A method and system for anti-counterfeiting detection of concrete test blocks based on the CoAtNet network

Abstract

The present invention relates to the technical field of specimen anti-counterfeiting detection, and discloses a concrete specimen anti-counterfeiting detection method and system based on the CoAtNet network, including: obtaining a concrete specimen, collecting images of the concrete specimen before and after testing to obtain the pre-test specimen image and the post-test specimen image, inputting the pre-test specimen image and the post-test specimen image into an initial convolutional layer module, a multi-level repeated convolutional module, a feature integration module, and feature fusion to obtain the pre-test initial shallow features and the post-test initial shallow features, reducing the feature dimension to obtain the pre-test feature vector and the post-test feature vector, performing similarity calculation based on the pre-test feature vector and the post-test feature vector, and if the two are similar, the concrete specimen before and after the experiment is the same concrete specimen. The present invention saves labor and time and avoids engineering quality problems caused by the replacement of specimens.

Description

Technical Field

[0001] The present invention relates to the technical field of test block anti-counterfeiting detection, and particularly relates to a method and system for anti-counterfeiting detection of concrete test blocks based on the CoAtNet network. Background Art

[0002] The quality inspection of concrete test blocks is the core link of building engineering quality control, and the reliability of its anti-counterfeiting technology is directly related to the safety assessment of engineering structures. Traditional anti-counterfeiting technologies mainly rely on physical identification means such as barcodes, two-dimensional codes or RFID tags. For example, the Chinese invention patent "A method for anti-counterfeiting traceability of concrete test blocks based on chip fingerprints" (CN202410603044.7) combines RFID chips with PUF technology to endow each test block with a unique identity identifier, improving anti-counterfeiting security. Another example is the Chinese invention patent "An AI-based method for anti-counterfeiting of concrete test pieces by image recognition" (CN202410742697.3), which proposes a multi-scale detection model combining YOLOv8 and BiFPN and a dual-branch ViT feature extraction network, and comprehensively determines the authenticity of test pieces through dry-wet state image alignment and deep learning similarity scores. The advantage of this scheme is to introduce an advanced multi-scale feature fusion mechanism, significantly improving the target recognition accuracy in complex backgrounds, and strengthening the local feature correlation through a bidirectional attention module. However, its limitations cannot be ignored, as it relies on prefabricated tags with two-dimensional codes as reference features.

[0003] Contactless anti-counterfeiting technology based on image recognition has gradually emerged. For example, the Chinese invention patent "A method for preventing counterfeiting of concrete test blocks based on image recognition" (CN202110459242.7) discloses a method for collecting on-site and laboratory test block images through a mobile terminal APP and performing similarity comparison. This method uses a template matching algorithm to improve the robustness of image matching through rotation and cropping operations. Its advantages are convenient operation and low cost, but there are some limitations. This method does not consider multi-angle feature fusion, which may lead to the omission of key texture information and reduce the comprehensive judgment ability of the anti-counterfeiting system.

[0004] In summary, the existing technologies are either limited by the inherent defects of physical tags or fail to fully explore the potential of image recognition in anti-counterfeiting comparison. How to perform fine anti-counterfeiting on non-contact concrete blocks is still an extremely difficult problem. Through domestic and foreign patent and literature searches, there are currently many studies on the recognition of physical identification means for concrete anti-counterfeiting, but the research on contactless anti-counterfeiting technology generally uses single-angle image detection, which is difficult to comprehensively capture the surface texture details of test blocks, and the classification accuracy may decrease when the size or lighting conditions of the test blocks change. There is no multi-modal comparison anti-counterfeiting method optimized for the surface characteristics of concrete test blocks using the CoAtNet model. Summary of the Invention

[0005] The purpose of the present invention is to overcome one or more of the above existing technical problems, and provide a method and system for anti-counterfeiting detection of concrete test blocks based on the CoAtNet network.

[0006] To achieve the above purpose, a method for anti-counterfeiting detection of concrete test blocks based on the CoAtNet network provided by the present invention includes:

[0007] Obtain a concrete test block, and perform image acquisition on the concrete test block before and after the test to obtain the pre-experiment test block image and the post-experiment test block image;

[0008] Input the pre-experiment test block image and the post-experiment test block image into the initial convolutional layer module to obtain the pre-experiment initial convolutional feature and the post-experiment initial convolutional feature;

[0009] Input the pre-experiment initial convolutional feature and the post-experiment initial convolutional feature into the multi-level repeated convolutional module to obtain the pre-experiment deepened convolutional feature and the post-experiment deepened convolutional feature;

[0010] Input the pre-experiment deepened convolutional feature and the post-experiment deepened convolutional feature into the feature integration module to obtain the pre-experiment high-level convolutional feature and the post-experiment high-level convolutional feature;

[0011] Based on the features of the pre-experiment image and the post-experiment image obtained, perform feature fusion on them respectively, gradually fuse deeper features into shallow features, and obtain the pre-experiment initial shallow feature and the post-experiment initial shallow feature;

[0012] Reduce the feature dimensions of the pre-experiment initial shallow feature and the post-experiment initial shallow feature to obtain the pre-experiment feature vector and the post-experiment feature vector;

[0013] Perform similarity calculation based on the pre-experiment feature vector and the post-experiment feature vector. If the two are similar, the concrete test block is the same concrete test block before and after the experiment.

