CN114998852A - Intelligent detection method for road pavement diseases based on deep learning - Google Patents
Intelligent detection method for road pavement diseases based on deep learning Download PDFInfo
- Publication number
- CN114998852A CN114998852A CN202110896211.8A CN202110896211A CN114998852A CN 114998852 A CN114998852 A CN 114998852A CN 202110896211 A CN202110896211 A CN 202110896211A CN 114998852 A CN114998852 A CN 114998852A
- Authority
- CN
- China
- Prior art keywords
- disease
- pavement
- diseases
- image
- detection
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention adopts a deep learning method, and develops a detection algorithm special for a specific target, namely the road disease, on the basis of a multi-class target detection algorithm. And parameters such as length, width, area and the like of the diseases are quantitatively identified, and detailed information of the diseases is described, so that intelligent assessment and targeted maintenance of the road diseases are facilitated. And intelligent pavement disease detection is realized. The main flow of the algorithm comprises the following steps: preparing pavement disease image data; classifying and calibrating the disease image; carrying out image normalization processing; dividing the image into a training set, a verification set and a test set according to a proportion; building a convolutional neural network model structure YOLO-Unet; according to the output pavement disease contour information, quantitatively identifying parameters such as length, width, area and the like of the diseases; training a data set, calculating a loss function and configuring parameters; performing parallel computation; testing and evaluating the pavement disease detection according to the training result; and the indexes such as the accuracy rate, the detection speed and the like of the field test are obtained.
Description
Technical Field
The invention uses deep learning and road engineering technology, is suitable for the field of road maintenance and management, and particularly relates to a road pavement disease detection algorithm based on deep learning.
Background
After the highway is built and put into operation, under the influence of long-term repeated automobile load and external natural environment factors, particularly, the accumulated axle load quantity is increased due to the increase of traffic volume, so that the operation of the road in an overload state is caused, the material property and the structure of the road are fatigue-decayed, and finally the service performance and the safety performance of the road are reduced, therefore, the road surface damage degree can be timely detected and the method is used for judging the road surface damage degree and accurately maintaining the damage.
The traditional detection mainly depends on manpower, the working efficiency is low, the detection time is long, and the normal operation of traffic is influenced during the detection. With the development of science and technology, people begin to research road disease detection methods based on computer vision, which can be mainly divided into two types, the first method is an edge detection algorithm based on large change of disease edge pixel values, and mainly comprises a Soble edge detection method, a Canny edge detection method and a wavelet analysis method; the second method is a method based on a large pixel value of a disease area, and mainly comprises an HOG histogram detection method, an anisotropy detection method, a threshold segmentation method, a minimum path method and the like.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide a road pavement disease intelligent detection method based on deep learning, and aims to solve the problems of low detection precision and efficiency, and poor robustness and generalization capability of the road disease detection method in the prior art.
