Method, device, electronic equipment and storage medium for detecting airport pavement defects

Abstract

The present invention discloses a method, device, electronic device and storage medium for detecting airport pavement defects, including: obtaining an airport pavement defect image to be detected; inputting the airport pavement defect image to be detected into a trained defect detection model to obtain extraction results of regions corresponding to different types of defects in the airport pavement defect image to be detected; wherein the trained defect detection model is obtained after training using different airport pavement defect images and semantic segmentation truth maps and semantic edge truth maps corresponding to different airport pavement defect images. The present invention realizes pixel-level segmentation of multiple defects in airport pavement images through the defect detection model, thereby improving the accuracy of segmentation.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method, a device, an electronic device and a storage medium for detecting airport pavement defects. Background Art

[0002] Airport pavement safety inspection is the top priority of civil aviation operation safety, but currently the inspection of airport pavement defects is still mainly done manually.

[0003] Some of the current defect detection methods propose a secondary threshold segmentation algorithm based on morphology to achieve the segmentation of cracks in images containing runway markings. Some use seed growth algorithms to detect cracks at multiple scales, and obtain crack segmentation results by matching and fusing crack areas at different scales. Most of these methods rely on manually designed parameters, have poor generalization capabilities, cannot handle complex and changeable airport pavement images, and the detection accuracy is greatly affected by the environment. With the rise of deep learning, more and more deep learning-based defect detection methods have been proposed. For example, the full convolutional network is applied to crack segmentation, and the network is trained by inputting multiple types of crack images to segment and predict cracks of different scales. However, this method only realizes defect detection based on rectangular boxes, and cannot achieve more refined pixel-level segmentation. In addition, the shapes of different types of defects cannot be reflected, and the position information still has a large deviation.

[0004] In summary, there is an urgent need for a method for detecting airport pavement defects to solve the problems existing in the above-mentioned prior art. Summary of the invention

[0005] Since the existing methods have the above-mentioned problems, the present invention proposes a method, device, electronic equipment and storage medium for detecting airport pavement defects.

[0006] In a first aspect, the present invention provides a method for detecting airport pavement defects, comprising:

[0007] Acquire the airport pavement disease image to be detected;

[0008] Inputting the airport pavement defect image to be detected into a trained defect detection model to obtain extraction results of regions corresponding to different types of defects in the airport pavement defect image to be detected;

[0009] The trained defect detection model is obtained by training with different airport pavement defect images and semantic segmentation truth maps and semantic edge truth maps corresponding to different airport pavement defect images.

[0010] Furthermore, before inputting the airport pavement defect image to be detected into the trained defect detection model to obtain the extraction result of the defect corresponding area in the airport pavement defect image to be detected, it also includes:

[0011] Obtain disease training images, semantic segmentation truth maps, and semantic edge truth maps;

[0012] Performing wavelet transform on the disease training image to obtain a low-frequency feature map;

[0013] Using a residual network to extract image features from the disease training image to obtain a semantic edge feature map;

[0014] Encoding the disease training image according to the low-frequency feature map and the semantic edge feature map based on a U-Net network to obtain a first feature map;

[0015] Decoding the first feature map according to the low-frequency feature map and the semantic edge feature map to obtain a semantic segmentation prediction map;

[0016] Determine a first loss function according to the semantic segmentation true value map and the semantic segmentation prediction map;

[0017] The parameters of the disease detection model are updated according to the first loss function to obtain a trained disease detection model.

[0018] Furthermore, the obtaining of the semantic segmentation truth map and the semantic edge truth map includes:

[0019] Extracting the outer contour of the disease in the disease training image to obtain image annotation information;

[0020] Determine a semantic segmentation truth map according to the image annotation information;

[0021] The semantic edge of the diseased area is extracted based on the semantic segmentation truth map to obtain a semantic edge truth map.

[0022] Furthermore, after extracting image features from the disease training image using the residual network to obtain a semantic edge feature map, the method further includes:

[0023] Using an attention mechanism to obtain a channel domain weight matrix of the semantic edge feature map;

[0024] Obtain a semantic edge prediction map according to the semantic edge feature map and the channel domain weight matrix;

[0025] Determine a second loss function according to the semantic edge truth map and the semantic edge prediction map;

[0026] The parameters of the disease detection model are updated according to the second loss function.

