Crack tip identification method based on joint view field segmentation network

Abstract

The invention belongs to the field of structural health detection, and particularly relates to a crack tip identification method based on a joint view field segmentation network. The method comprises the following steps: step 1, constructing a full-view-field crack data set; 2, constructing a local field crack tip data set; 3, according to the full-view-field crack data set and the local-field crack tip data set, constructing a combined view field segmentation network model; step 4, constructing a U-Net crack segmentation model according to the full-view-field crack data set; and 5, carrying out crack tip identification and positioning according to the combined view field segmentation network model and the U-Net crack segmentation model. Based on the deep neural network and the image processing technology, high-precision and automatic positioning of the crack tip in the full-size aircraft strength test is achieved, high-precision and automatic crack tip positioning can be achieved instead of manual vision, and a technical foundation is laid for subsequent accurate measurement of the crack length.

Description

Technical Field

[0001] This application belongs to the field of structural health monitoring, and particularly relates to a method for identifying crack tips based on a joint field-of-view segmentation network. Background Art

[0002] Full-scale aircraft strength tests are a key link in aircraft structural integrity design. The purpose is to expose weak parts of the structural design and support the improvement of structural design. Achieving accurate measurement of the length of crack damage in full-scale aircraft strength tests is crucial, which helps to understand the crack initiation and propagation laws of key aircraft structures, thus playing a guiding role in the optimization of aircraft structural design, the determination of service life, and subsequent maintenance and repair, and greatly improving the safety performance of the aircraft.

[0003] Obtaining the accurate position of the crack tip is a prerequisite for accurate measurement of the crack length. At present, crack tip positioning and crack length measurement in aircraft strength tests mainly rely on manual visual timing monitoring, which has disadvantages such as large crack tip positioning errors, low damage monitoring efficiency, and high labor costs.

[0004] In recent years, computer vision technology based on deep learning has developed rapidly and has been effectively applied in many fields such as industrial inspection, which can provide a reliable technical approach for high-precision and automated positioning of crack tips in full-scale aircraft strength tests. However, the crack tip belongs to a small target, and there are a large number of scratches, stains, etc. on the surface of the aircraft structure, resulting in great difficulty in identifying the crack tip.

[0005] Therefore, it is hoped that there is a technical solution to overcome or at least mitigate at least one of the above defects of the existing technology. Summary of the Invention

[0006] The purpose of this application is to provide a method for identifying crack tips based on a joint field-of-view segmentation network to solve at least one problem existing in the prior art.

[0007] The technical solution of this application is as follows:

[0008] A method for identifying crack tips based on a joint field-of-view segmentation network, comprising:

[0009] Step 1, constructing a full-field crack data set;

[0010] Step 2, constructing a local-field crack tip data set;

[0011] Step 3, constructing a joint field-of-view segmentation network model according to the full-field crack data set and the local-field crack tip data set;

[0012] Step 4, constructing a U-Net crack segmentation model according to the full-field crack data set;

[0013] Step 5: Identify and locate the crack tip according to the combined field-of-view segmentation network model and the U-Net crack segmentation model.

[0014] In at least one embodiment of the present application, in step 1, construct a full-field crack data set, including:

[0015] S101: Obtain the full-field crack images of the aircraft structure;

[0016] S102: Segment and label the full-field crack images and generate mask images to obtain a full-field crack data set.

[0017] In at least one embodiment of the present application, in step 2, construct a local-field crack tip data set, including:

[0018] S201: Select the full-field crack images and the corresponding mask images from the full-field crack data set;

[0019] S202: In the mask image, take out the endpoints of the largest connected component based on the opencv library, and the endpoint coordinates are the rough positioning coordinates of the crack tip in the full-field crack image;

[0020] S203: Take a picture of a predetermined size from the full-field crack image with the rough positioning coordinates of the crack tip as the center to obtain a local-field crack tip image;

[0021] Repeat steps S201 - S203 to obtain the local-field crack tip images of all the full-field crack images in the full-field crack data set, and obtain a local-field crack tip data set.