In order to solve the technical problems, the invention adopts the following technical scheme:
the invention provides a deep learning-based intelligent detection method for road pavement diseases, which comprises the following steps:
s10, acquiring a data set: collecting pavement disease images as a data set;
s20, image data classification and calibration: classifying and labeling the diseases according to the pavement disease data set obtained in the step S10;
s30, image preprocessing: preprocessing the marked pavement disease image;
s40, data set division: dividing the pavement damage data set into three data subsets which are respectively a training set, a verification set and a test set;
s50, building a network and detecting: building a core convolutional neural network model, namely a YOLO-Unet network for short, detecting the diseases by using the YOLO-Unet network, and outputting outline information of the diseases;
s60, disease quantitative identification: quantitatively identifying parameters such as length, width and area of the disease by using the contour information output in the step S50;
s70, model training: carrying a model of a YOLO-Unet network, initializing the model, importing a training set picture into the model, and training;
s80, loss function setting: the loss function during the training of the YOLO-Unet network consists of a bounding box regression loss Lciou, a confidence coefficient loss Lconf and a classification loss Lclass 3. If no target exists in a certain bounding box, only calculating confidence loss, and if the target exists, calculating 3 types of loss; confidence coefficient loss Lconf and classification loss Lclass are calculated through a cross entropy method, and boundary box regression loss Lciou is calculated through a CIoU loss function. The specific calculation formula is as follows:
L_ciou=∑_(i=0)^(S^2)▒∑_(j=0)^B▒I_(i,j)^obj [1-IoU+(ρ^2 (b,b^gt ))/c^2 +(16/π^4 (arctan w^gt/h^gt -arctan w/h)^4)/(1-IoU+4/π^2 (arctan w^gt/h^gt -arctan w/h)^2 )]；
L_conf=-∑_(i=0)^(S^2)▒∑_(j=0)^B▒I_(i,j)^obj [C ̂_i^j log〖(C_i^j )+〗 (1-C ̂_i^j ) log(〖1-C〗_i^j ) ]-λ_noobj ∑_(i=0)^(S^2)▒∑_(j=0)^B▒I_(i,j)^noobj [C ̂_i^j log〖(C_i^j )+〗 (1-C ̂_i^j ) log(〖1-C〗_i^j ) ]；
L_class=-∑_(i=0)^(S^2)▒〖I_(i,j)^obj ∑_cϵclasses▒[P ̂_i^j log〖(P_i^j )+〗 (1-P ̂_i^j ) log(〖1-P〗_i^j ) ] 〗；
in the formula: s2 and B are respectively a characteristic diagram scale and a prior frame; λ noobj is a weight coefficient; i _ (I, j) ^ obj, I _ (I, j) ^ nonobj mean that if the jth prior frame of the ith grid has a target, 1 and 0 are respectively taken, and if no target is present, 0 and 1 are respectively taken; ρ (.) is the Euclidean distance; c is the diagonal distance between the prediction frame and the actual frame closure area; b. w and h are respectively the central coordinate and the width and the height of the prediction frame; bgt, wgt and hgt are the center coordinates and width and height of the area frame where the actual target is located; c _ i ^ j, C ̂ _ i ^ j-confidence of the prediction frame and the labeling frame; p _ i ^ j and P ̂ _ i ^ j are the class probabilities of the prediction frame and the label frame;
s90, model test: introducing the test set picture to perform model test, outputting various detection numerical indexes, and judging whether the expected value is reached; and (4) shooting the road surface disease video by using the detection vehicle, transmitting the video back to the high-performance computer terminal for detection, and judging whether the detection efficiency and accuracy of the model can meet the required requirements.
Further, the step S10 specifically includes:
s11, the method for acquiring the data set comprises the following steps: the method comprises the steps of collecting the road surface disease data through a network, and collecting the road surface disease data through a crawler and a road detection device on site to establish a road surface disease data set with 10000 road surface disease images.
Further, the step S20 specifically includes:
s21, dividing the acquired images in the pavement disease data set into four types of disease types, namely transverse cracks, longitudinal cracks, pits and cracks;
and S22, marking the pavement diseases by using a visual image marking tool, namely a lableImg tool.
Further, the step S30 specifically includes:
s31, processing the images into a resolution of 600 multiplied by 600;
s32, a CLAHE (class-restricted contrast ratio) adaptive histogram equalization method is used for ensuring that the brightness of a darker area is greatly different from that of a bright area, the method is a calculation method for dividing an image into blocks, calculating a histogram by taking the blocks as a unit, then pruning the histogram and finally averaging a horizontal histogram.
Further, the proportion of the three data subsets in the data set division of step S40 is 8:1: 1.
