CN115861324B - Pavement damage maintenance cost prediction method and system - Google Patents
Pavement damage maintenance cost prediction method and system Download PDFInfo
- Publication number
- CN115861324B CN115861324B CN202310181026.XA CN202310181026A CN115861324B CN 115861324 B CN115861324 B CN 115861324B CN 202310181026 A CN202310181026 A CN 202310181026A CN 115861324 B CN115861324 B CN 115861324B
- Authority
- CN
- China
- Prior art keywords
- model
- pavement
- rdaacn
- convolution layer
- convolution
- 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.)
- Active
Links
Images
Landscapes
- Measuring And Recording Apparatus For Diagnosis (AREA)
- Image Analysis (AREA)
Abstract
A pavement damage maintenance cost prediction method and system, the method includes: step S1: acquiring a traffic pavement image dataset, constructing a training set and a testing set for training and testing a network model by utilizing semantic analysis, and preprocessing images in the training set and the testing set; step S2: constructing an RDAACN model; step S3: training the RDAACN model on a training set; step S4: testing the trained RDAACN model on the test set, and outputting a test result; step S5: calculating the area of a pavement damage area according to the extraction result of the RDAACN model on the pavement damage pixel level precision in the pavement image; step S6: and predicting the maintenance cost of the pavement damage according to the calculated area of the pavement damage area. The system is used for implementing the method. The method has the advantages of simple principle, high intelligent degree, capability of improving prediction efficiency and accuracy and the like.
Description
Technical Field
The invention mainly relates to the technical field of intelligent monitoring of pavement diseases, in particular to a pavement disease maintenance cost prediction method and system.
Background
The road transportation is flexible and convenient, the national road transportation system is continuously perfected, the economic high-speed development is promoted, and the requirements on road operation maintenance are also continuously increased. The pavement can generate diseases such as transverse cracks, longitudinal cracks, crazes and the like due to factors such as increased service time, climate influence, improper use and maintenance modes and the like, and if the pavement diseases are not maintained in time, the service performance and the safety performance of road transportation can be affected. Therefore, it is very important to extract road surface diseases rapidly and accurately, and to improve the reliability of road transportation.
Especially for smart cities, the key infrastructure components and services formed by cities such as urban management, education, medical treatment, real estate, transportation, public utilities, public safety and the like are more interconnected, efficient and intelligent by the application of intelligent computing technologies such as Internet of things, cloud computing, big data, space geographic information integration and the like in the fields of urban planning, design, construction, management, operation and the like through technical means, so that better life and work services are provided for citizens, a more favorable commercial development environment is created for enterprises, and a more efficient operation and management mechanism is provided for governments.
The method of manually extracting road surface diseases is time consuming and the results are susceptible to subjective experience. The image segmentation method based on the traditional features is used for segmenting the traffic road surface image according to the features such as threshold features, similarity features, regional edge features and the like extracted by an algorithm to extract road surface diseases, but the method generally needs prior information and has general segmentation accuracy for the traffic road surface image with a complex background, so that the method for extracting the road surface diseases has limitations. The semantic segmentation method based on the deep learning extracts the characteristics of the traffic road surface image according to the deep learning network model, and carries out category prediction on each pixel point in the image so as to segment the traffic road surface image, thereby extracting the road surface diseases in the image with pixel-level precision.
In recent years, semantic segmentation methods based on deep learning have been developed rapidly, and the full convolution network (Fully Convolutional Networks, FCN) proposed by Long et al and the SegNet proposed by U-Net, badrinarayanan et al proposed by Ronneberger et al have successively achieved excellent performances. However, the existing method has the problems of large model parameter and complex calculation.
For example, chinese patent application "method, apparatus and computer device for training a model for detecting road surface diseases" (CN 112966665 a), in which the technical scheme includes: acquiring a plurality of road surface images; detecting a disease image in the plurality of road surface images; the disease image is a pavement image with diseases; the number of the disease images is multiple; marking the disease image according to the disease, and generating a mask image of the disease image according to marking content; obtaining a training data set of the pavement disease detection model according to the disease image and the mask image of the disease image; training the pavement disease detection model based on the training data set to obtain a trained pavement disease detection model; the trained pavement disease detection model is used for pavement disease detection of the pavement image to be detected. By adopting the method, all road surface diseases in the road surface image to be detected can be accurately identified.