[0014] According to one aspect of the present invention, the method for obtaining the pre-experiment test block image and the post-experiment test block image is:

[0015] Obtain a concrete test block, perform the first image acquisition on the current concrete test block to obtain the first image of the concrete test block;

[0016] Perform an experiment on the current concrete test block, and perform the second image acquisition on the concrete test block after the experiment to obtain the second image of the concrete test block;

[0017] Perform central cropping on the first image of the concrete test block and the second image of the concrete test block, and scale the first image of the concrete test block and the second image of the concrete test block to a resolution size of 448×448 pixels;

[0018] Enhance the first image of the concrete specimen and the second image of the concrete specimen based on the application of data augmentation technology, increase the diversity of the first image of the concrete specimen and the second image of the concrete specimen, and simulate the situations encountered in the actual application scenario;

[0019] Normalize the first image of the concrete specimen and the second image of the concrete specimen to obtain the pre-experiment specimen image and the post-experiment specimen image.

[0020] According to one aspect of the present invention, the method for obtaining the pre-experiment initial convolution features and the post-experiment initial convolution features:

[0021] The initial convolution layer module includes a first convolution module and a second convolution module, and the first convolution module and the second convolution module are two 3×3 convolution kernels;

[0022] Input the pre-experiment specimen image and the post-experiment specimen image into the first convolution module to obtain the pre-experiment first convolution module features and the post-experiment first convolution module features, where the formula is,

[0023] ;

[0024] ;

[0025] Wherein, represents the pre-experiment specimen image;

[0026] represents the post-experiment specimen image;

[0027] represents a 3×3 convolution kernel;

[0028] represents the first convolution module;

[0029] represents the pre-experiment first convolution module features;

[0030] represents the post-experiment first convolution module features;

[0031] Input the pre-experiment first convolution module features and the post-experiment second convolution module features into the second convolution module to obtain the pre-experiment initial convolution features and the post-experiment initial convolution features, where the formula is,

[0032] ;

[0033] ;

[0034] Wherein, represents the second convolution module;

[0035] Represents the initial convolutional features before the experiment;

[0036] Represents the initial convolutional features after the experiment.

[0037] According to one aspect of the present invention, the method for obtaining the deepened convolutional features before and after the experiment is as follows:

[0038] The multi-level repeated convolution module includes a first repeated convolution module and a second repeated convolution module. The first repeated convolution module and the second repeated convolution module are composed of 2 1×1 convolutional kernels and 1 4×4 depthwise separable convolution;

[0039] Input the initial convolutional features before the experiment and the initial convolutional features after the experiment into the first repeated convolution module to obtain the features of the first repeated convolution module before the experiment and the features of the first repeated convolution module after the experiment. The formula is as follows,

[0040] ;

[0041] ;

[0042] Among them, Represents the first repeated convolution module;

[0043] Represents a 1×1 convolutional kernel;

[0044] Represents a depthwise separable convolution with a 4×4 convolutional kernel;

[0045] Represents the features of the first repeated convolution module before the experiment;

[0046] Represents the features of the first repeated convolution module after the experiment;

[0047] Input the features of the first repeated convolution module before the experiment and the features of the first repeated convolution module after the experiment into the second repeated convolution module to obtain the deepened convolutional features before the experiment and the deepened convolutional features after the experiment. The formula is as follows,

[0048] ;

[0049] ;

[0050] Among them, Represents the second repeated convolution module;

[0051] Represents the deepened convolutional features before the experiment;

[0052] Represents the deep convolutional features after the experiment.

[0053] According to one aspect of the present invention, the method for obtaining the pre-experiment high-level convolutional features and the post-experiment high-level convolutional features is as follows:

[0054] The feature integration module includes a first integration module and a second integration module, and the first integration module and the second integration module are repeated associative attention mechanisms and feed-forward neural networks;

[0055] Input the pre-experiment deep convolutional features and the post-experiment deep convolutional features into the first integration module to obtain the pre-experiment first integration module features and the post-experiment first integration module features, where the formula is,

[0056] ;

[0057] ;

[0058] Where, Represents the first integration module;

[0059] FFN represents the feed-forward neural network;

[0060] Represents the repeated associative attention mechanism;

[0061] Represents the pre-experiment first integration module features;

[0062] Represents the post-experiment first integration module features;

[0063] Input the pre-experiment first integration module features and the post-experiment first integration module features into the second integration module to obtain the pre-experiment high-level convolutional features and the post-experiment high-level convolutional features, where the formula is,

[0064] ;

[0065] ;

[0066] Where, Represents the second integration module;

[0067] Represents the pre-experiment high-level convolutional features;

[0068] Represents the post-experiment high-level convolutional features.