[0022] In at least one embodiment of the present application, in step 3, construct a combined field-of-view segmentation network model according to the full-field crack data set and the local-field crack tip data set, including:

[0023] S301: Construct a full-field downsampling network according to the full-field crack data set;

[0024] S302: Construct a local-field downsampling network according to the local-field crack tip data set;

[0025] S303: Construct multiple combined field-of-view segmentation modules according to the full-field downsampling network and the local-field downsampling network;

[0026] S304: Construct a combined field-of-view segmentation network based on multiple combined field-of-view segmentation modules;

[0027] S305: Input the full-field crack data set and the local-field crack tip data set into the combined field-of-view segmentation network for training to obtain a combined field-of-view segmentation network model.

[0028] In at least one embodiment of the present application, in S301, according to the full-field crack data set, a full-field downsampling network is constructed, including:

[0029] The full-field crack images with a size of 512*512*3 in the full-field crack data set are continuously input into four convolutional layers for downsampling, and feature maps of different sizes are obtained in sequence, including: the feature map Cr_d_a with a size of 256*256*16, the feature map Cr_d_b with a size of 128*128*32, the feature map Cr_d_c with a size of 64*64*64, and the feature map Cr_d_d with a size of 32*32*64.

[0030] In at least one embodiment of the present application, in S302, according to the local field crack tip data set, a local field downsampling network is constructed, including:

[0031] The local field crack tip images with a size of 64*64*3 in the local field crack tip data set are continuously input into four convolutional layers for downsampling, and feature maps of different sizes are obtained in sequence, including: the feature map Ti_a with a size of 32*32*16, the feature map Ti_b with a size of 16*16*32, the feature map Ti_c with a size of 8*8*64, and the feature map Ti_d with a size of 8*8*64.

[0032] In at least one embodiment of the present application, in S303, according to the full-field downsampling network and the local field downsampling network, a plurality of combined field-of-view segmentation modules are constructed, including:

[0033] The feature map Cr_d_d and the feature map Ti_d are input into the first combined field-of-view segmentation module to obtain the feature map Cr_u_a with a size of 32*32*128;

[0034] The feature map Cr_d_c, the feature map Cr_u_a, and the feature map Ti_c are input into the second combined field-of-view segmentation module to obtain the feature map Cr_u_b with a size of 64*64*64;

[0035] The feature map Cr_d_b, the feature map Cr_u_b, and the feature map Ti_b are input into the third combined field-of-view segmentation module to obtain the feature map Cr_u_c with a size of 128*128*32;

[0036] The feature map Cr_d_a, the feature map Cr_u_a, and the feature map Ti_a are input into the fourth combined field-of-view segmentation module to obtain the feature map Cr_u_d with a size of 256*256*16;

[0037] Input the full-field crack image and local field crack tip image after single-layer convolution, as well as the feature map Cr_u_d, into the fifth joint field-of-view segmentation module to obtain the feature map Cr_u_e with a size of 512*512*3.

[0038] In at least one embodiment of the present application, S304. Based on multiple said joint field-of-view segmentation modules, construct a joint field-of-view segmentation network, including:

[0039] Perform a convolution operation with a convolution kernel of 1*1 on the feature map Cr_u_e to obtain a crack segmentation result map with a size of 512*512*1.

[0040] In at least one embodiment of the present application, in step four, construct a U-Net crack segmentation model according to the full-field crack data set, including:

[0041] Input the full-field crack data set into the U-Net network for training to obtain a U-Net crack segmentation model.

