure a strong GPS signal.

[0060] 2. On-site inspection: Aim the inspection device at the suspected crack area, maintaining an appropriate distance and angle so that the camera can completely capture the crack. Press the photo button (or set to automatic acquisition mode) to complete one data acquisition. Repeat this operation along the preset inspection route to conduct a comprehensive inspection of all areas of the house.

[0061] 3. Data Upload: After confirming the data is accurate on your handheld device, click the upload button to upload the data to the cloud server. Ensure a stable network connection to guarantee complete data upload.

[0062] 4. Predictive Viewing: Inspection personnel or relevant management personnel can log in to the cloud platform with authorization and view the crack prediction results on the designated interface. The platform intuitively displays the prediction information in the form of charts, text, etc., such as crack width change curves over a future period and early warning information.

[0063] 5. Historical Data Query: Based on your needs, enter keywords such as house address and inspection time range in the data query interface of the cloud platform to quickly retrieve historical inspection data for comparative analysis and research on crack development trends.

[0064] 6. Data Export: When the test data is needed, the user can click the data export button on the touch screen, select the required data format (such as Excel, PDF, Word, etc.), specify the export path (external storage device or email, etc.) to complete the data export.

[0065] The detection end is tightly integrated with an optical module (an industrial camera with macro lenses and ≥5 million pixels, and an LED ring light), a positioning module (GPS, Bluetooth or WiFi positioning module can be selected), and a wireless communication module (supporting WiFi, 4G or 5G communication protocols).

[0066] Optical Module: Employs an industrial camera with a macro lens and ≥5 megapixels, capable of clearly capturing the minute features of cracks. To ensure high-quality images under varying lighting conditions, an LED ring light is included. In practice, inspectors place the inspection end close to the cracked area on walls, floors, or other surfaces. By maintaining a fixed distance between the camera and the wall, the camera can clearly capture the cracked area. For example, for wall cracks, the inspection end is kept perpendicular to the wall, approximately 2-3 cm away, ensuring the camera's field of view covers both ends of the crack and a certain area around it.

[0067] Positioning Module: This module determines the precise coordinates of the detection location. It can employ a GPS positioning system or indoor positioning technology (such as Bluetooth or WiFi positioning), depending on the environment of the construction area. When inspecting buildings surrounding outdoor construction sites, GPS positioning is prioritized to accurately obtain the geographical location information of the buildings. In indoor environments, such as basements where GPS signals are weak, Bluetooth or WiFi positioning can be switched to. The positioning information is recorded along with the crack image data, facilitating accurate marking of crack locations on maps or building floor plans later, providing convenience for subsequent analysis and management.

[0068] Wireless communication module: Responsible for transmitting image data acquired by the optical module and location information obtained by the positioning module to the handheld terminal in real time. It employs common wireless communication protocols such as WiFi, 4G, or 5G to ensure stable and high-speed data transmission.

[0069] The implementation steps of the detection end are as follows:

[0070] 1) Image acquisition: The inspector holds the inspection device close to areas such as the walls and floors of the house where cracks may appear. The industrial camera, under the uniform illumination provided by the LED ring light, clearly captures the image information of the cracks.

[0071] 2) Location Positioning: The positioning module automatically or manually switches between positioning modes depending on the detection environment. Outdoors, GPS positioning is activated to obtain an accurate geographical location; indoors, Bluetooth or WiFi positioning is used to determine the detection location coordinates.

[0072] 3) Data transmission: After the image data collected by the optical module is integrated with the location information obtained by the positioning module, it is transmitted to the handheld terminal in real time and stably through the wireless communication module according to the selected communication protocol.

[0073] The handheld terminal is equipped with a powerful AI processor and a touch display screen, and integrates multiple AI algorithm modules such as image preprocessing, crack recognition, width calculation, and feature analysis.

