Multi-depth model coupled fresh concrete quality evaluation method, equipment and medium

Abstract

The invention relates to the technical field of concrete quality evaluation, in particular to a multi-depth model coupled fresh concrete quality evaluation method, equipment and a medium, which can evaluate various characteristics such as workability and mechanical properties of fresh concrete. According to the method, the influence of environment illumination change on video data acquisition is effectively eliminated, the consistency and stability of image quality are ensured, a high-quality data basis is provided for subsequent feature extraction and classification tasks, and the accuracy and reliability of recognition are improved; in addition, the convolutional layer hyper-parameter optimization design of the convolutional neural network (CNN) is combined with a multi-classifier deep learning model, and key features, such as workability and mechanical properties, of fresh concrete can be captured on different scales. The multi-scale feature extraction mode not only improves the robustness of the model, but also can reflect the quality condition of the concrete more comprehensively.

Description

Technical Field

[0001] The present invention relates to the technical field of concrete quality evaluation, and particularly relates to a method, device and medium for evaluating the quality of fresh concrete by coupling multi-depth models. Background Art

[0002] As the largest bulk material in the construction industry, the quality control of concrete is particularly important. However, quality fluctuations often occur during the production process of concrete. These fluctuations may stem from factors such as inconsistencies in raw materials, changes in mix ratios, and metering errors in the mixing system. Traditional concrete quality evaluation mainly relies on manual sampling or visual inspection. This method is not only highly subjective, time-consuming and laborious, but also difficult to guarantee the accuracy of the evaluation results. Therefore, there is an urgent need in the market for a fast and accurate method for evaluating the quality of concrete.

[0003] Computer vision is an important branch of artificial intelligence. It combines knowledge in multiple fields such as image processing, pattern recognition, and machine learning, and has broad application prospects. The application of computer vision technology in quality evaluation in the industrial field is becoming increasingly widespread, such as the dimensional recognition of parts, defect detection, etc. Through image recognition and processing technology, computer vision can automatically detect and analyze product quality, significantly improving the efficiency and accuracy of detection.

[0004] CN 202311528509.9 discloses a method for online identification of the workability and quality of ready-mixed concrete, which classifies the workability of ready-mixed concrete through a deep learning network, but this method cannot evaluate other characteristics or quality indicators of ready-mixed concrete except workability. Summary of the Invention

[0005] In view of this, the present invention provides a method, device and medium for evaluating the quality of fresh concrete by coupling multi-depth models, which can evaluate various characteristics of fresh concrete including workability and mechanical properties.

[0006] To achieve the above object, the technical solution of the present invention is as follows:

[0007] A multi-label evaluation method for the quality of fresh concrete by coupling multi-depth models, comprising the following steps:

[0008] S1: Collect video data of fresh concrete, and eliminate environmental influencing factors through lighting equipment;

[0009] S2: Extract frames from the collected video data and perform image preprocessing;

[0010] S3: Set the hyperparameters of the convolutional layer of the convolutional neural network, and use this neural network to extract features from the preprocessed image, and then perform target recognition task training through at least one fully connected layer to identify the concrete region of interest that needs further evaluation and the rapid concrete flow region that occludes part of the region of interest;

[0011] S4: Combine the target recognition results, further crop the recognition target according to the requirements of fresh concrete recognition, retain the main concrete region of interest, and remove the blurred part of the rapid flow of ready-mixed concrete through the recognition results;

[0012] S5: Use the multi-classifier deep learning model to perform training on the multi-label classification task of concrete quality for the image obtained in S4;

[0013] S6: Evaluate the quality of fresh concrete and give early warnings for possible risks according to the recognition results of the deep learning models in steps S3 and S5;

[0014] S7: Optimize the classifier and model structure of the model with the image representation of fresh concrete and the accuracy and robustness of the quality evaluation model as indicators.

[0015] Among them, the multi-label quality evaluation includes the workability and mechanical property characteristics of fresh concrete.

[0016] Among them, the image preprocessing in step S2 includes histogram equalization, noise removal, and edge enhancement.

