A method and system for detecting the quality of concrete for hydraulic engineering

Abstract

This invention provides a method and system for quality inspection of concrete used in water conservancy projects, relating to the field of image processing technology. The method includes: acquiring multiple training samples; preprocessing each training sample; constructing a concrete quality inspection model based on a deep neural network algorithm, combined with a depth completion network module, a depth adaptive region proposal module, and a multi-resolution detection network module; inputting the preprocessed training samples into the concrete quality inspection model and outputting the concrete quality inspection results; training the concrete quality inspection model in a joint manner based on the concrete labels and quality inspection results; acquiring an RGB-D image of the concrete to be inspected; and inputting the RGB-D image of the concrete to be inspected into the trained concrete quality inspection model to complete the quality inspection of the concrete to be inspected.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and system for detecting the quality of concrete used in water conservancy projects. Background Technology

[0002] Water conservancy projects refer to engineering projects related to the development, utilization, protection, and prevention of water resources, such as reservoirs, dams, river management, and hydropower stations. Concrete is a building material made of cement, aggregates (such as sand and gravel), and water, and is widely used in building structures. Concrete quality inspection uses image processing, deep learning, and other methods to automatically and intelligently detect potential defects in concrete (such as cracks and spalling).

[0003] For water conservancy projects, the quality of concrete is related to the safety and durability of the project. Especially in infrastructure such as dams and canals, cracks and spalling in concrete can cause serious safety problems. Therefore, testing the quality of concrete in water conservancy projects helps to detect problems in time, repair them in advance, and ensure the safe operation of the project. This is of great significance for improving the safety and quality assurance of water conservancy projects.

[0004] However, traditional inspection methods rely on manual inspection, which is easily affected by human factors, leading to missed detections or misjudgments. Secondly, manual inspection is inefficient, especially in large-scale projects. The inspection process is time-consuming and labor-intensive, making it impossible to thoroughly assess internal defects. It often lacks real-time capability, failing to detect potential problems in a timely manner, thus affecting the later maintenance and safety assurance of the project. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a method for testing the quality of concrete used in water conservancy projects. This method solves the problems of traditional testing methods relying on manual inspection, which is easily affected by human factors, leading to missed detections or misjudgments. Furthermore, manual inspection is inefficient, especially in large-scale projects, where the process is time-consuming and labor-intensive, making it impossible to thoroughly assess internal defects. It also typically lacks real-time capability, failing to promptly identify potential problems and impacting the subsequent maintenance and safety of the project.

[0006] A first aspect of this invention provides a method for quality testing of concrete used in hydraulic engineering, comprising: S1: Obtain multiple training samples, where each training sample includes an RGB-D image set of concrete samples and a corresponding label set; S2: Preprocess each training sample; S3: Based on deep neural network algorithms, a concrete quality detection model is constructed by combining a deep completion network module, a deep adaptive region proposal module, and a multi-resolution detection network module. S4: Input the preprocessed training samples into the concrete quality inspection model and output the concrete quality inspection results; S5: Based on the concrete labels and quality inspection results, the concrete quality inspection model is trained in a combined manner; S6: Acquire the RGB-D image of the concrete to be inspected; S7: Input the RGB-D image of the concrete to be tested into the trained concrete quality inspection model to complete the quality inspection of the concrete to be tested.

[0007] A second aspect of the present invention provides a concrete quality testing system for hydraulic engineering projects, comprising: a processor and a memory; The memory stores programs or instructions that can run on the processor, and when the program or instructions are executed by the processor, they implement the steps of the concrete quality testing method for hydraulic engineering as described in the first aspect.

[0008] A third aspect of the present invention provides a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the concrete quality testing method for hydraulic engineering as described in the first aspect.

[0009] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this embodiment of the invention, a concrete quality inspection model is constructed by acquiring diverse RGB-D image sets and corresponding labels, and based on a deep neural network algorithm, combining a deep completion network module, a deep adaptive region proposal module, and a multi-resolution detection network module. Then, the preprocessed training samples are input into the concrete quality inspection model, which outputs the concrete quality inspection results. Based on the concrete labels and the quality inspection results, the concrete quality inspection model is trained in a joint manner, ensuring efficient defect identification. Finally, the RGB-D image of the concrete to be inspected is acquired and input into the trained concrete quality inspection model, achieving a comprehensive assessment of concrete quality, significantly improving detection accuracy and efficiency, and ensuring engineering safety by timely detection of potential defects. Attached Figure Description

[0010] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0011] Figure 1This is a flowchart illustrating a method for testing the quality of concrete used in water conservancy projects, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a concrete quality testing system for water conservancy projects provided in an embodiment of the present invention. Detailed Implementation

[0012] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope 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 should fall within the scope of protection of the present invention.