However, all the existing intelligent image recognition modes have the problems of low efficiency, low recognition accuracy and the like, and cannot realize real unmanned, intelligent and reliable recognition. Therefore, it is necessary to provide a method for calculating the area of the road surface defect region from the result of semantic segmentation of the traffic road surface image and predicting the maintenance cost of the road surface defect according to the area of the road surface defect region.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: aiming at the technical problems existing in the prior art, the invention provides the pavement disease maintenance cost prediction method and system which are simple in principle, high in intelligent degree and capable of improving prediction efficiency and accuracy.
In order to solve the technical problems, the invention adopts the following technical scheme:
a pavement damage repair cost prediction method, comprising:
step S1: acquiring a traffic pavement image dataset, firstly utilizing semantic analysis to construct a training set and a testing set for training and testing a network model, and preprocessing images in the training set and the testing set;
step S2: constructing an RDAACN model, wherein the RDAACN model is provided with an encoder structure and a decoder structure; the encoder structure is used for extracting the characteristics of the traffic pavement image, and the decoder is used for outputting the segmentation result of the image according to the characteristics of the traffic pavement image extracted by the encoder;
step S3: training the RDAACN model on a training set;
step S4: the trained RDAACN model is stored, the trained RDAACN model is tested on a test set, a test result is output, and road surface diseases in traffic road surface images are extracted with pixel-level precision;
step S5: calculating the area of a pavement disease area according to the extraction result of the RDAACN model on the pavement disease pixel level precision in the traffic pavement image;
step S6: and predicting the maintenance cost of the pavement damage according to the calculated area of the pavement damage area.
As a further improvement of the process of the invention: in the step S1, the traffic road surface image data set is a GAPs384 data set obtained by processing the asphalt road surface damage data set GAPs.
As a further improvement of the process of the invention: each pixel in each image in the GAPs384 dataset represents 1.2mm x 1.2mm; the image is cropped to 528 pixels in width and 432 pixels in height.
As a further improvement of the process of the invention: in the step S2, a constructed residual dense asymmetric cavity convolution network RDAACN model is constructed, wherein the RDAACN model comprises an encoder and a decoder; the encoder of the RDAACN model has four residual dense asymmetric hole convolution blocks RDAACM connected in sequence.
As a further improvement of the process of the invention: the RDAACN model adopts an asymmetric convolution kernel, and the feature map of the input RDAACM is processed by a 3X 3 convolution layer and then added into the RDAACM model in a residual connection mode.
As a further improvement of the process of the invention: the connection structures of the convolution layers, the hole convolution layers and the pooling layers in RDAACB1 to RDAACB4 in the RDAACB model are the same, but the number of convolution kernels is different.
As a further improvement of the process of the invention: the RDAACB model comprises the following structures:
(a) The original characteristic diagram is input into a convolution kernel with the quantity ofN e 2, expansion ratiod1×3 hole convolution layer using modified linear unit ReLU for 3, activation function, the output feature map of 1×3 hole convolution layer is input to a convolution kernel of numberN e 2, expansion ratiodFor 3, 3×1 hole convolution layer of ReLU is used as activation function, the output characteristic diagram of 3×1 hole convolution layer is input into a convolution kernel with the quantity ofN e 2, expansion ratiod3×3 hole convolution layer using ReLU for 6, activation function;
(b) The 3X 1 hole convolution layer and the 3X 3 hole convolution layer output feature images are subjected to feature fusion in a channel merging mode, and the feature images obtained after fusion are input into a convolution kernel with the quantity ofN e The activation function uses a 1 multiplied by 1 convolution layer of Sigmoid, and the output feature map of the 1 multiplied by 1 convolution layer is multiplied by a feature map obtained by feature fusion;
(c) The original characteristic diagram is input into a convolution kernel with the quantity ofN e The activation function uses a 3×3 convolution layer of the ReLU, and the output feature map of the 3×3 convolution layer is added to the feature map obtained by multiplying in (b);
(d) The feature map obtained by adding is input into a step length of a pooling window which is 2 multiplied by 2S p In the maximum pooling layer of 2, reducing the size of the feature map to 0.5 times of the original feature map to obtain an output feature map of RDAACB;
