Crack identification method and system based on image processing

Abstract

The application provides a crack identification method and system based on image processing, which comprises the following steps: image preprocessing is performed on each image data to obtain preprocessed images; a feature structure is constructed by using a skip connection and a depth separable convolution algorithm; a first identification model and the feature structure are fused based on each preprocessed image to obtain an identification optimization model; each preprocessed image is input into a second identification model for model training to obtain an identification improvement model; the identification optimization model and the identification improvement model are fused to obtain a crack identification model; a preset image acquisition device is used to collect real-time processing images, and the crack identification model is used for image processing on the processing images to identify crack information of the processing images. The identification optimization model and the identification improvement model are fused, and a multi-scale convolution attention mechanism is introduced, so that the extraction and discrimination ability of crack features is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a crack identification method and system based on image processing. BACKGROUND

[0002] In many fields such as building engineering, road and bridge, water conservancy facilities, etc., structural surface cracks are common safety hazards, and timely and accurate identification of cracks is crucial for structural health monitoring and safety evaluation.

[0003] Traditional crack detection mainly relies on manual visual inspection or simple measurement tools, which has the problems of low efficiency, strong subjectivity, high missed detection rate, high danger, etc., and is difficult to meet the rapid detection needs of large-scale infrastructure. In recent years, with the development of image processing and computer vision technology, image-based automatic crack identification methods have gradually been applied. However, the existing technology still has obvious deficiencies:

[0004] 1. Large image quality interference: the collected images are easily affected by uneven lighting, shadows, stains, texture interference, etc., making it difficult to extract crack features and reducing recognition accuracy.

[0005] 2. Weak model generalization ability: most methods rely on a single model structure, which has poor adaptability to cracks of different scales and shapes, especially unstable recognition effect in complex backgrounds.

[0006] 3. Incomplete feature fusion mechanism: traditional models perform poorly in multi-level feature fusion, making it difficult to capture both global context information and local detail features, affecting crack boundary positioning accuracy.

[0007] 4. Lack of dynamic adjustment in training process: the loss weights of each task in the model training process are fixed, and cannot be adjusted adaptively according to the training state, resulting in slow model convergence or overfitting. SUMMARY

[0008] Therefore, the purpose of the present application is to provide a crack identification method and system based on image processing to at least solve the above technical deficiencies.

[0009] The present application provides a crack identification method based on image processing, comprising:

[0010] Obtain a plurality of image data, and perform image preprocessing on each of the image data to obtain corresponding preprocessed images;

[0011] Construct a first identification model, and construct a feature structure using a skip connection and a depth separable convolution algorithm, and fuse the first identification model and the feature structure based on each of the preprocessed images to obtain an identification optimization model;

[0012] construct a second identification model, and input each of the preprocessed images into the second identification model for model training, dynamically adjust the weight according to the change of each task loss in the model training process, and form an improved identification model;

[0013] fuse the identification optimization model and the improved identification model, and introduce a multi-scale convolution attention mechanism in the model fusion process to obtain a crack identification model;

[0014] use a preset image acquisition device to collect real-time images to be processed, and use the crack identification model to process the images to be processed to identify the crack information of the images to be processed.

[0015] Further, the step of obtaining a plurality of image data and pre-processing each of the image data to obtain corresponding preprocessed images comprises:

[0016] each of the image data is subjected to mean filtering, median filtering and Gaussian filtering, and the preliminary images meeting the preset requirements are selected based on the processing results;

[0017] image enhancement is performed on the preliminary images, and residual optimization is performed on the preliminary images after image enhancement using an edge detection algorithm to obtain corresponding preprocessed images.

[0018] Further, the step of constructing a feature structure using a skip connection and a depth separable convolution algorithm, and fusing the first identification model and the feature structure based on the preprocessed images to obtain an identification optimization model comprises:

[0019] a feature structure is constructed using a skip connection and a depth separable convolution algorithm, and the feature structure is optimized using an architecture search algorithm to obtain an optimized feature structure;

[0020] each of the preprocessed images is input into the first identification model to generate low-level features of different sizes, and the low-level features of different sizes are fused, and the optimized feature structure is added in the fusion process to obtain an identification optimization model.