[0013] The concrete quality testing method for water conservancy projects provided by the present invention will be described in detail below with reference to the accompanying drawings, through specific embodiments and application scenarios.

[0014] Reference manual attached Figure 1 The diagram shows a flowchart of a method for testing the quality of concrete used in water conservancy projects, provided by an embodiment of the present invention.

[0015] This invention provides a method for testing the quality of concrete used in water conservancy projects, which may include the following steps: S1: Obtain multiple training samples, where each training sample includes a set of RGB-D images of concrete and a corresponding label set.

[0016] The training samples are the dataset used to train the model. Each sample includes an input RGB-D image and a corresponding label indicating the type or location of defects in the concrete. The RGB-D image set is an image dataset composed of RGB (color) images and D (depth) images. The RGB images provide color information of the scene, while the D images provide depth information of objects in the scene, enabling the system to obtain the three-dimensional structure of space.

[0017] The label set consists of the correct answers in the training samples corresponding to the input image. For concrete quality inspection, the label set usually includes information such as the type and location of defects.

[0018] Specifically, RGB-D images are acquired by drones equipped with RGB-D cameras, which can simultaneously provide color and depth images.

[0019] It should be noted that by using RGB-D image sets and label sets, comprehensive training data can be provided, helping the model accurately identify concrete defects in various complex scenarios. The combination of RGB-D images provides the model with rich spatial information, and depth images are particularly helpful in identifying surface and internal defects in concrete, such as cracks or spalling, thus achieving greater accuracy than traditional color images. Drones equipped with RGB-D cameras can efficiently acquire images at challenging hydraulic engineering sites, avoiding the tediousness and limitations of manual inspection.

[0020] S2: Preprocess each training sample.

[0021] Preprocessing refers to various data cleaning and optimization operations performed before the data is input into the model.

[0022] It should be noted that preprocessing is used to improve image quality, especially in complex environments where RGB-D images can be affected by noise and other interference. Subtraction preprocessing and median filtering effectively remove noise from the image, ensuring the clarity and accuracy of the image data.

[0023] In one possible implementation, the preprocessing is specifically: subtraction preprocessing.

[0024] Subtraction preprocessing specifically includes: S201: The RGB-D image is smoothed using a median filter to obtain a corrected image.

[0025] S202: Based on the corrected image, remove noise from the RGB-D image through subtraction preprocessing:

[0026] in, I s ( x i ) represents the _th ... i The value of each pixel, max indicates the maximum value. Represents the RGB-D image of the th j The brightness value of each pixel. Represents the RGB-D image of the th i The brightness value of each pixel. R i Indicates the first i The set of pixel values ​​in the region surrounding each pixel. median This indicates the median operation.

[0027] It should be noted that subtraction preprocessing preserves important structural information and removes irrelevant interference, providing more accurate input data for subsequent quality inspection models. Compared to the unprocessed original image, the preprocessed image is easier for the model to understand, thus improving detection accuracy and efficiency. In summary, by eliminating noise and improving image quality, more reliable data support is provided for concrete quality inspection, ensuring that the model can accurately identify various defects in concrete.

[0028] S3: Based on deep neural network algorithms, a concrete quality detection model is constructed by combining a deep completion network module, a deep adaptive region proposal module, and a multi-resolution detection network module.

[0029] Deep neural network algorithms are a type of artificial neural network that process input data through multi-layer nonlinear transformations and are widely used in tasks such as pattern recognition and image classification.

[0030] The depth completion network module is used to fill in blank areas in an image. Especially in RGB-D images, when depth information is missing or occluded, the completion network can infer and reconstruct this missing depth information, making subsequent detection more accurate.

[0031] The depth-adaptive region proposal module is a deep learning module that automatically extracts regions that may contain defects based on the depth and texture information of the image. In concrete quality inspection, the depth-adaptive region proposal module can accurately locate defect areas, avoiding the computational waste caused by full-image processing.