Further, the step S50 specifically includes:
s51, taking the original whole pavement damage image as input of a YOLO-Unet network, dividing the pavement damage image into grids with the size of 7 multiplied by 7, and for each grid, if the center of the damage appears in the grid, detecting a damage target in the grid;
s52, each grid needs to predict a plurality of candidate frames, and each candidate frame includes 5 predictors, namely (x, y, w, h, score), where the coordinates of the center point of the object are (x, y), the width and height of the candidate frame are (w, h), the confidence that the candidate frame contains the road surface damage is score, and the expression is:
score=Pr(object)×〖IoU〗_pred^truth
wherein, Pr (object) indicates whether the unit of the candidate frame contains a defect, if so, Pr (object) is set to 1, otherwise, it is set to 0. IoU, pred Truth is the ratio of the intersection area of the prediction frame and the actual frame to the joint area;
s53, carrying out multi-scale prediction from 3 scales, predicting 3 areas possibly containing detection targets in each scale, and finally carrying out prediction classification on pavement diseases and judging the types of the diseases;
s54, in order to quantitatively calculate indexes such as length and area of the road surface damage, aiming at the quantitative identification of the road surface damage, extracting accurate contour information of the road surface damage through a YOLO-Unet network, improving the anti-interference capability of fine pen marks and fine scratches through target identification and image segmentation by utilizing a multi-neural network, and outputting the contour information of the identified road surface damage.
Further, the step S60 specifically includes:
s61, refining a connected region into the width of a pixel through binary image refinement, continuously reducing peripheral pixels, and extracting an image pixel skeleton; and obtaining the length or width of the defect pixel through the expansion circle, converting the number of pixels corresponding to the actual size according to the distance or the reference object, and further converting the contour length, width or area of the pavement defect.
Further, the step S70 specifically includes:
s71, initializing parameters by adopting a random initialization method;
s72, training the YOLO-Unet network by using a gradient descent method.
Further, the detection value index in step S90 is an accuracy rate, and is a ratio of real targets in all prediction results.
Compared with the prior art, the intelligent detection method for the road pavement diseases based on deep learning at least has the following beneficial effects:
the method adopts a model of the YOLO-Unet network as a detection frame, and the model can learn characteristic expression with higher robustness by learning a large number of pavement diseases, so that generalization capability is increased, the identification accuracy is improved, meanwhile, accurate outline information of the diseases is extracted, the anti-interference capability on fine writing and fine scratches is improved by utilizing a multi-neural network through target identification and image segmentation, and the outline information of the identified pavement diseases is output, so that quantitative identification of the diseases is realized.
Description of the drawings in order to illustrate the solution of the invention more clearly, a brief description of the figures required for the description of the embodiments will be given below, it being clear that the figures in the following description are some embodiments of the invention, from which other figures can be derived by a person skilled in the art without inventive effort.
FIG. 1 is a general flow chart of an intelligent detection method for road pavement diseases based on deep learning provided by the invention
Fig. 2 is a general network structure diagram of the intelligent detection method for road pavement diseases based on deep learning provided by the invention.
Detailed Description
To facilitate an understanding of the invention, the invention will now be described more fully with reference to the accompanying drawings. Preferred embodiments of the present invention are shown in the drawings. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.
As shown in fig. 1, the intelligent detection method for road pavement diseases based on deep learning provided by the invention comprises the following steps:
s10, acquiring a data set: collecting a data set of a pavement disease image for pavement disease detection;
s11, the method for acquiring the data set comprises the following steps: the method comprises the steps of collecting the road surface disease data through a network, and collecting the road surface disease data through a crawler and a road detection device on site to establish a road surface disease data set with 10000 road surface disease images.
S12, selecting disease pictures by professionals
Specifically, the main bases for selecting pictures are:
1. the images are diversified and comprise cement roads, pedestrian road green belts, asphalt roads and the like;
2. the pictures have multiple visual angles, including front view, side view, oblique view and the like, and an image rotation algorithm can be adopted;
3. diseases should have obvious characteristics;
4. the sharpness of the image should be guaranteed.
S20, image data classification and calibration: and classifying and labeling the diseases according to the pavement disease data set obtained in the step S10.
S21, dividing the acquired image in the pavement disease data set into four types of disease types, namely transverse cracks, longitudinal cracks, pit grooves and cracks;
and S22, marking the pavement diseases by using a visual image marking tool, namely a lableImg tool.