[0027] Further, after obtaining the first feature map and before decoding the first feature map according to the low-frequency feature map and the semantic edge feature map, the method further includes:

[0028] Using the attention mechanism to obtain the spatial domain weight matrix of the first feature map;

[0029] A second feature map is determined according to the first feature map and the spatial domain weight matrix.

[0030] Furthermore, the categories of defects in the airport pavement defect image to be detected are cracks, repairs, patches, ground lights and board seams.

[0031] In a second aspect, the present invention provides a device for detecting airport pavement defects, comprising:

[0032] An acquisition module is used to acquire the airport pavement disease image to be detected;

[0033] A processing module is used to input the airport pavement defect image to be detected into a trained defect detection model to obtain the extraction results of the corresponding areas of different types of defects in the airport pavement defect image to be detected; wherein the trained defect detection model is obtained after training using different airport pavement defect images and the semantic segmentation truth maps and semantic edge truth maps corresponding to different airport pavement defect images.

[0034] Furthermore, the processing module is also used for:

[0035] Before inputting the airport pavement defect image to be detected into the trained defect detection model to obtain the extraction result of the defect corresponding area in the airport pavement defect image to be detected, obtaining the defect training image, the semantic segmentation truth map and the semantic edge truth map;

[0036] Performing wavelet transform on the disease training image to obtain a low-frequency feature map;

[0037] Using a residual network to extract image features from the disease training image to obtain a semantic edge feature map;

[0038] Encoding the disease training image according to the low-frequency feature map and the semantic edge feature map based on a U-Net network to obtain a first feature map;

[0039] Decoding the first feature map according to the low-frequency feature map and the semantic edge feature map to obtain a semantic segmentation prediction map;

[0040] Determine a first loss function according to the semantic segmentation true value map and the semantic segmentation prediction map;

[0041] The parameters of the disease detection model are updated according to the first loss function to obtain a trained disease detection model.

[0042] Furthermore, the processing module is specifically used for:

[0043] Extracting the outer contour of the disease in the disease training image to obtain image annotation information;

[0044] Determine a semantic segmentation truth map according to the image annotation information;

[0045] The semantic edge of the diseased area is extracted based on the semantic segmentation truth map to obtain a semantic edge truth map.

[0046] Furthermore, the processing module is also used for:

[0047] Using an attention mechanism to obtain a channel domain weight matrix of the semantic edge feature map;

[0048] Obtain a semantic edge prediction map according to the semantic edge feature map and the channel domain weight matrix;

[0049] Determine a second loss function according to the semantic edge truth map and the semantic edge prediction map;

[0050] The parameters of the disease detection model are updated according to the second loss function.

[0051] Furthermore, the processing module is also used for:

[0052] After obtaining the first feature map, and before decoding the first feature map according to the low-frequency feature map and the semantic edge feature map, obtaining a spatial domain weight matrix of the first feature map by using an attention mechanism;

[0053] A second feature map is determined according to the first feature map and the spatial domain weight matrix.

[0054] Furthermore, the processing module is specifically used for:

[0055] The categories of the defects in the airport pavement defect image to be detected are cracks, repairs, patches, ground lights and board seams.

[0056] In a third aspect, the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for detecting airport pavement defects as described in the first aspect is implemented.

[0057] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for detecting airport pavement defects as described in the first aspect.

[0058] It can be seen from the above technical scheme that the method, device, electronic device and storage medium for airport pavement defect detection provided by the present invention realize pixel-level segmentation of various defects in airport pavement images through the defect detection model, thereby improving the segmentation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0060] Figure 1 The system framework of the method for detecting airport pavement defects provided by the present invention;

[0061] Figure 2 A schematic diagram of the flow chart of the method for detecting airport pavement defects provided by the present invention;

[0062] Figure 3 A schematic diagram of the flow chart of the method for detecting airport pavement defects provided by the present invention;

[0063] Figure 4 A schematic diagram of wavelet transform provided by the present invention;

[0064] Figure 5 A schematic diagram of a residual network provided by the present invention;

[0065] Figure 6 A schematic diagram of a residual network provided by the present invention;

[0066] Figure 7 A schematic diagram of the U-Net network provided by the present invention;

[0067] Figure 8 A schematic diagram of the U-Net network provided by the present invention;

[0068] Fig. 9 A schematic diagram of a device for detecting airport pavement defects provided by the present invention;

[0069] Fig.10 This is a schematic structural diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0070] The specific implementation of the present invention is further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0071] The method for detecting airport pavement defects provided by the embodiment of the present invention can be applied to Figure 1 In the system architecture shown, the system architecture includes an image sensor 100 and a disease detection model 200 .

