Real-time detection system for metal plate crack based on infrared thermal imaging and machine learning

Abstract

The application provides a metal plate crack real-time detection system combining infrared thermal imaging and machine learning, and belongs to the technical field of nondestructive testing.The method comprises the following steps: acquiring an infrared thermal image of a metal plate surface; and pre-processing the infrared thermal image to obtain a pre-processed image capable of reflecting real temperature distribution.The application realizes non-contact, full-field and rapid detection of metal plate cracks, can adapt to harsh industrial environments such as high temperature and strong electromagnetic interference, can acquire surface temperature distribution information of the metal plate to be detected through infrared thermal imaging technology without direct contact with the metal plate to be detected, avoids damage to the surface of the workpiece and interruption of the production process in the traditional contact detection method, realizes online continuous monitoring, and is especially suitable for quality monitoring of the metal plate in high temperature or high speed motion.

Description

Technical Field

[0001] This application relates to the field of nondestructive testing technology, and in particular to a real-time detection system for cracks in metal plates that integrates infrared thermal imaging and machine learning. Background Technology

[0002] Metal sheets, as a basic industrial material, are widely used in key fields such as aerospace, automobile manufacturing, and pressure vessels. Their surface quality is directly related to the safety and reliability of the final product. Cracks are one of the most common and most harmful surface defects of metal sheets. Cracks in metal sheets can cause them to expand under stress and lead to structural failure. Therefore, in order to ensure industrial production safety and product quality, it is necessary to conduct real-time crack detection on metal sheets.

[0003] Currently, the main methods for detecting cracks in metal plates include penetrant testing, magnetic particle testing, eddy current testing, and ultrasonic testing. Although penetrant and magnetic particle testing are widely used, the process is relatively cumbersome and requires the use of chemical reagents. The actual operation requirements are high, making it difficult to achieve large-scale automated testing. Eddy current and ultrasonic testing have high detection sensitivity and accuracy, but they are mostly point or line scanning, resulting in low detection efficiency, making them unsuitable for rapid detection of large-area metal plates.

[0004] In recent years, machine vision-based inspection methods have been gradually applied to industrial inspection. However, their imaging quality is unstable under complex working conditions. Most traditional non-destructive testing methods are difficult to apply directly in high-temperature environments or cannot achieve rapid full-field scanning. In practical applications, optical imaging-based visual inspection technology is extremely sensitive to ambient lighting and the surface condition of objects, and its stability and reliability face challenges in actual industrial scenarios. In addition, most existing automated inspection systems rely on simple image processing algorithms and fixed threshold judgments, which have limited ability to identify micro-cracks with diverse shapes and are easily affected by background textures and noise, resulting in a high risk of false alarms and missed alarms. Summary of the Invention

[0005] This application provides a real-time crack detection system for metal plates that integrates infrared thermal imaging and machine learning. This system aims to improve upon conventional methods that rely on user experience and intuition for data interpretation, lack automated professional analysis capabilities, and are unable to independently analyze and enhance both textual and graphical data.

[0006] To achieve the above objectives, this application adopts the following technical solution:

[0007] In a first aspect, a real-time crack detection method for metal plates integrating infrared thermal imaging and machine learning is provided, the method comprising:

[0008] Acquire infrared thermal images of the surface of a metal plate;

[0009] The infrared thermal image is preprocessed to obtain a preprocessed image that reflects the true temperature distribution;

[0010] Extract multi-scale thermal features from the preprocessed image, including local temperature gradient features, hot spot morphology features, and spatiotemporal temperature change patterns.

[0011] The multi-scale thermal features are input into a lightweight convolutional neural network model trained with training data that can identify cracks, and the model outputs a judgment result on whether a crack exists and the confidence level of the result.

[0012] If the judgment result indicates the presence of a crack, the crack is analyzed based on the multi-scale thermal features to identify detailed crack information including crack location, length, and orientation.

[0013] Based on the judgment result and the detailed information of the crack, a detection report and an alarm signal are generated.