[0021] Further, the step of inputting each of the preprocessed images into the second identification model for model training, and dynamically adjusting the weight according to the change of each task loss in the model training process to form an improved identification model comprises:

[0022] define a target detection loss and a semantic segmentation loss, and dynamically monitor the convergence of the target detection loss and the semantic segmentation loss in the model training process, and adaptively adjust the weight according to the loss ratio to obtain a preliminary improved model;

[0023] The spatial attention mechanism is introduced into the preliminary improved model and fused with the backbone network of the preliminary improved model to obtain an identification improved model.

[0024] Further, the identification optimization model and the identification improved model are fused, and a multi-scale convolution attention mechanism is introduced in the model fusion process to obtain a crack identification model.

[0025] The multi-scale convolution attention mechanism is defined, wherein the multi-scale convolution attention mechanism adopts a plurality of bar-shaped convolution kernels of different sizes.

[0026] The multi-scale convolution attention mechanism is introduced in the model fusion process, so that the input feature map of the model is subjected to convolution operation in the vertical and horizontal axis directions to obtain a crack identification model.

[0027] The application also provides a crack identification system based on image processing, comprising:

[0028] An image preprocessing module is configured to acquire a plurality of image data and perform image preprocessing on each of the image data to obtain corresponding preprocessed images.

[0029] A first model construction module is configured to construct a first identification model, construct a feature structure using a skip connection and a depth separable convolution algorithm, and fuse the first identification model and the feature structure based on each of the preprocessed images to obtain an identification optimization model.

[0030] A second model construction module is configured to construct a second identification model, input each of the preprocessed images into the second identification model for model training, and dynamically adjust the weight according to the change of each task loss in the model training process to form an identification improved model.

[0031] A model fusion module is configured to fuse the identification optimization model and the identification improved model, and introduce a multi-scale convolution attention mechanism in the model fusion process to obtain a crack identification model.

[0032] An image processing module is configured to acquire a to-be-processed image in real time using a preset image acquisition device, and perform image processing on the to-be-processed image using the crack identification model to identify the crack information of the to-be-processed image.

[0033] Further, the image preprocessing module is specifically configured to:

[0034] Each of the image data is subjected to mean filtering processing, median filtering processing and Gaussian filtering processing, and a preliminary image meeting a preset requirement is selected based on the processing results.

[0035] The preliminary image is image-enhanced, and an edge detection algorithm is used to perform residual optimization on the image-enhanced preliminary image to obtain a corresponding preprocessed image.

[0036] Further, the first model construction module is specifically used for:

[0037] A skip connection and a depth separable convolution algorithm are used to construct a feature structure, and an architecture search algorithm is used to optimize the feature structure to obtain a feature optimization structure.

[0038] The preprocessed images are input into the first recognition model to generate low-level features of different sizes, and the low-level features of different sizes are fused, and the feature optimization structure is added in the fusion process to obtain a recognition optimization model.

[0039] Further, the second model construction module is specifically used for:

[0040] A target detection loss and a semantic segmentation loss are defined, and convergence of the target detection loss and the semantic segmentation loss is dynamically monitored during model training, and weights are adaptively adjusted according to loss ratios to obtain a preliminary improved model.

[0041] A spatial attention mechanism is introduced into the preliminary improved model, and the spatial attention mechanism is fused with a backbone network of the preliminary improved model to obtain a recognition improved model.

[0042] Further, the model fusion module is specifically used for:

[0043] A multi-scale convolution attention mechanism is defined, wherein the multi-scale convolution attention mechanism uses a plurality of bar-shaped convolution kernels of different sizes.

[0044] The multi-scale convolution attention mechanism is introduced in the model fusion process to enable input feature maps of the model to be subjected to convolution operations in vertical and horizontal axis directions to obtain a crack recognition model.

[0045] The application further provides a storage medium having a computer program stored thereon, and the program is executed by a processor to implement the crack recognition method based on image processing.

[0046] The application further provides a computer comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor implements the crack recognition method based on image processing when executing the computer program.