[0032] The multi-resolution detection network module refers to a network module that processes images at different resolutions. By analyzing images at different scales, it is possible to better detect defects of different sizes and scales, thereby improving the robustness and accuracy of the model.

[0033] The concrete quality inspection model is a model that integrates a depth completion network, a depth adaptive region proposal module, and a multi-resolution detection network module.

[0034] It should be noted that a powerful concrete quality inspection model was constructed by integrating a depth completion network, a depth adaptive region proposal module, and a multi-resolution detection network module. The depth completion network fills in blank areas in RGB-D images, ensuring that all image information is effectively utilized. The depth adaptive region proposal module accurately identifies and locates potential defect areas, significantly improving detection accuracy and reducing computational waste. The use of the multi-resolution detection network module enables the model to detect defects at different scales, especially when dealing with defects of varying sizes and shapes, improving the flexibility and comprehensiveness of detection.

[0035] S4: Input the preprocessed training samples into the concrete quality inspection model and output the concrete quality inspection results.

[0036] Among them, the quality inspection results are the conclusions drawn by the model after processing the input data. They usually include the type of defect (such as cracks, spalling, etc.) and the location of the defect. The quality inspection results are the output of the model and are used to evaluate the quality of concrete.

[0037] It should be noted that automated quality inspection can be achieved by inputting pre-processed training samples into a trained concrete quality inspection model. The pre-processed samples have had noise removed and image quality enhanced, ensuring the model can make accurate judgments based on high-quality data. After inputting this data into the trained model, the model can quickly process and provide quality inspection results, improving inspection efficiency and reducing human error.

[0038] In one possible implementation, the quality inspection results include: the type and location of defects in the concrete.

[0039] Among them, the defect category refers to the different types of defects found in concrete. Common concrete defect categories include cracks, spalling, porosity and corrosion. The defect location refers to the specific area or coordinate position of the defect in the concrete structure.

[0040] In one possible implementation, S4 specifically includes: S401: The blank areas of the RGB-D image are filled in using the depth completion network module.

[0041] S402: Based on the completed RGB-D image, a depth-adaptive region proposal is generated through the depth-adaptive region proposal module.

[0042] in, flag Indicates category labels, n ( T ) represents the input imageT The number of pixels located within the defect area. f1ag =0 indicates that the input image has no defective pixels and is classified as a background image. N ( T ) represents the input image T Total number of pixels, k Indicates the empirical percentage threshold. f1ag = discard This indicates that the input image does not meet the criteria for a defective image and should be discarded. f1ag =1 indicates that the defect type of the input image is marked as a crack. f1ag =2 indicates that the defect type of the input image is marked as peeling.

[0043] S403: Based on the depth-adaptive region proposal, the multi-resolution detection network module outputs the defect category and location of the concrete.

[0044] It should be noted that the depth completion network module fills in the blank areas of the RGB-D image, ensuring the complete depth data of each pixel and improving image accuracy. The depth-adaptive region proposal module automatically identifies and locates potential defect areas based on the completed image, reducing the processing of irrelevant areas and improving detection efficiency. During this process, the defect type and region of the image are accurately marked, such as cracks and spalling, ensuring more precise subsequent processing. Combined with the multi-resolution detection network module, the category and location of defects are accurately identified, improving the flexibility and accuracy of detection. The advantage of this series of operations is that it can automatically and efficiently identify concrete defects, significantly improving detection accuracy, processing speed, and system robustness, while reducing manual intervention and errors.

[0045] In one possible implementation, S401 specifically includes: S4011: Calculate the gradient of the depth map in an RGB-D image using a standalone U-Net framework.

[0046] Among them, △( x ) represents the gradient in the horizontal direction of the depth map, △( y ) represents the gradient in the vertical direction of the depth map.

[0047] It's worth noting that U-Net is a deep learning network architecture widely used in image segmentation tasks, particularly in medical image analysis. Its structure consists of an encoder and a decoder, enabling it to effectively extract features from images and perform accurate pixel-level predictions. In this scenario, U-Net is used to process the depth map in an RGB-D image and calculate the image's gradient.

[0048] S4012: Determine the surface normal of each pixel in the depth map based on the gradient:

[0049] in, The surface normals represent the depth map.

[0050] Surface normals are vectors perpendicular to the image surface, used to describe the surface's orientation. In 3D image processing, surface normals help determine the geometry and orientation of image surfaces; in depth maps, calculating normals is crucial for reconstructing blank areas in the image.

