Concrete performance identification system based on deep learning image processing technology

Abstract

The invention relates to the field of concrete detection, and discloses a concrete performance identification system based on a deep learning image processing technology, which is applied to detection of concrete of a building structure beam column and comprises an image processing module, a deep learning identification module and a performance prediction module. The image preprocessed by the image processing module has stronger edge information and clearer texture features, which is helpful for the deep learning model to more efficiently extract discrimination features of key defects such as cracks, holes and the like; after details such as crack edges and hole boundaries are accurately segmented through a deep learning recognition module and image segmentation is completed, intelligent prediction of concrete compressive strength and compactness grade is finally realized by combining a regression model.

Description

Technical Field

[0001] This invention relates to the field of concrete testing, and more particularly to a concrete performance recognition system based on deep learning image processing technology. Background Technology

[0002] In modern construction engineering, concrete is widely used as a core structural material in load-bearing components such as beams, columns, and slabs. Its performance stability directly affects the structural safety and service life of buildings. Therefore, the detection and evaluation of concrete performance has become an important means of structural quality control and subsequent safety assessment. Some studies have attempted to identify concrete surface defects through image acquisition, but image processing is mostly based on traditional algorithms (such as edge detection and filtering enhancement), which cannot deeply explore the potential microscopic features in the image, resulting in insufficient accuracy and generalization ability. In recent years, with the rapid development of artificial intelligence technology, deep learning-based image recognition technology has achieved remarkable results in fields such as medicine and industrial defect detection, possessing advantages such as high accuracy, self-learning, and large-scale adaptability. However, in the field of concrete performance evaluation, there is still a lack of systematic and integrated solutions for effectively introducing deep learning image processing technology into concrete component inspection to achieve automatic identification and prediction from image features to performance parameters. Especially in practical engineering applications, concrete components have various complex defects (such as micro-cracks and pores), and the image acquisition quality is greatly affected by the environment. Existing image recognition methods are unable to cope with these unstructured, non-uniform, and low-contrast image information, resulting in detection accuracy and performance prediction precision that cannot meet engineering requirements.

[0003] Therefore, there is an urgent need to provide a concrete performance recognition system based on deep learning image processing technology, which can perform image optimization processing, high-precision feature extraction and intelligent performance prediction for concrete components such as beams and columns of building structures, so as to improve detection efficiency, accuracy and applicability. Summary of the Invention

[0004] The purpose of this invention is to provide a concrete performance recognition system based on deep learning image processing technology to solve the problem that, due to defects such as cracks and pores in concrete components, the image acquisition quality is greatly affected by the environment, and existing image recognition methods are difficult to handle these unstructured, non-uniform, and low-contrast image information, resulting in detection accuracy and performance prediction precision that cannot meet engineering requirements. The specific technical solution is as follows: A concrete performance recognition system based on deep learning image processing technology is used to detect concrete in beams and columns of building structures. It includes an image processing module, a deep learning recognition module, and a performance prediction module. The image processing module divides the beams and columns into two areas to be detected and captures images simultaneously. The image processing module optimizes the captured images to highlight image features. The deep learning recognition module extracts and analyzes the image features and generates data. The performance prediction module receives the data and, in conjunction with its built-in regression model, calculates the performance indicators of the detected concrete.

[0005] As an improvement to the above technical solution, the image processing module includes an acquisition submodule, which includes at least two drones, each equipped with an industrial camera. The at least two drones fly along opposite sides of the building structure beams and columns and take pictures using the industrial camera.

[0006] As an improvement to the above technical solution, the image processing module includes a preprocessing submodule, which performs grayscale conversion, noise removal, edge enhancement, and image enhancement processing on the image acquired by the acquisition submodule.

[0007] As an improvement to the above technical solution, the deep learning recognition module includes a ResNet network and a UNet network. The ResNet network is used to extract features of cracks and holes in the image, and the UNet network is used to segment the boundaries of cracks and holes. The ResNet network is used as the backbone feature extraction network in conjunction with the UNet network to complete the composite recognition.

[0008] As an improvement to the above technical solution, the UNet network labels parameters such as crack size, pore size, distribution density, and pore shape after segmentation.

[0009] As an improvement to the above technical solution, the performance prediction module includes one or more of the multilayer perceptron, XGBoost regressor, or support vector regression for calculation. After receiving the data generated by the deep learning recognition module, the performance prediction module outputs one or more parameters, such as the predicted value of concrete compressive strength or the density grade.

[0010] As an improvement to the above technical solution, the output density level is defined as 0-100. A parameter value close to 100 indicates that the concrete structure is relatively dense and has low porosity; a parameter value close to 0 indicates that the concrete structure is loose, has high porosity, and poor density.