[0017] Among them, the hyperparameters of the convolutional layer of the convolutional neural network in step S3 include the convolutional kernel size, stride size, padding method, and activation function.

[0018] Among them, in the multi-label classification task in step S5, the image processing also includes standardizing and normalizing the image. The standardization formula is:

[0019]

[0020] Among them, I std represents the standardized image, X represents the image matrix, μ represents the image mean, σ represents the standard deviation of the image, and N represents the number of pixels in the image matrix X;

[0021] The normalization formula is:

[0022]

[0023] Among them, x i represents the image pixel value, min(x) and max(x) respectively represent the minimum and maximum values of the image pixels.

[0024] Among them, during the training process of the deep learning models in steps S3 and S5, the Adam optimizer is used for step size update, and an early stopping mechanism is set to prevent overfitting.

[0025] Among them, in the optimization of the classifier and model structure in step S7, the optimization methods include but are not limited to:

[0026] Using the cross-entropy loss function for end-to-end training;

[0027] Applying an attention mechanism to highlight the key features in the concrete images;

[0028] Adopting data augmentation with the model generalization ability including random cropping, rotation, and color transformation.

[0029] Among them, in the early warning step for possible risks, when the concrete quality evaluation result does not match the preset threshold, the early warning mechanism is automatically triggered. The early warning mechanism includes notifying by calling, sending text messages, emails, and popping up windows on the software interface to the quality control personnel; and recording the abnormal data in the system.

[0030] The present invention also provides a fresh concrete quality evaluation device with a multi-depth model coupling, including:

[0031] An image acquisition module: used to collect video data of fresh concrete and eliminate environmental influencing factors through lighting equipment;

[0032] An image preprocessing module: used to perform frame extraction and image preprocessing on the collected video data;

[0033] A feature extraction module: used to set the hyperparameters of the convolutional layer of the convolutional neural network and extract features from the preprocessed images;

[0034] A target recognition module: used to perform target recognition task training through at least one fully connected layer, and identify the concrete region of interest to be further evaluated and the concrete fast-flow region that occludes part of the region of interest;

[0035] An image cropping module: used to further crop the recognition target according to the fresh concrete recognition requirements in combination with the target recognition result, retain the main concrete region of interest, and remove the blurred part of the ready-mixed concrete fast flow through the recognition result;

[0036] A multi-label classification module: used to perform concrete quality multi-label classification task training on the cropped images using a multi-classifier deep learning model;

[0037] A quality evaluation and early warning module: used to evaluate the fresh concrete quality and give early warnings for possible risks according to the recognition results of the deep learning model;

[0038] Model optimization module: used to optimize the classifier and model structure of the model with the accuracy and robustness of the fresh concrete image characterization and quality evaluation model as the indicators.

[0039] The present invention also provides a computer-readable medium, on which a computer program is stored. When the computer program is executed, it causes the computer to execute the method of the present invention.

[0040] Beneficial effects:

[0041] 1. In the method of the present invention, the influence of environmental light changes on video data acquisition is effectively eliminated, ensuring the consistency and stability of image quality, providing a high-quality data basis for subsequent feature extraction and classification tasks, and improving the accuracy and reliability of recognition; in addition, the optimized design of the hyperparameters of the convolutional layer of the convolutional neural network (CNN), combined with the multi-classifier deep learning model, can capture the key features of fresh concrete at different scales, such as workability, mechanical properties, etc. This multi-scale feature extraction method not only improves the robustness of the model, but also can more comprehensively reflect the quality status of concrete.

[0042] 2. In the method of the present invention, by performing frame extraction and image preprocessing (such as histogram equalization, noise removal, edge enhancement) on video data, and using CNN for target recognition, the main interest area of concrete is automatically cropped, removing the blurred part of the fast flow. This process is completely automated, reducing manual intervention and improving work efficiency. When the concrete quality evaluation result does not match the preset threshold, the system will automatically trigger an early warning mechanism, and timely notify the quality control personnel through various notification methods (such as phone calls, text messages, emails, software interface pop-ups, etc.), and record the abnormal data in the system. This intelligent risk early warning mechanism can help the construction unit timely discover and handle potential problems, and avoid the expansion of quality problems.