as a further improvement of the process of the invention: the decoder of the RDAACN is used for outputting the segmentation result of the image according to the characteristics of the traffic pavement image extracted by the encoder, and the flow comprises the following steps:
step S10: the feature map output by RDAACB4 is input into a 3X 3 convolution layer with 512 convolution kernels and an activation function of ReLU, and 4 times up-sampling is carried out on the output feature map of the 3X 3 convolution layer by bilinear interpolation, so that the size of the feature map is enlarged by 4 times;
step S20: the feature images output by the first 4 times of up-sampling are input into a 3X 3 convolution layer with the number of convolution kernels being 64 and an activation function using ReLU, the output feature images of the convolution layer are added with the output feature images of RDAACB2, 4 times of up-sampling is carried out on the added feature images by using bilinear interpolation, and the size of the feature images is enlarged by 4 times;
step S30: the feature map output by the up sampling of the second time is input into a 3 multiplied by 3 convolution layer with the number of convolution kernels being 32 and the activation function using ReLU, the output of the 3 multiplied by 3 convolution layer is input into a 1 multiplied by 1 convolution layer with the number of convolution kernels being 2, and the result of dividing the traffic road surface image output by RDAACN is obtained.
As a further improvement of the process of the invention: in the step S5, the method for calculating the area of the pavement damage area includes:
step S501: calculating the number of pixels of the RDAACN model prediction belonging to the road surface disease category in one traffic road surface image according to the result of the RDAACN model segmentation of the traffic road surface imageM；
Step S502: obtaining the actual physical area corresponding to each pixel point in the traffic road surface imageL；
Step S503: calculating the area of a road surface disease area in a traffic road surface imageKThe calculation formula is as follows:
the invention further provides a pavement disease maintenance cost prediction system based on semantic segmentation, which comprises:
the image data acquisition unit is used for constructing a training set and a testing set for training and testing the network model by utilizing semantic analysis, and preprocessing images in the training set and the testing set;
a training unit to build a RDAACN model having an encoder structure and a decoder structure; the encoder structure is used for extracting the characteristics of the traffic pavement image, and the decoder is used for outputting the segmentation result of the image according to the characteristics of the traffic pavement image extracted by the encoder; training the RDAACN model on a training set;
the test unit is used for storing the trained RDAACN model, testing the trained RDAACN model on a test set, outputting a test result, and accurately extracting road surface diseases in the traffic road surface image at a pixel level;
the SVM model generating unit is used for calculating the area of the pavement disease area according to the extraction result of the RDAACN model on the pavement disease pixel level precision in the traffic pavement image;
the prediction unit is used for predicting the maintenance cost of the pavement diseases by using the trained SVM model according to the extraction result of the improved U-Net model on the pavement disease pixel level precision in the traffic pavement image of the test set;
and the output unit is used for outputting the maintenance cost of the pavement damage according to the calculated area of the pavement damage area.
Compared with the prior art, the invention has the advantages that:
the pavement damage maintenance cost prediction method and system are simple in principle, high in intelligent degree and capable of improving prediction efficiency and accuracy. The invention innovatively adopts the RDAACN model, the RDAACM constructed in the RDAACN model has dense hole convolution, the obtained receptive field is large, the pixel sampling is dense, and meanwhile, an asymmetric convolution kernel is used for reducing the quantity of parameters, and the RDAACM can retain the information of an input feature map; the decoder of RDAACN outputs an extraction result of the pixel-level precision of the road surface diseases in the image according to the image features extracted by RDAACB, thereby calculating the area of the road surface disease area and predicting the cost for maintaining the road surface diseases according to the area of the road surface disease area.
Drawings
FIG. 1 is a schematic flow chart of the method of the invention in a specific example.
Fig. 2 is a schematic diagram of the structural principle of the rdaac n model in a specific application example of the present invention.
Fig. 3 is a schematic diagram of the structural principle of the RDAACB proposed by the present invention.
Detailed Description
The invention will be described in further detail with reference to the drawings and the specific examples.