[0069] According to one aspect of the present invention, the method for obtaining the pre-experiment initial shallow features and the post-experiment initial shallow features is as follows:

[0070] Store the features of the pre-experiment images and the features of the post-experiment images extracted from each layer in the pre-experiment feature list and the post-experiment feature list;

[0071] Based on the FPN strategy, gradually fuse the deeper features into the shallow features to obtain the pre-experiment initial shallow features and the post-experiment initial shallow features. The formula is

[0072] ;

[0073] ;

[0074] where represents feature fusion;

[0075] represents the pre-experiment feature list;

[0076] represents the post-experiment feature list;

[0077] represents the pre-experiment initial shallow features;

[0078] represents the post-experiment initial shallow features.

[0079] According to one aspect of the present invention, the method for obtaining the pre-experiment feature vector and the post-experiment feature vector is as follows:

[0080] Apply adaptive average pooling to the pre-experiment initial shallow features and the post-experiment initial shallow features to obtain the pre-experiment compressed features and the post-experiment compressed features. The formula is

[0081] ;

[0082] ;

[0083] where represents feature fusion;

[0084] represents the pre-experiment feature list;

[0085] represents the post-experiment feature list;

[0086] Apply the fully connected layer to the pre-experiment compressed features and the post-experiment compressed features to obtain the pre-experiment feature vector and the post-experiment feature vector. The formula is

[0087] ;

[0088] ;

[0089] where Represents a fully connected layer;

[0090] Represents the pre - experiment feature vector;

[0091] Represents the post - experiment feature vector.

[0092] To achieve the above object, the present invention provides a concrete block anti - counterfeiting detection system based on the CoAtNet network, including:

[0093] Block image acquisition module: Acquire concrete blocks, and perform image acquisition on the concrete blocks before and after the test to obtain the pre - experiment block image and the post - experiment block image;

[0094] Initial convolution feature acquisition module: Input the pre - experiment block image and the post - experiment block image into the initial convolution layer module to obtain the pre - experiment initial convolution feature and the post - experiment initial convolution feature;

[0095] Deepening convolution feature acquisition module: Input the pre - experiment initial convolution feature and the post - experiment initial convolution feature into the multi - level repeated convolution module to obtain the pre - experiment deepening convolution feature and the post - experiment deepening convolution feature;

[0096] Advanced convolution feature acquisition module: Input the pre - experiment deepening convolution feature and the post - experiment deepening convolution feature into the feature integration module to obtain the pre - experiment advanced convolution feature and the post - experiment advanced convolution feature;

[0097] Initial shallow - layer feature acquisition module: Based on the features of the pre - experiment image and the post - experiment image obtained, perform feature fusion on them respectively, gradually integrating deeper - layer features into the shallow - layer features to obtain the pre - experiment initial shallow - layer feature and the post - experiment initial shallow - layer feature;

[0098] Feature vector acquisition module: Reduce the feature dimensions of the pre - experiment initial shallow - layer feature and the post - experiment initial shallow - layer feature to obtain the pre - experiment feature vector and the post - experiment feature vector;

[0099] Similarity calculation module: Calculate the similarity based on the pre - experiment feature vector and the post - experiment feature vector. If the two are similar, the concrete block is the same concrete block before and after the experiment.

[0100] To achieve the above object, the present invention provides an electronic device, including a processor, a memory, and a computer program stored on the memory and executable on the processor. When the computer program is executed by the processor, it implements the above - mentioned method for anti - counterfeiting detection of concrete blocks based on the CoAtNet network.

[0101] To achieve the above object, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it realizes the above-mentioned anti-counterfeiting detection method for concrete test blocks based on the CoAtNet network.