[0014] Optionally, the step of preprocessing the infrared thermal image to obtain a preprocessed image includes:

[0015] Non-uniformity correction is performed on the infrared thermal image. Using a standard blackbody radiation source, images are acquired at high temperature point T_H and low temperature point T_L, respectively, to obtain the response gray values ​​G_H(i,j) and G_L(i,j) of each pixel (i,j). Based on these values, the gain correction coefficient K(i,j) and bias correction coefficient B(i,j) of each pixel are calculated.

[0016] K(i,j)=(G_avg-G_min) / (G_H(i,j)-G_L(i,j));

[0017] B(i,j)=G_min-K(i,j)×G_L(i,j);

[0018] G_avg is the average grayscale value of all pixels at high and low temperatures, and G_min is the minimum grayscale value of all pixels at low temperature. In actual detection, the original grayscale value G_raw(i,j) of each pixel in the real-time acquired infrared thermal image is substituted into the correction formula G_corr(i,j)=K(i,j)×G_raw(i,j)+B(i,j) to obtain the corrected image after eliminating the inherent error of the device.

[0019] The corrected image is calibrated for temperature, and a temperature-grayscale response curve T=f(G_corr) is fitted. Then, the grayscale values ​​of each pixel in the corrected image are converted into actual temperature values ​​to obtain the calibrated image.

[0020] The calibrated image is then subjected to noise filtering to obtain a preprocessed image that suppresses environmental interference.

[0021] Optionally, the step of extracting multi-scale thermal features from the preprocessed image includes:

[0022] The gradient operator is used to calculate the temperature gradient components Gx(x,y) and Gy(x,y) of each pixel in the image in the horizontal and vertical directions, and the magnitude of the temperature gradient at each point is calculated accordingly. The gradient magnitudes of all pixels constitute the temperature change rate data, thus obtaining the local temperature gradient features;

[0023] Hot spot regions are identified by adaptive threshold segmentation, and connected component analysis is performed on each independent hot spot to extract morphological parameters including area, perimeter, eccentricity and principal axis direction, so as to obtain the morphological features of the hot spot region.

[0024] The preprocessed images of consecutive frames are processed to analyze the temperature change over time and obtain the spatiotemporal temperature change pattern.

[0025] The multi-scale thermal features are generated by integrating the local temperature gradient features, the hot spot region morphology features, and the spatiotemporal temperature change patterns.

[0026] Optionally, the lightweight convolutional neural network model is obtained through the following steps:

[0027] Obtain training data containing multi-scale thermal features and corresponding labels of cracked and non-cracked samples;

[0028] The training data is used to train a convolutional neural network with a simplified network structure. The network is constructed using depthwise separable convolution or model pruning techniques to reduce the number of parameters and computational complexity. During the training process, the backpropagation algorithm and gradient descent optimizer are used to continuously adjust the internal weight parameters of the model to minimize the difference between the model's prediction results and the true labels until the model reaches the preset accuracy standard on the validation set.

[0029] The parameters of the trained convolutional neural network are solidified to obtain the lightweight convolutional neural network model.

[0030] Optionally, the step of identifying detailed crack information includes:

[0031] Based on the local temperature gradient characteristics, non-maximum suppression and connection component analysis are applied to the temperature change rate data to extract candidate crack segments. The crack segments with the highest confidence are selected by combining the model judgment results, and their pixel coordinate set is recorded as crack location data.

[0032] Based on the morphological characteristics of the hot spot region, the skeleton of the hot spot region identified as a crack is extracted, and the cumulative Euclidean distance of the skeleton lines is calculated as an estimate of the crack length.

[0033] Based on the morphological characteristics of the hot spot region and the spatiotemporal temperature variation pattern, the direction of crack propagation is determined by using the hot spot principal axis direction and combining the consistency statistics of the temperature gradient direction of the crack region, and expressed as an angle or unit vector.

[0034] By integrating the crack location data, length estimation value, and extension direction information, a structured detailed crack information is generated.