[0051] S4013: Fill in the blank areas of the RGB-D image based on the surface normal.

[0052] It's worth noting that by combining the U-Net framework, gradient calculation, and surface normal reconstruction, the accuracy of filling in blank areas in RGB-D images is effectively improved. The U-Net framework uses deep learning to precisely process the depth map, calculating the image's gradient to effectively capture subtle changes in the depth map, especially in complex environments. Gradient calculation identifies areas with significant depth changes, which is crucial for determining concrete surface details (such as cracks or irregularities). Then, based on the gradient information, surface normal calculation helps accurately reconstruct blank areas in the depth map, ensuring complete depth information for all areas and avoiding false detections due to missing data.

[0053] In one possible implementation, S403 specifically includes: S4031: Adjust the size of the depth-adaptive region proposal to extract image features:

[0054] in, X 1 Representing feature maps, CNN Represents a convolutional neural network. w Represents the convolution kernel. b Indicates bias. T Indicates the input image. This represents the image after the input image has been reduced to half its original size. This represents the input image scaled down by 1 / 4. This represents the feature map of the input image after being scaled down by 1 / 2. This represents the feature map of the input image after being scaled down by 1 / 4.

[0055] Depth-adaptive region proposal refers to the adaptive selection of potentially defective regions in an image by analyzing depth image data. This proposal automatically determines which regions require more attention based on the image's depth information, thereby improving the accuracy of quality detection. Image features refer to key information extracted from the input image using deep learning algorithms that describes the image's content. Feature maps contain structural, textural, and other visual information about the image, aiding in subsequent classification and analysis.

[0056] S4032: Perform upsampling operation on image features.

[0057] Upsampling refers to enlarging the size of an image feature map to restore the image's spatial resolution. It is typically used to generate higher-resolution images for more accurate subsequent analysis.

[0058] S4033: Based on the image features after upsampling, combined with convolution and average pooling operations, output feature values:

[0059] in, f q This represents the first value obtained through average pooling operation. q Feature values ​​of the layer feature map C q Indicates the first q Feature map of the layer size Indicates the number of pixels. The summation symbol is used.

[0060] Among them, the convolution operation uses a convolution kernel to filter and extract features from the image, while the average pooling operation reduces the image resolution by calculating the average value of pixels in a small region of the image, thereby reducing the amount of computation and preserving important features.

[0061] S4034: Based on the eigenvalues, output the defect type and location of the concrete through the fully connected layer.

[0062] It should be noted that efficient and accurate concrete defect classification and localization are achieved through multi-layered image processing and feature extraction methods. First, image features are extracted by adjusting the size of depth-adaptive region proposals. This process accurately targets potential defect areas in the image, thereby improving subsequent detection performance. Upsampling operations enhance the resolution of the image feature map, helping to retain more image details and strengthening defect recognition. Combining convolution and average pooling operations further extracts and compresses image features, ensuring efficient feature representation and computational capabilities. Finally, a fully connected layer outputs the concrete defect category and location, achieving accurate defect classification and localization.

[0063] S5: Based on the concrete labels and quality test results, the concrete quality test model is trained in a combined manner.

[0064] The joint approach refers to the process of simultaneously optimizing multiple objectives or modules during model training. Joint training involves simultaneously optimizing the loss function to adjust multiple parts of the concrete quality inspection model, making it work more coordinated and efficient.

[0065] It should be noted that joint training effectively improves the overall performance of the concrete quality inspection model. By simultaneously considering concrete labels and the model's quality inspection results, joint training allows the model to optimize across multiple objectives, thereby enhancing its overall performance. Compared to optimizing each module individually, joint training more accurately balances the influence of different modules, preventing over-adjustment of certain modules and improving the model's stability and accuracy. In this way, the model can not only accurately identify defect categories but also accurately locate defect positions, thus providing more comprehensive and reliable quality inspection results.

[0066] In one possible implementation, S5 specifically includes: S501: Based on the error between the concrete label and the quality inspection results, determine the first loss function and the second loss function of the concrete quality inspection model.

[0067] The first loss function is as follows:

[0068] in, Loss 1 represents the loss function of the deep completion network module, i.e., the first loss function, and arg min represents minimizing it. L 1 represents L1 loss. The surface normals that represent the depth map. Represents the true normal of the depth map. Indicates the input image T Predicted depth Indicates the input image T The true depth.