[0102] Based on this, the beneficial effects of the present invention are as follows: By designing a concrete test block registration system, collecting high-resolution images, and constructing a full-surface feature database of the test blocks, an improved CoAtNet model is proposed to address the problem of detail loss caused by traditional model size scaling. An innovative bottom-up semantic information fusion mechanism is introduced to dynamically combine the deep global structure with the shallow texture features;

[0103] It can comprehensively capture the surface texture, edge deformation, and microscopic details of the test blocks, effectively avoid missing key features, provide a high-integrity data basis for subsequent comparative analysis, and at the same time achieve fully automated operation, reduce labor costs, and improve supervision efficiency. The model does not require manual preset labels and directly generates a comparison benchmark through multi-angle image features. It has a high accuracy and less calculation time in high-resolution image classification tasks, significantly improving the ability to capture features such as fine cracks and particle distribution on the surface of the test blocks, and solving the misjudgment problem caused by preprocessing distortion in the prior art. In addition, the system is compatible with different specifications of test blocks and complex environmental lighting conditions, has strong generalization ability, and provides a technical support with reliability, efficiency, and universality for concrete quality supervision. BRIEF DESCRIPTION OF THE DRAWINGS

[0104] Figure 1 is a flowchart of an anti-counterfeiting detection method for concrete test blocks based on the CoAtNet network shown according to an exemplary embodiment;

[0105] Figure 2 is a module structure diagram of an anti-counterfeiting detection method for concrete test blocks based on the CoAtNet network shown according to an exemplary embodiment;

[0106] Figure 3 is a flowchart of an anti-counterfeiting detection system for concrete test blocks based on the CoAtNet network shown according to an exemplary embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0107] Now, the content of the present invention will be described with reference to exemplary embodiments. It should be understood that the described embodiments are only for enabling those of ordinary skill in the art to better understand and thus implement the content of the present invention, rather than implying any limitation to the scope of the present invention.

[0108] As used herein, the term "comprising" and its variants are to be construed as open-ended terms meaning "including but not limited to". The term "based on" is to be construed as "at least partially based on", and the terms "one embodiment" and "an embodiment" are to be construed as "at least one embodiment".

[0109] According to one embodiment of the present invention, Figure 1 is a flowchart of a method for anti-counterfeiting detection of concrete test blocks based on the CoAtNet network shown according to an exemplary embodiment, as Figure 1 shown. To achieve the above object, a method for anti-counterfeiting detection of concrete test blocks based on the CoAtNet network provided by the present invention includes:

[0110] Obtain concrete test blocks, collect images of the concrete test blocks before and after the test, and obtain the pre-test block image and the post-test block image;

[0111] Figure 2 is a module structure diagram of a method for anti-counterfeiting detection of concrete test blocks based on the CoAtNet network shown according to an exemplary embodiment, as Figure 2 shown. Input the pre-test block image and the post-test block image into the initial convolutional layer module to obtain the pre-test initial convolutional features and the post-test initial convolutional features;

[0112] Input the pre-test initial convolutional features and the post-test initial convolutional features into the multi-level repeated convolutional module to obtain the pre-test deepened convolutional features and the post-test deepened convolutional features;

[0113] Input the pre-test deepened convolutional features and the post-test deepened convolutional features into the feature integration module to obtain the pre-test high-level convolutional features and the post-test high-level convolutional features;

[0114] Based on the features of the pre-test image and the post-test image obtained, perform feature fusion on them respectively, gradually fusing deeper features into shallower features to obtain the pre-test initial shallow features and the post-test initial shallow features;

[0115] Reduce the feature dimensions of the pre-test initial shallow features and the post-test initial shallow features to obtain the pre-test feature vector and the post-test feature vector;

[0116] Based on the pre-test feature vector and the post-test feature vector, perform similarity calculation. If the two are similar, the concrete test block is the same concrete test block before and after the test.

[0117] According to one embodiment of the present invention, the method for obtaining the pre-test block image and the post-test block image is:

[0118] Obtain concrete test blocks, conduct the first image acquisition on the current concrete test blocks to obtain the first image of the concrete test blocks;

[0119] Conduct experiments on the current concrete test blocks, conduct the second image acquisition on the concrete test blocks after the experiments to obtain the second image of the concrete test blocks;

[0120] The pixel sizes of the first image of the concrete test blocks and the second image of the concrete test blocks are 4000×3000. To meet the input requirements of the model, preprocess the acquired first image of the concrete test blocks and the second image of the concrete test blocks;

[0121] Perform central cropping on the first image of the concrete test blocks and the second image of the concrete test blocks to remove unnecessary background information, and scale the first image of the concrete test blocks and the second image of the concrete test blocks to a pixel resolution size of 448×448. This size can greatly reduce the impact of the compression size on the model performance;

[0122] Enhance the first image of the concrete test blocks and the second image of the concrete test blocks based on the application of data augmentation techniques, such as random rotation, translation, and color adjustment, to increase the diversity of the first image of the concrete test blocks and the second image of the concrete test blocks and simulate the situations encountered in the actual application scenarios;

[0123] Normalize the first image of the concrete test blocks and the second image of the concrete test blocks to have the same mean and standard deviation to ensure the consistency of model training, and obtain the pre-experiment test block images and the post-experiment test block images.