[0042] In at least one embodiment of the present application, in step five, perform crack tip identification and positioning according to the joint field-of-view segmentation network model and the U-Net crack segmentation model, including:

[0043] S501. Input the full-field crack image to be identified into the U-Net crack segmentation model to obtain a mask image, and according to the full-field crack image to be identified and the corresponding mask image, obtain a local field crack tip image;

[0044] S502. Input the full-field crack image to be identified and the corresponding local field crack tip image into the joint field-of-view segmentation network model to obtain a crack segmentation result map;

[0045] S503. In the crack segmentation result map, based on the opencv library, take out the endpoints of the largest connected domain. In the crack segmentation result map, based on the opencv library, take out the endpoints of the largest connected domain, and the coordinates of this endpoint are the accurate positioning coordinates of the crack tip in the full-field crack image to be identified.

[0046] The invention has at least the following beneficial technical effects:

[0047] The crack tip identification method based on the joint field-of-view segmentation network of the present application, based on deep neural network and image processing technology, realizes high-precision and automated positioning of crack tips in full-scale aircraft strength tests, can replace manual visual inspection to achieve high-precision and automated crack tip positioning, and lays a technical foundation for subsequent accurate measurement of crack lengths. Description of the Drawings

[0048] Figure 1Flowchart of a crack tip identification method based on a joint field-of-view segmentation network according to an embodiment of the present application;

[0049] Figure 2 Schematic diagram of the joint field-of-view segmentation network structure according to an embodiment of the present application;

[0050] Figure 3 Schematic diagram of the joint field-of-view segmentation module according to an embodiment of the present application;

[0051] Figure 4 Schematic diagram of the in-situ supplementary graph operation according to an embodiment of the present application. Detailed implementation manners

[0052] To make the purpose, technical solutions and advantages of the implementation of the present application clearer, the technical solutions in the embodiments of the present application will be described in more detail below with reference to the accompanying drawings in the embodiments of the present application. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are some, but not all, of the embodiments of the present application. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present application and should not be construed as limiting the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts fall within the scope of protection of the present application. The embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0053] In the description of the present application, it should be understood that the orientation or positional relationship indicated by the terms "center", "longitudinal", "transverse", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as limiting the scope of protection of the present application.

[0054] The following combines the attached Figures 1 to 4 A further detailed description of the present application will be made below.

[0055] The present application provides a crack tip identification method based on a joint field-of-view segmentation network, as Figure 1 shown, including the following steps:

[0056] Step 1: Construct a full-field crack data set;

[0057] Step 2: Construct a local-field crack tip data set;

[0058] Step 3: Construct a joint field-of-view segmentation network model according to the full-field crack data set and the local-field crack tip data set;

[0059] Step 4: Construct a U-Net crack segmentation model based on the full-field crack dataset;

[0060] Step 5: Identify and locate the crack tip according to the combined field-of-view segmentation network model and the U-Net crack segmentation model.

[0061] For the crack tip identification method based on the combined field-of-view segmentation network of the present application, first, in Step 1, construct a full-field crack dataset, specifically including:

[0062] S101: Obtain the full-field crack images of the aircraft structure;

[0063] Obtain a large number of crack damage images with a size of 512*512*3 on the aircraft structure as the full-field crack images;

[0064] S102: Segment and label the full-field crack images and generate mask images to obtain the full-field crack dataset.

[0065] Through two preprocessing operations of segmenting and labeling the obtained full-field crack images and generating mask pictures, the full-field crack dataset is obtained.

[0066] For the crack tip identification method based on the combined field-of-view segmentation network of the present application, in Step 2, the method for constructing the local field crack tip dataset is as follows:

[0067] Perform crack tip extraction operations:

[0068] S201: Select the full-field crack images and the corresponding mask images from the full-field crack dataset;

[0069] S202: In the mask image, take out the endpoints of the largest connected component based on the opencv library, and the coordinates of this endpoint are the rough positioning coordinates (x, y) of the crack tip in the full-field crack image;

[0070] S203: Take a picture of a predetermined size from the full-field crack image with the rough positioning coordinates of the crack tip as the center (x, y) to obtain the local field crack tip image;

[0071] In this embodiment, a small picture with a size of 64*64*3 is intercepted from the full-field crack image to obtain the local field crack tip image, and the name of the local field crack tip image should be "full-field crack image name + cracktip + x + y".