[0069] The second loss function is as follows:

[0070] in, Loss 2 represents the loss function of the multi-resolution detection network module, i.e., the second loss function. y r Indicates the first r The true labels of each training sample log Represents the logarithmic function. x r express r training samples,W AdaNet express AdaNet Model weights f ( ) indicates the probability of predicting the output category.

[0071] S502: If both the first loss function and the second loss function are less than the preset loss function value, complete the training of the concrete quality detection model.

[0072] In deep learning, the loss function is used to quantify the difference between the model's predictions and the actual labels. The first and second loss functions correspond to different model modules or tasks, and the model is trained and tuned by optimizing the loss functions.

[0073] Those skilled in the art can set the value of the preset loss function according to the actual situation, and the present invention does not limit it.

[0074] It should be noted that the first loss function focuses on optimizing the deep completion network module, while the second loss function focuses on the accuracy of the multi-resolution detection module, ensuring that the model's performance is optimized at different levels. Furthermore, through weight adjustment in the AdaNet model, the model can dynamically optimize weight allocation during training, improving its adaptability and performance. Ultimately, joint training enables the model to better capture the complex features of concrete defects, improving overall prediction accuracy and stability. In summary, the refined loss function design and joint training strategy improve the model's performance in concrete quality detection, providing a solid foundation for efficient and accurate detection.

[0075] S6: Obtain the RGB-D image of the concrete to be inspected.

[0076] S7: Input the RGB-D image of the concrete to be tested into the trained concrete quality inspection model to complete the quality inspection of the concrete to be tested.

[0077] The trained concrete quality inspection model refers to a deep learning model optimized through a prior training process. This model has learned how to identify and classify defects in concrete based on RGB-D images, and through training, it can accurately determine the type and location of defects in new images. Quality inspection refers to the inspection and analysis of concrete to determine whether defects or problems exist. In this context, quality inspection mainly refers to using a deep learning model to analyze the concrete images to be inspected and identify defects such as cracks and spalling.

[0078] It should be noted that by inputting the RGB-D image of the concrete to be inspected into the trained quality inspection model, quality inspection can be completed quickly and accurately. The trained model has learned the characteristics of concrete defects, thus enabling it to efficiently identify potential defects when faced with new images, reducing manual intervention. The depth information of the RGB-D image enhances the spatial understanding of quality inspection, allowing the model to not only identify surface defects but also analyze the deep structure within the image, showing a significant advantage, especially in detecting complex defects such as cracks and spalling. Furthermore, the automated inspection process significantly improves inspection efficiency, shortens inspection time, provides engineers with more timely feedback, and effectively improves the level of concrete quality control. In summary, the intelligent and automated inspection process ensures both accuracy and efficiency in the inspection process.

[0079] In one possible implementation, the process after S7 includes: S8: When the defect type is crack, determine the width of the crack.

[0080] Cracks are a common defect in concrete, usually caused by external forces, temperature changes, or material problems. The presence of cracks can affect the structural strength and durability of concrete, therefore accurate measurement of their width and other characteristics is necessary. Width is a key parameter of a crack, representing the distance from one side to the other. In concrete quality inspection, determining the width of cracks helps assess their impact on structural safety and, consequently, whether repair is required.

[0081] It's important to note that accurately measuring crack width provides a more detailed analysis of concrete defects, especially since width is a critical indicator in crack assessment. Crack width directly impacts the structural strength and safety of concrete; therefore, automated and accurate crack width measurement enables the timely detection of potential risk areas. This method not only improves detection accuracy but also reduces the bias of manual measurements, allowing engineers to obtain reliable crack data and make informed decisions.

[0082] In one possible implementation, S8 specifically includes: S801: Determine the coordinates of the crack's centerline:

[0083] in, Indicates the first l The pixel at the th point f The coordinates on the central axis of the crack. Indicates the first f The center of the crack, h l Indicates the first l The scaling factor of each pixel Indicates the first f The edge of the crack.

[0084] The central axis coordinates refer to the positions of various points along the center line of the crack. The central axis of the crack is usually the center line of symmetry of the crack. By calculating the position of each point, the shape of the crack can be described more accurately.

[0085] S802: Calculate the brightness value of each point on the central axis of the crack based on the coordinates of the central axis:

[0086] in, Indicates the first f The first crack l The brightness value of each pixel. g 0 represents the brightness of the image.