[0011] As an improvement to the above technical solution, the acquisition submodule also includes an ultrasonic detection module for obtaining longitudinal depth information of cracks or holes. The ultrasonic detection module includes an ultrasonic transmitting unit, a receiving unit, and a data acquisition and processing unit.

[0012] As an improvement to the above technical solution, the ultrasonic detection module generates ultrasonic acoustic feature vector parameters based on the concrete being tested, inputs these parameters into the performance prediction module, and the ultrasonic detection module and the deep learning recognition module respectively calculate the performance parameters and then perform a weighted average.

[0013] The beneficial effects of this invention are as follows: The image preprocessed by the image processing module has stronger edge information and clearer texture features, which helps the deep learning model to extract the discrimination features of key defects such as cracks and holes more efficiently, thereby improving the recognition accuracy. After the deep learning recognition module achieves accurate segmentation of details such as crack edges and hole boundaries and completes image segmentation, it further automatically labels the size information such as crack length, width, and direction angle based on the segmentation results. Finally, by combining multiple regression models such as multilayer perceptron, XGBoost regressor and support vector regression, intelligent prediction of concrete compressive strength and density level is realized.

[0014] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Of course, implementing any product or method of this application does not necessarily require achieving all of the advantages described above at the same time. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a system framework diagram of the present invention. Detailed Implementation

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0018] Due to defects such as cracks and pores in concrete components, and the significant environmental impact on image acquisition quality, existing image recognition methods struggle to handle unstructured, non-uniform, and low-contrast image information. This results in detection accuracy and performance prediction precision failing to meet engineering requirements. Please refer to [link to relevant documentation]. Figure 1 The present invention provides some embodiments to solve the above problems. A concrete performance recognition system based on deep learning image processing technology is applied to detect concrete in beams and columns of building structures. The system includes an image processing module, a deep learning recognition module, and a performance prediction module. The image processing module optimizes the acquired images to highlight image features. The deep learning recognition module extracts and analyzes the image features and generates data. The performance prediction module receives the data and calculates the performance indicators of the detected concrete by combining the regression model built into the performance prediction module.

[0019] Currently, image acquisition of concrete components in the field of construction engineering mostly relies on manual handheld devices or fixed cameras. This method has problems such as limited coverage, single shooting angle, poor environmental adaptability, and poor image consistency. To address this, the present invention provides some embodiments. Specifically, the image processing module includes an acquisition submodule, which includes at least two drones, each equipped with an industrial camera. The at least two drones fly along opposite sides of the beams and columns of the building structure and take pictures using the industrial camera. During the flight of the drones, the camera can adjust the shooting angle according to the position. By introducing drones for data acquisition, high-altitude access can be quickly achieved, handling complex structural areas with strong adaptability. Secondly, equipped with professional industrial cameras and an image stabilization system, image clarity is guaranteed. This invention specifically proposes a data acquisition method using at least two drones flying synchronously along opposite sides of the beam and column. This is mainly to avoid visual blind spots. Understandably, a single drone can only acquire images of one side of the component, easily missing defects on the other side. Symmetrical flight of two drones can achieve full-circumference image acquisition of the component, effectively avoiding missed detections. Secondly, it improves image matching and stitching efficiency. Images from both sides have high geometric symmetry and spatial consistency, facilitating subsequent image stitching, registration, and fusion processing, thereby improving overall image quality.

[0020] Preferably, the image processing module includes a preprocessing submodule, which performs grayscale conversion, noise removal, edge enhancement, and image enhancement processing on the images acquired by the acquisition submodule; this significantly improves the overall clarity and defect visibility of the images, facilitating accurate identification by subsequent models.

[0021] Specifically, color RGB images can be converted to grayscale images to reduce image dimensionality and computational complexity while preserving the main structural information of the image; Gaussian filtering is used to effectively suppress random noise interference in the image and improve the image signal-to-noise ratio; and the Canny edge detector is used to enhance the edge features of defects such as cracks and holes, thereby improving the recognizability of the target area. The preprocessed image has stronger edge information and clearer texture features, which helps deep learning models extract the discrimination features of key defects such as cracks and holes more efficiently, thereby improving the recognition accuracy.

[0022] It reduces the difficulty of model training and improves generalization ability; it reduces image fluctuations caused by environmental differences, making deep learning models more adaptable and stable when facing images in different scenarios; grayscale processing reduces the number of image channels and reduces the amount of data for subsequent convolution operations. This invention, by setting up an image preprocessing submodule, performs a series of processing operations on the acquired image, including grayscale conversion, noise removal, edge enhancement, and image enhancement. This not only significantly improves image quality but also enhances the expressive ability of defect features, providing high-quality input data for the subsequent deep learning recognition module. This design effectively solves the recognition difficulties of traditional image recognition methods when processing low-contrast, non-uniform, and blurry images, demonstrating good practicality and applicability to intelligent detection tasks of various concrete structures such as bridges, tunnels, and high-rise buildings.