[0074] AI Processor: Equipped with a powerful AI processor, it utilizes various advanced AI algorithms to achieve precise analysis of cracks, including:

[0075] 1) Image preprocessing algorithm: Gaussian filtering is used to reduce noise in the acquired image. For each pixel (x, y) in the image, the formula for calculating the pixel value G(x, y) after Gaussian filtering is:

[0076]

[0077] Where I(x+m, y+n) is the pixel value at coordinates (x+m, y+n) in the original image, w(m, n) is the weight value of the Gaussian weight matrix at position (m, n), and k determines the size of the Gaussian filter kernel. For example, for a 3×3 Gaussian filter kernel, its weight matrix is:

[0078]

[0079] 2) Crack Detection Algorithm: Crack detection is based on Convolutional Neural Networks (CNN). The convolutional layers of a CNN extract image features through convolution operations. For an input image I and a convolutional kernel K, the calculation formula for the convolution operation is:

[0080]

[0081] Where O(i,j) is the value of the convolutional output feature map at position (i,j), and M and N are the height and width of the convolutional kernel, respectively. Pooling layers generally use max pooling or average pooling. Taking max pooling as an example, assuming the pooling window size is s×s, for the input feature map F, the calculation formula for max pooling is:

[0082]

[0083] Where P(i,j) is the value of the pooled output feature map at position (i,j). The fully connected layer classifies the pooled features and transforms the network output into a probability distribution using the softmax function. The formula for calculating the softmax function is:

[0084]

[0085] Where σ(z) j z is the probability of category j. j K is the network's output for category j, and K is the total number of categories.

[0086] 3) Crack width calculation algorithm: Edge detection is performed using the Canny algorithm. First, the image is smoothed using Gaussian filtering. Then, the image gradient magnitude G and direction θ are calculated using the following formulas:

[0087]

[0088] Among them G x (x,y) and G y (x, y) represent the gradients of the image in the x and y directions, respectively. Next, non-maximum suppression is performed to refine the edges, and finally, double threshold detection and edge connection are used to obtain the crack edge image. Based on the camera's imaging parameters, assuming the camera focal length is f, the image sensor pixel size is d, and the pixel spacing of the crack edge measured in the image is p, the formula for calculating the actual crack width W is:

[0089]

[0090] 4) Crack Feature Analysis Algorithm: A morphological analysis algorithm is used to analyze the morphology, orientation, and other features of the cracks. Taking the corrosion operation as an example, for the binarized crack image B and the structuring element S, the calculation formula for the corrosion operation is:

[0091]

[0092] in This is the value of the eroded image at position (x, y). The dilation operation is similar to the erosion operation, except that `min` is replaced with `max`. This is achieved by adjusting the value of the crack pixel at position (x, y). i ,y i By fitting a straight line to the coordinates of the crack, we can obtain the linear equation y = ax + b, where... n is the number of crack pixels.

[0093] The implementation steps of an AI processor are as follows:

[0094] 1) Image preprocessing: The AI ​​processor receives image data from the detection end and uses algorithms such as Gaussian filtering to reduce noise in the image, enhance the image clarity and contrast, and provide high-quality images for subsequent analysis.

[0095] 2) Crack identification: Based on the convolutional neural network (CNN) algorithm, feature extraction and pattern recognition are performed on the preprocessed image to automatically determine whether there are cracks in the image and accurately outline the general contour of the crack.

[0096] 3) Width calculation: Using edge detection technologies such as the Canny algorithm, the crack edge is further refined. Combined with the camera's imaging parameters and the preset calibration model, the actual width of the crack is accurately calculated.

[0097] The handheld terminal is equipped with a touch screen, integrating an operation interface and data export function module, and supports exporting multiple data formats.

[0098] Touchscreen display: Used to display inspection data, operating interface, and analysis results. Inspectors can intuitively view real-time acquired crack images, measured crack width values, crack location information, etc., through the touchscreen display. Simultaneously, on the operating interface, inspectors can set inspection parameters, such as image resolution, supplementary lighting brightness, and positioning mode. Analysis results, such as crack development trend prediction reports and comparisons with historical data, are clearly displayed in the form of charts and text. Furthermore, the touchscreen display supports one-click data export; inspectors can simply click the corresponding button to export the required data in Excel, PDF, Word, or other formats to external storage devices or send it to relevant personnel via email.

[0099] The steps to implement a touch screen display are as follows:

[0100] 1. Interface Display: The touch screen displays the detection data (such as crack images, width values, and location information), analysis results (such as crack morphology, orientation, and development trend prediction), and operation menus in real time, presenting them to the user in an intuitive graphical interface.