Specifically, when marking diseases, the diseases should be divided along the edges as much as possible, and for the diseases with inconspicuous boundaries or the areas with overlapped diseases, the diseases should be identified by experienced professionals.
S30, image preprocessing: and preprocessing the marked pavement disease image.
S31, processing the images into a resolution of 600 x 600;
s32, a CLAHE (class-restricted contrast ratio) adaptive histogram equalization method is used for ensuring that the brightness of a darker area is greatly different from that of a bright area, the method is a calculation method for dividing an image into blocks, calculating a histogram by taking the blocks as a unit, then pruning the histogram and finally averaging a horizontal histogram.
S40, data set division: and dividing the data set into three data subsets, namely a training set, a verification set and a test set.
Specifically, the ratio of the three data subsets in the data set division in step S40 is 8:1: 1.
S50, building a network and detecting: and (3) building a core convolutional neural network model structure, namely a YOLO-Unet network for short, detecting the diseases by using the YOLO-Unet network, and outputting the outline information of the diseases.
Specifically, the general structure diagram of the convolutional network model constructed in step S50 is shown in fig. 2, YOLOv5 is combined with Unet, YOLOv5 is used to learn and recognize road surface defect images, Unet is used to extract precise defect contour information at the same time, the multi-neural network is used to improve the anti-interference capability on fine writing and fine scratches through target recognition and image segmentation, and the contour information of road surface defects recognized through the YOLOv-Unet network is output, so that quantitative recognition of the defects is realized.
S51, taking the original whole pavement damage image as input of a YOLO-Unet network, dividing the pavement damage image into grids with the size of 7 multiplied by 7, and for each grid, if the center of the damage appears in the grid, detecting a damage target in the grid;
s52, each grid needs to predict a plurality of candidate frames, and each candidate frame includes 5 predictors, namely (x, y, w, h, score), where the coordinates of the center point of the object are (x, y), the width and height of the candidate frame are (w, h), the confidence that the candidate frame contains the road surface damage is score, and the expression is:
score=Pr(object)×〖IoU〗_pred^truth
wherein, Pr (object) indicates whether the unit of the candidate frame contains a defect, if so, Pr (object) is set to 1, otherwise, it is set to 0. IoU, pred Truth is the ratio of the intersection area of the prediction frame and the actual frame to the joint area;
s53, carrying out multi-scale prediction from 3 scales, predicting 3 areas possibly containing detection targets in each scale, and finally carrying out prediction classification on pavement diseases and judging the types of the diseases;
s54, in order to quantitatively calculate indexes such as length and area of the road surface damage, aiming at the quantitative identification of the road surface damage, extracting accurate contour information of the road surface damage through a YOLO-Unet network, improving the anti-interference capability of fine pen marks and fine scratches through target identification and image segmentation by utilizing a multi-neural network, and outputting the contour information of the identified road surface damage.
S60, disease quantitative identification: the parameters such as the length, width, and area of the defect are quantitatively identified using the contour information output in step S50.
S61, thinning a connected region into the width of a pixel through binary image thinning, continuously reducing peripheral pixels, and extracting an image pixel skeleton; the length or width of the damaged pixel is obtained through the expansion circle, the number of pixels corresponding to the actual size is converted according to the distance or the reference object, and further the contour length, the width or the area of the road surface damage is converted.
S70, model training: and carrying out model loading of a YOLO-Unet network, initializing the model, importing the training set picture into the model, and training.
S71, initializing parameters by adopting a random initialization method;
s72, training the YOLO-Unet network by using a gradient descent method.
Specifically, a small-batch gradient descent method is used for training the YOLO-Unet network, loss value function images are observed, model convergence is achieved when the loss value function images are stable, and training is stopped at the moment.
