CN117670855A - RoadU-Net-based intelligent recognition and classification method for asphalt pavement diseases - Google Patents
RoadU-Net-based intelligent recognition and classification method for asphalt pavement diseases Download PDFInfo
- Publication number
- CN117670855A CN117670855A CN202311735898.2A CN202311735898A CN117670855A CN 117670855 A CN117670855 A CN 117670855A CN 202311735898 A CN202311735898 A CN 202311735898A CN 117670855 A CN117670855 A CN 117670855A
- Authority
- CN
- China
- Prior art keywords
- roadu
- image
- crack
- diseases
- net
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 201000010099 disease Diseases 0.000 title claims abstract description 93
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 title claims abstract description 93
- 239000010426 asphalt Substances 0.000 title claims abstract description 29
- 238000000034 method Methods 0.000 title claims abstract description 24
- 238000012549 training Methods 0.000 claims abstract description 36
- 230000006870 function Effects 0.000 claims abstract description 23
- 238000013136 deep learning model Methods 0.000 claims abstract description 22
- 238000012795 verification Methods 0.000 claims abstract description 21
- 230000008439 repair process Effects 0.000 claims abstract description 16
- 238000011158 quantitative evaluation Methods 0.000 claims abstract description 11
- 238000007781 pre-processing Methods 0.000 claims abstract description 10
- 238000000926 separation method Methods 0.000 claims abstract description 7
- 239000003086 colorant Substances 0.000 claims abstract description 4
- 238000004364 calculation method Methods 0.000 claims description 15
- 238000012937 correction Methods 0.000 claims description 6
- 230000007797 corrosion Effects 0.000 claims description 6
- 238000005260 corrosion Methods 0.000 claims description 6
- 238000001514 detection method Methods 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 5
- 230000004927 fusion Effects 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 238000007635 classification algorithm Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 claims description 3
- 230000003902 lesion Effects 0.000 claims description 3
- 230000000877 morphologic effect Effects 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention relates to an intelligent identification and classification method for asphalt pavement diseases based on RoadU-Net, which comprises the following steps: marking corresponding colors for crack diseases, strip repairs and pavement markings in the pavement image, and forming a training set and a verification set by using the marked images; preprocessing an image; setting a loss function, an optimizer and super parameters, constructing a RoadU-Net deep learning model, taking a training set sample as input, and finishing model training; preprocessing a pavement image, and then putting the pavement image into a trained RoadU-Net deep learning model to identify asphalt pavement diseases; regarding the prediction result of each channel as a binary image, and extracting a connected domain in the binary image by using a connected domain separation function aiming at crack diseases; when the circumscribed rectangles of the two connected domains belong to a crossing or containing relation, fusing the two circumscribed rectangles into a rectangle; classifying crack diseases in the processed image into longitudinal cracks, transverse cracks and crazes; and carrying out quantitative evaluation on different diseases.
Description
Technical Field
The invention belongs to the technical field of intelligent identification and classification of asphalt pavement diseases, and relates to an intelligent identification and classification method of asphalt pavement diseases based on RoadU-Net.
Background
The traditional detection method of the asphalt pavement diseases at present is to periodically carry out manual inspection, and has the defects of poor safety, strong subjectivity, low detection speed, low precision and the like. Along with the wide application of the multifunctional road surface detection vehicle, most of road management departments can collect road surface images at the running speed at present, and the road surface diseases are manually treated, extracted and counted. But the manual work in the industry is tedious, time consuming and prone to error.
Disclosure of Invention
In order to solve the technical problems, the invention aims to provide an intelligent identification and classification method for asphalt pavement diseases based on RoadU-Net.
The invention discloses an intelligent identification and classification method for asphalt pavement diseases based on RoadU-Net, which comprises the following steps:
step 1: respectively marking corresponding colors for crack diseases, strip repairs and pavement markings in the pavement image, and forming a training set and a verification set by using the marked images;
step 2: preprocessing images in the training set and the verification set, reconstructing the image size and adjusting the pixel value of the pixel point;
step 3: setting a loss function, an optimizer and super parameters, constructing a RoadU-Net deep learning model, taking a training set sample as input, and finishing model training;
step 4: preprocessing a pavement image, putting the preprocessed pavement image into a trained RoadU-Net deep learning model, and identifying asphalt pavement diseases to obtain a multichannel prediction result;
step 5: regarding the prediction result of each channel as a binary image, extracting a connected domain in each binary image by using a connected domain separation function aiming at crack diseases, and removing crack disease identification results with the number of connected domain pixels being less than 3000;
step 6: judging the position relation of the crack disease connected domains, and fusing two circumscribed rectangles into a rectangle when the circumscribed rectangles of the two connected domains belong to a crossed or contained relation;
step 7: classifying the crack diseases in the image processed in the step 6 into longitudinal cracks, transverse cracks and cracks by adopting a crack disease classification algorithm;
step 8: and carrying out quantitative evaluation on different types of crack diseases and strip repair diseases.