[0072] Specifically, the image sensor 100 is used to obtain an image of airport pavement defects to be detected.

[0073] The defect detection model 200 is used to obtain the extraction result of the defect corresponding area in the airport pavement defect image to be detected after inputting the airport pavement defect image to be detected.

[0074] It should be noted that the trained defect detection model is obtained by training with different airport pavement defect images and semantic segmentation truth maps and semantic edge truth maps corresponding to different airport pavement defect images.

[0075] It should be noted that Figure 1 This is only an example of the system architecture of the embodiment of the present invention, and the present invention does not make any specific limitation to this.

[0076] Based on the system architecture shown above, Figure 2 A schematic diagram of a flow chart corresponding to a method for detecting airport pavement defects provided by an embodiment of the present invention is shown in FIG. Figure 2 As shown, the method includes:

[0077] Step 201, obtaining an image of the airport pavement to be detected.

[0078] Specifically, an image sensor such as a visible light camera is used to capture images of airport pavement defects in a real environment.

[0079] Step 202: Input the airport pavement defect image to be detected into the trained defect detection model to obtain the extraction result of the defect corresponding area in the airport pavement defect image to be detected.

[0080] It should be noted that the trained defect detection model is obtained by training with different airport pavement defect images and semantic segmentation truth maps and semantic edge truth maps corresponding to different airport pavement defect images.

[0081] The above scheme realizes pixel-level segmentation of various defects in airport pavement images through the defect detection model, thereby improving the segmentation accuracy.

[0082] In the embodiment of the present invention, before step 202, the steps are as follows: Figure 3As shown, the details are as follows:

[0083] Step 301, obtaining a disease training image, a semantic segmentation truth map, and a semantic edge truth map.

[0084] Specifically, the outer contour of the disease in the disease training image is extracted to obtain image annotation information;

[0085] Determine the semantic segmentation truth map based on the image annotation information;

[0086] The semantic edge of the diseased area is extracted based on the semantic segmentation truth map to obtain the semantic edge truth map.

[0087] In a possible implementation, LabelMe is used to manually annotate the defect training image to obtain a semantic segmentation truth map of the airport pavement defect image, and then Matlab is used to extract the corresponding semantic edge truth map therefrom.

[0088] In the embodiment of the present invention, the categories of the defects in the airport pavement defect image to be detected are cracks, repairs, patches, ground lights and board seams.

[0089] In addition to the three types of defects, namely cracks, repairs and patches, the above scheme also considers two common typical categories of airport pavements, namely ground lights and plate seams, to better detect airport pavements and provide more detailed information.

[0090] Step 302: Perform wavelet transform on the disease training image to obtain a low-frequency feature map.

[0091] In a possible implementation, Haar wavelet transform is performed on the disease training image.

[0092] Specifically, Figure 4 As shown, the input disease training image is subjected to a four-level Haar wavelet transform, and the low-frequency part generated by each level of wavelet transform is used as the effective frequency domain information and as the input of the next level of wavelet transform, thereby obtaining LL1, LL2, LL3, and LL4.

[0093] In the above scheme, the extraction of multi-scale low-frequency information reduces the noise in the complex background of the disease training image.

[0094] Step 303: Use a residual network to extract image features from the disease training image to obtain a semantic edge feature map.

[0095] In a possible implementation, a residual network ResNet101 is used to extract image features from disease training images.

[0096] Furthermore, the attention mechanism is used to obtain the channel domain weight matrix of the semantic edge feature map;

[0097] Obtain a semantic edge prediction map based on the semantic edge feature map and the channel domain weight matrix;

[0098] Determine a second loss function according to the semantic edge truth map and the semantic edge prediction map;

[0099] The parameters of the disease detection model are updated according to the second loss function.