S80, loss function setting: the loss function in the process of YOLO-Unet network training is composed of a bounding box regression loss Lciou, a confidence coefficient loss Lconf and a classification loss Lclass 3. If no target exists in a certain bounding box, only calculating confidence loss, and if the target exists, calculating 3 types of loss; confidence coefficient loss Lconf and classification loss Lclass are calculated through a cross entropy method, and boundary box regression loss Lciou is calculated through a CIoU loss function. The specific calculation formula is as follows:
L_ciou=∑_(i=0)^(S^2)▒∑_(j=0)^B▒I_(i,j)^obj [1-IoU+(ρ^2 (b,b^gt ))/c^2 +(16/π^4 (arctan w^gt/h^gt -arctan w/h)^4)/(1-IoU+4/π^2 (arctan w^gt/h^gt -arctan w/h)^2 )]；
L_conf=-∑_(i=0)^(S^2)▒∑_(j=0)^B▒I_(i,j)^obj [C ̂_i^j log〖(C_i^j )+〗 (1-C ̂_i^j ) log(〖1-C〗_i^j ) ]-λ_noobj ∑_(i=0)^(S^2)▒∑_(j=0)^B▒I_(i,j)^noobj [C ̂_i^j log〖(C_i^j )+〗 (1-C ̂_i^j ) log(〖1-C〗_i^j ) ]；
L_class=-∑_(i=0)^(S^2)▒〖I_(i,j)^obj ∑_cϵclasses▒[P ̂_i^j log〖(P_i^j )+〗 (1-P ̂_i^j ) log(〖1-P〗_i^j ) ] 〗；
in the formula: s2 and B are respectively a characteristic diagram scale and a prior frame; λ noobj is a weight coefficient; i _ (I, j) obj, I _ (I, j) nonobj mean that if there is a target at the jth prior frame of the ith grid, 1 and 0 are respectively taken, and if there is no target, 0 and 1 are respectively taken; ρ (.) is the Euclidean distance; c is the diagonal distance between the prediction frame and the actual frame closure area; b. w and h are respectively the central coordinate and the width and the height of the prediction frame; bgt, wgt and hgt are the center coordinates and width and height of the area frame where the actual target is located; c _ i ^ j, C ̂ _ i ^ j-confidence of the prediction frame and the labeling frame; p _ i ^ j and P ̂ _ i ^ j are the class probability of the prediction frame and the label frame.
S90, model test: introducing the test set picture to perform model test, outputting various detection numerical indexes, and judging whether the expected value is reached; and (4) shooting a road surface disease video by using a detection vehicle, transmitting the video back to a high-performance computer terminal for detection, and judging whether the detection efficiency and accuracy of the model can meet the required requirements.
According to the intelligent detection method for the road pavement diseases based on the deep learning in the embodiment, the model of the YOLO-Unet network is used as a detection frame, the model can learn the characteristic expression with higher robustness through learning a large number of road pavement diseases, so that the generalization capability is increased, the recognition accuracy is improved, meanwhile, the Unet is used for extracting the accurate outline information of the diseases, the anti-interference capability on fine traces and fine scratches is improved through target recognition and image segmentation by utilizing the multi-neural network, and the outline information of the road pavement diseases recognized through the YOLO is output, so that the quantitative recognition of the diseases is realized.
It is to be understood that the above-described embodiments are merely preferred embodiments of the present invention, and not all embodiments are shown in the drawings, which are set forth to limit the scope of the invention. All equivalent structures made by using the contents of the specification and the attached drawings of the invention can be directly or indirectly applied to other related technical fields, and are also within the protection scope of the patent of the invention.