[0072] Perform crack tip extraction operations on each group of full-field crack images and the corresponding mask images in the full-field crack dataset:

[0073] Repeat steps S201 - S203 to obtain the local field crack tip images of all full - field crack images in the full - field crack dataset, and obtain the local field crack tip dataset.

[0074] The crack tip recognition method based on the joint field segmentation network of the present application, as Figure 2 shown, in step three, the construction method of the joint field segmentation network model is specifically as follows:

[0075] S301. Construct a full - field downsampling network according to the full - field crack dataset;

[0076] Continuously input the full - field crack images with a size of 512 * 512 * 3 in the full - field crack dataset into four convolutional layers for downsampling, and sequentially obtain feature maps of different sizes, including: the feature map Cr_d_a with a size of 256 * 256 * 16, the feature map Cr_d_b with a size of 128 * 128 * 32, the feature map Cr_d_c with a size of 64 * 64 * 64, and the feature map Cr_d_d with a size of 32 * 32 * 64.

[0077] S302. Construct a local - field downsampling network according to the local - field crack tip dataset;

[0078] Continuously input the local - field crack tip images with a size of 64 * 64 * 3 in the local - field crack tip dataset into four convolutional layers for downsampling, and sequentially obtain feature maps of different sizes, including: the feature map Ti_a with a size of 32 * 32 * 16, the feature map Ti_b with a size of 16 * 16 * 32, the feature map Ti_c with a size of 8 * 8 * 64, and the feature map Ti_d with a size of 8 * 8 * 64.

[0079] It can be understood that to avoid the loss of crack tip detail information caused by excessive downsampling, in the local - field downsampling network, the stride of the fourth convolutional layer is set to 1, that is, the size of the feature map does not change before and after the convolution operation.

[0080] S303. Construct multiple joint field segmentation modules according to the full - field downsampling network and the local - field downsampling network;

[0081] Input the feature map Cr_d_d and the feature map Ti_d into the first joint field segmentation module to obtain the feature map Cr_u_a with a size of 32 * 32 * 128;

[0082] Input the feature map Cr_d_c, the feature map Cr_u_a, and the feature map Ti_c into the second joint field segmentation module to obtain the feature map Cr_u_b with a size of 64 * 64 * 64;

[0083] Input the feature maps Cr_d_b, Cr_u_b, and Ti_b into the third joint field-of-view segmentation module to obtain a feature map Cr_u_c with a size of 128*128*32;

[0084] Input the feature maps Cr_d_a, Cr_u_a, and Ti_a into the fourth joint field-of-view segmentation module to obtain a feature map Cr_u_d with a size of 256*256*16;

[0085] Input the full-field crack image and local-field crack tip image after single-layer convolution, as well as the feature map Cr_u_d, into the fifth joint field-of-view segmentation module to obtain a feature map Cr_u_e with a size of 512*512*3.

[0086] Specifically, in this embodiment, as Figure 3 shown, the specific process of constructing the first joint field-of-view segmentation module is as follows:

[0087] The first joint field-of-view segmentation module has two inputs, namely the feature map Cr_d_d (with a size of 32*32*64) and the feature map Ti_d (with a size of 8*8*64).

[0088] Perform an in-situ padding operation on the feature map Ti_d, that is, restore the feature map Ti_d (as Figure 4 (a) left shows) to its relative position in the original image (as Figure 4 (c) shows), as Figure 4 (a) right shows;

[0089] According to the crack tip extraction operation in step two, the coordinates of the center point of the feature map Ti_d in the original image are (x, y); after successive convolutions, the feature map Ti_d (with a size of 8*8*64) is reduced by n times (here n = 8) relative to the local-field crack tip map (with a size of 64*64*3).