As shown in fig. 1, the pavement damage repair cost prediction method of the present invention is a pavement damage repair cost prediction method based on semantic segmentation and an rdacn model, and the flow of the method includes:
step S1: acquiring a traffic pavement image dataset, firstly utilizing semantic analysis to construct a training set and a testing set for training and testing a network model, and preprocessing images in the training set and the testing set;
step S2: constructing an RDAACN model, wherein the RDAACN model is provided with an encoder structure and a decoder structure; the encoder structure is used for extracting the characteristics of the traffic pavement image, and the decoder is used for outputting the segmentation result of the image according to the characteristics of the traffic pavement image extracted by the encoder;
step S3: training the RDAACN model on a training set;
step S4: the trained RDAACN model is stored, the trained RDAACN model is tested on a test set, a test result is output, and road surface diseases in traffic road surface images are extracted with pixel-level precision;
step S5: calculating the area of a pavement disease area according to the extraction result of the RDAACN model on the pavement disease pixel level precision in the traffic pavement image;
step S6: and predicting the maintenance cost of the pavement damage according to the calculated area of the pavement damage area.
In a specific application example, in the step S1, the traffic road surface image data set may be a GAPs384 data set obtained by processing an asphalt road surface damage data set (German Asphalt Pavement Distress, GAPs) according to actual needs.
Further, the GAPs384 dataset has 509 road surface disease images and corresponding image annotations. As a preferred embodiment, each pixel point in each image of the present invention represents 1.2mm×1.2mm in order to improve accuracy and effect of recognition.
As a preferred embodiment, for training and testing of the RDAACN model, 470 images in the GAPs384 data set are selected as a training set, and 39 images are selected as a testing set. Further, to improve the efficiency and accuracy of downsampling and upsampling of images by the network model, the present invention also clips the images to 528 pixels in width and 432 pixels in height.
In a specific application example, in the step S2, the structure of the residual dense asymmetric cavity convolution network (Residual DenseAsymmetric Atrous Convolution Network, rdaac n) further constructed by the present invention is shown in fig. 2. In the present invention, the RDAACN model includes an encoder and a decoder. The encoder of the RDAACN model has four serially connected residual dense asymmetric hole convolution blocks (Residual DenseAsymmetric Atrous Convolution Block, RDAACB); with this design, the RDAACB model has dense hole convolution, and the obtained receptive field is large and the pixels are densely sampled.
Furthermore, the invention also uses an asymmetric convolution kernel to reduce the parameter quantity, and adds the feature map input into the RDAACB model in a residual connection mode after the feature map input into the RDAACB is processed by a 3X 3 convolution layer to keep the image features ignored by the cavity convolution.
Further, as a preferred example, in order to improve accuracy of recognition and processing efficiency, the connection structures of the convolution layers, the hole convolution layers, and the pooling layers in the rdaac b1 to rdaac b4 in the present invention are the same, but the number of convolution kernels is different.
Further, as a preferred embodiment, as shown in fig. 3, the structure of the rdaac b includes:
(a) The original characteristic diagram is input into a convolution kernel with the quantity ofN e 2, expansion ratiod1×3 hole convolution layer using modified linear units (Rectified Linear Unit, reLU) for 3, activation function, the output feature map of the 1×3 hole convolution layer being input to a convolution kernel of numberN e 2, expansion ratiodFor 3, 3×1 hole convolution layer of ReLU is used as activation function, the output characteristic diagram of 3×1 hole convolution layer is input into a convolution kernel with the quantity ofN e 2, expansion ratiod3×3 hole convolution layer using ReLU for 6, activation function;
(b) The 3X 1 hole convolution layer and the 3X 3 hole convolution layer output feature images are subjected to feature fusion in a channel merging mode, and the feature images obtained after fusion are input into a convolution kernel with the quantity ofN e The activation function uses a 1 multiplied by 1 convolution layer of Sigmoid, and the output feature map of the 1 multiplied by 1 convolution layer is multiplied by a feature map obtained by feature fusion;
(c) The original characteristic diagram is input into a convolution kernel with the quantity ofN e The activation function uses a 3×3 convolution layer of the ReLU, and the output feature map of the 3×3 convolution layer is added to the feature map obtained by multiplying in (b);
(d) The feature map obtained by adding is input into a step length of a pooling window which is 2 multiplied by 2S p In the maximum pooling layer of 2, reducing the size of the feature map to 0.5 times of the original feature map to obtain an output feature map of RDAACB;
further, in RDAACB, the data of the data,N e different parameters of the rdaac b1 to rdaac b4 set in fig. 2 are represented.