Further, the step 1 specifically includes:
step 1.1: acquiring a road surface image acquired by a road detection vehicle;
step 1.2: manually marking the identification target in each image, marking crack diseases as red, marking strip-shaped repair as blue, and marking pavement markings as green;
step 1.3: and constructing a database by the marked images, and screening the images from the database to respectively serve as a training set and a verification set.
Further, the step 2 specifically includes:
and reconstructing the image size of the images in the training set and the verification set to 2048×1024, adjusting the dynamic range of the images, and adjusting the pixel value of each pixel point to be between 0 and 1.
Further, the step 3 specifically includes:
step 3.1: building a RoadU-Net deep learning model;
step 3.2: setting a loss function: the cross entropy loss function is adopted, and specifically:
wherein y is i Representing the true class of the pixel, p i Representing the prediction probability of the pixel, n representing the total number of pixels;
step 3.3: setting an optimizer and super parameters: adopting an Adam optimizer, setting the size of a super-parameter batch to be 4, the learning rate to be 1e-3, the iteration number to be 300, and adopting default values for weight attenuation and momentum initial values;
step 3.4: training the RoadU-Net deep learning model by taking the training set as input according to the setting of the super parameters, calculating training errors by using label tensors and network output tensors according to the loss function, updating model parameters by using the training errors by an optimizer, verifying the identification effect of the round of model on the verification set, and storing the RoadU-Net model parameters with the best performance on the verification set when the performance of the model reaches a stable state, namely the loss function value and the identification accuracy of the verification set tend to converge.
Further, the RoadU-Net deep learning model built in the step 3.1 specifically includes:
the downsampling path of the model adopts a classical structure of a convolution network, two convolution kernels of 3×3 are repeatedly applied, and each convolution kernel is followed by a linear correction unit ReLU and a 2×2 maximum pooling layer to extract the characteristics of an input pavement image; each step in the upsampling path of the model includes: convolving the feature image using two 3 x 3 convolutions, each convolution kernel being followed by a linear correction unit ReLU, and deconvolving the feature image using a 2 x 2 deconvolution kernel; to obtain multi-scale features, after each deconvolution, the deconvolution feature layer is combined with the feature layer stored in the downsampling path as input to the next deconvolution, which facilitates minimization of back propagation loss during training; the feature vectors of 64 components are mapped into 4 channels using a 1×1 convolution kernel in the last layer, and finally a binarized lesion image having 4 channels, which is the same size as the input image, is output.
Further, the step 4 specifically includes:
and (3) preprocessing the road surface image, putting the road surface image into a trained RoadU-Net deep learning model, and obtaining prediction output through reasoning to obtain a binary image with four channels, wherein the 0 th channel, the 1 st channel, the 2 nd channel and the 3 rd channel are respectively the prediction results of the background, the crack diseases, the strip repair and the road surface marking.
Further, the step 6 specifically includes:
step 6.1: judging the position relation of the communicating domains of the crack diseases, and if the position relation is a cross relation, realizing the fusion of the communicating domains, wherein the height and width calculation method of the fused communicating domains is shown as the following formula:
H merge ＝H 1 +H 2 -ΔH
W merge ＝W 1 +W 2 -ΔW
wherein H is merge To merge the height of rectangle, H 1 Height of rectangle one, H 2 The height of the rectangle II is the height value of the overlapping of the rectangle I and the rectangle II; w (W) merge To merge the width of rectangle, W 1 Width of rectangle one, W 2 The width of the rectangle II is the width value of overlapping of the rectangle I and the rectangle delta W;
step 6.2: if the position relationship of the crack disease connected domains is an inclusion relationship, the connected domains are fused, and the height and width calculation method of the fused connected domains is shown as follows:
H merge ＝max(H 1 ,H 2 )
W merge ＝max(W 1 ,W 2 )
step 6.3: if the position relationship of the crack disease connected domain is a separation relationship, the connected domain is not treated.