[0100] like Figure 5 As shown in Figure 1, the outputs of the five parts of the residual network are feature maps c1, c2, c3, c4, and c5 of different scales.

[0101] Specifically, the output result of the first branch is obtained by convolution operation on c5. The second branch deconvolutions the four feature maps c1, c2, c3, and c5 to the same size, and splices them together, and finally outputs a rough semantic edge prediction result through convolution operation.

[0102] In a possible implementation, the semantic edge prediction map is compared with the semantic edge truth map to calculate the weighted cross entropy loss and perform back propagation.

[0103] In the embodiment of the present invention, the specific calculation formula of the second loss function is as follows:

[0104]

[0105] Where M represents the number of categories; w c represents the weight of category c; y ic is the symbol function; N is the number of samples; if the true category of sample i is equal to c, it takes 1, otherwise it takes 0; p ic It represents the predicted probability that the observed sample i belongs to category c.

[0106] The above scheme can accelerate the convergence of the network and reduce the influence of the noise contained in c1, c2, c3, and c4 on the prediction branch corresponding to c5.

[0107] The above solution alleviates the problem that semantic edges are small and few in number, and the proportion of foreground and background in the dataset is extremely unbalanced.

[0108] Furthermore, if Figure 6 As shown, since the diseased edges of the collected images account for a very small proportion and generally only occupy one pixel, in order to improve the prediction effect of the semantic edge, an embodiment of the present invention adds a channel domain attention mechanism after the c5 prediction branch, that is, the channel domain weight matrix of the feature map c5 is obtained through the global pooling layer and the fully connected layer, and the output after multiplying c5 by the weight matrix is ​​convolved to obtain the semantic edge segmentation result.

[0109] The above solution, based on the residual network, fuses the underlying features of the residual network, making the semantic segmentation results less affected by the underlying feature noise. The addition of the attention mechanism enables the network to better capture small edge features, thereby obtaining more accurate prediction results.

[0110] Step 304: Encode the disease training image based on the U-Net network according to the low-frequency feature map and the semantic edge feature map to obtain a first feature map.

[0111] like Figure 7 As shown in the figure, the encoding part is composed of four convolution blocks, where c1, c2, c3, and c5 are semantic edge feature maps, and LL1, LL2, LL3, and LL4 are low-frequency feature maps.

[0112] It should be noted that each convolution operation reduces the side length of the feature map by half.

[0113] For example, the input image size is 512*512, and the feature map size after four convolution operations is 32*32.

[0114] Step 305: decode the first feature map according to the low-frequency feature map and the semantic edge feature map to obtain a semantic segmentation prediction map.

[0115] Furthermore, the decoding part contains four deconvolution layers, and each deconvolution operation doubles the side length of the feature map. For example, if the input feature map size is 64*64, the feature map size after four deconvolution operations is 512*512.

[0116] Based on this, the disease training image passes through the U-Net network to obtain a pixel-level semantic segmentation prediction map of the same size.

[0117] Step 306, determining a first loss function according to the semantic segmentation truth map and the semantic segmentation prediction map.

[0118] In one possible implementation, a weighted cross entropy loss is calculated.

[0119] Step 307: Update the parameters of the disease detection model according to the first loss function to obtain a trained disease detection model.

[0120] The above scheme uses wavelet transform to extract low-frequency information of the image and fuses it through the U-Net network, which helps to eliminate noise and improve the accuracy of semantic segmentation. Based on the U-Net network, multi-scale frequency domain information features, semantic edge features and image encoder extraction features are fused to achieve high-precision pixel-level segmentation of various diseases in airport pavement images, improving the accuracy of semantic segmentation results and the generalization performance of the model.

[0121] Furthermore, in the embodiment of the present invention, before step 305, an attention mechanism is used to obtain a spatial domain weight matrix of the first feature map;

[0122] A second feature map is determined according to the first feature map and the spatial domain weight matrix.

[0123] Specifically, Figure 8 As shown in the figure, the encoding part is followed by a spatial domain attention mechanism, and the spatial domain weight matrix of the first feature map is obtained by convolution operation. After the first feature map is multiplied by the weight matrix, the weighted feature map is output, which is the second feature map.

[0124] The above scheme adds an attention mechanism between the encoder and the decoder to weight the fusion features, thereby improving the accuracy of the semantic segmentation results.