Further, in the present invention, the decoder of the rdaac n is configured to output a segmentation result of an image according to the features of the traffic road image extracted by the encoder, and the process includes the steps of:
step S10: the feature map output by RDAACB4 is input into a 3X 3 convolution layer with 512 convolution kernels and an activation function of ReLU, and 4 times up-sampling is carried out on the output feature map of the 3X 3 convolution layer by bilinear interpolation, so that the size of the feature map is enlarged by 4 times;
step S20: the feature images output by the first 4 times of up-sampling are input into a 3X 3 convolution layer with the number of convolution kernels being 64 and an activation function using ReLU, the output feature images of the convolution layer are added with the output feature images of RDAACB2, 4 times of up-sampling is carried out on the added feature images by using bilinear interpolation, and the size of the feature images is enlarged by 4 times;
step S30: the feature map output by the up sampling of the second time is input into a 3 multiplied by 3 convolution layer with the number of convolution kernels being 32 and the activation function using ReLU, the output of the 3 multiplied by 3 convolution layer is input into a 1 multiplied by 1 convolution layer with the number of convolution kernels being 2, and the result of dividing the traffic road surface image output by RDAACN is obtained.
In the specific application example, in the step S3, when training the rdaac n model, the optimizer uses Adam, the learning rate is set to 0.001, the batch size is set to 5, and training is performed on the training set for 100 rounds.
In a specific application example, in the step S3, the loss function is used when training the rdaac n modelLUsing cross entropy loss functionL CE And a noise-robust Dice loss functionL NR-Dice The calculation formula is as follows:
in the method, in the process of the invention,
the setting is made to be 0.8,Nrepresenting the number of pixels in the traffic road image, < >>
Representing the +.>
Values of corresponding labels of the pixel points, < >>
Representing the +.f. in RDAACN model pair image>
The prediction of the individual pixels is a value after the Softmax function.
In a specific application example, in the step S5, the method for calculating the area of the pavement damage area includes:
step S501: calculating the number of pixels of the RDAACN model prediction belonging to the road surface disease category in one traffic road surface image according to the result of the RDAACN model segmentation of the traffic road surface imageM；
Step S502: obtaining the actual physical area corresponding to each pixel point in the traffic road surface imageL；
Step S503: calculating the area of a road surface disease area in a traffic road surface imageKThe calculation formula is as follows:
in a specific application example, in the step S6, the process of predicting the maintenance cost of the pavement damage includes:
step S601: obtaining the correlation coefficient of the road surface maintenance cost, which are respectively the material coefficientT 1 Coefficient of road ageT 2 Coefficient of labor costT 3 And climate coefficientT 4 ；
Step S602: predicting the maintenance cost of the pavement damage according to the calculated area of the pavement damage area and the correlation coefficient of the pavement maintenance costPThe calculation formula is as follows:
the invention further provides a pavement disease maintenance cost prediction system based on semantic segmentation, which is used for implementing the method of the invention, and comprises the following steps:
the image data acquisition unit is used for constructing a training set and a testing set for training and testing the network model by utilizing semantic analysis, and preprocessing images in the training set and the testing set;
a training unit to build a RDAACN model having an encoder structure and a decoder structure; the encoder structure is used for extracting the characteristics of the traffic pavement image, and the decoder is used for outputting the segmentation result of the image according to the characteristics of the traffic pavement image extracted by the encoder; training the RDAACN model on a training set;
the test unit is used for storing the trained RDAACN model, testing the trained RDAACN model on a test set, outputting a test result, and accurately extracting road surface diseases in the traffic road surface image at a pixel level;
the SVM model generating unit is used for calculating the area of the pavement disease area according to the extraction result of the RDAACN model on the pavement disease pixel level precision in the traffic pavement image;
the prediction unit is used for predicting the maintenance cost of the pavement diseases by using the trained SVM model according to the extraction result of the improved U-Net model on the pavement disease pixel level precision in the traffic pavement image of the test set;
and the output unit is used for outputting the maintenance cost of the pavement damage according to the calculated area of the pavement damage area.