[0023] The deep learning recognition module includes a ResNet network and a UNet network. The ResNet network extracts features of cracks and holes in the image, and the UNet network segments the boundaries of cracks and holes. The ResNet network serves as the backbone feature extraction network, working in conjunction with the UNet network to complete composite recognition. Preferably, after segmentation, the UNet network labels parameters such as crack size, hole diameter, distribution density, and hole shape. Specifically, the workflow of the deep learning recognition module is as follows: First, a ResNet network is used as the backbone feature extractor to perform multi-level feature extraction on the preprocessed image. ResNet effectively alleviates the gradient vanishing problem in deep networks through the residual connection mechanism, and can extract high-dimensional semantic features with strong discriminative power from the image. It is especially suitable for the initial localization and classification of slender, low-contrast targets such as cracks and holes. Based on the high-level features extracted by ResNet, a UNet network is introduced to complete the pixel-level semantic segmentation task. UNet adopts an encoder-decoder structure and combines skip connections, which can restore image resolution while preserving spatial information, thereby achieving accurate segmentation of details such as crack edges and hole boundaries. After image segmentation, the system further automatically labels the crack length, width, direction angle and other size information based on the segmentation results. For hole-type defects, key parameters such as pore size, distribution density and shape contour (such as circle, ellipse, irregular shape) are extracted and the above information is output in a structured manner for use by the subsequent performance prediction module.

[0024] The performance prediction module of the present invention is used to receive image feature data and structured defect parameters (such as crack length, width, hole area, distribution density, etc.) output by the deep learning recognition module, and to calculate the key performance indicators of concrete components based on machine learning or deep learning regression models. The performance prediction module includes one or more of the following for calculation: multilayer perceptron, XGBoost regressor, or support vector regression. After receiving the data generated by the deep learning recognition module, the performance prediction module outputs one or more parameters, such as the predicted value of concrete compressive strength or density grade.

[0025] The performance prediction module outputs at least one concrete performance index, including but not limited to: Predicted compressive strength of concrete (unit: MPa); The density grade of concrete is defined as ranging from 0 to 100, where: A value close to 100 indicates that the concrete structure is dense, has low porosity, and is of excellent quality. A value close to 0 indicates that the concrete structure is loose, has high porosity, and poor density; Through this quantitative output method, the system can intuitively reflect the internal structural state of concrete, assisting engineers in quickly judging the quality and durability level of components; the performance prediction module also supports historical data backtracking and online model update mechanism, which can continuously optimize the prediction model based on newly collected data, and improve the accuracy and adaptability of long-term prediction. This invention achieves intelligent prediction of concrete compressive strength and density grade by setting up a performance prediction module and combining multiple regression models such as multilayer perceptron, XGBoost regressor, and support vector regression. This module not only effectively utilizes image features and defect parameters output by the deep learning recognition module, but also provides structured numerical feedback on the concrete performance status, significantly improving the intelligence, automation, and engineering practicality of the detection system.

[0026] In some embodiments, the acquisition submodule further includes an ultrasonic detection module for longitudinal depth information of cracks or holes. The ultrasonic detection module includes an ultrasonic transmitting unit, a receiving unit, and a data acquisition and processing unit. Specifically, the ultrasonic detection module generates ultrasonic acoustic feature vector parameters based on the concrete being tested, inputs the parameters into the performance prediction module, and the ultrasonic detection module and the deep learning recognition module respectively calculate the performance parameters and then perform a weighted average. The ultrasonic detection module of this invention is used to acquire longitudinal depth information and other internal structural features of cracks or holes in concrete components. This module works in conjunction with a deep learning recognition module to form an intelligent detection system that integrates visual and acoustic multimodal data.

[0027] Specifically: Ultrasonic transmitting unit: used to transmit high-frequency ultrasonic pulse signals to the concrete component being tested; Ultrasonic receiving unit: used to receive ultrasonic signals that penetrate or reflect from the internal structure of concrete; Data acquisition and processing unit: digitizes the received ultrasonic signals, extracts key parameters such as propagation time, amplitude attenuation, and spectral changes, and generates ultrasonic acoustic feature vectors.

[0028] In practical applications, the ultrasonic detection module can be deployed synchronously with the UAV platform to perform non-contact ultrasonic detection while completing image acquisition, and obtain information on internal defects such as crack depth and cavity distribution.