Further, the step 7 specifically includes:
step 7.1: inputting the width and height of the fused rectangle, which are W and H respectively;
step 7.2: the aspect ratio is calculated, namely: ratio=w/H;
step 7.3: if Ratio >2.0 and H <0.5, defining the crack as a transverse crack;
step 7.4: if Ratio >0.5 and W <0.5, defining the crack as a longitudinal crack;
step 7.5: other cracks are defined as crazes.
Further, the step 8 specifically includes:
step 8.1: extracting the skeleton length of the longitudinal cracks, the transverse cracks and the strip-shaped repair for quantitative evaluation;
step 8.2: for cracks, quantitative evaluation was performed by obtaining the area of the connected-domain rectangular frame.
Further, the specific steps of extracting the skeleton length in the step 8.1 are as follows:
step 8.1.1: the connected domains in the longitudinal crack, the transverse crack and the strip-shaped repairing binary image can be regarded as a set A, and morphological calculation is carried out on each connected domain according to the following formula to obtain the complete skeleton of the disease:
wherein,for corrosion operation->For the operation of the open operation, B is a structural element and +.>Representing a succession of k times corrosion to A, i.e.>K is the last iteration step before a is eroded to empty set: />
Step 8.1.2: and taking one connected domain in the disease binary image as input, outputting the connected domain as a skeleton of the connected domain through the calculation of a skeleton extraction function, and calculating the skeleton length.
The intelligent identification and classification method for asphalt pavement diseases based on RoadU-Net has the following advantages:
1. the invention uses the RoadU-Net deep learning model to realize the pixel-level extraction of multi-category diseases of the asphalt pavement, can realize the rapid identification of the diseases of the asphalt pavement, and lightens the pressure of manual operation in the industry.
2. The method can realize classification of crack diseases, acquire characteristic information according to the crack types, and provide structural data for calculation of subsequent pavement technical condition indexes.
3. The method creatively uses the method of fusing the crack disease connected domain, so that the machine identification result is similar to the manual marking disease result of a road engineer, and the manual internal industry pressure is further reduced.
4. The method provided by the invention extracts and quantitatively evaluates the geometric information of the strip-shaped repairing diseases for the first time, and provides important experience for the subsequent identification research of the strip-shaped repairing diseases.
Drawings
FIG. 1 is a flow chart of an intelligent identification and classification method for asphalt pavement diseases based on RoadU-Net;
FIG. 2 is an exemplary diagram of a RoadU-Net deep learning model of the present invention;
FIG. 3 is a schematic representation of the present invention for communicating domain fusion with crack disease.
Detailed Description
As shown in FIG. 1, the intelligent identification and classification method for asphalt pavement diseases based on RoadU-Net comprises the following steps:
step 1: the method comprises the steps of respectively marking corresponding colors for crack diseases, strip repairs and pavement markings in pavement images, and forming a training set and a verification set by using the marked images, wherein the step 1 specifically comprises the following steps:
step 1.1: acquiring a road surface image acquired by a road detection vehicle;
step 1.2: manually marking the identification target in each image, marking crack diseases as red, marking strip-shaped repair as blue, and marking pavement markings as green;
step 1.3: and constructing a database by the marked images, and screening the images from the database to respectively serve as a training set and a verification set.
Step 2: the high-resolution large-size asphalt pavement image needs more calculation resources, and the wide dynamic range of the pavement image can cause difficulty in developing an intelligent recognition algorithm. For both reasons, each of the asphalt pavement images in the training set and the validation set needs to be preprocessed. And (2) reconstructing the image size and adjusting the pixel value of the pixel point, wherein the step (2) specifically comprises the following steps:
and reconstructing the image size of the images in the training set and the verification set to 2048×1024, adjusting the dynamic range of the images, and adjusting the pixel value of each pixel point to be between 0 and 1.