[0035] Optionally, the physical interpretability of the multi-scale thermal features is utilized in conjunction with the low computational cost of the lightweight convolutional neural network model, wherein the indirect processing includes:

[0036] The multi-scale thermal features provide highly discriminative input to the lightweight convolutional neural network model, reducing the model's learning complexity and improving the accuracy of the judgment results.

[0037] The efficient inference capability of the lightweight convolutional neural network model is utilized to ensure the real-time processing of the multi-scale thermal features;

[0038] By combining the confidence level output by the model, the analysis strategy for the multi-scale thermal features is dynamically adjusted to achieve a synergistic improvement in accuracy and real-time performance.

[0039] Optionally, the step of generating an alarm signal includes:

[0040] Obtain the confidence level when the judgment result indicates the presence of a crack;

[0041] The confidence level is compared with a confidence threshold used to characterize the alarm risk level;

[0042] If the confidence level is higher than the confidence level threshold, a high-level alarm signal is generated; otherwise, a low-level re-examination signal is generated.

[0043] Secondly, a real-time crack detection system for metal plates integrating infrared thermal imaging and machine learning includes:

[0044] An infrared image acquisition device is used to acquire infrared thermal images of the surface of a metal plate.

[0045] The image preprocessing module is used to preprocess the infrared thermal image to obtain a preprocessed image;

[0046] A multi-scale thermal feature extraction module is used to extract multi-scale thermal features from the preprocessed image;

[0047] The crack detection module has a lightweight convolutional neural network model deployed inside it, which is used to receive the multi-scale thermal features and output the judgment result and confidence level of whether the crack exists.

[0048] A crack information identification module is used to identify detailed crack information based on the multi-scale thermal features when the judgment result indicates that a crack exists.

[0049] The report and alarm generation module is used to generate a detection report and alarm signal based on the judgment result and the detailed information of the crack;

[0050] The data storage module is used to store the test report and historical test data;

[0051] The communication module is used to enable remote data interaction;

[0052] Human-computer interaction interface, used for data display and operation control.

[0053] A third aspect of this invention provides an electronic device, which includes:

[0054] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method proposed in the first aspect of the present invention.

[0055] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in the first aspect of the present invention.

[0056] In summary, the above method and apparatus have the following technical effects:

[0057] 1. This invention enables non-contact, full-field, and rapid detection of cracks in metal plates. It can adapt to harsh industrial environments such as high temperatures and strong electromagnetic interference. Furthermore, by using infrared thermal imaging technology, it can obtain surface temperature distribution information of the metal plate without direct contact with it, avoiding damage to the workpiece surface and interruption of the production process caused by traditional contact detection methods. It achieves online continuous monitoring and is especially suitable for quality monitoring of metal plates in high-temperature or high-speed motion.

[0058] 2. This invention significantly improves the accuracy and robustness of crack identification by deeply fusing multi-scale thermal features with lightweight neural networks. This method can not only analyze local temperature anomalies at cracks, but also comprehensively consider their macroscopic morphology and time dynamic characteristics. By constructing a more comprehensive and reliable crack characterization and combining the powerful pattern recognition capabilities of machine learning models, it effectively overcomes interference factors such as background noise and surface stains, reduces the false alarm rate and the missed detection rate, and improves the accuracy of detection results.

[0059] 3. This invention can not only determine the presence or absence of cracks, but also provide detailed diagnostic information such as the location, length, and direction of cracks. It also has automated data management and intelligent alarm functions. The system can automatically generate a structured inspection report containing detailed crack information and implement graded alarms through confidence assessment, providing accurate data support for subsequent maintenance. The integrated remote access and historical query functions improve the automation level and management efficiency of the inspection process. Attached Figure Description

[0060] Figure 1 This is a flowchart of a real-time crack detection system for metal plates that integrates infrared thermal imaging and machine learning. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] Example 1

[0063] like Figure 1 As shown, a real-time crack detection method for metal plates integrating infrared thermal imaging and machine learning is described, the method comprising:

[0064] Acquire infrared thermal images of the surface of a metal plate;

[0065] The infrared thermal image is preprocessed to obtain a preprocessed image that reflects the true temperature distribution;

[0066] Extract multi-scale thermal features from the preprocessed image, including local temperature gradient features, hot spot morphology features, and spatiotemporal temperature change patterns.