[0087] In this context, brightness value refers to the brightness of each pixel in an image, and is usually related to the pixel's color or depth. In crack detection, brightness value can help distinguish cracks from their surrounding areas.

[0088] S803: The edge brightness variation of the crack is smoothed using Cardan spline interpolation. in, Indicates the first f The first crack l Each pixel in the parameter t The smooth brightness value below, Indicates the first f The first crack l + j The brightness value of each pixel. Indicates the first j Each pixel in the parameter t The basis functions of Cardan spline interpolation.

[0089] Cardan spline interpolation is a mathematical method used to smooth and approximate the relationship between data points. It is often used in image processing to smooth brightness variations, remove noise, and preserve the overall shape of the data.

[0090] S804: Differentiate the edge brightness change of the smoothed crack to determine the edge position of the crack.

[0091] The edge point of a crack refers to the extreme positions at both ends of the crack, and is usually used to define the start and end positions of a crack.

[0092] S805: Determine the width of the crack based on its edge location:

[0093] in, Indicates the first f The width of the crack, Indicates the first f The edge of the crack, Indicates the first f The edge point at the other end of the crack.

[0094] It is important to note that by determining the coordinates of the crack's central axis and its brightness value, the center location and shape of the crack can be accurately identified, providing crucial geometric references for subsequent analysis. Smoothing the crack edge brightness using Cardan spline interpolation not only eliminates errors caused by noise but also ensures the accuracy of the crack edge shape. Differentiating the smoothed data allows for more precise positioning of the crack edge, leading to a more accurate calculation of the crack width. Crack width is a crucial indicator for assessing the degree of damage to concrete structures; precise width measurement helps engineers determine the impact of cracks on structural safety. The advantage of this process lies in its ability to acquire accurate crack geometric information through a highly automated method, reducing manual intervention and improving the accuracy and efficiency of detection, particularly in large-scale concrete inspections, where it significantly saves time and costs. Furthermore, the use of mathematical interpolation and derivative methods better addresses the analysis of complex cracks, providing more accurate data support for concrete quality control.

[0095] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this embodiment of the invention, a concrete quality inspection model is constructed by acquiring diverse RGB-D image sets and corresponding labels, and based on a deep neural network algorithm, combining a deep completion network module, a deep adaptive region proposal module, and a multi-resolution detection network module. Then, the preprocessed training samples are input into the concrete quality inspection model, which outputs the concrete quality inspection results. Based on the concrete labels and the quality inspection results, the concrete quality inspection model is trained in a joint manner, ensuring efficient defect identification. Finally, the RGB-D image of the concrete to be inspected is acquired and input into the trained concrete quality inspection model, achieving a comprehensive assessment of concrete quality, significantly improving detection accuracy and efficiency, and ensuring engineering safety by timely detection of potential defects.

[0096] Reference manual attached Figure 2 The diagram shows a structural schematic of a concrete quality testing system for water conservancy projects provided by an embodiment of the present invention.

[0097] This invention provides a concrete quality testing system 20 for hydraulic engineering, comprising: a processor 201 and a memory 202; The memory 202 stores programs or instructions that can run on the processor 201. When the program or instructions are executed by the processor 201, they implement the steps of the above-described concrete quality testing method for water conservancy projects and achieve the same technical effect. To avoid repetition, the present invention will not repeat the above-described steps.

[0098] It should be understood that the processor 201 in this embodiment of the invention may be a central processing unit (CPU), or it may 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 may be a microprocessor or any conventional processor.

[0099] It should also be understood that the memory 202 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).

[0100] 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. A 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 flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, 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. A 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. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0101] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers 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.

[0102] 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.

[0103] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0104] In the embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0105] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0106] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0107] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0108] This invention provides a readable storage medium that stores a program or instructions on the medium. When the program or instructions are executed by a processor, they implement the steps of the above-described method for detecting the quality of concrete used in water conservancy projects and achieve the same technical effect. To avoid repetition, this invention will not elaborate further.