[0047] The crack recognition method and system based on image processing in the application effectively suppress noise and enhance crack region features by combining various filtering methods and edge enhancement strategies, providing high-quality input for subsequent model recognition; the recognition optimization model and the recognition improvement model are fused, and a multi-scale convolution attention mechanism is introduced, effectively improving the extraction and discrimination ability of crack features, significantly improving the recognition accuracy and boundary positioning accuracy; a multi-scale bar convolution kernel and a spatial attention mechanism are used to enhance the perception ability of the model to cracks of different sizes and directions, adapting to the multi-morphology crack recognition requirements in complex scenes; the loss weights of the target detection and semantic segmentation tasks are dynamically adjusted to optimize the model training process, speed up the convergence speed, and improve the model robustness and generalization ability. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 a flow chart of the crack recognition method based on image processing in the first embodiment of the application;

[0049] Figure 2 a structural block diagram of the crack recognition system based on image processing in the second embodiment of the application;

[0050] Figure 3 a structural block diagram of the computer in the third embodiment of the application.

[0051] The following specific embodiments will further illustrate the application in conjunction with the above drawings. DETAILED DESCRIPTION

[0052] In order to facilitate the understanding of the application, the application will be described more fully below with reference to the accompanying drawings. The drawings show several embodiments of the application. However, the application can be realized in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the application more thorough and comprehensive.

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the application belongs. The terms used in the specification of the application herein are only for the purpose of describing the specific embodiments and are not intended to limit the application. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0054] Embodiment one

[0055] Please refer to Figure 1 , which shows the crack recognition method based on image processing in the first embodiment of the application, which specifically includes steps S101 to S105:

[0056] S101, acquire a plurality of image data, and perform image preprocessing on each of the image data to obtain a corresponding preprocessed image;

[0057] Further, the step S101 specifically comprises steps S1011-S1012:

[0058] S1011, respectively perform mean filtering processing, median filtering processing and Gaussian filtering processing on each of the image data, and filter out a preliminary image meeting a preset requirement based on the processing results;

[0059] S1012, perform image enhancement on the preliminary image, and perform residual optimization on the preliminary image after image enhancement by using an edge detection algorithm to obtain a corresponding preprocessed image.

[0060] In a specific implementation, a plurality of image data is acquired, wherein the image data is obtained by a preset image acquisition device collecting a detection area, the image acquisition device includes a camera, a drone, and other devices with image acquisition function; however, during image acquisition, the image acquisition process is affected by environmental light, device performance and other factors, resulting in noise in the collected image;

[0061] In the embodiment, the above image data is respectively subjected to mean filtering processing, median filtering processing and Gaussian filtering processing, wherein the mean filtering calculates a new pixel value by averaging the gray values of each pixel and its surrounding pixels, selects a 3X3 window neighborhood for each pixel, calculates the average value of all pixel gray values in the neighborhood, and assigns the average value to the pixel, thereby smoothing the image;

[0062] Further, the median filtering sorts the window neighborhood of each pixel point in the image as the center, updates the original value of the pixel point by using the sorted median, to protect the image edge and reduce the artifacts in the image. The Gaussian filtering reduces the noise and details in the image by superimposing a Gaussian function on the image, and performs local averaging by standard deviation, thereby preserving the original details of the image.

[0063] Specifically, after the above processing, a threshold value corresponding to each filtering processing is defined, the processing result is compared with the corresponding threshold value, and the processing result meeting the threshold value is marked as a preliminary image meeting the requirement;

[0064] In this embodiment, after filtering the image, the preliminary image is optimized by using an image enhancement method to improve the image quality and establish a high-quality data basis for subsequent model training. A residual image of the preliminary image is generated, a key area is distinguished by an edge detection algorithm, an edge mask is generated, a crack detail is strengthened in the edge area residual by using an enhancement strategy, a non-edge area is suppressed by using a suppression strategy, and an image is reconstructed to obtain a corresponding preprocessed image.

[0065] S102, a first recognition model is constructed, a feature structure is constructed by using a skip connection and a depth separable convolution algorithm, and the first recognition model and the feature structure are fused based on each of the preprocessed images to obtain a recognition optimization model.

[0066] Further, the step S102 specifically includes steps S1021-S1022:

[0067] S1021, a feature structure is constructed by using a skip connection and a depth separable convolution algorithm, and the feature structure is optimized by using an architecture search algorithm to obtain a feature optimization structure.