[0043] 3. The present invention can not only evaluate the workability of fresh concrete, but also comprehensively evaluate its mechanical properties, durability and other characteristics. Through the multi-label classification task, multiple quality indicators can be output simultaneously, providing a more comprehensive concrete quality evaluation result to meet the needs of different application scenarios. During the model training process, the cross-entropy loss function is used for end-to-end training, and the attention mechanism is applied to highlight the key features in the concrete image. In addition, data augmentation techniques (such as random cropping, rotation, color transformation, etc.) are also adopted to improve the generalization ability of the model. These optimization measures enable the model to maintain high accuracy and stability under different working conditions.

[0044] 4. In the method of the present invention, by performing standardization and normalization processing on the images, the consistency and standardization of the input data are ensured, the training speed and inference efficiency of the model are accelerated, the image data is made more suitable for processing by the deep learning model, and the performance of the model is further improved. In addition, this method supports real-time monitoring and dynamic adjustment, can continuously collect data and perform quality evaluation during the concrete production process, and can promptly detect and correct potential quality problems. This real-time feedback mechanism helps to improve production efficiency and reduce rework and waste.

[0045] 5. Traditional concrete quality evaluation relies on manual sampling and visual inspection. This method is not only time-consuming and laborious, but also easily affected by subjective factors. The present invention significantly reduces manual intervention, lowers labor costs, and improves the objectivity and accuracy of the evaluation through automated and intelligent technical means. The entire evaluation process, from data collection, preprocessing, feature extraction to quality evaluation and risk warning, is automatically completed by the system, greatly shortening the evaluation cycle and improving work efficiency. Especially in large-scale concrete production, the advantages of this method are particularly obvious.

[0046] 6. The method of the present invention is not only applicable to the laboratory environment, but also can be applied to real-time monitoring at the construction site. By optimizing the model structure and algorithm, it is ensured that it can work stably under complex environments such as different lighting conditions and temperature changes, and has strong adaptability. Specifically, it can not only be used for concrete quality control in large-scale infrastructure construction such as bridges and tunnels, but also be extended to other fields, such as precast component production, building construction, etc. Its multi-label evaluation function can meet the personalized needs of different engineering projects and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic flow chart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] The present invention will be described in detail below.

[0049] The present invention provides a fresh concrete quality evaluation method with multi-depth model coupling. The evaluation is a quality multi-label evaluation, and the quality multi-labels include various characteristics such as the workability and mechanical properties of fresh concrete. The evaluation method includes the following steps:

[0050] Step S1, using a high-precision camera to collect video data of fresh concrete, and eliminating environmental influencing factors through lighting devices such as fill lights.

[0051] Step S2, performing frame extraction and image preprocessing on the collected video data.

[0052] Specifically, the image preprocessing includes histogram equalization, noise removal, and edge enhancement.

[0053] Step S3, set the hyperparameters of the convolutional layer of the convolutional neural network (CNN), and use this neural network to extract features from the preprocessed image, and then perform target recognition task training through at least one fully connected layer to identify the concrete region of interest that needs further evaluation and interference regions such as the concrete rapid flow region that occludes part of the region of interest; among them, the hyperparameters of the convolutional layer of the convolutional neural network include but are not limited to the convolutional kernel size, stride size, padding method, activation function, etc.

[0054] Furthermore, during the training process of the deep learning model in this embodiment, the Adam optimizer is used to update the step size, and an early stopping mechanism is set to prevent overfitting.

[0055] Step S4, in combination with the target recognition result, further crop the recognition target according to the requirements of fresh concrete recognition, retain the main body region of interest of the concrete, and remove the blurred part of the rapid flow of the ready-mixed concrete through the recognition result.

[0056] Step S5, use the multi-classifier deep learning model to perform training on the multi-label classification task of concrete quality for the image obtained in S4; during the training process of the deep learning model, the Adam optimizer can be used to update the step size, and an early stopping mechanism is set to prevent overfitting.