Step 3: setting a loss function, an optimizer and super parameters, constructing a RoadU-Net deep learning model, taking a training set sample as input, and finishing model training, wherein the step 3 specifically comprises the following steps:
step 3.1: building a RoadU-Net deep learning model as shown in FIG. 2;
the RoadU-Net deep learning model specifically comprises the following steps: the downsampling path of the model adopts a classical structure of a convolution network, two convolution kernels of 3×3 are repeatedly applied, and each convolution kernel is followed by a linear correction unit ReLU and a 2×2 maximum pooling layer to extract the characteristics of an input pavement image; each step in the upsampling path of the model includes: convolving the feature image using two 3 x 3 convolutions, each convolution kernel being followed by a linear correction unit ReLU, and deconvolving the feature image using a 2 x 2 deconvolution kernel; to obtain multi-scale features, after each deconvolution, the deconvolution feature layer is combined with the feature layer stored in the downsampling path as input to the next deconvolution, which facilitates minimization of back propagation loss during training; the feature vectors of 64 components are mapped into 4 channels using a 1×1 convolution kernel in the last layer, and finally a binarized lesion image having 4 channels, which is the same size as the input image, is output.
Step 3.2: setting a loss function: the cross entropy loss function is adopted, and specifically:
wherein y is i Representing the true class of the pixel, p i Representing the prediction probability of the pixel, n representing the total number of pixels;
step 3.3: setting an optimizer and super parameters: an Adam optimizer is employed that automatically adjusts the learning rate (LearnRate) by estimating the first and second moments of the previous gradient. The Adam optimizer would maintain an average gradient for each parameter and an exponentially weighted average of its squares, and update the learning rate with the above information. Setting the size of the super-parameter batch to be 4, the learning rate to be 1e-3, the iteration number to be 300, and adopting default values for weight attenuation and momentum initial values;
step 3.4: training the RoadU-Net deep learning model by taking the training set as input according to the setting of the super parameters, calculating training errors by using label tensors and network output tensors according to the loss function, updating model parameters by using the training errors by an optimizer, verifying the identification effect of the round of model on the verification set, and storing the RoadU-Net model parameters with the best performance on the verification set when the performance of the model reaches a stable state, namely the loss function value and the identification accuracy of the verification set tend to converge.
Step 4: preprocessing a pavement image, putting the pavement image into a trained RoadU-Net deep learning model, and identifying asphalt pavement diseases to obtain a multi-channel prediction result, wherein the step 4 specifically comprises the following steps:
and (3) preprocessing the road surface image, putting the road surface image into a trained RoadU-Net deep learning model, and obtaining prediction output through reasoning to obtain a binary image with four channels, wherein the 0 th channel, the 1 st channel, the 2 nd channel and the 3 rd channel are respectively the prediction results of the background, the crack diseases, the strip repair and the road surface marking.
Step 5: regarding the prediction result of each channel as a binary image, extracting a connected domain in each binary image by using a connected domain separation function aiming at crack diseases, and removing crack disease recognition results with the number of pixel points of the connected domain being less than 3000 in order to improve the crack disease recognition accuracy;
step 6: judging the position relation of the crack disease connected domains, and fusing two circumscribed rectangles into a rectangle when the circumscribed rectangles of the two connected domains belong to a crossed or contained relation, wherein the step 6 specifically comprises:
step 6.1: judging the position relation of the communicating domains of the crack diseases, and if the position relation is a cross relation, realizing the fusion of the communicating domains, wherein the height and width calculation method of the fused communicating domains is shown as the following formula:
H merge ＝H 1 +H 2 -ΔH
W merge ＝W 1 +W 2 -ΔW
wherein H is merge To merge the height of rectangle, H 1 Height of rectangle one, H 2 The height of the rectangle II is the height value of the overlapping of the rectangle I and the rectangle II; w (W) merge To merge the width of rectangle, W 1 Width of rectangle one, W 2 The width of the rectangle II is the width value of overlapping of the rectangle I and the rectangle delta W;
step 6.2: if the position relationship of the crack disease connected domains is an inclusion relationship, the connected domains are fused, and the height and width calculation method of the fused connected domains is shown as follows:
H merge ＝max(H 1 ,H 2 ) (3)
W merge ＝max(W 1 ,W 2 )
step 6.3: if the position relationship of the crack disease connected domain is a separation relationship, the connected domain is not treated.
The fused result is shown in fig. 3, and the result after rectangular fusion calculation is closer to the subjective judgment result of the road engineer.