[0124] According to an embodiment of the present invention, the method for obtaining the pre-experiment initial convolution features and the post-experiment initial convolution features:

[0125] The initial convolution layer module includes a first convolution module and a second convolution module, and the first convolution module and the second convolution module are 2 3×3 convolution kernels;

[0126] Input the pre-experiment test block images and the post-experiment test block images into the first convolution module to obtain the pre-experiment first convolution module features and the post-experiment first convolution module features, and for the first time reduce the image size to 224×224, where the formula is,

[0127] ;

[0128] ;

[0129] Among them, represents the pre-experiment test block image;

[0130] represents the post-experiment test block image;

[0131] Represents a 3×3 convolutional kernel;

[0132] Represents the first convolutional module;

[0133] Represents the features of the first convolutional module before the experiment;

[0134] Represents the features of the first convolutional module after the experiment;

[0135] Input the features of the first convolutional module before the experiment and the features of the second convolutional module after the experiment into the second convolutional module to obtain the initial convolutional features before the experiment and the initial convolutional features after the experiment, reducing the image size to 112×112, where the formula is,

[0136] ;

[0137] ;

[0138] Among them, Represents the second convolutional module;

[0139] Represents the initial convolutional features before the experiment;

[0140] Represents the initial convolutional features after the experiment.

[0141] According to an embodiment of the present invention, the method for obtaining the deepened convolutional features before the experiment and the deepened convolutional features after the experiment is:

[0142] The multi-level repeated convolutional module includes a first repeated convolutional module and a second repeated convolutional module. The first repeated convolutional module and the second repeated convolutional module are 2 1×1 convolutional kernels and 1 4×4 depthwise separable convolution, gradually reducing the image size and deepening the features;

[0143] Input the initial convolutional features before the experiment and the initial convolutional features after the experiment into the first repeated convolutional module to obtain the features of the first repeated convolutional module before the experiment and the features of the first repeated convolutional module after the experiment, reducing the image size to 56×56, where the formula is,

[0144] ;

[0145] ;

[0146] Among them, Represents the first repeated convolutional module;

[0147] Represents a 1×1 convolutional kernel;

[0148] Represents depthwise separable convolution with a convolution kernel of 4×4;

[0149] Represents the features of the first repeated convolution module before the experiment;

[0150] Represents the features of the first repeated convolution module after the experiment;

[0151] Input the features of the first repeated convolution module before the experiment and the features of the first repeated convolution module after the experiment into the second repeated convolution module to obtain the deepened convolution features before the experiment and the deepened convolution features after the experiment, reducing the image size to 28×28. The formula is as follows:

[0152] ;

[0153] ;

[0154] Among them, Represents the second repeated convolution module;

[0155] Represents the deepened convolution features before the experiment;

[0156] Represents the deepened convolution features after the experiment.

[0157] According to an embodiment of the present invention, the method for obtaining the high-level convolution features before the experiment and the high-level convolution features after the experiment is as follows:

[0158] After the deep processing is completed, the model enters a more complex high-level feature integration stage. The key in this stage is to use a repeated correlation attention mechanism and a feed-forward neural network to further refine and fuse features, strengthening the model's ability to learn the relationships between features of different scales. The feature integration module includes a first integration module and a second integration module. The first integration module and the second integration module are repeated correlation attention mechanisms and feed-forward neural networks;

[0159] Input the deepened convolution features before the experiment and the deepened convolution features after the experiment into the first integration module to obtain the features of the first integration module before the experiment and the features of the first integration module after the experiment, reducing the image size to 14×14. The formula is as follows:

[0160] ;

[0161] ;

[0162] Among them, Represents the first integration module;

[0163] FFN represents a feed-forward neural network;

[0164] Represents the repeated associated attention mechanism;

[0165] Represents the features of the first integration module before the experiment;

[0166] Represents the features of the first integration module after the experiment;

[0167] Input the features of the first integration module before the experiment and the features of the first integration module after the experiment into the second integration module to obtain the high-level convolutional features before the experiment and the high-level convolutional features after the experiment, reducing the image size to 7×7, where the formula is,

[0168] ;

[0169] ;

[0170] Among them, Represents the second integration module;

[0171] Represents the high-level convolutional features before the experiment;

[0172] Represents the high-level convolutional features after the experiment.

[0173] According to an embodiment of the present invention, the method for obtaining the initial shallow features before the experiment and the initial shallow features after the experiment is:

[0174] Store the features of the images before the experiment and the features of the images after the experiment extracted from each layer in the feature list before the experiment and the feature list after the experiment. These features of different levels of feature maps have a progressive change from high-resolution shallow information to low-resolution deep semantic information;

[0175] Optimize through the concept of the Feature Pyramid Network (FPN) to make it more effectively fuse features of different scales and improve the performance in similarity calculation for concrete specimen images. Based on the FPN strategy, gradually fuse deeper features into shallow features to obtain the initial shallow features before the experiment and the initial shallow features after the experiment, where the formula is,

[0176] ;

[0177] ;

[0178] Among them, Represents feature fusion;

[0179] Represents the feature list before the experiment;

[0180] Represents the feature list after the experiment;

[0181] Represents the initial shallow features before the experiment;

[0182] Represents the initial shallow features after the experiment.