[0029] Regarding the module collaborative working mechanism: During operation, the performance prediction module receives data input from two independent sources: image features and defect parameters (such as crack length, width, pore area, distribution density, etc.) output by the deep learning recognition module; and ultrasonic acoustic feature vectors (such as sound velocity, first wave arrival time, energy attenuation rate, frequency response, etc.) output by the ultrasonic detection module. Subsequently, the performance prediction module independently calculates the key performance indicators of concrete (such as compressive strength, density grade, etc.) based on the above two types of input data, and fuses and optimizes the results through a weighted average algorithm, finally outputting a comprehensive evaluation result. Preferably, the weighting coefficients can be dynamically adjusted according to the confidence level of different models, historical prediction errors, and engineering experience to improve the overall prediction accuracy and stability.

[0030] This invention introduces an ultrasonic detection module into the acquisition submodule and combines it with a deep learning recognition module to construct a multi-source information fusion system of "image recognition + ultrasonic detection." This overcomes the limitations of single image recognition and improves the completeness of detection. Image recognition can only reflect the state of defects on the concrete surface and cannot obtain key information such as crack depth and internal voids. In contrast, the ultrasonic detection module can penetrate the concrete surface and obtain the acoustic response of the internal structure, thereby achieving three-dimensional perception and quantitative analysis of defects. The performance prediction module performs calculations based on data from both image recognition and ultrasonic detection, and obtains the final result through a weighted fusion strategy. Compared with single-data-source prediction methods, it has higher prediction accuracy and robustness, and is particularly suitable for evaluating the performance of concrete under complex damage conditions. The ultrasonic acoustic feature vector provides information dimensions that are complementary to image features, helping machine learning or neural network models to more comprehensively understand the physical properties of concrete materials and improve the adaptability and generalization ability of the model under different environmental conditions.

[0031] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of equivalents of the claims be included within the present invention.

Claims

1. A concrete performance recognition system based on deep learning image processing technology, applied to the detection of concrete in beams and columns of building structures, characterized in that, The system includes an image processing module, a deep learning recognition module, and a performance prediction module. The image processing module divides the beam and column into two areas to be detected and captures images simultaneously. The image processing module optimizes the captured images to highlight image features. The deep learning recognition module extracts and analyzes the image features and generates data. The performance prediction module receives the data and, in conjunction with its built-in regression model, calculates the performance indicators of the detected concrete.

2. The concrete performance identification system based on deep learning image processing technology according to claim 1, characterized in that: The image processing module includes an acquisition submodule, which includes at least two drones, each equipped with an industrial camera. The at least two drones fly along opposite sides of the building structure beams and columns and take pictures using the industrial camera.

3. The concrete performance identification system based on deep learning image processing technology according to claim 2, characterized in that: The image processing module includes a preprocessing submodule, which performs grayscale conversion, noise removal, edge enhancement, and image enhancement processing on the images acquired by the acquisition submodule.

4. The concrete performance identification system based on deep learning image processing technology according to claim 3, characterized in that: The deep learning recognition module includes a ResNet network and a UNet network. The ResNet network is used to extract features of cracks and holes in the image, and the UNet network is used to segment the boundaries of cracks and holes. The ResNet network is used as the backbone feature extraction network in conjunction with the UNet network to complete the composite recognition.

5. The concrete performance identification system based on deep learning image processing technology according to claim 4, characterized in that: After segmentation, the UNet network is labeled with parameters such as crack size, pore size, distribution density, and pore shape.

6. The concrete performance identification system based on deep learning image processing technology according to claim 5, characterized in that: The performance prediction module includes one or more of the following for calculation: multilayer perceptron, XGBoost regressor, or support vector regression. After receiving the data generated by the deep learning recognition module, the performance prediction module outputs one or more parameters, such as the predicted value of concrete compressive strength or the density grade.

7. The concrete performance identification system based on deep learning image processing technology according to claim 6, characterized in that: The output density level is defined as 0-100. An output parameter value close to 100 indicates that the concrete structure is relatively dense and has low porosity; an output parameter value close to 0 indicates that the concrete structure is loose, has high porosity, and has poor density.

8. The concrete performance identification system based on deep learning image processing technology according to claim 7, characterized in that: The acquisition submodule also includes an ultrasonic detection module for obtaining longitudinal depth information of cracks or holes. The ultrasonic detection module includes an ultrasonic transmitting unit, a receiving unit, and a data acquisition and processing unit.

9. The concrete performance identification system based on deep learning image processing technology according to claim 8, characterized in that: The ultrasonic testing module generates ultrasonic acoustic feature vector parameters based on the concrete being tested, and inputs these parameters into the performance prediction module. The ultrasonic testing module and the deep learning recognition module respectively calculate the performance parameters and then perform a weighted average.

Images