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A method for quality testing of concrete used in water conservancy projects, characterized in that, include: S1: Obtain multiple training samples, wherein each training sample includes an RGB-D image set of concrete samples and a corresponding label set; S2: Preprocess each of the training samples; S3: Based on deep neural network algorithms, a concrete quality detection model is constructed by combining a deep completion network module, a deep adaptive region proposal module, and a multi-resolution detection network module. S4: Input the preprocessed training samples into the concrete quality detection model and output the quality detection results of the concrete samples; S4 specifically includes: S401: The blank areas of the RGB-D images in the RGB-D image set are filled in using the depth completion network module; Specifically, S401 includes: S4011: Calculate the gradient of the depth map in the RGB-D images of the RGB-D image set using an independent U-Net framework; S4012: Determine the surface normal of each pixel in the depth map based on the gradient; S4013: Fill in the blank areas of the RGB-D images in the RGB-D image set according to the surface normal; S402: Based on the completed RGB-D image, a depth-adaptive region proposal is generated through the depth-adaptive region proposal module: in, flag Indicates category labels, n ( T ) represents the input image T The number of pixels located within the defect area. flag =0 indicates that the input image has no defective pixels and is classified as a background image. N ( T ) represents the input image T Total number of pixels, k Indicates the empirical percentage threshold. flag = discard This indicates that the input image does not meet the criteria for a defective image and should be discarded. flag =1 indicates that the defect type of the input image is marked as a crack. flag =2 indicates that the defect type of the input image is marked as peeling; S403: Based on the depth adaptive region proposal, the quality detection result of the concrete sample is output through the multi-resolution detection network module. The multi-resolution detection network module refers to a network module that processes images at different resolutions. Specifically, S403 includes: S4031: Adjust the size of the depth-adaptive region proposal to extract image features: Extract feature maps from the input image T, the image scaled down by 1 / 2, and the image scaled down by 1 / 4 using a convolutional neural network (CNN). S4032: Perform an upsampling operation on the image features; S4033: Based on the image features after upsampling, combine convolution and average pooling operations to output feature values; S4034: Based on the aforementioned feature values, the quality test results of the concrete sample are output through the fully connected layer; S5: Based on the label of the concrete sample and the quality test results, the concrete quality test model is trained in a combined manner; S5 specifically includes: S501: Based on the error between the concrete label and the quality inspection results, determine the first loss function and the second loss function of the concrete quality inspection model; The first loss function is as follows: in, Loss 1 represents the loss function of the deep completion network module, i.e., the first loss function, and arg min represents minimizing it. L 1 represents L1 loss. The surface normals that represent the depth map. Represents the true normal of the depth map. Indicates the input image T Predicted depth Indicates the input image T The true depth; The second loss function is as follows: in, Loss 2 represents the loss function of the multi-resolution detection network module, i.e., the second loss function. y r Indicates the first r The true labels of each training sample log Represents the logarithmic function. x r express r training samples, W AdaNet express AdaNet Model weights f ( ) indicates the probability of predicting the output category; S502: When the function values ​​of the first loss function and the second loss function are both less than the preset loss function value, the training of the concrete quality detection model is completed; S6: Acquire the RGB-D image of the concrete to be inspected; S7: Input the RGB-D image of the concrete to be tested into the trained concrete quality inspection model, output the quality inspection result of the concrete to be tested, and complete the quality inspection of the concrete to be tested.

2. The method for testing the quality of concrete used in water conservancy projects according to claim 1, characterized in that, The preprocessing specifically includes: subtraction preprocessing; The subtraction preprocessing specifically includes: S201: The RGB-D image is smoothed using a median filter to obtain a corrected image; S202: Based on the corrected image, noise in the RGB-D image is removed through the subtraction preprocessing.

3. The method for testing the quality of concrete used in water conservancy projects according to claim 1, characterized in that, The quality inspection results include: the type and location of defects in the concrete.

4. The method for testing the quality of concrete used in water conservancy projects according to claim 1, characterized in that, Following S7, it also includes: S8: When the defect category of the concrete to be tested is crack, determine the width of the crack.

5. The method for testing the quality of concrete used in water conservancy projects according to claim 4, characterized in that, S8 specifically includes: S801: Determine the coordinates of the central axis of the crack; S802: Calculate the brightness value of each point on the central axis of the crack based on the central axis coordinates; S803: The edge brightness variation of the crack is smoothed by Cardan spline interpolation. S804: Differentiate the edge brightness change of the smoothed crack to determine the edge position of the crack; S805: Determine the width of the crack based on the edge position.

6. A concrete quality testing system for hydraulic engineering projects, characterized in that, include: processor; A memory storing computer-readable instructions, which, when executed by the processor, implement the method for testing the quality of concrete for water conservancy projects as described in any one of claims 1 to 5.

Images