[0067] The multi-scale thermal features are input into a lightweight convolutional neural network model trained with training data that can identify cracks, and the model outputs a judgment result on whether a crack exists and the confidence level of the result.

[0068] If the judgment result indicates the presence of a crack, the crack is analyzed based on the multi-scale thermal features to identify detailed crack information including crack location, length, and orientation.

[0069] Based on the judgment result and the detailed information of the crack, a detection report and an alarm signal are generated.

[0070] By employing an innovative collaborative processing mechanism combining multi-scale thermal features and lightweight neural networks, this method significantly improves the accuracy and reliability of crack identification, effectively reducing missed detections and false alarms caused by environmental interference and the diversity of crack morphologies. Furthermore, the system can not only determine the presence of cracks but also provide detailed diagnostic information such as crack location, length, and orientation, providing precise data support for subsequent maintenance and quality control. Combined with automated report generation and alarm functions, this invention greatly enhances the efficiency and safety of quality monitoring during the manufacturing and use of metal plates, promoting the development of industrial inspection towards intelligence and unmanned operation, and possessing significant industrial application value and technological advancement significance.

[0071] Optionally, the step of preprocessing the infrared thermal image to obtain a preprocessed image includes:

[0072] Non-uniformity correction is performed on the infrared thermal image. Using a standard blackbody radiation source, images are acquired at high temperature point T_H and low temperature point T_L, respectively, to obtain the response gray values ​​G_H(i,j) and G_L(i,j) of each pixel (i,j). Based on these values, the gain correction coefficient K(i,j) and bias correction coefficient B(i,j) of each pixel are calculated.

[0073] K(i,j)=(G_avg-G_min) / (G_H(i,j)-G_L(i,j)),

[0074] B(i,j)=G_min-K(i,j)×G_L(i,j),

[0075] G_avg is the average grayscale value of all pixels at high and low temperatures, and G_min is the minimum grayscale value of all pixels at low temperature. In actual detection, the original grayscale value G_raw(i,j) of each pixel in the real-time acquired infrared thermal image is substituted into the correction formula G_corr(i,j)=K(i,j)×G_raw(i,j)+B(i,j) to obtain the corrected image after eliminating the inherent error of the device.

[0076] The corrected image is calibrated for temperature, and a temperature-grayscale response curve T=f(G_corr) is fitted. Then, the grayscale values ​​of each pixel in the corrected image are converted into actual temperature values ​​to obtain the calibrated image.

[0077] The calibrated image is then subjected to noise filtering to obtain a preprocessed image that suppresses environmental interference.

[0078] Non-uniformity correction provides a unified, noise-free benchmark for subsequent temperature calibration, ensuring global consistency and accuracy. Without this step, the temperature-grayscale response curve of each pixel would be different, making accurate temperature measurement impossible. Accurate temperature calibration also ensures that noise filtering operates on a temperature field with clear physical meaning. This allows the filtering algorithm to more effectively identify and remove random noise that does not conform to the true temperature fluctuation pattern, rather than simply smoothing grayscale data without physical meaning. Therefore, this preprocessing process is not a simple superposition of three independent steps, but an interconnected and progressively optimized processing chain. It systematically eliminates various interferences from the sensor itself to the external environment. The final preprocessed image has a signal-to-noise ratio and temperature measurement accuracy far exceeding the level achievable by any single processing step. This lays a solid foundation for the accurate extraction of crack thermal features and reliable judgment by machine learning models, thereby significantly improving the robustness and sensitivity of the crack detection system as a whole.