[0125] Based on the same inventive concept, Fig. 9 An airport pavement defect detection device provided by an embodiment of the present invention is exemplarily shown, and the device may be a process of an airport pavement defect detection method.

[0126] The device comprises:

[0127] An acquisition module 901 is used to acquire an airport pavement defect image to be detected;

[0128] Processing module 902 is used to input the airport pavement defect image to be detected into a trained defect detection model to obtain the extraction results of the corresponding areas of different types of defects in the airport pavement defect image to be detected; wherein the trained defect detection model is obtained after training using different airport pavement defect images and the semantic segmentation truth maps and semantic edge truth maps corresponding to different airport pavement defect images.

[0129] Furthermore, the processing module 902 is also used for:

[0130] Before inputting the airport pavement defect image to be detected into the trained defect detection model to obtain the extraction result of the defect corresponding area in the airport pavement defect image to be detected, obtaining the defect training image, the semantic segmentation truth map and the semantic edge truth map;

[0131] Performing wavelet transform on the disease training image to obtain a low-frequency feature map;

[0132] Using a residual network to extract image features from the disease training image to obtain a semantic edge feature map;

[0133] Encoding the disease training image according to the low-frequency feature map and the semantic edge feature map based on a U-Net network to obtain a first feature map;

[0134] Decoding the first feature map according to the low-frequency feature map and the semantic edge feature map to obtain a semantic segmentation prediction map;

[0135] Determine a first loss function according to the semantic segmentation true value map and the semantic segmentation prediction map;

[0136] The parameters of the disease detection model are updated according to the first loss function to obtain a trained disease detection model.

[0137] Furthermore, the processing module 902 is specifically used for:

[0138] Extracting the outer contour of the disease in the disease training image to obtain image annotation information;

[0139] Determine a semantic segmentation truth map according to the image annotation information;

[0140] The semantic edge of the diseased area is extracted based on the semantic segmentation truth map to obtain a semantic edge truth map.

[0141] Furthermore, the processing module 902 is also used for:

[0142] After extracting image features from the disease training image using the residual network to obtain a semantic edge feature map, an attention mechanism is used to obtain a channel domain weight matrix of the semantic edge feature map;

[0143] Obtain a semantic edge prediction map according to the semantic edge feature map and the channel domain weight matrix;

[0144] Determine a second loss function according to the semantic edge truth map and the semantic edge prediction map;

[0145] The parameters of the disease detection model are updated according to the second loss function.

[0146] Furthermore, the processing module 902 is also used for:

[0147] After obtaining the first feature map, and before decoding the first feature map according to the low-frequency feature map and the semantic edge feature map, obtaining a spatial domain weight matrix of the first feature map by using an attention mechanism;

[0148] A second feature map is determined according to the first feature map and the spatial domain weight matrix.

[0149] Furthermore, the processing module 902 is specifically used for:

[0150] The categories of the defects in the airport pavement defect image to be detected are cracks, repairs, patches, ground lights and board seams.

[0151] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, see Fig.10 , the electronic device specifically includes the following contents: a processor 1001, a memory 1002, a communication interface 1003 and a communication bus 1004;

[0152] The processor 1001, the memory 1002, and the communication interface 1003 communicate with each other through the communication bus 1004; the communication interface 1003 is used to realize information transmission between various devices;

[0153] The processor 1001 is used to call the computer program in the memory 1002. When the processor executes the computer program, all steps of the above-mentioned method for detecting airport pavement defects are implemented. For example, when the processor executes the computer program, the following steps are implemented: obtaining an airport pavement defect image to be detected; inputting the airport pavement defect image to be detected into a trained defect detection model to obtain an extraction result of the defect corresponding area in the airport pavement defect image to be detected; wherein the trained defect detection model is obtained after training using different airport pavement defect images and semantic segmentation truth maps and semantic edge truth maps corresponding to different airport pavement defect images.

[0154] Based on the same inventive concept, another embodiment of the present invention provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, all the steps of the above-mentioned method for detecting airport pavement defects are implemented. For example, when the processor executes the computer program, the following steps are implemented: obtaining an airport pavement defect image to be detected; inputting the airport pavement defect image to be detected into a trained defect detection model to obtain an extraction result of the defect corresponding area in the airport pavement defect image to be detected; wherein the trained defect detection model is obtained after training using different airport pavement defect images and semantic segmentation truth maps and semantic edge truth maps corresponding to different airport pavement defect images.