[0090] Therefore, the number of pixel points to be expanded on the left, right, top, and bottom edges of the feature map Ti_d are:

[0091] Left edge: (x - 64*0.5) / n Right edge: (512 - x - 64*0.5) / n

[0092] Top edge: (y - 64*0.5) / n Bottom edge: (512 - y - 64*0.5) / n

[0093] At the same time, the in-situ padding operation uses the background pixel value 0 for image expansion.

[0094] Then the padded image is as Figure 4 (b) shows.

[0095] After the original image is padded to the feature map Ti_d, its size is 64*64*64; it is input into a convolutional layer for size adjustment to obtain a feature map with a size of 32*32*64; then this feature map is concatenated with the feature map Cr_d_d (with a size of 32*32*64) in the channel dimension, and feature abstraction is performed through a convolutional layer to obtain the feature map Cr_u_a (with a size of 32*32*128).

[0096] The above is the construction process of the first combined field-of-view segmentation module.

[0097] As Figure 3 shown, the specific process of constructing the second combined field-of-view segmentation module is as follows:

[0098] The construction principles of the third, fourth, and fifth combined field-of-view segmentation modules are the same as those of the second combined field-of-view segmentation module, and they are all constructed based on feature maps at three different levels. Therefore, taking the second combined field-of-view segmentation module as an example, the construction methods of the remaining combined field-of-view segmentation modules are demonstrated.

[0099] The second combined field-of-view segmentation module has three inputs, namely the feature map Cr_d_c (with a size of 64*64*64), the feature map Cr_u_a (with a size of 32*32*128), and the feature map Ti_c (with a size of 8*8*64).

[0100] Perform a transposed convolution operation on the feature map Cr_u_a to increase the size to 64*64*128; perform an in-situ padding operation on the feature map Ti_c to adjust the size to 64*64*64; concatenate the two processed feature maps with Cr_d_c in the channel dimension to obtain a feature map with a size of 64*64*256; then reduce the number of channels and perform feature abstraction on this feature map through a convolutional layer to obtain the feature map Cr_u_b (with a size of 64*64*64);

[0101] As Figure 2 shown, based on the above steps, input the feature maps Cr_d_b, Cr_u_b, and Ti_b into the third combined field-of-view segmentation module to obtain the feature map Cr_u_c (with a size of 128*128*32); input the feature maps Cr_d_a, Cr_u_a, and Ti_a into the fourth combined field-of-view segmentation module to obtain the feature map Cr_u_d (with a size of 256*256*16); input the full-field crack image and local-field crack image after single-layer convolution operation and the feature map Cr_u_d into the fifth combined field-of-view segmentation module to obtain the feature map Cr_u_e (with a size of 512*512*3).

[0102] S304. Based on multiple combined field-of-view segmentation modules, construct a combined field-of-view segmentation network;

[0103] Perform a convolution operation with a 1×1 convolution kernel on the feature map Cr_u_e to obtain a crack segmentation result map with a size of 512×512×1.

[0104] S305. Input the full-field crack data set and the local-field crack tip data set into the joint field-of-view segmentation network for training to obtain a joint field-of-view segmentation network model.

[0105] In the crack tip recognition method based on the joint field-of-view segmentation network of the present application, in step four, input the full-field crack data set into the U-Net network for training. After the training is completed, a U-Net crack segmentation model is obtained.

[0106] Finally, in step five, construct a crack tip recognition algorithm based on the U-Net crack segmentation model and the joint field-of-view segmentation network model to achieve crack tip recognition and positioning. Specifically:

[0107] S501. Input the full-field crack image to be recognized into the U-Net crack segmentation model to obtain a mask image, and based on the full-field crack image to be recognized and the corresponding mask image, obtain a local-field crack tip image (the detailed operation is as shown in step two);

[0108] S502. Input the full-field crack image to be recognized and the corresponding local-field crack tip image into the joint field-of-view segmentation network model to obtain a crack segmentation result map;

[0109] S503. In the crack segmentation result map, based on the opencv library, extract the endpoints of the largest connected domain. In the crack segmentation result map, based on the opencv library, extract the endpoints of the largest connected domain. The coordinates of this endpoint are the accurate positioning coordinates (a, b) of the crack tip in the full-field crack image to be recognized.