[0183] This process ensures that the deep semantic information is effectively incorporated into the shallow features, thereby ensuring that information at different scales is fully utilized. The initial shallow features before the experiment and the initial shallow features after the experiment of the final shallow features will contain rich details and semantic information, which are ideal feature representations for image comparison;

[0184] According to an embodiment of the present invention, the method for obtaining the feature vector before the experiment and the feature vector after the experiment is as follows:

[0185] Apply adaptive average pooling to the initial shallow features before the experiment and the initial shallow features after the experiment to obtain the compressed features before the experiment and the compressed features after the experiment, where the formula is,

[0186] ;

[0187] ;

[0188] Where, Represents feature fusion;

[0189] Represents the feature list before the experiment;

[0190] Represents the feature list after the experiment;

[0191] Apply the fully connected layer to the compressed features before the experiment and the compressed features after the experiment to obtain the feature vector before the experiment and the feature vector after the experiment, where the formula is,

[0192] ;

[0193] ;

[0194] Where, Represents the fully connected layer;

[0195] Represents the feature vector before the experiment;

[0196] Represents the feature vector after the experiment.

[0197] According to an embodiment of the present invention, using the obtained pre-experiment feature vectors and post-experiment feature vectors, it will be trained by a cosine similarity loss function to ensure that similar images are closer in the feature space, while dissimilar images are farther apart. This loss function not only optimizes the expression ability of the features but also ensures the accuracy and robustness in practical applications.

[0198] According to an embodiment of the present invention, based on the similarity score, a decision is made whether to accept or reject the authenticity of the concrete specimen. If the similarity exceeds a preset threshold (for example = 0.85), it is considered that the two concrete specimens have the same origin; otherwise, they are considered to have different origins. It is formally expressed as

[0199] ;

[0200] where represents the decision;

[0201] represents acceptance;

[0202] represents rejection;

[0203] represents the similarity;

[0204] represents the threshold.

[0205] According to an embodiment of the present invention, for regions where the comparison features do not match, the abnormal regions are further screened through statistical analysis. The detection regions that are too small or significantly do not conform to the feature morphology are filtered out according to the area or shape of the region. After rechecking and confirming that there is no error, it is reported to the system platform to warn of the behavior of replacing the specimen.

[0206] Moreover, to achieve the above invention purpose, the present invention also provides a concrete specimen anti-counterfeiting detection system based on the CoAtNet network. Figure 3 It is a flowchart of a concrete specimen anti-counterfeiting detection system based on the CoAtNet network shown according to an exemplary embodiment. As Figure 3 shown, a concrete specimen anti-counterfeiting detection system based on the CoAtNet network in the present invention includes:

[0207] Specimen image acquisition module: Acquire the concrete specimen, and perform image acquisition on the concrete specimen before the test and after the test to obtain the pre-experiment specimen image and the post-experiment specimen image;

[0208] Initial Convolutional Feature Acquisition Module: Input the pre-experiment specimen image and the post-experiment specimen image into the initial convolutional layer module to obtain the pre-experiment initial convolutional features and the post-experiment initial convolutional features;

[0209] Deepening Convolutional Feature Acquisition Module: Input the pre-experiment initial convolutional features and the post-experiment initial convolutional features into the multi-level repeated convolution module to obtain the pre-experiment deepening convolutional features and the post-experiment deepening convolutional features;

[0210] Advanced Convolutional Feature Acquisition Module: Input the pre-experiment deepening convolutional features and the post-experiment deepening convolutional features into the feature integration module to obtain the pre-experiment advanced convolutional features and the post-experiment advanced convolutional features;

[0211] Initial Shallow Feature Acquisition Module: Based on the features of the pre-experiment image and the post-experiment image obtained, perform feature fusion on them respectively, and gradually fuse deeper features into the shallow features to obtain the pre-experiment initial shallow features and the post-experiment initial shallow features;

[0212] Feature Vector Acquisition Module: Reduce the feature dimensions of the pre-experiment initial shallow features and the post-experiment initial shallow features to obtain the pre-experiment feature vector and the post-experiment feature vector;

[0213] Similarity Calculation Module: Perform similarity calculation based on the pre-experiment feature vector and the post-experiment feature vector. If the two are similar, the concrete specimen before and after the experiment is the same concrete specimen.

[0214] To achieve the above invention purpose, the present invention also provides an electronic device, which includes: a processor, a memory, and a computer program stored on the memory and executable on the processor. When the computer program is executed by the processor, it realizes the above method for anti-counterfeiting detection of concrete specimens based on the CoAtNet network.