Step 7: classifying the crack diseases in the image processed in the step 6 into longitudinal cracks, transverse cracks and crazes by adopting a crack disease classification algorithm, wherein the step 7 specifically comprises the following steps:
step 7.1: inputting the width and height of the fused rectangle, which are W and H respectively;
step 7.2: the aspect ratio is calculated, namely: ratio=w/H;
step 7.3: if Ratio >2.0 and H <0.5, defining the crack as a transverse crack;
step 7.4: if Ratio >0.5 and W <0.5, defining the crack as a longitudinal crack;
step 7.5: other cracks are defined as crazes.
Step 8: based on the disease classification result, quantitative evaluation is carried out on different types of crack diseases and strip-shaped repairing diseases, wherein the step 8 specifically comprises the following steps:
step 8.1: extracting the skeleton length of the longitudinal cracks, the transverse cracks and the strip-shaped repair for quantitative evaluation;
in specific implementation, the specific extraction of the skeleton length is as follows:
step 8.1.1: the connected domains in the longitudinal crack, the transverse crack and the strip-shaped repairing binary image can be regarded as a set A, and morphological calculation is carried out on each connected domain according to the following formula to obtain the complete skeleton of the disease:
wherein,for corrosion operation->For the operation of the open operation, B is a structural element and +.>Representing a succession of k times corrosion to A, i.e.>K is the last iteration step before a is eroded to empty set: />
Step 8.1.2: and taking one connected domain in the disease binary image as input, outputting the connected domain as a skeleton of the connected domain through the calculation of a skeleton extraction function, and calculating the skeleton length.
Step 8.2: for cracks, quantitative evaluation was performed by obtaining the area of the connected-domain rectangular frame.
And finally, carrying out structural storage on intelligent identification results of all diseases. The recognition result in the text file includes six parts: 1) The type of road surface; 2) Disease type; 3) Disease severity; 4) Corner information of the disease circumscribed rectangular frame; 5) Disease quantitative evaluation information (transverse and longitudinal cracks and strip repair are skeleton length, and cracks are area); 6) The picture name where the disease is located. And outputting a disease identification result image. Finally, the data structuring and visualization of the disease recognition result are realized, and a refined data support is provided for subsequent road surface technical condition evaluation.
The foregoing description of the preferred embodiments of the invention is not intended to limit the scope of the invention, but rather to enable any modification, equivalent replacement, improvement or the like to be made without departing from the spirit and principles of the invention.
Claims (10)
1. The intelligent identification and classification method for asphalt pavement diseases based on RoadU-Net is characterized by comprising the following steps:
step 1: respectively marking corresponding colors for crack diseases, strip repairs and pavement markings in the pavement image, and forming a training set and a verification set by using the marked images;
step 2: preprocessing images in the training set and the verification set, reconstructing the image size and adjusting the pixel value of the pixel point;
step 3: setting a loss function, an optimizer and super parameters, constructing a RoadU-Net deep learning model, taking a training set sample as input, and finishing model training;
step 4: preprocessing a pavement image, putting the preprocessed pavement image into a trained RoadU-Net deep learning model, and identifying asphalt pavement diseases to obtain a multichannel prediction result;
step 5: regarding the prediction result of each channel as a binary image, extracting a connected domain in each binary image by using a connected domain separation function aiming at crack diseases, and removing crack disease identification results with the number of connected domain pixels being less than 3000;
step 6: judging the position relation of the crack disease connected domains, and fusing two circumscribed rectangles into a rectangle when the circumscribed rectangles of the two connected domains belong to a crossed or contained relation;
step 7: classifying the crack diseases in the image processed in the step 6 into longitudinal cracks, transverse cracks and cracks by adopting a crack disease classification algorithm;
step 8: and carrying out quantitative evaluation on different types of crack diseases and strip repair diseases.
2. The intelligent identification and classification method for asphalt pavement diseases based on RoadU-Net according to claim 1, wherein the step 1 is specifically as follows:
step 1.1: acquiring a road surface image acquired by a road detection vehicle;
step 1.2: manually marking the identification target in each image, marking crack diseases as red, marking strip-shaped repair as blue, and marking pavement markings as green;
step 1.3: and constructing a database by the marked images, and screening the images from the database to respectively serve as a training set and a verification set.
3. The intelligent identification and classification method for asphalt pavement diseases based on RoadU-Net according to claim 1, wherein the step 2 is specifically as follows:
and reconstructing the image size of the images in the training set and the verification set to 2048×1024, adjusting the dynamic range of the images, and adjusting the pixel value of each pixel point to be between 0 and 1.