[0101] 2. Parameter settings: Inspectors can set inspection parameters, such as camera resolution, fill light brightness, and positioning mode, through touch operation in the interface to meet the needs of different inspection scenarios.

[0102] The general usage method of the handheld terminal is as follows:

[0103] 1. Data reception: Keep the handheld terminal turned on to ensure a normal wireless connection with the detection terminal and receive data transmitted by the detection terminal in real time.

[0104] 2. Automated Analysis: After receiving the data, the AI ​​processor automatically calls various algorithm modules to process and analyze the data, requiring minimal manual intervention. Inspectors can view the analysis progress in real time on the handheld terminal's display screen.

[0105] 3. Results viewing: After the analysis is completed, the inspectors can view the crack identification results, width values, shape and direction and other detailed information through the touch screen.

[0106] The cloud server constructs a distributed storage system and incorporates a crack prediction model based on Long Short-Term Memory (LSTM) and Support Vector Regression (SVR) algorithms.

[0107] 1) Data Storage: The system receives and stores all detection data uploaded from handheld terminals, including crack images, measurement values, location information, and analysis reports. Distributed storage technology is employed, storing data across multiple server nodes to improve data security and reliability, preventing data loss due to single server failures. Simultaneously, the data is categorized and indexed according to dimensions such as house address, detection time, and crack type, facilitating quick querying and retrieval of historical data. When needing to view the crack development of a specific house over the past year, users can quickly obtain relevant data by entering the house address and time range.

[0108] 2) Crack Prediction Model: A built-in crack prediction model based on machine learning algorithms is included. This model employs a Long Short-Term Memory (LSTM) network combined with a Support Vector Regression (SVR) algorithm. The core structure of the LSTM includes the input gate i... t Forgotten Gate t Output gate o t and memory unit C t The calculation formula is as follows:

[0109] i t =σ(W ii x t +b ii +W hi h t-1 +b hi )

[0110] f t =σ(W if x t +b if +W hf h t-1 +b hf )

[0111] o t =σ(W io x t +b io +W ho h t-1 +b ho )

[0112]

[0113] h t =o t ×tanh(C t )

[0114] Where σ is the sigmoid function, tanh is the hyperbolic tangent function, W and b are the weight matrix and bias vector, respectively, and x... t It is the input at the current moment, h t-1It is the hidden state from the previous moment, C t-1 This is the memory unit from the previous time step. The feature vector output by the LSTM serves as the input to the SVR. The SVR predicts the crack width by finding an optimal hyperplane, and its objective function is: Extraclose brace or missing open brace.

[0115]

[0116] Where ω is the weight vector, b is the bias term, and ξ and ξ * Here, y is a slack variable, C is a penalty parameter, ∈ is a parameter of the insensitive loss function, and y is a slack variable. i This is the actual crack width value, φ(x) i ) is the input data x i A function mapped to a high-dimensional feature space. The prediction results are fed back to handheld terminals and client devices of relevant personnel, providing a scientific basis for construction teams to take targeted protective measures. For example, if it is predicted that the width of a crack may increase beyond a safe threshold within the next month, the construction team can take reinforcement measures or adjust the construction plan in advance to avoid structural damage to the building due to further crack development.

[0117] The steps to implement a clientless server are as follows:

[0118] 1. Data Storage: The handheld terminal uploads the processed detection data (including crack images, measurement values, location information, analysis reports, etc.) to the cloud server. The distributed storage system categorizes and stores the data according to multiple dimensions such as building address, detection time, and crack type, and establishes an efficient index.

[0119] 2. Model Training: Using historical detection data, the crack prediction model based on LSTM and SVR is trained. LSTM learns the long-term dependence of crack width changes over time, and SVR performs regression analysis on the features output by LSTM. The model parameters are continuously adjusted to enable the model to accurately capture the crack development pattern.

[0120] The following example illustrates this through a specific implementation scenario.

[0121] During the inspection of cracks in buildings surrounding a construction project, inspectors carried out on-site operations using a detection device. For outdoor wall cracks, the inspectors activated the GPS positioning module of the detection device, ensuring a good GPS signal. They then held the device perpendicular to the wall, approximately 2-3 centimeters away, ensuring the industrial camera with a macro lens (5 megapixels) covered both ends of the crack and the surrounding area. Simultaneously, an LED ring light provided uniform illumination. The inspector pressed the shutter button to capture an image of the crack. The positioning module obtained the crack's GPS coordinates, and the wireless communication module transmitted the image data and location information to a handheld terminal in real time via a 4G network.