Claims (9)
1. An intelligent detection method for road pavement diseases based on deep learning is characterized by comprising the following steps: s10, acquiring a data set: collecting pavement disease images as a data set;
s20, image data classification and calibration: classifying and labeling the diseases according to the data set obtained in the step S10;
s30, image preprocessing: preprocessing the marked pavement disease image;
s40, dividing the data set: dividing the pavement damage data set into three data subsets which are respectively a training set, a verification set and a test set;
s50, building a network and detecting: building a core convolutional neural network model, namely a YOLO-Unet network for short, detecting the diseases by using the YOLO-Unet network, and outputting outline information of the diseases;
s60, disease quantitative identification: quantitatively identifying parameters such as length, width and area of the disease by using the contour information output in the step S50;
s70, model training: carrying a model of a YOLO-Unet network, initializing the model, importing a training set picture into the model, and training;
s80, loss function setting: the loss function during the YOLO-Unet network training consists of a bounding box regression loss Lciou, a confidence coefficient loss Lconf and a classification loss Lclass 3;
if no target exists in a certain bounding box, only calculating confidence loss, and if the target exists, calculating 3 types of loss; confidence coefficient loss Lconf and classification loss Lclass are calculated through a cross entropy method, and boundary box regression loss Lciou is calculated through a CIoU loss function;
the specific calculation formula is as follows:
in the formula: s2 and B are respectively a characteristic diagram scale and a prior frame; λ noobj is a weight coefficient;
、
meaning that if there is an object at the jth prior frame of the ith grid, respectivelyTaking 1 and 0, if no target exists, respectively taking 0 and 1; ρ (-) is the Euclidean distance; c is the diagonal distance between the predicted frame and the actual frame closure area; b. w and h are respectively the central coordinate and the width and the height of the prediction frame;b gt 、w gt 、h gt the center coordinates and the width and the height of the area frame where the actual target is located;
-confidence of the prediction box and the annotation box;
is the class probability of the prediction box and the label box;
s90, model test: introducing a test set picture to perform model test, outputting various detection numerical indexes, and judging whether the expected value is reached; and (4) shooting the road surface disease video by using the detection vehicle, transmitting the video back to the high-performance computer terminal for detection, and judging whether the detection efficiency and accuracy of the model can meet the required requirements.
2. The method for intelligently detecting the road pavement diseases based on the deep learning of claim 1, wherein the step S10 specifically includes:
s11, the method for acquiring the data set comprises the following steps: the method comprises the steps of collecting the road surface disease data through a network, and collecting the road surface disease data through a crawler and a road detection device on site to establish a road surface disease data set with 10000 road surface disease images.
3. The method for intelligently detecting the road pavement diseases based on the deep learning of claim 1, wherein the step S20 specifically includes:
s21, dividing the pavement damage data set image into four types of damage, namely a transverse crack, a longitudinal crack, a pit slot and a crack;
and S22, marking the pavement diseases by using a visual image marking tool, namely a lableImg tool.
4. The method for intelligently detecting the road pavement diseases based on the deep learning of claim 1, wherein the step S30 specifically includes:
s31, processing the images into a resolution of 600 x 600;
s32, a CLAHE (class-restricted contrast ratio) adaptive histogram equalization method is used for ensuring that the brightness of a darker area is greatly different from that of a bright area, the method is a calculation method for dividing an image into blocks, calculating a histogram by taking the blocks as a unit, then pruning the histogram and finally averaging a horizontal histogram.
5. The method for intelligently detecting the road pavement diseases based on the deep learning of claim 1, wherein the proportion of the three data subsets in the data set division in the step S40 is 8:1: 1.
6. The intelligent road pavement disease detection method based on deep learning of claim 1, wherein the steps of detection and quantitative identification in step S50 include:
s51, taking the original whole pavement damage image as input of a YOLO-Unet network, dividing the pavement damage image into grids with the size of 7 multiplied by 7, and for each grid, if the center of the damage appears in the grid, detecting a damage target in the grid;
s52, each grid needs to predict a plurality of candidate frames, and each candidate frame includes 5 predictors, namely (x, y, w, h, score), where the coordinates of the center point of the object are (x, y), the width and height of the candidate frame are (w, h), the confidence that the candidate frame contains the road surface damage is score, and the expression is:
wherein, Pr (object) indicates whether the unit of the candidate frame contains diseases, if yes, the Pr (object) is set to 1, otherwise, the Pr (object) is set to 0;
s53, carrying out multi-scale prediction from 3 scales, predicting 3 areas possibly containing detection targets in each scale, and finally carrying out prediction classification on pavement diseases and judging the types of the diseases;
s54, in order to quantitatively calculate indexes such as length and area of the road surface damage, aiming at the quantitative identification of the road surface damage, extracting accurate contour information of the road surface damage through a YOLO-Unet network, improving the anti-interference capability of fine pen marks and fine scratches through target identification and image segmentation by utilizing a multi-neural network, and outputting the contour information of the identified road surface damage.