[0215] To achieve the above invention purpose, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by the processor, it realizes the above method for anti-counterfeiting detection of concrete specimens based on the CoAtNet network.

[0216] Those of ordinary skill in the art can realize that the modules and algorithm steps described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0217] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the devices and equipment described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0218] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the couplings or direct couplings or communication connections shown or discussed with each other can be through some interfaces. The indirect couplings or communication connections of the devices or modules can be in electrical, mechanical or other forms.

[0219] The modules described as separate components may or may not be physically separated. The components shown as modules may or may not be physical modules, that is, they can be located in one place or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of the embodiments of the present invention.

[0220] In addition, in the embodiments of the present invention, the various functional modules can be integrated in a processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0221] If the function is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or a part of this 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 for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method for sending / receiving energy-saving signals in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, ROM, RAM, magnetic disks, or optical discs that can store program codes.

[0222] The above description is only a preferred embodiment of the present application and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solutions formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) disclosed in the present application that have similar functions.

[0223] It should be understood that the magnitudes of the sequence numbers of the steps in the invention content and embodiments of the present invention do not absolutely mean 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 to the implementation process of the embodiments of the present invention.

Claims

1. A method for anti-counterfeiting detection of concrete test blocks based on the CoAtNet network, characterized in that Including: Obtain concrete test blocks, collect images of the concrete test blocks before and after the test to obtain the pre-test block image and the post-test block image; Input the pre-test block image and the post-test block image into the initial convolutional layer module to obtain the pre-test initial convolutional features and the post-test initial convolutional features; The initial convolutional layer module includes a first convolutional module and a second convolutional module, and the first convolutional module and the second convolutional module are 2 3×3 convolutional kernels; Input the pre-test initial convolutional features and the post-test initial convolutional features into the multi-level repeated convolutional module to obtain the pre-test deepened convolutional features and the post-test deepened convolutional features; The multi-level repeated convolutional module includes a first repeated convolutional module and a second repeated convolutional module, and the first repeated convolutional module and the second repeated convolutional module are 2 1×1 convolutional kernels and 1 4×4 depthwise separable convolution; Input the pre-test deepened convolutional features and the post-test deepened convolutional features into the feature integration module to obtain the pre-test high-level convolutional features and the post-test high-level convolutional features; The feature integration module includes a first integration module and a second integration module, and the first integration module and the second integration module are repeated correlation attention mechanisms and feed-forward neural networks; Input the pre-test deepened convolutional features and the post-test deepened convolutional features into the first integration module to obtain the pre-test first integration module features and the post-test first integration module features, where the formula is, ； ； Among them, represents the first integration module; FFN represents a feed-forward neural network; Represents a repeated associative attention mechanism; Indicates the features of the first integration module before the experiment; Indicates the features of the first integration module after the experiment; Input the pre-test first integration module features and the post-test first integration module features into the second integration module to obtain the pre-test high-level convolutional features and the post-test high-level convolutional features, where the formula is, ； ； Among them, represents the second integration module; Represents the high-level convolutional features before the experiment; Represents the high-level convolutional features after the experiment; Based on the features of the pre-test image and the features of the post-test image obtained, perform feature fusion on them respectively, gradually fuse deeper features into shallower features to obtain the pre-test initial shallow features and the post-test initial shallow features; Store the features of the pre-test image and the features of the post-test image extracted in each layer in the pre-test feature list and the post-test feature list; Based on the FPN strategy, gradually fuse deeper features into shallower features to obtain the pre-test initial shallow features and the post-test initial shallow features, where the formula is, ； ； Among them, represents feature fusion; Indicates the feature list before the experiment; Indicates the list of features after the experiment; Represents the initial shallow features before the experiment; Indicates the initial shallow features after the experiment; Reduce the feature dimensions of the pre-test initial shallow features and the post-test initial shallow features to obtain the pre-test feature vector and the post-test feature vector; Perform similarity calculation based on the pre-test feature vector and the post-test feature vector. If the two are similar, the concrete test block before and after the test is the same concrete test block.

2. The anti-counterfeiting detection method for concrete test blocks based on the CoAtNet network according to claim 1, wherein The method for obtaining the pre-test block image and the post-test block image is as follows: Obtain concrete test blocks, perform the first image collection on the current concrete test block to obtain the first concrete test block image; Perform an experiment on the current concrete test block, and perform the second image collection on the concrete test block after the experiment to obtain the second concrete test block image; Perform central cropping on the first concrete test block image and the second concrete test block image, and scale the first concrete test block image and the second concrete test block image to a resolution size of 448×448 pixels; Enhance the first image of the concrete specimen and the second image of the concrete specimen based on the application data augmentation technology, increase the diversity of the first image of the concrete specimen and the second image of the concrete specimen, and simulate the situations encountered in the actual application scenario; Standardize the first image of the concrete specimen and the second image of the concrete specimen to obtain the pre-experiment specimen image and the post-experiment specimen image.