[0122] After receiving the data transmitted from the detection terminal, the AI ​​processor immediately preprocesses the image using a 3×3 Gaussian filter kernel, with the following weight matrix:

[0123]

[0124] Noise reduction is performed to enhance image clarity and contrast. Next, a CNN algorithm is used to identify cracks in the preprocessed image. Features are extracted using convolutional layers, dimensionality is reduced using max-pooling layers (2×2 pooling windows), and fully connected layers combined with the softmax function determine the presence of cracks and delineate their outlines. Then, the Canny algorithm is used for edge detection, calculating gradient magnitude and direction. After non-maximum suppression and double threshold detection, the crack edge image is obtained. Considering the camera focal length f = 10mm, image sensor pixel size d = 2.2μm, and measured pixel pitch p = 50 pixels, the actual crack width is calculated as 0.011mm using W = (50×2.2) / 10 = 11μm. Simultaneously, a straight line is fitted to the crack pixel coordinates to obtain the equation of the directional line, analyzing the crack morphology. Inspectors can view the detection data and analysis results in real time on a touchscreen display, set parameters such as camera resolution 1920×1080 and supplementary lighting brightness 70%, and export the results in PDF format.

[0125] The handheld terminal uploads the processed detection data (including crack images, a width of 0.011 mm, GPS coordinates, analysis reports, etc.) to the cloud server. The distributed storage system stores and indexes the data according to the building address ("Building XX, No. XX, XX Road, XX City"), the detection date ("Year XX Month XX Day"), and the crack type ("Wall Crack"). The cloud server trains an LSTM+SVR crack prediction model using historical detection data. The LSTM learns the historical crack width variation patterns over time, while the SVR performs regression analysis on the LSTM output features to adjust the model parameters. After training, the model predicts that the crack width may increase to 0.015 mm within the next month, which is below the safety threshold of 0.02 mm. The cloud server feeds back the prediction results to the handheld terminal and relevant personnel's clients, providing a reference for subsequent detection.

[0126] The above description is only a specific embodiment of the present invention. Various examples and illustrations do not constitute a limitation on the substantive content of the present invention. Those skilled in the art can make modifications or variations to the above-described specific embodiments after reading the specification without departing from the substance and scope of the invention.

Claims

1. A wireless crack width measurement method based on image recognition, characterized in that: Includes the following steps, S1. Detection preparation: Ensure that the detection terminal has sufficient power and turn on the corresponding positioning module according to the detection environment; S2. On-site inspection: Align the inspection end with the suspected crack area, maintain the preset distance and vertical angle to ensure that the industrial camera completely covers the crack area, and collect crack images and location information; S3. Data Processing: The handheld terminal receives data transmitted from the detection end and automatically calls the algorithm module to process the data through the AI ​​processor, including: using Gaussian filtering algorithm to reduce noise and enhance image quality; identifying cracks and outlining their contours based on convolutional neural network (CNN); detecting edges through Canny algorithm and calculating the actual width by combining camera parameters and calibration model; and using morphological algorithm to analyze crack morphology and determine the direction by pixel line fitting. S4. Data Upload and Storage: The handheld terminal uploads the processed data to the cloud server, which then categorizes, stores, and indexes the data. S5. Crack Prediction: The cloud server uses a crack prediction model based on Long Short-Term Memory Network (LSTM) and Support Vector Regression (SVR) algorithm to train on historical detection data to predict the development trend of cracks. LSTM learns the long-term dependence of crack width on time, and SVR performs regression analysis on the features output by LSTM. S6. Data Interaction: View detection results and prediction information via handheld terminal or cloud platform, and support historical data query and data export.

2. The wireless crack width measurement method based on image recognition as described in claim 1, characterized in that: In step S3, the calculation formula for the Gaussian filtering algorithm is as follows: Where I(x+m,y+n) is the pixel value at coordinates (x+m,y+n) in the original image, w(m,n) is the weight value of the Gaussian weight matrix at position (m,n), and k determines the size of the Gaussian filter kernel.