[0079] Optionally, the step of extracting multi-scale thermal features from the preprocessed image includes:

[0080] The gradient operator is used to calculate the temperature gradient components Gx(x,y) and Gy(x,y) of each pixel in the image in the horizontal and vertical directions, and the magnitude of the temperature gradient at each point is calculated accordingly. The gradient magnitudes of all pixels constitute the temperature change rate data, thus obtaining the local temperature gradient features;

[0081] Hot spot regions are identified by adaptive threshold segmentation, and connected component analysis is performed on each independent hot spot to extract morphological parameters including area, perimeter, eccentricity and principal axis direction, so as to obtain the morphological features of the hot spot region.

[0082] The preprocessed images of consecutive frames are processed to analyze the temperature change over time and obtain the spatiotemporal temperature change pattern.

[0083] The multi-scale thermal features are generated by integrating the local temperature gradient features, the hot spot region morphology features, and the spatiotemporal temperature change patterns.

[0084] By organically combining three different scales of thermal features—local temperature gradient features, hotspot region morphology features, and spatiotemporal temperature variation patterns—a significant synergistic enhancement effect is achieved. While a single local temperature gradient feature is sensitive to crack edges, it easily misjudges abrupt temperature changes in other non-crack areas as cracks. A single hotspot region morphology feature can identify the linear geometry of cracks, but it is ineffective for microcracks that do not produce obvious large-area hotspots. A single spatiotemporal temperature variation pattern can eliminate transient interference but cannot provide information on the location and morphology of cracks. This method combines these three features to form a complementary, multi-level verification mechanism, with the gradient feature responsible for... The initial "discovery" of all possible temperature anomalies is then performed; morphological features "screen" these anomalies, retaining regions that conform to the geometric characteristics of cracks; and spatiotemporal pattern features finally "confirm" the screened regions, eliminating dynamic interference and ensuring that they are stable physical defects. This fusion of multi-scale features enables the final multi-scale thermal features to simultaneously take into account the microscopic details, macroscopic shape, and dynamic stability of cracks. Its crack characterization ability far exceeds that of any single feature or their simple combination, thus providing unprecedented information richness and reliability for subsequent accurate identification, achieving a fundamental improvement in the accuracy and robustness of crack detection.

[0085] Optionally, the lightweight convolutional neural network model is obtained through the following steps:

[0086] Obtain training data containing multi-scale thermal features and corresponding labels of cracked and non-cracked samples;

[0087] The training data is used to train a convolutional neural network with a simplified network structure. The network is constructed using depthwise separable convolution or model pruning techniques to reduce the number of parameters and computational complexity. During the training process, the backpropagation algorithm and gradient descent optimizer are used to continuously adjust the internal weight parameters of the model to minimize the difference between the model's prediction results and the true labels until the model reaches the preset accuracy standard on the validation set.

[0088] The parameters of the trained convolutional neural network are solidified to obtain the lightweight convolutional neural network model.

[0089] It can effectively learn and utilize the complex, nonlinear relationships between various components within multi-scale thermal features. These relationships cannot be captured by traditional threshold-based or simple rule-based methods. At the same time, the design of the output confidence level provides an uncertainty measure for system decision-making, enabling the system to not only provide a "yes" or "no" answer, but also to inform the reliability of the answer. This provides a more intelligent basis for subsequent alarm strategies and manual intervention. This deep coupling of feature engineering and model design jointly achieves a balance and optimization of the detection system in the three key indicators of high accuracy, high real-time performance, and high reliability. Its overall performance far exceeds the effect of applying advanced feature extraction techniques or advanced models in isolation.

[0090] Optionally, the step of identifying detailed crack information includes:

[0091] Based on the local temperature gradient characteristics, non-maximum suppression and connection component analysis are applied to the temperature change rate data to extract candidate crack segments. The crack segments with the highest confidence are selected by combining the model judgment results, and their pixel coordinate set is recorded as crack location data.