7. The intelligent road pavement disease detection method based on deep learning of claim 1, wherein the disease quantitative identification step in step S60 comprises:
s61, thinning a connected region into the width of a pixel through binary image thinning, continuously reducing peripheral pixels, and extracting an image pixel skeleton; and obtaining the length or width of the defect pixel through the expansion circle, converting the number of pixels corresponding to the actual size according to the distance or the reference object, and further converting the contour length, width or area of the pavement defect.
8. The intelligent road pavement disease detection method based on deep learning of claim 1, wherein the specific steps in step S70 include:
s71, initializing parameters by adopting a random initialization method;
s72, training the YOLO-Unet network by using a gradient descent method.
9. The method for intelligently detecting the road pavement diseases based on the deep learning of claim 1, wherein the detection numerical index in the step S90 is an accuracy rate, which is a ratio of real targets in all prediction results.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110896211.8A CN114998852A (en) | 2021-08-05 | 2021-08-05 | Intelligent detection method for road pavement diseases based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110896211.8A CN114998852A (en) | 2021-08-05 | 2021-08-05 | Intelligent detection method for road pavement diseases based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114998852A true CN114998852A (en) | 2022-09-02 |
Family
ID=83018039
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110896211.8A Pending CN114998852A (en) | 2021-08-05 | 2021-08-05 | Intelligent detection method for road pavement diseases based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114998852A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115271565A (en) * | 2022-09-29 | 2022-11-01 | 中南大学 | DEA-based method, device and equipment for evaluating highway pavement maintenance measures |
| CN115345881A (en) * | 2022-10-18 | 2022-11-15 | 上海交强国通智能科技有限公司 | Pavement disease detection method based on computer vision |
| CN115424232A (en) * | 2022-11-04 | 2022-12-02 | 深圳市城市交通规划设计研究中心股份有限公司 | Method for identifying and evaluating pavement pit, electronic equipment and storage medium |
| CN116152211A (en) * | 2023-02-28 | 2023-05-23 | 哈尔滨市科佳通用机电股份有限公司 | Identification method for brake shoe abrasion overrun fault |
| CN116229396A (en) * | 2023-02-17 | 2023-06-06 | 广州丰石科技有限公司 | High-speed pavement disease identification and warning method |
| CN116797436A (en) * | 2023-08-29 | 2023-09-22 | 北京道仪数慧科技有限公司 | Processing system for carrying out road disease inspection by utilizing bus |
| CN117078235A (en) * | 2023-10-17 | 2023-11-17 | 深圳市城市交通规划设计研究中心股份有限公司 | Road network maintenance method, electronic equipment and storage medium for comprehensive evaluation |
| CN117237925A (en) * | 2023-11-16 | 2023-12-15 | 南京萨利智能科技有限公司 | Intelligent road disease inspection method and system based on computer vision |
| CN117474912A (en) * | 2023-12-27 | 2024-01-30 | 浪潮软件科技有限公司 | Road section gap analysis method and model based on computer vision |
-
2021
- 2021-08-05 CN CN202110896211.8A patent/CN114998852A/en active Pending
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115271565B (en) * | 2022-09-29 | 2023-01-24 | 中南大学 | DEA-based method, device and equipment for evaluating highway pavement maintenance measures |
| CN115271565A (en) * | 2022-09-29 | 2022-11-01 | 中南大学 | DEA-based method, device and equipment for evaluating highway pavement maintenance measures |
| CN115345881A (en) * | 2022-10-18 | 2022-11-15 | 上海交强国通智能科技有限公司 | Pavement disease detection method based on computer vision |
| CN115424232A (en) * | 2022-11-04 | 2022-12-02 | 深圳市城市交通规划设计研究中心股份有限公司 | Method for identifying and evaluating pavement pit, electronic equipment and storage medium |
| CN116229396B (en) * | 2023-02-17 | 2023-11-03 | 广州丰石科技有限公司 | High-speed pavement disease identification and warning method |