It will be appreciated by those skilled in the art that the above-described embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein. The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks. These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above examples, and all technical solutions belonging to the concept of the present invention belong to the protection scope of the present invention. It should be noted that modifications and adaptations to the invention without departing from the principles thereof are intended to be within the scope of the invention as set forth in the following claims.
Claims (8)
1. The pavement damage maintenance cost prediction method is characterized by comprising the following steps of:
step S1: acquiring a traffic pavement image dataset, firstly utilizing semantic analysis to construct a training set and a testing set for training and testing a network model, and preprocessing images in the training set and the testing set;
step S2: constructing an RDAACN model, wherein the RDAACN model is provided with an encoder structure and a decoder structure; the encoder structure is used for extracting the characteristics of the traffic pavement image, and the decoder is used for outputting the segmentation result of the image according to the characteristics of the traffic pavement image extracted by the encoder; in the step S2, a constructed residual dense asymmetric cavity convolution network RDAACN model is constructed, wherein the RDAACN model comprises an encoder and a decoder; the encoder of the RDAACN model is provided with four residual dense asymmetric cavity convolution blocks RDAACM which are connected in sequence;
step S3: training the RDAACN model on a training set;
step S4: the trained RDAACN model is stored, the trained RDAACN model is tested on a test set, a test result is output, and road surface diseases in traffic road surface images are extracted with pixel-level precision;
step S5: calculating the area of a pavement disease area according to the extraction result of the RDAACN model on the pavement disease pixel level precision in the traffic pavement image;
step S6: predicting maintenance cost of the pavement damage according to the calculated area of the pavement damage area;
the RDAACB model comprises the following structures:
(a) The original characteristic diagram is input into a convolution kernel with the quantity ofN e 2, expansion ratiod1×3 hole convolution layer using modified linear unit ReLU for 3, activation function, the output feature map of 1×3 hole convolution layer is input to a convolution kernel of numberN e 2, expansion ratiodFor 3, 3×1 hole convolution layer of ReLU is used as activation function, the output characteristic diagram of 3×1 hole convolution layer is input into a convolution kernel with the quantity ofN e 2, expansion ratiod3×3 hole convolution layer using ReLU for 6, activation function;
(b) The 3X 1 hole convolution layer and the 3X 3 hole convolution layer output feature images are subjected to feature fusion in a channel merging mode, and the feature images obtained after fusion are input into a convolution kernel with the quantity ofN e The activation function uses a 1 multiplied by 1 convolution layer of Sigmoid, and the output feature map of the 1 multiplied by 1 convolution layer is multiplied by a feature map obtained by feature fusion;
(c) The original characteristic diagram is input into a convolution kernel with the quantity ofN e Activating the function to causeAdding the characteristic map obtained by multiplying the 3×3 convolution layer with the 3×3 convolution layer of ReLU to the characteristic map obtained by multiplying in (b);
(d) The feature map obtained by adding is input into a step length of a pooling window which is 2 multiplied by 2S p And in the maximum pooling layer of 2, reducing the size of the feature map to 0.5 times of the original feature map to obtain an output feature map of RDAACB.
2. The method according to claim 1, wherein in the step S1, the traffic road surface image data set is a GAPs384 data set obtained by processing a asphalt road surface damage data set GAPs.
3. The method of claim 2, wherein each pixel in each image in the GAPs384 dataset represents 1.2mm x 1.2mm; the image is cropped to 528 pixels in width and 432 pixels in height.
4. The pavement damage repair cost prediction method according to claim 1, wherein an asymmetric convolution kernel is adopted in the rdaac n model, and the feature map input into the rdaac b model is added into the rdaac b model by using a residual connection mode after being processed by a 3×3 convolution layer.
5. The method for predicting the maintenance cost of pavement damage according to claim 4, wherein the connection structures of the convolution layers, the cavity convolution layers and the pooling layers in the rdaab 1 to rdaab 4 in the rdaab model are the same but the number of convolution kernels is different.