[0068] S1022, each of the preprocessed images is input into the first recognition model to generate low-level features of different sizes, and the low-level features of different sizes are fused, and the feature optimization structure is added in the fusion process to obtain a recognition optimization model.

[0069] In specific implementation, a first recognition model (which can use any one of DeepLabV3+, PSPNet, U-Net, and DenseNet models) is constructed, a feature structure is constructed by using a skip connection and a depth separable convolution algorithm, the feature structure is optimized by using an architecture search algorithm to improve the efficiency of model training, and the combination strategy of the feature structure is optimized.

[0070] Further, each of the preprocessed images is input into the first recognition model to generate low-level features of different sizes (such as 1 / 2, 1 / 4, and 1 / 8), the low-level features are fused, 1 / 4 feature maps are used as a fusion standard to perform down-sampling and bilinear interpolation processing on the other two low-level features in the fusion process, three low-level feature maps of 1 / 4 size are finally obtained, the three low-level feature maps are fused, and the feature optimization structure obtained above is added in the fusion process to obtain a recognition optimization model by replacing an original backbone network.

[0071] S103, construct a second recognition model, and input each of the preprocessed images into the second recognition model for model training. Dynamically adjust the weights according to the changes in the loss of each task during model training to form an improved recognition model.

[0072] Furthermore, step S103 specifically includes steps S1031 to S1032:

[0073] S1031, Define object detection loss and semantic segmentation loss, and dynamically monitor the convergence of the object detection loss and semantic segmentation loss during model training. Adaptively adjust the weights according to the loss ratio to obtain a preliminary improved model.

[0074] S1032, a spatial attention mechanism is introduced into the preliminary improved model, and the spatial attention mechanism is fused with the backbone network of the preliminary improved model to obtain an improved recognition model.

[0075] In practical implementation, the target detection loss is defined as follows:

[0076] ;

[0077] In the formula, Indicates the total number of pixels. Indicates the total number of categories. Indicates the first The pixel belongs to the first Real labels for each category Indicates real label The predicted value,

[0078] Define semantic segmentation loss:

[0079] ；

[0080] In the formula, Indicates the bounding box loss. Indicates confidence loss;

[0081] Based on the object detection loss and semantic segmentation loss described above, the total loss function of the model is defined as follows:

[0082] ;

[0083] ;

[0084] ;

[0085] In the formula, and They represent the first The dynamic weights of the two losses during each training round. temperature hyper-parameter, used to control the sensitivity of weight change, 、 respectively represent the size of the two losses in the current stage, and the weight is adaptively adjusted using the loss ratio to obtain a preliminary improved model.

[0086] Further, a spatial attention mechanism (in this embodiment, the spatial attention mechanism adopts any one of DSA, NSA, and BSA attention mechanisms) is introduced into the preliminary improved model, and the spatial attention mechanism is fused with the backbone network of the preliminary improved model to obtain a recognition improved model.

[0087] In this embodiment, the spatial attention mechanism is combined through dynamic convolution and Swish activation function, each branch respectively performs average pooling and maximum pooling operations on the input tensor in the channel dimension, the obtained results are spliced in the channel dimension to form a new tensor, and the obtained new tensor is subjected to dynamic convolution operation, so that the model can obtain global distribution information and capture significant features.

[0088] S104, model fusion is performed on the recognition optimization model and the recognition improved model, and a multi-scale convolution attention mechanism is introduced in the model fusion process to obtain a crack recognition model;

[0089] Further, the step S104 specifically includes steps S1041-S1042:

[0090] S1041, defining a multi-scale convolution attention mechanism, wherein the multi-scale convolution attention mechanism adopts a plurality of bar-shaped convolution kernels of different sizes;

[0091] S1042, introducing the multi-scale convolution attention mechanism in the model fusion process to perform convolution operation on the input feature map of the model in the vertical axis and horizontal axis directions to obtain a crack recognition model.

[0092] In specific implementation, the recognition optimization model and the recognition improved model are model fused, and a multi-scale convolution attention mechanism is defined, a plurality of bar-shaped convolution kernels of different sizes (in this embodiment, the convolution kernel size is 3x1, 5x1, and 7x1) are adopted, and convolution operation is performed on the input feature map of the model in the vertical axis and horizontal axis directions to obtain a crack recognition model. The multi-scale convolution attention mechanism is used to improve the capture ability of different size features, so as to more effectively cope with the size change of the crack target and improve the overall perception and boundary positioning ability of the model for the crack region.