[0092] Based on the morphological characteristics of the hot spot region, the skeleton of the hot spot region identified as a crack is extracted, and the cumulative Euclidean distance of the skeleton lines is calculated as an estimate of the crack length.

[0093] Based on the morphological characteristics of the hot spot region and the spatiotemporal temperature variation pattern, the direction of crack propagation is determined by using the hot spot principal axis direction and combining the consistency statistics of the temperature gradient direction of the crack region, and expressed as an angle or unit vector.

[0094] By integrating the crack location data, length estimation value, and extension direction information, a structured detailed crack information is generated.

[0095] The three sub-tasks of crack location, length estimation, and orientation determination are precisely matched and deeply utilized with the three different scales of thermal features extracted earlier, demonstrating a significant technological synergy. The high sensitivity of local temperature gradient features to the edges makes them the optimal choice for locating micro-cracks, while the morphological features of hot spot regions provide the most direct physical basis for length estimation from a macro-geometric perspective. The combination of spatiotemporal temperature change patterns and gradient directions provides dual verification for orientation determination from both dynamic and physical heat transfer laws, eliminating the interference of static artifacts.

[0096] Optionally, the physical interpretability of the multi-scale thermal features is utilized in conjunction with the low computational cost of the lightweight convolutional neural network model, wherein the indirect processing includes:

[0097] The multi-scale thermal features provide highly discriminative input to the lightweight convolutional neural network model, reducing the model's learning complexity and improving the accuracy of the judgment results.

[0098] The efficient inference capability of the lightweight convolutional neural network model is utilized to ensure the real-time processing of the multi-scale thermal features;

[0099] By combining the confidence level output by the model, the analysis strategy for the multi-scale thermal features is dynamically adjusted to achieve a synergistic improvement in accuracy and real-time performance.

[0100] By integrating the three major functional modules of data storage, remote communication, and human-machine interaction with the core testing process, a system-level synergistic effect is generated, which greatly enhances the practical value and intelligence level of the testing system. Data storage is not just a simple record, but also forms the foundation for data-driven quality control and predictive maintenance of equipment, providing valuable data assets for subsequent process optimization and failure mode analysis.

[0101] Optionally, the step of generating an alarm signal includes:

[0102] Obtain the confidence level when the judgment result indicates the presence of a crack;

[0103] The confidence level is compared with a confidence threshold used to characterize the alarm risk level;

[0104] If the confidence level is higher than the confidence level threshold, a high-level alarm signal is generated; otherwise, a low-level re-examination signal is generated.

[0105] The model's output is transformed from a simple binary judgment (yes / no) into a probabilistic output that incorporates uncertainty information, and this output is effectively applied to alarm decisions. This enables the system to respond adaptively based on risk and cost, avoiding the "one-size-fits-all" drawbacks of traditional fixed threshold alarm strategies.

[0106] Example 2

[0107] A real-time crack detection system for metal plates integrating infrared thermal imaging and machine learning includes:

[0108] An infrared image acquisition device is used to acquire infrared thermal images of the surface of a metal plate.

[0109] The image preprocessing module is used to preprocess the infrared thermal image to obtain a preprocessed image;

[0110] A multi-scale thermal feature extraction module is used to extract multi-scale thermal features from the preprocessed image;

[0111] The crack detection module has a lightweight convolutional neural network model deployed inside it, which is used to receive the multi-scale thermal features and output the judgment result and confidence level of whether the crack exists.

[0112] A crack information identification module is used to identify detailed crack information based on the multi-scale thermal features when the judgment result indicates that a crack exists.

[0113] The report and alarm generation module is used to generate a detection report and alarm signal based on the judgment result and the detailed information of the crack;

[0114] The data storage module is used to store the test report and historical test data;

[0115] The communication module is used to enable remote data interaction;

[0116] Human-computer interaction interface, used for data display and operation control.

[0117] Example 3

[0118] Based on the same inventive concept, embodiments of this application also propose an electronic device, which includes:

[0119] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to execute the real-time detection system for metal plate cracks fused with infrared thermal imaging and machine learning according to embodiments of this application.