| CN116229396A (en) * | 2023-02-17 | 2023-06-06 | 广州丰石科技有限公司 | High-speed pavement disease identification and warning method |
| CN116152211A (en) * | 2023-02-28 | 2023-05-23 | 哈尔滨市科佳通用机电股份有限公司 | Identification method for brake shoe abrasion overrun fault |
| CN116797436A (en) * | 2023-08-29 | 2023-09-22 | 北京道仪数慧科技有限公司 | Processing system for carrying out road disease inspection by utilizing bus |
| CN116797436B (en) * | 2023-08-29 | 2023-10-31 | 北京道仪数慧科技有限公司 | Processing system for carrying out road disease inspection by utilizing bus |
| CN117078235A (en) * | 2023-10-17 | 2023-11-17 | 深圳市城市交通规划设计研究中心股份有限公司 | Road network maintenance method, electronic equipment and storage medium for comprehensive evaluation |
| CN117078235B (en) * | 2023-10-17 | 2024-03-26 | 深圳市城市交通规划设计研究中心股份有限公司 | Road network maintenance method, electronic equipment and storage medium for comprehensive evaluation |
| CN117237925A (en) * | 2023-11-16 | 2023-12-15 | 南京萨利智能科技有限公司 | Intelligent road disease inspection method and system based on computer vision |
| CN117237925B (en) * | 2023-11-16 | 2024-02-13 | 南京萨利智能科技有限公司 | Intelligent road disease inspection method and system based on computer vision |
| CN117474912A (en) * | 2023-12-27 | 2024-01-30 | 浪潮软件科技有限公司 | Road section gap analysis method and model based on computer vision |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114998852A (en) | Intelligent detection method for road pavement diseases based on deep learning | |
| CN108830188B (en) | Vehicle detection method based on deep learning | |
| CN113160192B (en) | Visual sense-based snow pressing vehicle appearance defect detection method and device under complex background | |
| Chen et al. | A self organizing map optimization based image recognition and processing model for bridge crack inspection | |
| Varadharajan et al. | Vision for road inspection | |
| CN103049763B (en) | Context-constraint-based target identification method | |
| 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 | |
| CN107316064B (en) | Asphalt pavement crack classification and identification method based on convolutional neural network | |
| CN103020605B (en) | Bridge identification method based on decision-making layer fusion | |
| CN108960055B (en) | Lane line detection method based on local line segment mode characteristics | |
| CN102867183B (en) | Method and device for detecting littered objects of vehicle and intelligent traffic monitoring system | |
| CN111274926B (en) | Image data screening method, device, computer equipment and storage medium | |
| CN115393727B (en) | Pavement linear crack identification method, electronic equipment and storage medium | |
| CN104978567A (en) | Vehicle detection method based on scenario classification | |
| US20220315243A1 (en) | Method for identification and recognition of aircraft take-off and landing runway based on pspnet network | |
| CN110956207B (en) | Method for detecting full-element change of optical remote sensing image | |
| CN106372624A (en) | Human face recognition method and human face recognition system | |
| CN110751619A (en) | Insulator defect detection method | |
| CN103413131A (en) | Tower crane recognition method based on spectral and geometric characteristics | |
| CN113240623A (en) | Pavement disease detection method and device | |
| Han et al. | Vision-based crack detection of asphalt pavement using deep convolutional neural network | |
| CN113158954B (en) | Automatic detection method for zebra crossing region based on AI technology in traffic offsite | |
| CN105469099B (en) | Pavement crack detection and identification method based on sparse representation classification | |
| CN106548195A (en) | A kind of object detection method based on modified model HOG ULBP feature operators | |
| Tsai et al. | Automatic detection of deficient video log images using a histogram equity index and an adaptive Gaussian mixture model |
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 |