4. The intelligent identification and classification method for asphalt pavement diseases based on RoadU-Net according to claim 1, wherein the step 3 is specifically as follows:
step 3.1: building a RoadU-Net deep learning model;
step 3.2: setting a loss function: the cross entropy loss function is adopted, and specifically:
wherein y is i Representing the true class of the pixel, p i Representing the prediction probability of the pixel, n representing the total number of pixels;
step 3.3: setting an optimizer and super parameters: adopting an Adam optimizer, setting the size of a super-parameter batch to be 4, the learning rate to be 1e-3, the iteration number to be 300, and adopting default values for weight attenuation and momentum initial values;
step 3.4: training the RoadU-Net deep learning model by taking the training set as input according to the setting of the super parameters, calculating training errors by using label tensors and network output tensors according to the loss function, updating model parameters by using the training errors by an optimizer, verifying the identification effect of the round of model on the verification set, and storing the RoadU-Net model parameters with the best performance on the verification set when the performance of the model reaches a stable state, namely the loss function value and the identification accuracy of the verification set tend to converge.
5. The intelligent identification and classification method for asphalt pavement diseases based on RoadU-Net according to claim 4, wherein the RoadU-Net deep learning model built in the step 3.1 is specifically:
the downsampling path of the model adopts a classical structure of a convolution network, two convolution kernels of 3×3 are repeatedly applied, and each convolution kernel is followed by a linear correction unit ReLU and a 2×2 maximum pooling layer to extract the characteristics of an input pavement image; each step in the upsampling path of the model includes: convolving the feature image using two 3 x 3 convolutions, each convolution kernel being followed by a linear correction unit ReLU, and deconvolving the feature image using a 2 x 2 deconvolution kernel; to obtain multi-scale features, after each deconvolution, the deconvolution feature layer is combined with the feature layer stored in the downsampling path as input to the next deconvolution, which facilitates minimization of back propagation loss during training; the feature vectors of 64 components are mapped into 4 channels using a 1×1 convolution kernel in the last layer, and finally a binarized lesion image having 4 channels, which is the same size as the input image, is output.
6. The intelligent identification and classification method for asphalt pavement diseases based on RoadU-Net according to claim 1, wherein the step 4 is specifically:
and (3) preprocessing the road surface image, putting the road surface image into a trained RoadU-Net deep learning model, and obtaining prediction output through reasoning to obtain a binary image with four channels, wherein the 0 th channel, the 1 st channel, the 2 nd channel and the 3 rd channel are respectively the prediction results of the background, the crack diseases, the strip repair and the road surface marking.
7. The intelligent identification and classification method for asphalt pavement diseases based on RoadU-Net according to claim 1, wherein the step 6 is specifically:
step 6.1: judging the position relation of the communicating domains of the crack diseases, and if the position relation is a cross relation, realizing the fusion of the communicating domains, wherein the height and width calculation method of the fused communicating domains is shown as the following formula:
H merge ＝H 1 +H 2 -ΔH
W merge ＝W 1 +W 2 -ΔW
wherein H is merge To merge the height of rectangle, H 1 Height of rectangle one, H 2 The height of the rectangle II is the height value of the overlapping of the rectangle I and the rectangle II; w (W) merge To merge the width of rectangle, W 1 Width of rectangle one, W 2 The width of the rectangle II is the width value of overlapping of the rectangle I and the rectangle delta W;
step 6.2: if the position relationship of the crack disease connected domains is an inclusion relationship, the connected domains are fused, and the height and width calculation method of the fused connected domains is shown as follows:
H merge ＝max(H 1 ,H 2 )
W merge ＝max(W 1 ,W 2 )
step 6.3: if the position relationship of the crack disease connected domain is a separation relationship, the connected domain is not treated.
8. The intelligent identification and classification method for asphalt pavement diseases based on RoadU-Net according to claim 1, wherein the step 7 is specifically:
step 7.1: inputting the width and height of the fused rectangle, which are W and H respectively;
step 7.2: the aspect ratio is calculated, namely: ratio=w/H;
step 7.3: if Ratio >2.0 and H <0.5, defining the crack as a transverse crack;
step 7.4: if Ratio >0.5 and W <0.5, defining the crack as a longitudinal crack;
step 7.5: other cracks are defined as crazes.