[0120] Furthermore, to achieve the above objectives, embodiments of this application also propose a computer-readable storage medium storing a computer program that, when executed by a processor, implements the real-time metal plate crack detection system that integrates infrared thermal imaging and machine learning, as described in the embodiments of this application.

[0121] The following is a detailed introduction to the various components of the electronic device:

[0122] In this context, the processor is the control center of the electronic device. It can be a single processor or a collective term for multiple processing elements. For example, a processor can be one or more central processing units (CPUs), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0123] Alternatively, the processor can perform various functions of the electronic device by running or executing software programs stored in memory and by calling data stored in memory.

[0124] The memory is used to store the software program that executes the solution of the present invention, and the execution is controlled by the processor. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0125] Optionally, the memory can be read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions, random access memory (RAM) or other types of dynamic storage devices capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory can be integrated with the processor or exist independently and coupled to the processor through the interface circuit of the electronic device; the embodiments of the present invention do not specifically limit this.

[0126] A transceiver is used to communicate with network devices or with terminal devices.

[0127] Optionally, the transceiver may include a receiver and a transmitter. The receiver is used to implement the receiving function, and the transmitter is used to implement the sending function.

[0128] Optionally, the transceiver can be integrated with the processor or exist independently and coupled to the processor through the router's interface circuit. This embodiment of the invention does not specifically limit this.

[0129] Furthermore, the technical effects of the electronic device can be referred to the technical effects of the data transmission method described in the above method embodiments, and will not be repeated here.

[0130] It should be understood that the processor in the embodiments of the present invention can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0131] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0132] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0133] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0134] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0135] It should be understood that, in various embodiments of the present invention, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0136] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

Claims

1. A real-time crack detection method for metal plates integrating infrared thermal imaging and machine learning, characterized in that, The method includes: S1 acquires an infrared thermal image of the metal plate surface; S2 preprocesses the infrared thermal image to obtain a preprocessed image that reflects the true temperature distribution; S3 extracts multi-scale thermal features from the preprocessed image, including local temperature gradient features, hot spot morphology features, and spatiotemporal temperature change patterns; the step of extracting multi-scale thermal features from the preprocessed image includes: The gradient operator is used to calculate the temperature gradient components Gx(x,y) and Gy(x,y) of each pixel in the image in the horizontal and vertical directions, and the magnitude of the temperature gradient at each point is calculated accordingly. The gradient magnitudes of all pixels constitute the temperature change rate data, thus obtaining the local temperature gradient features; Hot spot regions are identified by adaptive threshold segmentation, and connected component analysis is performed on each independent hot spot to extract morphological parameters including area, perimeter, eccentricity and principal axis direction, so as to obtain the morphological features of the hot spot region. The preprocessed images of consecutive frames are processed to analyze the temperature change over time and obtain the spatiotemporal temperature change pattern. The multi-scale thermal features are generated by integrating the local temperature gradient features, the hot spot region morphology features, and the spatiotemporal temperature change patterns. S4 inputs the multi-scale thermal features into a lightweight convolutional neural network model trained with training data that can identify cracks, and outputs the judgment result of whether a crack exists and the confidence level of the result. S5 If the judgment result indicates the presence of a crack, then the crack is analyzed based on the multi-scale thermal features to identify detailed crack information including crack location, length, and orientation; the step of identifying the detailed crack information includes: Based on the local temperature gradient characteristics, non-maximum suppression and connection component analysis are applied to the temperature change rate data to extract candidate crack segments. The crack segments with the highest confidence are selected by combining the model judgment results, and their pixel coordinate set is recorded as crack location data. Based on the morphological characteristics of the hot spot region, the skeleton of the hot spot region identified as a crack is extracted, and the cumulative Euclidean distance of the skeleton lines is calculated as an estimate of the crack length. Based on the morphological characteristics of the hot spot region and the spatiotemporal temperature variation pattern, the direction of crack propagation is determined by using the hot spot principal axis direction and combining the consistency statistics of the temperature gradient direction of the crack region, and expressed as an angle or unit vector. By integrating the crack location data, length estimation value, and extension direction information, a structured detailed crack information is generated. S6 generates a detection report and an alarm signal based on the judgment result and the detailed information of the crack.