3. The wireless crack width measurement method based on image recognition as described in claim 1, characterized in that: In step S3, crack identification uses the convolutional layer of a convolutional neural network (CNN) to extract image features through convolution operations. The calculation formula is as follows: Where O(i,j) is the value of the convolutional output feature map at position (i,j), M and N are the height and width of the convolutional kernel, respectively, I is the input image, and K is the convolutional kernel; the pooling layer uses max pooling, and the calculation formula is: Where P(i,j) is the value of the pooling output feature map at position (i,j), s is the pooling window size, and F is the input feature map; Fully connected layers use the softmax function to convert the network output into a probability distribution, calculated as follows: Where σ(z)j is the probability of category j, zj is the network output for category j, and K is the total number of categories.

4. The wireless crack width measurement method based on image recognition as described in claim 1, characterized in that: In step S3, when the Canny algorithm performs edge detection, it first calculates the image gradient magnitude G and direction θ, using the following formulas: Where Gx(x,y) and Gy(x,y) are the gradients of the image in the x and y directions, respectively; the formula for calculating the actual crack width W is: Where p is the pixel spacing of the crack edge measured in the image, d is the pixel size of the image sensor, and f is the camera focal length.

5. The wireless crack width measurement method based on image recognition as described in claim 1, characterized in that: In step S3, the crack morphology analysis uses an corrosion operation formula to analyze the crack morphology. For the binarized crack image B and the structuring element S, the calculation formula for the corrosion operation is: in, It is the value of the eroded image at position (x,y).

6. The wireless crack width measurement method based on image recognition as described in claim 1, characterized in that: In step S5, the input gate i of the LSTM t Forgotten Gate t Output gate o t and memory unit C t The calculation formulas are as follows: i t =σ(W ii x t +b ii +W hi h t-1 +b hi ) f t =σ(W if x t +b if +W hf h t-1 +b hf ) o t =σ(W io x t +b io +W ho h t-1 +b ho ) h t =o t ×tanh(C t ) Where σ is the sigmoid function, tanh is the hyperbolic tangent function, W and b are the weight matrix and bias vector, respectively, and x... t It is the input at the current moment, h t-1 It is the hidden state from the previous moment, C t-1 It is the memory unit from the previous moment; The feature vector output by the LSTM is used as the input to the SVR, which predicts the crack width by finding an optimal hyperplane.

7. A system used in the wireless crack width measurement method based on image recognition as described in claim 1, characterized in that: include, The detection end is used to acquire crack images and location information. The detection end includes an optical module, a positioning module, and a wireless communication module. A handheld terminal, wirelessly connected to the detection terminal, is used to receive and process data transmitted by the detection terminal. The handheld terminal includes an AI processor and a touch screen. A cloud server, which communicates with the handheld terminal, is used to store detection data and predict crack development trends.

8. The wireless crack width measurement system based on image recognition as described in claim 7, characterized in that: The optical module includes an industrial camera with macro lenses and ≥5 million pixels and an LED ring light; the positioning module includes at least one of a GPS positioning module, a Bluetooth positioning module and a WiFi positioning module, used to automatically or manually switch the positioning mode according to the detection environment; the wireless communication module supports WiFi, 4G or 5G communication protocols, used to integrate the image data collected by the optical module with the location information obtained by the positioning module and transmit it to the handheld terminal in real time.

9. The wireless crack width measurement system based on image recognition as described in claim 7, characterized in that: The AI ​​processor incorporates image preprocessing, crack recognition, width calculation, and feature analysis algorithms. The image preprocessing algorithm uses Gaussian filtering to reduce image noise. The crack recognition algorithm is based on a convolutional neural network (CNN). The width calculation algorithm uses the Canny algorithm for edge detection and combines camera imaging parameters and a preset calibration model to calculate the actual crack width. The touch screen is used to display detection data, analysis results and operation menus in real time, and supports detection parameter settings and export of multiple data formats.

10. The wireless crack width measurement system based on image recognition as described in claim 7, characterized in that: The cloud server constructs a distributed storage system to classify and store detection data and establish indexes according to multiple dimensions such as house address, detection time, and crack type; the cloud server also has a built-in crack prediction model based on Long Short-Term Memory (LSTM) network and Support Vector Regression (SVR) algorithm, which is used to predict crack development trends based on historical detection data.

Images