[0057] Among them, in the multi-label classification task, the image processing also includes performing standardization and normalization processing on the image. The standardization formula is:

[0058]

[0059] Where, I std represents the standardized image, X represents the image matrix, μ represents the image mean, σ represents the standard deviation of the image, and N represents the number of pixels of the image matrix X;

[0060] The normalization formula is:

[0061]

[0062] Where, x i represents the image pixel point value, min(x) and max(x) respectively represent the minimum and maximum values of the image pixels.

[0063] Step S6, according to the recognition results of the deep learning models in steps S3 and S5, evaluate the quality of the fresh concrete and give early warnings about possible risks.

[0064] Step S7, taking the image representation of the fresh concrete and the accuracy and robustness of the quality evaluation model as indicators, optimize the classifier and model structure of the model.

[0065] Furthermore, in the optimization of the classifier and model structure, the optimization methods include, but are not limited to: performing end-to-end training using the cross-entropy loss function; applying the attention mechanism to highlight key features in the concrete images; and adopting data augmentation techniques, including techniques related to the generalization ability of the model such as random cropping, rotation, and color transformation.

[0066] The present invention also provides a fresh concrete quality evaluation device with multi-depth model coupling, which can be implemented by software and / or hardware and is generally integrated in an electronic device. The image acquisition module of the device: is used to collect video data of fresh concrete using a high-precision camera and eliminate environmental influencing factors through lighting devices such as fill lights;

[0067] The image preprocessing module: is used to perform frame extraction and image preprocessing on the collected video data;

[0068] The feature extraction module: is used to set the hyperparameters of the convolutional layer of the convolutional neural network (CNN) and extract features from the preprocessed images;

[0069] The target recognition module: is used to perform target recognition task training through at least one fully connected layer, and identify the concrete region of interest that needs further evaluation and interference regions such as the concrete fast-flow region that occludes part of the region of interest;

[0070] The image cropping module: is used to further crop the recognition target according to the fresh concrete recognition requirements in combination with the target recognition result, retain the main body region of interest of the concrete, and remove the blurred part of the ready-mixed concrete fast-flow through the recognition result;

[0071] The multi-label classification module: is used to perform concrete quality multi-label classification task training on the cropped images using a multi-classifier deep learning model;

[0072] The quality evaluation and early warning module: is used to evaluate the quality of fresh concrete and give early warnings about possible risks according to the recognition results of the deep learning model;

[0073] The model optimization module: is used to optimize the classifier and model structure of the model with the accuracy and robustness of the image representation and quality evaluation model of fresh concrete as indicators.

[0074] The fresh concrete quality evaluation device with multi-depth model coupling provided by the embodiments of the present invention can execute the fresh concrete quality evaluation method with multi-depth model coupling provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0075] As another aspect, the present application also provides a computer-readable storage medium, which may be the computer-readable storage medium included in the device described in the above embodiments; or it may exist separately and be a computer-readable storage medium not assembled into the device. The computer-readable storage medium may be a tangible storage medium, such as a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, floppy disk, hard disk, removable storage disk, CD-ROM, or any other form of storage medium well known in the art. The computer-readable storage medium stores one or more programs, and the one or more programs are used by one or more processors to execute the fresh concrete quality evaluation method of multi-depth model coupling described in the present application.

[0076] In summary, the above are only the preferred embodiments of the present invention, and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-label evaluation method for fresh concrete quality coupled with multiple depth models, characterized in that: The following steps are involved: S1: Collect video data of fresh concrete and eliminate environmental factors through lighting equipment; S2: Frame extraction and image preprocessing of the collected video data; S3: setting the convolutional layer hyperparameters of the convolutional neural network, and using the neural network to extract features from the preprocessed image, and then training the target recognition task through at least one fully connected layer to identify the concrete interest area that needs to be further evaluated and the concrete fast circulation area that occludes part of the interest area; S4: Combined with the target recognition results, the recognition targets are further cropped according to the recognition requirements of fresh concrete, the main concrete area of ​​interest is retained, and the blurred parts of the ready-mixed concrete that circulate quickly are removed through the recognition results; S5: Use a multi-classifier deep learning model to train the concrete quality multi-label classification task on the images obtained in S4; S6: Based on the recognition results of the deep learning model in steps S3 and S5, the quality of fresh concrete is evaluated and possible risks are warned; S7: Based on the accuracy and robustness of the image representation and quality evaluation model of fresh concrete, the classifier and model structure of the model are optimized.