[0093] wherein the expression of the multi-scale convolution attention mechanism is:

[0094] ;

[0095] In the formula, represents a feature combination function.

[0096] S105, using a preset image acquisition device to collect a to-be-processed image in real time, and using the crack identification model to perform image processing on the to-be-processed image to identify crack information of the to-be-processed image.

[0097] In specific implementation, a to-be-processed image of a to-be-identified region is collected in real time using a preset image acquisition device, which includes a camera, a drone, or other devices with image acquisition functions, and the crack identification model obtained above is used to perform image processing on the to-be-processed image to identify crack information in the to-be-processed image, so as to facilitate health detection of the region by engineering personnel, thereby realizing rapid identification of cracks with higher work efficiency.

[0098] To sum up, the crack identification method based on image processing in the above embodiments effectively suppresses noise and enhances crack region features by combining various filtering methods and edge enhancement strategies, providing high-quality input for subsequent model identification; the identification optimization model and the identification improvement model are fused, and a multi-scale convolution attention mechanism is introduced, which effectively improves the extraction and discrimination ability of crack features, significantly improves the identification accuracy and boundary positioning accuracy; the multi-scale strip convolution kernel and the spatial attention mechanism are used to enhance the perception ability of the model to cracks of different sizes and directions, and to adapt to the multi-morphology crack identification requirements in complex scenes; the loss weights of the target detection and semantic segmentation tasks are dynamically adjusted to optimize the model training process, speed up the convergence speed, and improve the model robustness and generalization ability.

[0099] Embodiment two

[0100] Another aspect of the present application also provides a crack identification system based on image processing, please refer to Figure 2 , which is a crack identification system based on image processing in the second embodiment of the present application, the system comprises:

[0101] An image preprocessing module 11 is configured to obtain a plurality of image data and perform image preprocessing on each of the image data to obtain corresponding preprocessed images.

[0102] A first model construction module 12 is configured to construct a first identification model and construct a feature structure using a skip connection and a depth separable convolution algorithm, and fuse the first identification model and the feature structure based on each of the preprocessed images to obtain an identification optimization model.

[0103] The second model construction module 13 is configured to construct a second identification model, input each of the preprocessed images into the second identification model for model training, dynamically adjust weights according to changes of each task loss in the model training process, and form an identification improved model.

[0104] The model fusion module 14 is configured to fuse the identification optimized model and the identification improved model, introduce a multi-scale convolution attention mechanism in the model fusion process, and obtain a crack identification model.

[0105] The image processing module 15 is configured to collect a to-be-processed image in real time by using a preset image collection device, and process the to-be-processed image by using the crack identification model to identify crack information of the to-be-processed image.

[0106] Further, the image preprocessing module 11 is specifically configured to:

[0107] The mean filtering processing, the median filtering processing and the Gaussian filtering processing are performed on each of the image data respectively, and a preliminary image meeting a preset requirement is selected based on a processing result.

[0108] The preliminary image is subjected to image enhancement, and a residual error of the preliminary image after the image enhancement is optimized by using an edge detection algorithm to obtain a corresponding preprocessed image.

[0109] Further, the first model construction module 12 is specifically configured to:

[0110] A feature structure is constructed by using a skip connection and a depth separable convolution algorithm, and the feature structure is optimized by using an architecture search algorithm to obtain a feature optimized structure.

[0111] Each of the preprocessed images is input into the first identification model to generate low-level features of different sizes, and the low-level features of different sizes are fused, and the feature optimized structure is added in the fusion process to obtain an identification optimized model.

[0112] Further, the second model construction module 13 is specifically configured to:

[0113] A target detection loss and a semantic segmentation loss are defined, and convergence of the target detection loss and the semantic segmentation loss is dynamically monitored in the model training process, and weights are adaptively adjusted according to a loss ratio to obtain a preliminary improved model.

[0114] A spatial attention mechanism is introduced into the preliminary improved model, and the spatial attention mechanism is fused with a backbone network of the preliminary improved model to obtain an identification improved model.

[0115] Further, the model fusion module 14 is specifically used for:

[0116] defining a multi-scale convolution attention mechanism, wherein the multi-scale convolution attention mechanism adopts a plurality of bar convolution kernels of different sizes;

[0117] introducing the multi-scale convolution attention mechanism in the model fusion process, so that the input feature map of the model is subjected to convolution operation in the vertical axis and horizontal axis directions, to obtain a crack identification model.

[0118] The functions or operation steps realized when the above modules and units are executed are generally the same as those of the above method embodiments, and will not be described here again.

[0119] The crack identification system based on image processing provided by the embodiment of the present application has the same implementation principle and technical effects as the above method embodiments, and for the sake of brevity, the part of the system embodiment not mentioned can be referred to the corresponding content in the above method embodiments.

[0120] Embodiment three

[0121] The present application also proposes a computer, please refer to Figure 3 , which is a computer in the third embodiment of the present application, comprising a memory 10, a processor 20, and a computer program 30 stored on the memory 10 and executable on the processor 20, wherein the processor 20 executes the computer program 30 to realize the above-mentioned crack identification method based on image processing.

[0122] The memory 10 includes at least one type of storage medium, including flash memory, hard disk, multimedia card, card type memory (such as SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. The memory 10 can be an internal storage unit of the computer in some embodiments, such as the hard disk of the computer. The memory 10 can also be an external storage device in other embodiments, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the memory 10 can include both the internal storage unit and the external storage device of the computer. The memory 10 can be used not only to store application software and various data installed on the computer, but also to temporarily store data that has been output or will be output.

[0123] The processor 20 may, in some embodiments, be an electronic control unit (ECU), a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip, for running program codes or processing data stored in the memory 10, such as executing the access restriction program.

[0124] It should be noted that, Figure 3 The illustrated structure does not constitute a limitation on the computer, which may, in other embodiments, include fewer or more components than shown, or combine certain components, or arrange the components differently.

[0125] The embodiments of the present application also propose a storage medium having a computer program stored thereon, which, when executed by a processor, implements the image processing-based crack identification method as described above.

[0126] Those skilled in the art can understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a list of executable instructions for implementing the logic function, which can be specifically implemented in any computer readable medium for use by or in conjunction with an instruction execution system, device or apparatus, such as a computer-based system, a system including a processor or other system that can fetch and execute instructions from the instruction execution system, device or apparatus. For the present specification, the "computer readable medium" can be any device that can contain, store, communicate, propagate or transport programs for use by or in conjunction with the instruction execution system, device or apparatus, or in conjunction with these instruction execution systems, devices or apparatus.

[0127] More specific examples (a non-exhaustive list) of the computer readable medium include the following: an electrical connection having one or more wires (electrical devices), a portable computer diskette (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer readable medium can even be paper or other suitable medium on which the program can be printed, as the program can be electronically obtained, for example, by optical scanning of the paper or other medium, followed by electronic conversion, interpretation or processing, if necessary, in other suitable ways, and then stored in a computer memory.

[0128] It should be understood that portions of the present application can be implemented with hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented with software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any of the following technologies, known in the art, or a combination thereof, can be used: discrete logic circuitry having logic gates for implementing logic functions upon an application of data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like.

[0129] The technical features of the above-described embodiments can be combined in any manner. For the sake of brevity, not all possible combinations of the technical features in the above-described embodiments are described, however, as long as the combination of the technical features does not result in a contradiction, it should be considered as within the scope of the present disclosure.

[0130] The above-described embodiments only express several implementation manners of the present application, and the description is specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of the patent of the present application should be subject to the appended claims.

Claims

1. An image processing-based crack identification method, characterized by, The method comprises the following steps: acquiring a plurality of image data and performing image preprocessing on each of the image data to obtain corresponding preprocessed images; constructing a first recognition model and constructing a feature structure using a skip connection and a depth separable convolution algorithm, and fusing the first recognition model and the feature structure based on each of the preprocessed images to obtain a recognition optimization model; constructing a second recognition model and inputting each of the preprocessed images into the second recognition model for model training, and dynamically adjusting the weight according to the change of each task loss in the model training process to form a recognition improvement model; fusing the recognition optimization model and the recognition improvement model, and introducing a multi-scale convolution attention mechanism in the model fusion process to obtain a crack recognition model; using a pre-set image acquisition device to collect real-time processing images, and using the crack recognition model to process the processing images to identify the crack information of the processing images; wherein the step of constructing a feature structure using a skip connection and a depth separable convolution algorithm, and fusing the first recognition model and the feature structure based on each of the preprocessed images to obtain a recognition optimization model comprises: constructing a feature structure using a skip connection and a depth separable convolution algorithm, and optimizing the feature structure using an architecture search algorithm to obtain a feature optimization structure; inputting each of the preprocessed images into the first recognition model to generate low-level features of different sizes, and fusing the low-level features of different sizes and adding the feature optimization structure in the fusion process to obtain a recognition optimization model.

2. The image processing-based crack identification method of claim 1, wherein, The step of acquiring a plurality of image data and performing image preprocessing on each of the image data to obtain corresponding preprocessed images comprises: respectively performing mean filtering, median filtering and Gaussian filtering on each of the image data, and screening out preliminary images meeting the pre-set requirements based on the processing results; performing image enhancement on the preliminary images, and using an edge detection algorithm to optimize the residual error of the image-enhanced preliminary images to obtain corresponding preprocessed images.

3. The image processing-based crack identification method of claim 1, wherein, The step of inputting each of the preprocessed images into the second recognition model for model training, and dynamically adjusting the weight according to the change of each task loss in the model training process to form a recognition improvement model comprises: defining a target detection loss and a semantic segmentation loss, and dynamically monitoring the convergence of the target detection loss and the semantic segmentation loss in the model training process, and adaptively adjusting the weight according to the loss ratio to obtain a preliminary improvement model; introducing a spatial attention mechanism into the preliminary improvement model, and fusing the spatial attention mechanism with the backbone network of the preliminary improvement model to obtain a recognition improvement model.

4. The image processing-based crack identification method of claim 1, wherein, The step of fusing the recognition optimization model and the recognition improvement model, and introducing a multi-scale convolution attention mechanism in the model fusion process to obtain a crack recognition model comprises: defining a multi-scale convolution attention mechanism, wherein the multi-scale convolution attention mechanism uses a plurality of bar-shaped convolution kernels of different sizes; The multiscale convolution attention mechanism is introduced in the model fusion process, so that the input feature map of the model is subjected to convolution operation in the vertical axis and horizontal axis directions respectively to obtain a crack identification model.

5. An image processing based crack identification system characterized in that, The method comprises the following steps: An image preprocessing module is configured to acquire a plurality of image data and perform image preprocessing on each of the image data to obtain corresponding preprocessed images. A first model construction module is configured to construct a first identification model, construct a feature structure using a skip connection and a depth separable convolution algorithm, and fuse the first identification model and the feature structure based on each of the preprocessed images to obtain an identification optimization model. A second model construction module is configured to construct a second identification model, input each of the preprocessed images into the second identification model for model training, and dynamically adjust a weight according to the change of each task loss in the model training process to form an identification improvement model. A model fusion module is configured to fuse the identification optimization model and the identification improvement model, and introduce a multiscale convolution attention mechanism in the model fusion process to obtain a crack identification model. An image processing module is configured to acquire a to-be-processed image in real time using a preconfigured image acquisition device, perform image processing on the to-be-processed image using the crack identification model, and identify crack information of the to-be-processed image. The first model construction module is specifically configured to: construct a feature structure using a skip connection and a depth separable convolution algorithm, optimize the feature structure using an architecture search algorithm, and obtain a feature optimization structure. input each of the preprocessed images into the first identification model to generate low-level features of different sizes, fuse the low-level features of different sizes, and add the feature optimization structure in the fusion process to obtain an identification optimization model.

6. The image processing based crack identification system of claim 5, wherein, The image preprocessing module is specifically configured to: perform mean filtering, median filtering, and Gaussian filtering on each of the image data, and select a preliminary image meeting a preconfigured requirement based on the processing results. perform image enhancement on the preliminary image, and perform residual optimization on the preliminary image after image enhancement using an edge detection algorithm to obtain a corresponding preprocessed image.

7. A storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement the crack identification method based on image processing.

8. A computer comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the crack identification method based on image processing.

Images