3. The anti-counterfeiting detection method for concrete test blocks based on the CoAtNet network according to claim 2, characterized in that, The method for obtaining the pre-experiment initial convolution features and the post-experiment initial convolution features: Input the pre-experiment specimen image and the post-experiment specimen image into the first convolution module to obtain the pre-experiment first convolution module features and the post-experiment first convolution module features, where the formula is, ； ； Among them, represents the image of the test block before the experiment; Indicates the image of the test block after the test; Represents a 3×3 convolutional kernel; denotes the first convolutional module; Indicates the features of the first convolutional module before the experiment; Represents the features of the first convolutional module after the experiment; Input the pre-experiment first convolution module features and the post-experiment second convolution module features into the second convolution module to obtain the pre-experiment initial convolution features and the post-experiment initial convolution features, where the formula is, ； ； Among them, represents the second convolutional module; Indicates the initial convolutional features before the experiment; Represents the initial convolutional features after the experiment.

4. The anti-counterfeiting detection method for concrete test blocks based on the CoAtNet network according to claim 3, wherein, The method for obtaining the pre-experiment deepened convolution features and the post-experiment deepened convolution features is: Input the pre-experiment initial convolution features and the post-experiment initial convolution features into the first repeated convolution module to obtain the pre-experiment first repeated convolution module features and the post-experiment first repeated convolution module features, where the formula is, ； ； Among them, represents the first repeated convolution module; Represents a 1×1 convolutional kernel; Indicates depthwise separable convolution with a convolution kernel of 4×4; Indicates the features of the first repeated convolution module before the experiment; Indicates the features of the first repeated convolutional module after the experiment; Input the pre-experiment first repeated convolution module features and the post-experiment first repeated convolution module features into the second repeated convolution module to obtain the pre-experiment deepened convolution features and the post-experiment deepened convolution features, where the formula is, ； ； Among them, represents a second repeated convolution module; Indicates the deep convolutional features before the experiment; Indicates the deep convolutional features after the experiment.

5. The anti-counterfeiting detection method for concrete test blocks based on the CoAtNet network according to claim 4, wherein The method for obtaining the pre-experiment feature vector and the post-experiment feature vector is: Apply adaptive average pooling to the pre-experiment initial shallow features and the post-experiment initial shallow features to obtain the pre-experiment compressed features and the post-experiment compressed features, where the formula is, ； ； Among them, represents feature fusion; Indicates the feature list before the experiment; Indicates the list of features after the experiment; Apply the fully connected layer to the pre-experiment compressed features and the post-experiment compressed features to obtain the pre-experiment feature vector and the post-experiment feature vector, where the formula is, ； ； Among them, represents a fully connected layer; Indicates the feature vector before the experiment; Represents the feature vector after the test.

6. A concrete test block anti-counterfeiting detection system based on the CoAtNet network, characterized in that, Include: Specimen image acquisition module: Acquire the concrete specimen, and perform image acquisition on the concrete specimen before the test and the concrete specimen after the test to obtain the pre-experiment specimen image and the post-experiment specimen image; Initial convolution feature acquisition module: Input the pre-experiment specimen image and the post-experiment specimen image into the initial convolution layer module to obtain the pre-experiment initial convolution features and the post-experiment initial convolution features; Deepened convolution feature acquisition module: Input the pre-experiment initial convolution features and the post-experiment initial convolution features into the multi-level repeated convolution module to obtain the pre-experiment deepened convolution features and the post-experiment deepened convolution features; Advanced convolution feature acquisition module: Input the pre-experiment deepened convolution features and the post-experiment deepened convolution features into the feature integration module to obtain the pre-experiment advanced convolution features and the post-experiment advanced convolution features; Initial shallow feature acquisition module: Based on the features of the pre-experiment image and the post-experiment image obtained, perform feature fusion on them respectively, and gradually fuse deeper features into the shallow features to obtain the pre-experiment initial shallow features and the post-experiment initial shallow features; Feature vector acquisition module: Reduce the feature dimension of the pre-experiment initial shallow features and the post-experiment initial shallow features to obtain the pre-experiment feature vector and the post-experiment feature vector; Similarity calculation module: Calculate the similarity based on the pre-experiment feature vector and the post-experiment feature vector. If the two are similar, the concrete specimen before and after the experiment is the same concrete specimen.

7. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored on the memory and executable on the processor. When the computer program is executed by the processor, it implements a method for anti-counterfeiting detection of concrete specimens based on the CoAtNet network as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium. When the computer program is executed by the processor, it implements a method for anti-counterfeiting detection of concrete specimens based on the CoAtNet network as described in any one of claims 1 to 5.

Images