2. The multi-label evaluation method for fresh concrete quality coupled with multiple depth models according to claim 1 is characterized in that: The quality multi-label evaluation includes workability and mechanical performance characteristics of fresh concrete.

3. The multi-label evaluation method for fresh concrete quality coupled with multi-depth models according to claim 1 is characterized in that: The image preprocessing in step S2 includes histogram equalization, noise removal and edge enhancement.

4. The multi-label evaluation method for fresh concrete quality coupled with multiple depth models according to claim 1 is characterized in that: The convolution layer hyperparameters of the convolutional neural network in step S3 include convolution kernel size, step size, padding method and activation function.

5. A multi-label evaluation method for fresh concrete quality coupled with multiple depth models according to any one of claims 1 to 4, characterized in that: In the multi-label classification task of step S5, the image processing also includes standardizing and normalizing the image, and the standardization formula is: Among them, I std represents the standardized image, X represents the image matrix, μ represents the image mean, σ represents the standard deviation of the image, and N represents the number of pixels in the image matrix X; The normalization formula is: Among them, x i Represents the image pixel value, min(x) and max(x) represent the minimum and maximum values ​​of the image pixels respectively.

6. The multi-label evaluation method for fresh concrete quality coupled with multiple depth models according to claim 5 is characterized in that: During the deep learning model training process of step S3 and step S5, the Adam optimizer is used to update the step size, and an early stopping mechanism is set to prevent overfitting.

7. The multi-label evaluation method for fresh concrete quality coupled with multiple depth models according to claim 6 is characterized in that: In the classifier and model structure optimization of step S7, the optimization method includes but is not limited to: End-to-end training using cross entropy loss function; Apply attention mechanism to highlight key features in concrete images; Data augmentation using model generalization capabilities including random crops, rotations, and color transformations.

8. A multi-label evaluation method for fresh concrete quality coupled with multiple depth models according to claim 6 or 7, characterized in that: In the early warning step of possible risks, when the concrete quality evaluation results do not meet the preset threshold, the early warning mechanism is automatically triggered. The early warning mechanism includes making a phone call to the quality control personnel, sending text messages, emails, and pop-up windows on the software interface to notify them; And record abnormal data in the system.

9. A device for implementing the method according to any one of claims 1 to 8, characterized in that: include: Image acquisition module: used to collect video data of fresh concrete and eliminate environmental factors through lighting equipment; Image preprocessing module: used to extract frames and preprocess images of collected video data; Feature extraction module: used to set the convolution layer hyperparameters of the convolutional neural network and extract features from the preprocessed image; Target recognition module: used to perform target recognition task training through at least one fully connected layer to identify concrete interest areas that need to be further evaluated and concrete fast circulation areas that block part of the interest area; Image cropping module: used to combine the target recognition results and further crop the identified targets according to the recognition requirements of fresh concrete, retain the main area of ​​interest of concrete and remove the blurred parts of ready-mixed concrete that circulate quickly through the recognition results; Multi-label classification module: used to train the cropped images using a multi-classifier deep learning model for multi-label classification of concrete quality; Quality evaluation and early warning module: used to evaluate the quality of fresh concrete and issue early warnings on possible risks based on the recognition results of the deep learning model; Model optimization module: used to optimize the model classifier and model structure based on the accuracy and robustness of the image representation and quality evaluation model of fresh concrete.

10. A computer readable medium having a computer program stored thereon, characterized in that: When the computer program is executed, the computer is caused to perform the method according to any one of claims 1 to 8.

Images