9. The intelligent identification and classification method for asphalt pavement diseases based on RoadU-Net according to claim 1, wherein the step 8 is specifically:
step 8.1: extracting the skeleton length of the longitudinal cracks, the transverse cracks and the strip-shaped repair for quantitative evaluation;
step 8.2: for cracks, quantitative evaluation was performed by obtaining the area of the connected-domain rectangular frame.
10. The intelligent identification and classification method for asphalt pavement diseases based on RoadU-Net according to claim 9, wherein the specific steps of extracting the skeleton length in the step 8.1 are as follows:
step 8.1.1: the connected domains in the longitudinal crack, the transverse crack and the strip-shaped repairing binary image can be regarded as a set A, and morphological calculation is carried out on each connected domain according to the following formula to obtain the complete skeleton of the disease:
wherein,for corrosion operation->For the operation of the open operation, B is a structural element and +.>Representing a succession of k times corrosion to A, i.e.>K is the last iteration step before a is eroded to empty set:
step 8.1.2: and taking one connected domain in the disease binary image as input, outputting the connected domain as a skeleton of the connected domain through the calculation of a skeleton extraction function, and calculating the skeleton length.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311735898.2A CN117670855A (en) | 2023-12-18 | 2023-12-18 | RoadU-Net-based intelligent recognition and classification method for asphalt pavement diseases |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311735898.2A CN117670855A (en) | 2023-12-18 | 2023-12-18 | RoadU-Net-based intelligent recognition and classification method for asphalt pavement diseases |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117670855A true CN117670855A (en) | 2024-03-08 |
Family
ID=90064034
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311735898.2A Pending CN117670855A (en) | 2023-12-18 | 2023-12-18 | RoadU-Net-based intelligent recognition and classification method for asphalt pavement diseases |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117670855A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118038283A (en) * | 2024-04-15 | 2024-05-14 | 贵州黔通工程技术有限公司 | Method and equipment for detecting hidden diseases of asphalt pavement |
-
2023
- 2023-12-18 CN CN202311735898.2A patent/CN117670855A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118038283A (en) * | 2024-04-15 | 2024-05-14 | 贵州黔通工程技术有限公司 | Method and equipment for detecting hidden diseases of asphalt pavement |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109977793B (en) | Roadside image pedestrian segmentation method based on variable-scale multi-feature fusion convolutional network | |
| CN109446992B (en) | Remote sensing image building extraction method and system based on deep learning, storage medium and electronic equipment | |
| CN111325203B (en) | American license plate recognition method and system based on image correction | |
| Li et al. | Automatic crack recognition for concrete bridges using a fully convolutional neural network and naive Bayes data fusion based on a visual detection system | |
| CN110070008B (en) | Bridge disease identification method adopting unmanned aerial vehicle image | |
| CN109035273B (en) | Image signal fast segmentation method of immunochromatography test paper card | |
| CN111091541B (en) | Method for identifying fault of missing nut in cross beam assembly of railway wagon | |
| CN114998852A (en) | Intelligent detection method for road pavement diseases based on deep learning | |
| CN110751644B (en) | Road surface crack detection method | |
| CN117670855A (en) | RoadU-Net-based intelligent recognition and classification method for asphalt pavement diseases | |
| CN112819748B (en) | Training method and device for strip steel surface defect recognition model | |
| CN116029980A (en) | Asphalt pavement damage detection and evaluation method based on improved SegNet | |
| CN113762265B (en) | Classified segmentation method and system for pneumonia | |
| CN114627106A (en) | Weld defect detection method based on Cascade Mask R-CNN model | |
| CN113240623B (en) | Pavement disease detection method and device | |
| CN111882620A (en) | Road drivable area segmentation method based on multi-scale information | |
| CN111126127A (en) | High-resolution remote sensing image classification method guided by multi-level spatial context characteristics | |
| CN115546768A (en) | Pavement marking identification method and system based on multi-scale mechanism and attention mechanism | |
| CN106570503A (en) | Method and system for identifying vehicle body color | |
| CN114049538A (en) | Airport crack image confrontation generation method based on UDWGAN + + network | |
| CN115170479A (en) | Automatic extraction method for asphalt pavement repairing diseases | |
| CN116012291A (en) | Industrial part image defect detection method and system, electronic equipment and storage medium | |
| CN113326846A (en) | Rapid bridge apparent disease detection method based on machine vision | |
| CN112489026A (en) | Asphalt pavement disease detection method based on multi-branch parallel convolution neural network | |
| CN112270370B (en) | Vehicle apparent damage assessment method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |