Lightweight concrete structure crack automatic detection method

Abstract

The invention provides a lightweight concrete structure crack automatic detection method. The lightweight concrete structure crack automatic detection method comprises the following steps: S1, capturing images of a concrete structure under various illumination conditions by using at least one high-resolution camera, S2, preprocessing the captured images to enhance the crack visibility under a low illumination condition, and S3, detecting the crack visibility under the low illumination condition. S3, carrying out the feature extraction of the image through employing a backbone network based on the crack automatic detection network V3 and a Bottleneck Transform; and S4, carrying out the multi-scale fusion of features through employing a feature pyramid network and a path aggregation network. According to the lightweight concrete structure crack automatic detection method, a lightweight crack feature automatic extraction network is established, the network uses a crack automatic detection network V3 and a Transform, the addition of the Transform well enhances the effect of crack feature extraction, and the detection accuracy of the lightweight concrete structure crack is improved. The introduction of a channel attention mechanism makes the network pay more attention to crack characteristics.

Description

Technical Field

[0001] The invention relates to the technical field of automatic detection of cracks in concrete structures, and in particular to an automatic detection method for cracks in lightweight concrete structures. Background Art

[0002] One of the common forms of failure in transportation infrastructure is cracks. After long-term or excessive use of the road surface, most road surface diseases first appear as cracks. If not repaired in time, these cracks will continue to expand and extend due to external environment or human factors. Driving at high speed under such road conditions not only affects driving comfort, but may even cause serious traffic accidents. The continued expansion of tiny cracks may lead to potential dangers. Repairing cracks before they deteriorate can not only significantly reduce maintenance budgets, but also avoid accidents and extend the service life of roads. In order to evaluate the safety and durability of concrete structures, crack detection has become one of the important links in concrete structure maintenance. In traditional crack detection, it relies on manual visual inspection, or with the help of electromagnetic detection, infrared detection, ultrasonic detection or radioactive detection. The manual visual inspection method not only has low detection efficiency and high labor cost, but also has certain geographical limitations. For example, places such as elevated bridge pavement and bridge piers are not suitable for manual inspection. With the continuous improvement of detection methods with the help of external equipment, certain expected detection effects can be achieved, but these detection methods are still accompanied by some subjective factors.

[0003] The current image-based deep learning crack detection method requires a large amount of data to train the algorithm model. Collecting crack data and uploading it to the computing end is a time-consuming process, and cloud computing has a high latency and cannot respond quickly to accidents caused by cracks. Summary of the invention

[0004] 1. Technical issues to be solved

[0005] In view of the shortcomings of the prior art, the present invention provides a method for automatic crack detection in lightweight concrete structures, which solves the problem that collecting crack data and then uploading it to the computing end is a time-consuming process, cloud computing has high latency and cannot respond quickly to accidents caused by cracks, and crack features can still be extracted under poor lighting conditions.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: 1. A lightweight concrete structure crack automatic detection method, characterized in that it includes:

[0008] S1. capturing images of the concrete structure under various lighting conditions using at least one high-resolution camera;

[0009] S2, preprocessing the captured images to enhance crack visibility under low light conditions;

[0010] S3. Image feature extraction is performed using the backbone network based on the crack automatic detection network V3 and Bottleneck Transformer. By adopting the V3 network and Bottleneck Transformer structure, this solution achieves lightweight and efficient feature extraction, so that crack features can be better preserved in the deep network structure. The self-attention mechanism of Bottleneck Transformer improves the accuracy of feature extraction and reduces the interference of redundant features.

[0011] S4. Feature pyramid network and path aggregation network are used for multi-scale fusion of features. A parameter-free attention mechanism, SimAM, is inserted into the Neck network. After ablation experiments, it is proved that this module has a certain improvement on feature fusion. Finally, the fused feature map is transmitted to the prediction end to generate many candidate anchor boxes. Non-maximum suppression is used to eliminate redundant prediction boxes. Through threshold setting, non-maximum suppression selects the prediction box with the maximum score and deletes those prediction boxes that have significant overlap with this prediction box. The score value is measured by IoU. In the process of generating candidate boxes, non-maximum suppression is combined with the joint intersection algorithm for precise positioning. This combination effectively eliminates redundant prediction boxes while controlling the accuracy of crack recognition through the IoU threshold, making the final detection result more reliable.

[0012] S5, using non-maximum suppression and joint intersection indicators for accurate crack location and identification;

[0013] S6. Deploy the model on edge computing devices to achieve real-time crack detection.

[0014] Preferably, the image preprocessing step includes automatic exposure adjustment and contrast enhancement, and the feature extraction step includes applying a multi-head attention mechanism to emphasize crack features. The model deployed on the edge computing device has real-time data processing capabilities, and the response delay of crack detection is less than 50ms, ensuring rapid identification and response in harsh environments, thereby achieving more efficient crack detection under field conditions.

[0015] Preferably, the feature fusion step includes using a parameter-free attention mechanism to enhance the expression of key crack features.

[0016] Preferably, the crack location and identification step comprises applying continuous intersection super-union as a location loss function.

[0017] Preferably, the attention mechanism, to enhance the expression of key crack features, is defined as:

[0018]

[0019] Where S is the attention map, xi j is the pixel in the feature map, and f represents the transformation function applied to each pixel.

[0020] Preferably, the feature fusion strategy utilizes a combination of a feature pyramid network and a path aggregation network. The feature pyramid network is responsible for extracting features at different depth levels, and the path aggregation network is responsible for more efficiently fusing and transmitting these features to ensure that information flow from lower layers to higher layers is not hindered, thereby optimizing the entire network's ability to detect cracks.

[0021] Preferably, the model deployed on the edge computing device has real-time data processing capabilities, rapid response and instant update capabilities.

[0022] (III) Beneficial effects

[0023] The present invention provides a lightweight concrete structure crack automatic detection method. It has the following beneficial effects:

[0024] This lightweight concrete structure crack automatic detection method builds a lightweight crack feature automatic extraction network, which uses the crack automatic detection network V3 and Transformer. The addition of Transformer greatly enhances the effect of crack feature extraction. The introduction of the channel attention mechanism makes the network pay more attention to crack features. The path aggregation network is improved using Transformer, which further improves the feature fusion effect. Pictures containing pavement cracks are tested. The test results show that the inference time of a single picture is only 11.3ms. The lightweight crack automatic detection network not only reduces the burden on the computer, but also improves the speed and accuracy of crack detection, verifying the robust performance of the algorithm model. The crack automatic detection model can not only detect pavement cracks under normal conditions, but also automatically detect cracks under poor lighting conditions and rainy weather conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic diagram of the fusion process of the feature pyramid and the path aggregation network of the present invention;

[0026] Figure 2 It is the SimAM attention mechanism. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] like Figure 1-Figure 2 As shown, an embodiment of the present invention provides a method for automatic crack detection of a lightweight concrete structure, comprising: S1, using at least one high-resolution camera to capture images of the concrete structure under a variety of lighting conditions.

[0029] S2. The captured images are preprocessed to enhance the visibility of cracks in low-light conditions. The image preprocessing step includes automatic exposure adjustment and contrast enhancement. The feature extraction step includes applying a multi-head attention mechanism to emphasize crack features.

[0030] S3. The image features are extracted using the backbone network based on the crack automatic detection network V3 and Bott l eneck Transformer. The feature fusion strategy uses the combination of the feature pyramid network and the path aggregation network. The feature pyramid network is responsible for extracting features at different depth levels, and the path aggregation network is responsible for more efficient fusion and transmission of these features to ensure that the information flow from the lower layer to the higher layer is not hindered, thereby optimizing the entire network's ability to detect cracks. In the feature pyramid, since it is a process from a large feature map to a small feature map and then from a small feature map to a large feature map, after multiple layers of transmission, the shallow feature information is seriously lost, so a bottom-up path aggregation network is added after the feature pyramid. The i-th feature map of the path aggregation network is fused with the i+1-th feature map of the feature pyramid to obtain a new fused feature map. This method enhances the feature representation ability. By adopting the V3 network and Bott l eneck Transformer structure, this scheme achieves lightweight and efficient feature extraction, so that crack features can be better retained in the deep network structure. The self-attention mechanism of Bott l eneck Transformer improves the accuracy of feature extraction and reduces the interference of redundant features.

[0031] S4. The feature pyramid network and path aggregation network are used for multi-scale fusion of features. A parameter-free attention mechanism, SimAM, is inserted into the Neck network. After ablation experiments, it is proved that this module has a certain improvement on feature fusion. Finally, the fused feature map is passed to the prediction end to generate many candidate anchor boxes. Non-maximum suppression is used to eliminate redundant prediction boxes. Through threshold setting, non-maximum suppression selects the prediction box with the maximum score, and deletes those prediction boxes that have significant overlap with this prediction box. The score value is measured by I oU. The feature fusion step includes the use of a parameter-free attention mechanism to enhance the expression of key crack features. The attention mechanism is defined as:

[0032]

[0033] Where S is the attention map, xi j is the pixel point in the feature map, and f represents the conversion function applied to each pixel. In the process of generating candidate boxes, non-maximum suppression combined with the joint intersection algorithm is used for precise positioning. This combination effectively eliminates redundant prediction boxes and controls the accuracy of crack recognition through the I oU threshold, making the final detection result more reliable.

[0034] S5. Using non-maximum suppression and union intersection indicators for accurate crack positioning and identification, the crack positioning and identification steps include applying continuous intersection super-union as the positioning loss function.

[0035] S6. Deploy the model on the edge computing device to achieve real-time crack detection. The model deployed on the edge computing device has real-time data processing capabilities, fast response and instant update capabilities. The model deployed on the edge computing device has real-time data processing capabilities, and the response delay of crack detection is less than 50ms, ensuring rapid identification and response in harsh environments, thereby achieving more efficient crack detection under field conditions.

[0036] Training model and validation:

[0037] The model training epoch is set to 600, the batch-size is set to 8, the initial learning rate is set to 0.01, the weight decay is set to 0.0005, and we use the stochastic gradient descent optimization strategy. Since the given initial learning rate is small, the cosine annealing learning rate decay strategy is also used until the network parameters are stable during the training process and then the training is accelerated. As the training process progresses, the model gradually approaches the optimal solution. At this time, the learning rate update is reduced, so that the training results are infinitely close to the optimal solution. As the number of training times increases, the positioning loss and confidence loss are constantly decreasing and gradually stabilizing until the loss value hardly changes and the model converges. The final training positioning loss and confidence loss are close to 0.045 and 0.072. In the field of machine learning and deep learning, epoch refers to the process in which the entire training data set is used completely once during the training process. When the model is trained, it performs forward and backward propagation on the training set to adjust the model weights, which is called a training step or iteration.

[0038] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A lightweight concrete structure crack automatic detection method, characterized in that: include: S1. capturing images of the concrete structure under various lighting conditions using at least one high-resolution camera; S2, preprocessing the captured images to enhance crack visibility under low light conditions; S3, using the backbone network based on the crack automatic detection network V3 and Bottleneck Transformer to extract image features. By adopting the V3 network and Bottleneck Transformer structure, this solution achieves lightweight and efficient feature extraction, so that crack features can be better preserved in the deep network structure. The self-attention mechanism of Bottleneck Transformer improves the accuracy of feature extraction and reduces the interference of redundant features. S4. Feature pyramid network and path aggregation network are used for multi-scale fusion of features. A parameter-free attention mechanism, SimAM, is inserted into the Neck network. After ablation experiments, it is proved that this module has a certain improvement on feature fusion. Finally, the fused feature map is passed to the prediction end to generate many candidate anchor boxes. Non-maximum suppression is used to eliminate redundant prediction boxes. Through threshold setting, non-maximum suppression selects the prediction box with the maximum score and deletes those prediction boxes that have significant overlap with this prediction box. The score value is measured by IoU. In the process of generating candidate boxes, non-maximum suppression combined with joint intersection algorithm is used for precise positioning. This combination effectively eliminates redundant prediction boxes and controls the accuracy of crack recognition through IoU threshold, making the final detection result more reliable. S5, using non-maximum suppression and joint intersection indicators for accurate crack location and identification; S6. Deploy the model on edge computing devices to achieve real-time crack detection. The model deployed on edge computing devices has real-time data processing capabilities, and the response delay of crack detection is less than 50ms, ensuring rapid identification and response in harsh environments, thereby achieving more efficient crack detection under field conditions.

2. The method for automatically detecting cracks in a lightweight concrete structure according to claim 1, characterized in that: The image preprocessing step includes automatic exposure adjustment and contrast enhancement, and the feature extraction step includes applying a multi-head attention mechanism to emphasize crack features.

3. The method for automatically detecting cracks in a lightweight concrete structure according to claim 1, characterized in that: The feature fusion step includes using a parameter-free attention mechanism to enhance the representation of key crack features.

4. The method for automatically detecting cracks in a lightweight concrete structure according to claim 1, characterized in that: The crack localization and identification step includes applying continuous intersection super-union as a localization loss function.

5. The method for automatically detecting cracks in a lightweight concrete structure according to claim 3, characterized in that: The attention mechanism is used to enhance the expression of key crack features and is defined as: Where S is the attention map, xij is the pixel in the feature map, and f represents the transformation function applied to each pixel.

6. The method for automatically detecting cracks in a lightweight concrete structure according to claim 3, characterized in that: The feature fusion strategy utilizes the combination of a feature pyramid network and a path aggregation network. The feature pyramid network is responsible for extracting features at different depth levels, while the path aggregation network is responsible for more efficiently fusing and transmitting these features to ensure that the flow of information from lower layers to higher layers is not hindered, thereby optimizing the entire network's ability to detect cracks.

7. The method for automatically detecting cracks in a lightweight concrete structure according to claim 1, characterized in that: The model deployed on the edge computing device has the ability to process data in real time, respond quickly and update instantly.

Images