2. The real-time detection method for metal plate cracks fused with infrared thermal imaging and machine learning according to claim 1, characterized in that, The step of preprocessing the infrared thermal image to obtain the preprocessed image includes: The infrared thermal image is subjected to non-uniformity correction to obtain a corrected image that eliminates inherent equipment errors; The corrected image is calibrated for temperature, and a temperature-grayscale response curve T=f(G_corr) is fitted. Then, the grayscale values ​​of each pixel in the corrected image are converted into actual temperature values ​​to obtain the calibrated image. The calibrated image is then subjected to noise filtering to obtain a preprocessed image that suppresses environmental interference.

3. The real-time detection method for metal plate cracks fused with infrared thermal imaging and machine learning according to claim 2, characterized in that, The lightweight convolutional neural network model is obtained through the following steps: Obtain training data containing multi-scale thermal features and corresponding labels of cracked and non-cracked samples; The training data is used to train a convolutional neural network with a simplified network structure. The network is constructed using depthwise separable convolution or model pruning techniques to reduce the number of parameters and computational complexity. During the training process, the backpropagation algorithm and gradient descent optimizer are used to continuously adjust the internal weight parameters of the model to minimize the difference between the model's prediction results and the true labels until the model reaches the preset accuracy standard on the validation set. The parameters of the trained convolutional neural network are solidified to obtain the lightweight convolutional neural network model.

4. The real-time detection method for metal plate cracks fused with infrared thermal imaging and machine learning according to claim 3, characterized in that, The physical interpretability of the multi-scale thermal features is utilized in conjunction with the low computational cost of the lightweight convolutional neural network model. This joint processing includes: The multi-scale thermal features provide highly discriminative input to the lightweight convolutional neural network model, reducing the model's learning complexity and improving the accuracy of the judgment results. The efficient inference capability of the lightweight convolutional neural network model is utilized to ensure the real-time processing of the multi-scale thermal features; By combining the confidence level output by the model, the analysis strategy for the multi-scale thermal features is dynamically adjusted to achieve a synergistic improvement in accuracy and real-time performance.

5. The real-time detection method for metal plate cracks fused with infrared thermal imaging and machine learning according to claim 4, characterized in that, The step of generating the alarm signal includes: Obtain the confidence level when the judgment result indicates the presence of a crack; The confidence level is compared with a confidence threshold used to characterize the alarm risk level; If the confidence level is higher than the confidence level threshold, a high-level alarm signal is generated; otherwise, a low-level re-examination signal is generated.

6. A real-time crack detection system for metal plates integrating infrared thermal imaging and machine learning, characterized in that, The real-time crack detection method for metal plates, which integrates infrared thermal imaging and machine learning according to any of claims 1-5, includes: An infrared image acquisition device is used to acquire infrared thermal images of the surface of a metal plate. The image preprocessing module is used to preprocess the infrared thermal image to obtain a preprocessed image; A multi-scale thermal feature extraction module is used to extract multi-scale thermal features from the preprocessed image; The crack detection module has a lightweight convolutional neural network model deployed inside it, which is used to receive the multi-scale thermal features and output the judgment result and confidence level of whether the crack exists. A crack information identification module is used to identify detailed crack information based on the multi-scale thermal features when the judgment result indicates that a crack exists. The report and alarm generation module is used to generate a detection report and alarm signal based on the judgment result and the detailed information of the crack; The data storage module is used to store the test report and historical test data; The communication module is used to enable remote data interaction; Human-computer interaction interface, used for data display and operation control.

7. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the program, implements the steps of the method as described in claim 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in claim 5.

Images