[0155] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, an airport pavement disease detection device, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0156] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of the present invention. Those of ordinary skill in the art may understand and implement them without creative effort.

[0157] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, an airport pavement disease detection device, or a network device, etc.) to execute the airport pavement disease detection method described in each embodiment or some parts of the embodiment.

[0158] In addition, in the present invention, terms such as "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0159] In addition, in the present invention, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0160] In addition, in the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0161] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting airport pavement defects, characterized in that: include: Acquire the airport pavement disease image to be detected; Acquire a disease training image, a semantic segmentation truth map, and a semantic edge truth map; perform wavelet transform on the disease training image to obtain a low-frequency feature map; use a residual network to extract image features on the disease training image to obtain a semantic edge feature map; encode the disease training image based on the low-frequency feature map and the semantic edge feature map based on a U-Net network to obtain a first feature map; decode the first feature map based on the low-frequency feature map and the semantic edge feature map to obtain a semantic segmentation prediction map; determine a first loss function based on the semantic segmentation truth map and the semantic segmentation prediction map; update the parameters of the disease detection model based on the first loss function to obtain a trained disease detection model; Inputting the airport pavement defect image to be detected into a trained defect detection model to obtain extraction results of regions corresponding to different types of defects in the airport pavement defect image to be detected; The trained defect detection model is obtained by training with different airport pavement defect images and semantic segmentation truth maps and semantic edge truth maps corresponding to different airport pavement defect images.

2. The method for detecting airport pavement defects according to claim 1, characterized in that: The obtaining of the semantic segmentation truth map and the semantic edge truth map includes: Extracting the outer contour of the disease in the disease training image to obtain image annotation information; Determine a semantic segmentation truth map according to the image annotation information; The semantic edge of the diseased area is extracted based on the semantic segmentation truth map to obtain a semantic edge truth map.

3. The method for detecting airport pavement defects according to claim 1, characterized in that: After extracting image features from the disease training image using the residual network to obtain a semantic edge feature map, the method further includes: Using an attention mechanism to obtain a channel domain weight matrix of the semantic edge feature map; Obtain a semantic edge prediction map according to the semantic edge feature map and the channel domain weight matrix; Determine a second loss function according to the semantic edge truth map and the semantic edge prediction map; The parameters of the disease detection model are updated according to the second loss function.

4. The method for detecting airport pavement defects according to claim 1, characterized in that: After obtaining the first feature map and before decoding the first feature map according to the low-frequency feature map and the semantic edge feature map, the method further includes: Using the attention mechanism to obtain the spatial domain weight matrix of the first feature map; A second feature map is determined according to the first feature map and the spatial domain weight matrix.

5. The method for detecting airport pavement defects according to claim 1, characterized in that: The categories of the defects in the airport pavement defect image to be detected are cracks, repairs, patches, ground lights and board seams.

6. A device for detecting airport pavement defects, characterized in that: include: An acquisition module is used to acquire the airport pavement disease image to be detected; Acquire a disease training image, a semantic segmentation truth map, and a semantic edge truth map; perform wavelet transform on the disease training image to obtain a low-frequency feature map; use a residual network to extract image features on the disease training image to obtain a semantic edge feature map; encode the disease training image based on the low-frequency feature map and the semantic edge feature map based on a U-Net network to obtain a first feature map; decode the first feature map based on the low-frequency feature map and the semantic edge feature map to obtain a semantic segmentation prediction map; determine a first loss function based on the semantic segmentation truth map and the semantic segmentation prediction map; update the parameters of the disease detection model based on the first loss function to obtain a trained disease detection model; A processing module is used to input the airport pavement defect image to be detected into a trained defect detection model to obtain the extraction results of the corresponding areas of different types of defects in the airport pavement defect image to be detected; wherein the trained defect detection model is obtained after training using different airport pavement defect images and the semantic segmentation truth maps and semantic edge truth maps corresponding to different airport pavement defect images.

7. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method according to any one of claims 1 to 5 are implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images