6. The pavement damage repair cost prediction method according to claim 1, wherein the decoder of the rdacn is configured to output a segmentation result of an image according to the features of the traffic pavement image extracted by the encoder, and the flow includes the steps of:
step S10: the feature map output by RDAACB4 is input into a 3X 3 convolution layer with 512 convolution kernels and an activation function of ReLU, and 4 times up-sampling is carried out on the output feature map of the 3X 3 convolution layer by bilinear interpolation, so that the size of the feature map is enlarged by 4 times;
step S20: the feature images output by the first 4 times of up-sampling are input into a 3X 3 convolution layer with the number of convolution kernels being 64 and an activation function using ReLU, the output feature images of the convolution layer are added with the output feature images of RDAACB2, 4 times of up-sampling is carried out on the added feature images by using bilinear interpolation, and the size of the feature images is enlarged by 4 times;
step S30: the feature map output by the up sampling of the second time is input into a 3 multiplied by 3 convolution layer with the number of convolution kernels being 32 and the activation function using ReLU, the output of the 3 multiplied by 3 convolution layer is input into a 1 multiplied by 1 convolution layer with the number of convolution kernels being 2, and the result of dividing the traffic road surface image output by RDAACN is obtained.
7. The method for predicting the maintenance cost of pavement damage according to any one of claims 1 to 3, wherein in the step S5, the method for calculating the area of the pavement damage area comprises the steps of:
step S501: calculating the number of pixels of the RDAACN model prediction belonging to the road surface disease category in one traffic road surface image according to the result of the RDAACN model segmentation of the traffic road surface imageM；
Step S502: obtaining the actual physical area corresponding to each pixel point in the traffic road surface imageL；
Step S503: calculating the area of a road surface disease area in a traffic road surface imageKThe calculation formula is as follows:
8. a pavement damage-based repair cost prediction system, comprising:
the image data acquisition unit is used for constructing a training set and a testing set for training and testing the network model by utilizing semantic analysis, and preprocessing images in the training set and the testing set;
a training unit to build a RDAACN model having an encoder structure and a decoder structure; the encoder structure is used for extracting the characteristics of the traffic pavement image, and the decoder is used for outputting the segmentation result of the image according to the characteristics of the traffic pavement image extracted by the encoder; training the RDAACN model on a training set;
the test unit is used for storing the trained RDAACN model, testing the trained RDAACN model on a test set, outputting a test result, and accurately extracting road surface diseases in the traffic road surface image at a pixel level;
the SVM model generating unit is used for calculating the area of the pavement disease area according to the extraction result of the RDAACN model on the pavement disease pixel level precision in the traffic pavement image;
the prediction unit is used for predicting the maintenance cost of the pavement diseases by using the trained SVM model according to the extraction result of the improved U-Net model on the pavement disease pixel level precision in the traffic pavement image of the test set;
the output unit is used for outputting maintenance cost of the pavement damage according to the calculated area of the pavement damage area;
the RDAACN model is a residual dense asymmetric cavity convolution network RDAACN model, and comprises an encoder and a decoder; the encoder of the RDAACN model is provided with four residual dense asymmetric cavity convolution blocks RDAACM which are connected in sequence; the RDAACB model comprises the following structures:
(a) The original characteristic diagram is input into a convolution kernel with the quantity ofN e 2, expansion ratiod1×3 hole convolution layer using modified linear unit ReLU for 3, activation function, the output feature map of 1×3 hole convolution layer is input to a convolution kernel of numberN e 2, expansion ratiodFor 3, 3×1 hole convolution layer of ReLU is used as activation function, the output characteristic diagram of 3×1 hole convolution layer is input into a convolution kernel with the quantity ofN e 2, expansion ratiod3×3 hole convolution layer using ReLU for 6, activation function;
(b) The 3X 1 hole convolution layer and the 3X 3 hole convolution layer output feature images are subjected to feature fusion in a channel merging mode, and the feature images obtained after fusion are input into a convolution kernel with the quantity ofN e The activation function uses a 1 multiplied by 1 convolution layer of Sigmoid, and the output feature map of the 1 multiplied by 1 convolution layer is multiplied by a feature map obtained by feature fusion;