[0110] The crack tip recognition method based on the joint field-of-view segmentation network of the present application has the following beneficial effects:

[0111] (1) The crack tip belongs to a tiny target, and there are a large number of scratches, stains, etc. on the surface of the aircraft structure, resulting in great difficulty in recognizing the crack tip. In a conventional deep neural network, continuous downsampling will result in a very low resolution of the final feature map, losing a large amount of detailed information, which is particularly unfavorable for semantic segmentation of small targets such as crack tips; and it is easy to misjudge interference noise such as scratches and stains as crack tips.

[0112] To address the above problems, the method of the present application performs rough positioning of the crack tip area based on the U-Net segmentation network and the full-field crack image, which can filter out a large amount of background noise interference similar to the crack tip and avoid misjudgment of the crack tip position (see step S501 in detail);

[0113] By introducing a local field crack tip image downsampling path, the network focuses on the feature learning of the crack tip region (see step 302 for details);

[0114] And through the layer-by-layer joint field-of-view segmentation module, the global crack features are tightly coupled with the crack tip features, strengthening the feature information of the crack tip region, forming a more complete and detailed crack feature representation, and effectively avoiding the recognition delay problem caused by the loss of crack tip information. (See steps S303 - S304 for details)

[0115] (2) The high-precision and automated crack tip recognition and positioning method can effectively solve the problems existing in the manual visual method, such as large positioning error and low efficiency, and lay a technical foundation for the crack length expansion monitoring in the full-scale aircraft strength test.

[0116] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A crack tip identification method based on a joint field of view segmentation network, characterized in that: include: Step 1: Construct a full-field crack dataset; Step 2: construct a local field crack tip dataset; Step 3: constructing a joint field of view segmentation network model according to the full field of view crack data set and the local field crack tip data set; Step 4: construct a U-Net crack segmentation model based on the full-field crack data set; Step 5: Identify and locate the crack tip based on the joint field of view segmentation network model and the U-Net crack segmentation model.

2. The crack tip identification method based on the joint field of view segmentation network according to claim 1 is characterized in that: In step 1, a full-field crack dataset is constructed, including: S101, obtaining a full-field crack image of the aircraft structure; S102, segmenting and annotating the full-field-of-view crack image and generating a mask image to obtain a full-field-of-view crack data set.

3. The crack tip identification method based on the joint field of view segmentation network according to claim 2 is characterized in that: In step 2, a local field crack tip dataset is constructed, including: S201, selecting a full-field-of-view crack image and a corresponding mask image from the full-field-of-view crack data set; S202, in the mask image, based on the opencv library, extracting the endpoint of the largest connected domain, the coordinates of the endpoint being the coarse positioning coordinates of the crack tip in the full-field crack image; S203, taking the coarse positioning coordinates of the crack tip as the center, intercepting a picture of a predetermined size from the full-field crack image to obtain a local-field crack tip image; Repeat steps S201-S203 to obtain local field crack tip images of all the full field crack images in the full field crack data set to obtain a local field crack tip data set.

4. The crack tip identification method based on the joint field of view segmentation network according to claim 3 is characterized in that: In step three, a joint field of view segmentation network model is constructed based on the full field of view crack dataset and the local field crack tip dataset, including: S301, constructing a full-field-of-view downsampling network according to the full-field-of-view crack data set; S302, constructing a local field downsampling network according to the local field crack tip data set; S303, constructing a plurality of joint field of view segmentation modules according to the full field of view downsampling network and the local field of view downsampling network; S304, constructing a joint field of view segmentation network based on the plurality of joint field of view segmentation modules; S305, inputting the full-field-of-view crack data set and the local-field crack tip data set into the joint-field-of-view segmentation network for training to obtain a joint-field-of-view segmentation network model.

5. The crack tip identification method based on the joint field of view segmentation network according to claim 4 is characterized in that: In S301, a full-field downsampling network is constructed according to the full-field crack data set, including: The full-field crack image with a size of 512*512*3 in the full-field crack data set is continuously input into four convolutional layers for downsampling, and feature maps of different sizes are obtained in turn, including: a feature map Cr_d_a with a size of 256*256*16, a feature map Cr_d_b with a size of 128*128*32, a feature map Cr_d_c with a size of 64*64*64, and a feature map Cr_d_d with a size of 32*32*64.

6. The crack tip identification method based on the joint field of view segmentation network according to claim 5 is characterized in that: In S302, a local field downsampling network is constructed according to the local field crack tip data set, including: The local field crack tip images with a size of 64*64*3 in the local field crack tip dataset are continuously input into four convolutional layers for down-sampling, and feature maps of different sizes are obtained in turn, including: a feature map Ti_a with a size of 32*32*16, a feature map Ti_b with a size of 16*16*32, a feature map Ti_c with a size of 8*8*64, and a feature map Ti_d with a size of 8*8*64.

7. The crack tip identification method based on the joint field of view segmentation network according to claim 6 is characterized in that: In S303, a plurality of joint field of view segmentation modules are constructed according to the full field of view downsampling network and the local field of view downsampling network, including: Input the feature map Cr_d_d and the feature map Ti_d into the first joint field of view segmentation module to obtain a feature map Cr_u_a with a size of 32*32*128; Input the feature map Cr_d_c, the feature map Cr_u_a and the feature map Ti_c into the second joint field of view segmentation module to obtain a feature map Cr_u_b with a size of 64*64*64; Input the feature map Cr_d_b, the feature map Cr_u_b and the feature map Ti_b into the third joint field of view segmentation module to obtain a feature map Cr_u_c with a size of 128*128*32; Input the feature map Cr_d_a, the feature map Cr_u_a and the feature map Ti_a into the fourth joint field of view segmentation module to obtain a feature map Cr_u_d with a size of 256*256*16; The full-field crack image, local-field crack tip image and feature map Cr_u_d after single-layer convolution are input into the fifth joint field of view segmentation module to obtain a feature map Cr_u_e with a size of 512*512*3.

8. The crack tip identification method based on a joint field of view segmentation network according to claim 7 is characterized in that: S304: constructing a joint field of view segmentation network based on the plurality of joint field of view segmentation modules, including: A convolution operation with a convolution kernel of 1*1 is performed on the feature map Cr_u_e to obtain a crack segmentation result map with a size of 512*512*1.

9. The crack tip identification method based on the joint field of view segmentation network according to claim 8 is characterized in that: In step 4, a U-Net crack segmentation model is constructed based on the full-field crack data set, including: The full-field crack data set is input into a U-Net network for training to obtain a U-Net crack segmentation model.

10. The crack tip identification method based on a joint field of view segmentation network according to claim 9, characterized in that: In step 5, crack tip identification and positioning are performed according to the joint field of view segmentation network model and the U-Net crack segmentation model, including: S501, inputting the full-field crack image to be identified into the U-Net crack segmentation model to obtain a mask image, and obtaining a local field crack tip image according to the full-field crack image to be identified and the corresponding mask image; S502, inputting the full-field crack image to be identified and the corresponding local-field crack tip image into the joint field segmentation network model to obtain a crack segmentation result image; S503. In the crack segmentation result map, the endpoints of the largest connected domain are taken out based on the OpenCV library. In the crack segmentation result map, the endpoints of the largest connected domain are taken out based on the OpenCV library. The coordinates of the endpoints are the precise positioning coordinates of the crack tip in the full-field crack image to be identified.

Images