Concrete surface feature-based vibration degree determination method and related product
Patent Information
- Application Number
- CN202510930832.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-07
- Publication Date
- 2026-08-11
- Estimated Expiration
- 2045-07-07
AI Technical Summary
[0004]本发明提供一种基于混凝土表面特征的振捣程度判定方法及相关产品,用以解决现有技术中振捣完成度受人为主观因素影响而准确度不高的缺陷,实现对混凝土振捣不充分区域的准确判定及定位
[0017]本发明提供的基于混凝土表面特征的振捣程度判定方法及相关产品,通过振捣面网格化的局部图像以及振捣程度判定模型,识别局部图像中各网格的振捣程度,基于振捣面全局图像与局部图像的映射关系,将网格的振捣程度映射于所述全局图像中,实现振捣程度在全局图像中的定位。整个振捣面振捣程度的判定过程中无人工介入,避免了主观因素对准确性的影响,保障了振捣程度判定的客观准确性。
Smart Images

Figure CN121073866B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of concrete vibration technology, and in particular to a method for determining the degree of vibration based on the surface characteristics of concrete and related products. Background Technology
[0002] Concrete vibration is a crucial step in the concrete construction process, playing a vital role in the quality and performance of building structures. When concrete is thoroughly vibrated, a large amount of air is effectively expelled, resulting in a dense and well-bonded structure. This process significantly improves the strength and durability of the concrete, providing a solid guarantee for the safety and long-term use of the building structure.
[0003] Current concrete vibration processes primarily rely on worker experience to judge the degree of vibration, such as by observing air bubbles on the concrete surface or listening to the sound of the vibrator. These methods are often influenced by subjective factors and lack accuracy. While some vibrators are equipped with sensors, these are mainly used to monitor the worker's grip position or insertion depth, and cannot effectively determine the degree of vibration completion. Therefore, there is an urgent need for an objective and accurate automatic method for judging the degree of vibration completion on the vibrated surface to guide supplementary vibration work in areas with insufficient vibration. Summary of the Invention
[0004] This invention provides a method and related products for determining the degree of vibration based on the surface characteristics of concrete, in order to solve the defects of the existing technology where the degree of vibration completion is affected by subjective human factors and the accuracy is not high, and to achieve accurate determination and location of areas where concrete vibration is insufficient.
[0005] This invention provides a method for determining the degree of vibration based on the surface characteristics of concrete, comprising the following steps.
[0006] Acquire a global image and multiple local images of the concrete vibration surface; the multiple local images, when stitched together, can cover the entire vibration surface; The global image and the local image are registered and aligned to obtain the mapping relationship between the global image and the local image; The local image is divided into grids to obtain a gridded image; The degree of vibration in each grid of the gridded image is determined based on the vibration degree determination model; the vibration degree determination model is trained based on sample images with aggregate labels. Based on the mapping relationship, the degree of vibration of the grid is mapped onto the global image.
[0007] According to the present invention, a method for determining the degree of vibration based on concrete surface characteristics, based on a vibration degree determination model, determines the degree of vibration of each grid in the gridded image, including: The gridded image is input into the vibration degree determination model, and the aggregate area ratio of each grid is output. The degree of vibration for each grid is determined based on the aggregate ratio of each grid.
[0008] According to the present invention, a method for determining the degree of vibration based on concrete surface characteristics, based on a vibration degree determination model, determines the degree of vibration of each grid in the gridded image, including: The gridded image is input into the vibration degree determination model, and the vibration degree of each grid is output.
[0009] According to the present invention, a method for determining the degree of vibration based on concrete surface features, before performing mesh division on the local image, further includes: The local image is processed by vibration region segmentation, illumination compensation, geometric correction, and edge sharpening.
[0010] According to the present invention, a method for determining the degree of vibration based on concrete surface features is provided. The vibration degree determination model includes a backbone feature extraction network and a feature fusion module. The backbone feature extraction network extracts aggregate features from the gridded image, and the feature fusion module fuses the features at various scales extracted by the backbone feature extraction network.
[0011] According to the present invention, a method for determining the degree of vibration based on concrete surface features is provided. The backbone network of the vibration degree determination model is a hybrid framework of ConvNeXt and Swin Transformer. ConvNeXt and Swin Transformer respectively extract features from the gridded image, and the corresponding scale feature maps of ConvNeXt and Swin Transformer are fused through the CrossAttentionFusion module to obtain a multi-scale fused feature map.
[0012] According to the present invention, a method for determining the degree of vibration based on concrete surface features is provided. The feature fusion module of the vibration degree determination model includes a feature pyramid sub-module, a void space pyramid pooling sub-module, and a cross-layer attention mechanism. Specifically, the feature pyramid submodule performs feature fusion on the multi-scale fusion feature map of the backbone network; the hollow spatial pyramid pooling submodule performs multi-scale feature fusion on the output of the feature pyramid submodule; and the cross-layer attention mechanism dynamically adjusts the weights of each channel and spatial position in the multi-scale feature map.
[0013] The present invention also provides a device for determining the degree of vibration based on the surface characteristics of concrete, comprising the following modules: The image acquisition module is used to acquire a global image and multiple local images of the concrete vibration surface; the multiple local images can be stitched together to cover the entire vibration surface. An image registration module is used to register and align the global image and the local image to obtain the mapping relationship between the global image and the local image. A grid division module is used to divide the local image into grids to obtain a gridded image; The vibration degree determination module is used to determine the vibration degree of each grid in the gridded image based on the vibration degree determination model; the vibration degree determination model is trained based on sample images with aggregate labels. The under-vibration location module is used to map the vibration degree of the grid onto the global image based on the mapping relationship.
[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the vibration degree determination method based on concrete surface characteristics as described above.
[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the vibration degree determination method based on concrete surface characteristics as described above.
[0016] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the vibration degree determination method based on concrete surface characteristics as described above.
[0017] The present invention provides a method and related products for determining the degree of vibration of concrete based on surface features. By using a gridded local image of the vibration surface and a vibration degree determination model, the method identifies the degree of vibration of each grid in the local image. Based on the mapping relationship between the global image and the local image of the vibration surface, the degree of vibration of each grid is mapped onto the global image, thus achieving the localization of the degree of vibration in the global image. The entire process of determining the degree of vibration of the vibration surface is conducted without human intervention, avoiding the influence of subjective factors on accuracy and ensuring the objective accuracy of the determination. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in this 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0019] Figure 1 This is a flowchart illustrating the method for determining the degree of vibration based on the surface characteristics of concrete provided in this embodiment of the invention.
[0020] Figure 2 This is a structural framework diagram of the vibration degree determination model provided in the embodiment of the present invention.
[0021] Figure 3 This is a schematic diagram of the structure of the vibration degree determination device based on concrete surface characteristics provided in an embodiment of the present invention.
[0022] Figure 4 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation
[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0024] The method for determining the degree of vibration based on concrete surface characteristics provided in this invention is mainly for flat or floor slab components. As is well known, during the vibration process of poured concrete components, it is necessary to determine whether the vibration is sufficient. Existing methods rely on construction workers' observation of the vibration surface and their experience to judge whether the vibration is sufficient. This introduces uncertainties due to subjective factors; for example, different construction workers may give different judgments, and even the same construction worker may give different judgments under different individual conditions. To solve this technical problem, this invention provides a method for objectively determining whether the vibration surface is sufficiently vibrated.
[0025] The following is combined Figures 1-2 This invention describes a method for determining the degree of vibration based on the surface characteristics of concrete, as provided in an embodiment of the invention.
[0026] Figure 1 This is a flowchart illustrating the method for determining the degree of vibration based on concrete surface characteristics provided in this embodiment of the invention. Figure 1 As shown, the method includes the following: Step 101: Obtain a global image and multiple local images of the concrete vibration surface; the multiple local images, when stitched together, can cover the entire vibration surface.
[0027] Step 102: Register and align the global image and the local image to obtain the mapping relationship between the global image and the local image.
[0028] Step 103: Divide the local image into a grid to obtain a gridded image.
[0029] Step 104: Determine the degree of vibration of each grid in the gridded image based on the vibration degree determination model; the vibration degree determination model is trained based on sample images with aggregate labels.
[0030] Step 105: Based on the mapping relationship, map the vibration degree of the grid onto the global image. There is no order requirement between steps 102 and steps 103 and 104; step 102 only needs to be completed before step 105.
[0031] This embodiment determines whether vibration is sufficient based on images of the vibrating surface. Specifically, it identifies the vibration level of each grid region in the local image using a gridded local image of the vibrating surface and a vibration level determination model. By mapping the global image and local images of the vibrating surface, it maps or locates the vibration level of each grid region in the global image. This embodiment of the invention uses the above method to determine the vibration level of each region of the vibrating surface, allowing construction personnel to perform supplementary vibration operations on insufficiently vibrated areas. The entire determination and location process is conducted without human intervention, avoiding the influence of subjective factors on accuracy and ensuring the accuracy of the vibration level determination.
[0032] It should be noted that cameras can be used to acquire global and local images of the vibratory compaction surface via video or high-speed photography. Specifically, multiple cameras can be installed at high locations on the construction site, such as tower cranes or building tops. Some cameras (close-up cameras) use a close-up view to capture detailed features of the vibratory compaction surface, while others (far-view cameras) use a far-view view to acquire global information about the entire surface. For example, cameras installed on tower cranes or adjacent tall buildings can acquire global images of the surface, while cameras mounted on supports near the surface can capture local details. Of course, in other embodiments, the far-view cameras for acquiring global images and the close-up cameras for acquiring local images can have other installation methods or locations, as long as global and local image acquisition of the vibratory compaction surface is achieved; this invention does not impose any limitations on this. Furthermore, the close-up cameras can be mounted on an electric pan-tilt head, and the camera angle can be changed by rotating the pan-tilt head to capture images of different areas.
[0033] In some embodiments of the present invention, a threshold for the proportion of aggregate area is set (e.g., 0%-10%). When the proportion of aggregate area in the grid exceeds this threshold, an alarm or adjustment suggestion is triggered. Additionally, modules such as a visual dashboard can be used to display real-time heat maps of the vibration area (aggregate area proportion distribution), send SMS alerts for abnormalities, recommend adjustments to vibration parameters, interface with the BIM system, mark areas with substandard vibration quality, generate digital construction logs, and associate with acceptance criteria, among other functions.
[0034] In practical applications, after vibration is completed and construction personnel have withdrawn from the vibratory area, the camera can capture local and global images of the vibrated surface. The vibration degree determination model identifies the vibration degree of each grid area in the local image and, based on the mapping relationship between the local and global images, locates areas that are not sufficiently vibrated. Based on this location information, construction personnel can then perform corresponding supplementary vibration operations.
[0035] In some embodiments of the present invention, before step 103 divides the local image into grids or before step 104 determines the degree of vibration of each grid in the gridded image based on the vibration degree determination model, a preprocessing step of the image is also included.
[0036] (1) Image segmentation Image segmentation can be performed to distinguish the vibrating surface area from other areas based on factors such as grayscale, color, or texture. Specifically, threshold-based segmentation methods can be used, which calculate one or more grayscale thresholds based on the image's grayscale features, compare the grayscale value of each pixel in the image with the threshold, and finally classify the pixels into the appropriate category based on the comparison result. Alternatively, edge-based segmentation methods can be used, such as edge detection based on grayscale values, to identify the set of consecutive pixels (i.e., edges) on the boundary lines between two different regions in the image.
[0037] The above are just two simple examples of image segmentation methods. Other existing image segmentation methods can also be used to distinguish the vibrating surface area from other areas. This invention does not limit this.
[0038] (2) Illumination compensation Illumination compensation is applied to the image to eliminate the impact of uneven illumination on feature extraction. Specifically, illumination compensation methods such as global illumination compensation and local illumination compensation can be used to compensate for the image's illumination.
[0039] Global illumination compensation adjusts image contrast through histogram equalization, resulting in a more uniform brightness distribution. Local illumination compensation improves the illumination performance of local areas through adaptive histogram equalization. Illumination-invariant features are used for illumination compensation; this method reduces the impact of illumination variations by converting the RGB image to different color spaces. Since illumination compensation is an existing method, it will not be described further in this invention.
[0040] (3) Geometric correction Geometric correction is performed on the image to eliminate image distortion caused by differences in camera viewing angles. Specifically, the intrinsic parameters of the lens (focal length, principal point, distortion coefficient) are obtained using a checkerboard method or other calibration methods. These intrinsic parameters are then applied to the image for distortion correction, ensuring that images taken from different perspectives can be compared and analyzed within the same coordinate system. Since illumination compensation is a pre-existing method, it will not be elaborated upon further in this invention.
[0041] (4) Sharpening Image sharpening can be performed using sharpening filters (such as Laplacian or high-pass filters) to enhance image details and allow for better observation of the characteristics of the concrete compaction surface.
[0042] It should be noted that image preprocessing includes, but is not limited to, one or more of the preprocessing methods mentioned above. Preprocessing steps may be performed only on a local part of the image, or on both a local and global part of the image.
[0043] In some embodiments of the present invention, the preprocessing of the local image further includes image fusion with the global image. Specifically, the fusion of the local image and the global image can be achieved by using a pixel-weighted average.
[0044] In some embodiments of the present invention, step 102 involves registering and aligning the global image and the local image to obtain the mapping relationship between the global image and the local image. Specifically, feature extraction algorithms, such as SIFT (Scale Invariant Feature Transform) and ORB (Oriented Fast and Rotated BRIEF), and matching techniques can be used to determine the correspondence between the near-view image (local image) and the far-view image (global image). Based on the feature points obtained from the matching, a homography matrix is calculated to perform perspective transformation, mapping the detailed image from the near view to the view from the far view.
[0045] It should be noted that the vibration degree determination model in this embodiment can be used only to output the aggregate and its area ratio in the grid, or only to output the vibration degree of each grid, or it can output both the area ratio of the aggregate in the grid and the vibration degree of the grid. It is understood that if the vibration degree determination model is only used to output the aggregate and its area ratio in the grid, then the vibration degree of the grid needs to be calculated based on the area ratio of the aggregate in the grid.
[0046] See Figure 2 The vibration degree determination model mainly consists of three parts: the backbone feature extraction network, the feature fusion module, and the decoder.
[0047] (1) Backbone Feature Extraction Network The backbone feature extraction network adopts a hybrid architecture of ConvNeXt + Swin Transformer, which has the advantage of simultaneously capturing local texture features and global contextual relationships. ConvNeXt enhances the local receptive field through large convolutional kernels and depthwise separable convolutions, effectively extracting detailed features such as image edges and textures. Swin Transformer is based on window self-attention mechanism and shifted window design, dynamically establishing long-distance dependencies, capturing global semantics of the image, and gradually expanding the receptive field through a hierarchical structure (stage partitioning) to adapt to contextual needs of different granularities.
[0048] The gridded image is input into ConvNeXt and Swin Transformer respectively, and ConvNeXt and Swin Transformer perform feature extraction on the input image in parallel.
[0049] The ConvNeXt network is divided into different stages, which correspond to different feature extraction levels and scales. For example, C1 to C5 represent the feature maps output by five different stages from shallow to deep.
[0050] Shallow features (C1 - C2): C1 is the feature map output from the initial stage of the network. It retains more of the original information of the input image, such as local details like edges and textures. These features have relatively small receptive fields and are sensitive to smaller objects or local features in the image. Compared to C1, the feature maps in stage C2 have undergone more layers of convolution operations, resulting in features with a certain degree of abstraction. It can capture a wider range of information than C1, enhancing the semantic information of the features while still retaining more detailed information.
[0051] Intermediate-level features (C3): The C3 feature map is at an intermediate level, integrating detailed information from shallow features with some abstract semantic information. Features at this stage are effective in identifying medium-sized targets or local regions with certain structures, providing rich intermediate representations for subsequent feature processing.
[0052] Deep Features (C4 - C5): The feature maps in the C4 stage have higher abstraction and a further expanded receptive field, enabling them to capture more global information and more complex semantic features in the image. These features are crucial for recognizing large targets or targets with complex structures. C5 is the feature map output from the deepest layer of the network, possessing the highest level of abstraction and the largest receptive field. The C5 feature map primarily focuses on the overall semantic information of the image and plays a key role in high-level decision-making in tasks such as image classification and object detection.
[0053] Similarly, the Swin Transformer network is also divided into multiple feature extraction stages, such as S1-S5, to gradually extract feature information at different levels and scales.
[0054] S1: This stage mainly involves preliminary coarse feature extraction from the input image to obtain some basic, local feature information.
[0055] S2-S4: The features are further refined and enriched through progressive processing via attention enhancement modules, lossless feature transfer modules, and fine feature extraction modules. These stages may introduce different operations, such as window self-attention mechanisms and modal interactions, to continuously mine feature information in the image and improve the expressive power of the features.
[0056] S5: The output features may have integrated information from the previous stages and have a high degree of abstraction and representativeness.
[0057] The aggregate feature maps output by the corresponding layers of ConvNeXt and the Swing Transformer need to maintain the same scale, and feature fusion is performed through the CrossAttentionFusion module. The CrossAttentionFusion module uses an attention mechanism to better combine features from different networks.
[0058] (2) Feature fusion module The feature fusion module includes Multi-Scale Feature Pyramid (FPN), Spatial Pyramid Pooling with Dirt (ASPP), and Cross-Layer Attention Mechanism (CBAM). Among them, FPN is used to fuse multi-layer features, preserving high-resolution details and deep semantic information; ASPP is used for multi-scale context capture, enhancing the model's semantic understanding ability; and CBAM is used to enhance feature selection ability and improve the model's attention mechanism performance.
[0059] The Feature Pyramid Network (FPN) receives feature maps from different levels and fuses them through top-down and lateral connections. Specifically, the FPN takes multi-scale feature maps from ConvNeXt and Swin Transformers as input and fuses information from different levels through its structure: Bottom-up feature extraction, a forward propagation process, involves passing feature maps layer by layer through the convolutional neural network, each layer containing feature information at different scales and levels. During this process, the resolution of the feature maps gradually decreases while the semantic information becomes richer; each layer represents an abstract representation of the input image at different scales. The top-down feature fusion path first upsamples higher-level feature maps (e.g., through bilinear interpolation or deconvolution) to match their resolution to that of lower-level feature maps. Then, the upsampled higher-level feature maps are fused with the corresponding lower-level feature maps. The top-down feature fusion path and the bottom-up feature extraction path are connected by lateral connections. These lateral connections combine the high-resolution information of lower-level feature maps with the rich semantic information of higher-level feature maps to form new feature maps. Lateral connections not only help to convey detailed information from lower-level feature maps, but also enhance the localization capabilities of higher-level feature maps.
[0060] Atrous Spatial Pyramid Pooling (ASPP) is placed after the Feature Mapping Network (FPN). It receives the feature map output from the FPN and continues multi-scale feature processing to further refine the fused features. It captures multi-scale contextual information without reducing the feature map resolution through dilated convolutions with different dilation rates. ASPP's dilated convolutional layers, through different padding and dilation, can acquire receptive fields of different scales, extracting multi-scale information, avoiding the loss of detail caused by downsampling, and helping to capture information about objects of different sizes in a scene, thus improving feature representation capabilities.
[0061] CBAM (Convolutional Block Attention Module) is used to dynamically adjust the weights of each channel and spatial location in the fused feature map. It can be used at any stage to help focus on important features and suppress background noise.
[0062] This invention addresses the multi-scale characteristics of concrete aggregate distribution by utilizing Feature Fusion Network (FPN) and Advanced Feature Processing (ASPP) to capture features at different scales. FPN fuses deep and shallow features through top-down and lateral connections, preserving high-resolution details while utilizing deep semantic information. ASPP, through parallel use of dilated convolutions with different dilation rates, captures multi-scale contextual information without reducing feature map resolution. The distinction between aggregate regions and the background (cement paste) on the concrete surface can be low, especially under uneven lighting or wet conditions. Therefore, CBAM and progressive upsampling are used to balance shallow and deep features. CBAM dynamically adjusts the weights of each channel and spatial location in the feature map through channel attention and spatial attention mechanisms, helping the model focus on key features of the aggregate region and suppress background noise. Shallow features preserve details, while deep features provide semantic information. Finally, the location of the aggregate and / or its proportion in the grid are determined based on the output of the feature fusion module.
[0063] (3) Decoder The decoder adopts a progressive upsampling structure, which gradually restores the resolution by upsampling in multiple small steps, avoiding information loss, detail loss or chessboard artifacts caused by one-time upsampling. While maintaining detailed information, it restores high-resolution features. Combined with skip connections, it can better integrate shallow detailed features and deep semantic features.
[0064] The model for determining the degree of vibration uses a composite loss function during training, as follows: Composite loss function: L = λ1BCE + λ2Dice + λ3SSIM In the formula, BCE is the binary cross-entropy loss, Dice is the Dice loss, SSIM is the structural similarity index loss, and λ1, λ2, and λ3 are the weights of the binary cross-entropy loss, Dice loss, and structural similarity index loss, respectively.
[0065] Specifically, the preferred values for the weights λ1, λ2, and λ3 in the above formula are 0.4, 0.4, and 0.2, respectively.
[0066] Concrete aggregate recognition requires simultaneous attention to pixel-level accuracy and region-level (i.e., grid-level) consistency. Binary Cross-Entropy Loss (BCE) provides pixel-level supervision signals, suitable for binary classification tasks. Dice Loss optimizes the Intersection over Union (IoU), and compared to the cross-entropy loss function, Dice Loss better handles class imbalance (aggregate regions typically occupy a smaller proportion). Structural Similarity Index Loss (SSIM) captures structural information from the image, improving the naturalness of segmentation boundaries. The weights of BCE, Dice, and SSIM are assigned coefficients of 0.4, 0.4, and 0.2, respectively, to balance pixel-level accuracy and region-level consistency. This approach improves the model's segmentation accuracy and boundary quality, reducing false positives and false negatives.
[0067] This invention also enhances edge loss. Since errors in the boundary region can significantly affect the calculation accuracy of the aggregate area ratio of the vibrating surface, a weight of 3 times is applied to the boundary region to force the model to pay more attention to the boundary region. Canny edge detection or morphological operations are used to generate boundary masks to solve the part of the boundary region between aggregate and background that is difficult to separate, thereby improving the segmentation accuracy of the boundary region and reducing boundary blur.
[0068] The AdamW optimizer is chosen to adjust the model parameters based on the magnitude of the loss, ensuring that the predictions in the next iteration are closer to the actual values. Traditional Adam optimizers may result in unsatisfactory weight decay; AdamW decouples weight decay from gradient updates, avoiding overfitting and making it suitable for handling large-scale data and complex models. Preferably, its parameters are set to lr=3e-4 and weight_decay=0.05.
[0069] The CosineAnnealingWarmRestarts scheduler is used to dynamically adjust the learning rate during the training phase. By periodically adjusting the learning rate, CosineAnnealingWarmRestarts helps the model escape local optima, avoiding the possibility that a fixed learning rate might cause the model to get stuck in local optima, thus improving the model's convergence speed and final performance.
[0070] This invention also introduces an early stopping mechanism to avoid the risk of model overfitting. Based on the 10-cycle patience value of the validation set IoU, that is, when the IoU has not improved within 10 consecutive training cycles (e.g., it has not reached a higher IoU value), the early stopping mechanism is triggered to stop the training of the model, automatically save the best model, reduce manual intervention, improve the generalization ability of the model, and avoid overfitting.
[0071] The following describes the concrete vibration degree determination device provided by the present invention. The concrete vibration degree determination device described below can be referred to in correspondence with the vibration degree determination method based on concrete surface characteristics described above.
[0072] Figure 3 This is a schematic diagram of the vibration degree determination device based on concrete surface characteristics provided in an embodiment of the present invention, as shown below. Figure 3 As shown, the vibration degree determination device based on concrete surface characteristics includes: The image acquisition module 301 is used to acquire a global image and multiple local images of the concrete vibration surface; the multiple local images can be stitched together to cover the entire vibration surface.
[0073] The image registration module 302 is used to register and align the global image and the local image to obtain the mapping relationship between the global image and the local image.
[0074] The grid division module 303 is used to divide the local image into grids to obtain a gridded image.
[0075] The vibration degree determination module 304 is used to determine the vibration degree of each grid in the gridded image based on the vibration degree determination model; the vibration degree determination model is trained based on sample images with aggregate labels.
[0076] The vibration degree mapping module 305 is used to map the vibration degree of the grid onto the global image based on the mapping relationship.
[0077] In some embodiments of the present invention, the vibration degree determination module 304 is specifically used to input the gridded image into the vibration degree determination model, output the aggregate area ratio of each grid, and determine the vibration degree of each grid based on the aggregate ratio of each grid.
[0078] In some embodiments of the present invention, the vibration degree determination module 304 is specifically used to input the gridded image into the vibration degree determination model and output the vibration degree of each grid.
[0079] In some embodiments of the present invention, the vibration degree determination device based on concrete surface characteristics further includes: The image preprocessing module is used to perform vibration region segmentation, illumination compensation, geometric correction, and edge sharpening on local images.
[0080] For more specific functions of the module, please refer to the method for determining the degree of vibration based on the characteristics of concrete surface described above, which will not be repeated here.
[0081] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include a processor 410, a communication interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communication interface 420, and the memory 430 communicate with each other through the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute a method for determining the degree of vibration based on concrete surface features. This method includes: acquiring a global image and multiple local images of the concrete vibration surface; the multiple local images, after being stitched together, can cover the entire vibration surface; registering and aligning the global image and the local images to obtain a mapping relationship between the global image and the local images; dividing the local images into grids to obtain a gridded image; determining the degree of vibration of each grid in the gridded image based on a vibration degree determination model; the vibration degree determination model is trained based on sample images with aggregate labels; and mapping the degree of vibration of the grids to the global image based on the mapping relationship. For more specific methods, please refer to the method for determining the degree of vibration based on the characteristics of concrete surface described above, which will not be repeated here.
[0082] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0083] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the vibration degree determination method based on concrete surface features provided by the above methods. This method includes: acquiring a global image and multiple local images of the concrete vibration surface; the multiple local images being stitched together to cover the entire vibration surface; registering and aligning the global image and the local images to obtain a mapping relationship between the global image and the local images; dividing the local images into grids to obtain a gridded image; determining the vibration degree of each grid in the gridded image based on a vibration degree determination model; the vibration degree determination model being trained based on sample images with aggregate labels; and mapping the vibration degree of the grids to the global image based on the mapping relationship. For more specific methods, please refer to the vibration degree determination method based on concrete surface features described above, which will not be repeated here.
[0084] Furthermore, this invention also provides a non-transitory computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for determining the degree of vibration based on concrete surface features provided by the methods described above. This method includes: acquiring a global image and multiple local images of the concrete vibration surface; stitching the multiple local images together to cover the entire vibration surface; registering and aligning the global image and the local images to obtain a mapping relationship between the global image and the local images; dividing the local images into grids to obtain a gridded image; determining the degree of vibration of each grid in the gridded image based on a vibration degree determination model; the vibration degree determination model being trained based on sample images with aggregate labels; and mapping the degree of vibration of the grids to the global image based on the mapping relationship. For more specific methods, refer to the method for determining the degree of vibration based on concrete surface features described above; further details are omitted here.
[0085] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0086] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for determining the degree of vibration based on concrete surface characteristics, characterized in that, include: Acquire a global image and multiple local images of the concrete vibration surface; the multiple local images, when stitched together, can cover the entire vibration surface; The global image and the local image are registered and aligned to obtain the mapping relationship between the global image and the local image; The local image is divided into grids to obtain a gridded image; Based on the vibration degree determination model, the vibration degree of each grid in the gridded image is determined, including: inputting the gridded image into the vibration degree determination model and outputting the aggregate area ratio of each grid; determining the vibration degree of each grid based on the aggregate ratio of each grid; the vibration degree determination model is trained based on sample images with aggregate labels, and includes a backbone feature extraction network and a feature fusion module. The backbone feature extraction network extracts aggregate features from the gridded image, and the feature fusion module fuses the features at various scales extracted by the backbone feature extraction network; the backbone network of the vibration degree determination model is a hybrid framework of ConvNeXt and Swin Transformer; wherein ConvNeXt and Swin Transformer respectively extract features from the gridded image, and ConvNeXt and Swin... The Transformer's scale feature maps are fused using the CrossAttentionFusion module to obtain a multi-scale fused feature map. The feature fusion module of the vibration degree determination model includes a feature pyramid sub-module, a hollow space pyramid pooling sub-module, and a cross-layer attention mechanism. The feature pyramid sub-module performs feature fusion on the multi-scale fused feature map of the backbone network. The hollow space pyramid pooling sub-module performs multi-scale feature fusion on the output of the feature pyramid sub-module. The cross-layer attention mechanism dynamically adjusts the weights of each channel and spatial location in the multi-scale feature map. Based on the mapping relationship, the degree of vibration of the grid is mapped onto the global image.
2. The method for determining the degree of vibration based on concrete surface characteristics according to claim 1, characterized in that, Based on the vibration degree determination model, the vibration degree of each grid in the gridded image is determined, including: The gridded image is input into the vibration degree determination model, and the vibration degree of each grid is output.
3. The method for determining the degree of vibration based on concrete surface characteristics according to claim 1, characterized in that, Before performing grid division on the local image, the method further includes: The local image is processed by vibration region segmentation, illumination compensation, geometric correction, and edge sharpening.
4. A device for determining the degree of vibration based on concrete surface characteristics, comprising: The image acquisition module is used to acquire a global image and multiple local images of the concrete vibration surface; the multiple local images can be stitched together to cover the entire vibration surface. An image registration module is used to register and align the global image and the local image to obtain the mapping relationship between the global image and the local image. A grid division module is used to divide the local image into grids to obtain a gridded image; A vibration degree determination module is used to determine the vibration degree of each grid in the gridded image based on a vibration degree determination model. This includes: inputting the gridded image into the vibration degree determination model and outputting the aggregate area ratio of each grid; determining the vibration degree of each grid based on the aggregate ratio; the vibration degree determination model is trained based on sample images with aggregate labels, and includes a backbone feature extraction network and a feature fusion module. The backbone feature extraction network extracts aggregate features from the gridded image, and the feature fusion module fuses the features extracted by the backbone feature extraction network at various scales; the backbone network of the vibration degree determination model is a hybrid framework of ConvNeXt and Swin Transformer; wherein ConvNeXt and Swin Transformer respectively extract features from the gridded image, and ConvNeXt and Swin... The Transformer's scale feature maps are fused using the CrossAttentionFusion module to obtain a multi-scale fused feature map. The feature fusion module of the vibration degree determination model includes a feature pyramid sub-module, a hollow space pyramid pooling sub-module, and a cross-layer attention mechanism. The feature pyramid sub-module performs feature fusion on the multi-scale fused feature map of the backbone network. The hollow space pyramid pooling sub-module performs multi-scale feature fusion on the output of the feature pyramid sub-module. The cross-layer attention mechanism dynamically adjusts the weights of each channel and spatial location in the multi-scale feature map. The under-vibration location module is used to map the vibration degree of the grid onto the global image based on the mapping relationship.
5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the program, it implements the vibration degree determination method based on concrete surface characteristics as described in any one of claims 1 to 3.
6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for determining the degree of vibration based on the surface characteristics of concrete as described in any one of claims 1 to 3.
Citation Information
Patent Citations
Automatic concrete vibrating method and system based on machine vision
CN114648637A
Concrete vibration quality real-time detection method based on image recognition
CN118262293A