(c) The original characteristic diagram is input into a convolution kernel with the quantity ofN e The activation function uses a 3×3 convolution layer of the ReLU, and the output feature map of the 3×3 convolution layer is added to the feature map obtained by multiplying in (b);
(d) The feature map obtained by adding is input into a step length of a pooling window which is 2 multiplied by 2S p And in the maximum pooling layer of 2, reducing the size of the feature map to 0.5 times of the original feature map to obtain an output feature map of RDAACB.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310181026.XA CN115861324B (en) | 2023-03-01 | 2023-03-01 | Pavement damage maintenance cost prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310181026.XA CN115861324B (en) | 2023-03-01 | 2023-03-01 | Pavement damage maintenance cost prediction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115861324A CN115861324A (en) | 2023-03-28 |
| CN115861324B true CN115861324B (en) | 2023-05-12 |
Family
ID=85659418
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310181026.XA Active CN115861324B (en) | 2023-03-01 | 2023-03-01 | Pavement damage maintenance cost prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115861324B (en) |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210372938A1 (en) * | 2020-05-29 | 2021-12-02 | The Board Of Trustee Of The University Of Alabama | Deep learning-based crack segmentation through heterogeneous image fusion |
| CN115170479A (en) * | 2022-06-15 | 2022-10-11 | 兰州交通大学 | Automatic extraction method for asphalt pavement repairing diseases |
| CN114937033A (en) * | 2022-06-27 | 2022-08-23 | 辽宁工程技术大学 | Rural highway pavement disease intelligent detection method based on deep convolutional neural network |
| CN115345881B (en) * | 2022-10-18 | 2023-01-13 | 上海交强国通智能科技有限公司 | Pavement disease detection method based on computer vision |
-
2023
- 2023-03-01 CN CN202310181026.XA patent/CN115861324B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN115861324A (en) | 2023-03-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114120102A (en) | Boundary-optimized remote sensing image semantic segmentation method, device, equipment and medium | |
| CN104992223B (en) | Intensive population estimation method based on deep learning | |
| CN111104903B (en) | Depth perception traffic scene multi-target detection method and system | |
| CN110853057B (en) | Aerial image segmentation method based on global and multi-scale full-convolution network | |
| CN112232237B (en) | Method, system, computer device and storage medium for monitoring vehicle flow | |
| CN111027377B (en) | Double-flow neural network time sequence action positioning method | |
| CN105608454A (en) | Text structure part detection neural network based text detection method and system | |
| CN113792606B (en) | Low-cost self-supervision pedestrian re-identification model construction method based on multi-target tracking | |
| CN113762265B (en) | Classified segmentation method and system for pneumonia | |
| CN116206112A (en) | Remote sensing image semantic segmentation method based on multi-scale feature fusion and SAM | |
| CN114360030A (en) | Face recognition method based on convolutional neural network | |
| CN112819000A (en) | Streetscape image semantic segmentation system, streetscape image semantic segmentation method, electronic equipment and computer readable medium | |
| Zhu et al. | Towards automatic wild animal detection in low quality camera-trap images using two-channeled perceiving residual pyramid networks | |
| CN114332473A (en) | Object detection method, object detection device, computer equipment, storage medium and program product | |
| CN116597270A (en) | Road damage target detection method based on attention mechanism integrated learning network | |
| CN114463340B (en) | Agile remote sensing image semantic segmentation method guided by edge information | |
| Yang et al. | C-RPNs: Promoting object detection in real world via a cascade structure of Region Proposal Networks | |
| CN112580616B (en) | Crowd quantity determination method, device, equipment and storage medium | |
| CN114612803A (en) | Transmission line insulator defect detection method for improving CenterNet | |
| CN111881914B (en) | License plate character segmentation method and system based on self-learning threshold | |
| CN113192076A (en) | MRI brain tumor image segmentation method combining classification prediction and multi-scale feature extraction | |
| CN112597996A (en) | Task-driven natural scene-based traffic sign significance detection method | |
| CN117274355A (en) | Drainage pipeline flow intelligent measurement method based on acceleration guidance area convolutional neural network and parallel multi-scale unified network | |
| CN115861324B (en) | Pavement damage maintenance cost prediction method and system | |
| CN115456957B (en) | Method for detecting change of remote sensing image by full-scale feature aggregation |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |