CN117788409B - Pavement crack detection method based on data enhancement and multi-scale feature learning - Google Patents
Pavement crack detection method based on data enhancement and multi-scale feature learning Download PDFInfo
- Publication number
- CN117788409B CN117788409B CN202311811088.0A CN202311811088A CN117788409B CN 117788409 B CN117788409 B CN 117788409B CN 202311811088 A CN202311811088 A CN 202311811088A CN 117788409 B CN117788409 B CN 117788409B
- Authority
- CN
- China
- Prior art keywords
- point cloud
- layer
- side comprises
- full
- scale feature
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 14
- 238000012549 training Methods 0.000 claims abstract description 19
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 10
- 238000005070 sampling Methods 0.000 claims description 52
- 238000004220 aggregation Methods 0.000 claims description 24
- 230000002776 aggregation Effects 0.000 claims description 24
- 238000000034 method Methods 0.000 claims description 21
- 238000004364 calculation method Methods 0.000 claims description 13
- 238000000605 extraction Methods 0.000 claims description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000007689 inspection Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 230000035515 penetration Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Landscapes
- Image Analysis (AREA)
- Road Repair (AREA)
Abstract
The invention discloses a pavement crack detection method based on data enhancement and multi-scale feature learning, which relates to the technical field of three-dimensional computer vision and comprises the following steps: s1, carrying out data enhancement on original road point cloud data; s2, constructing a convolutional neural network CrackU 2 Net based on multi-scale feature learning; s3, training a convolutional neural network based on multi-scale feature learning by utilizing the enhanced data, and detecting pavement crack points from the pavement point cloud.
Description
Technical Field
The invention belongs to the technical field of three-dimensional computer vision, and particularly relates to a pavement crack detection method based on data enhancement and multi-scale feature learning.
Background
The pavement crack is one of the most common pavement damages as early performance of the pavement damage, and is also an important point of attention of pavement maintenance, if the early-appearing crack is not repaired in time, the pavement crack condition can be rapidly deteriorated and even the pavement collapse is developed along with the penetration of rainwater into the pavement foundation along the crack, so that the traffic safety of the pavement is seriously threatened.
The traditional pavement crack detection method mainly relies on manual inspection and visual observation. However, the pavement crack inspection technique based on artificial vision has the following disadvantages: (1) strong subjectivity; (2) low efficiency; (3) incomplete; (4) Imprecise, in recent years, with the development of three-dimensional data acquisition technology, an MLS system has been widely used for generating pavement data containing accurate and reliable three-dimensional coordinate information, however, a depth learning-based MLS point cloud pavement crack detection algorithm still has some difficulties and challenges: (1) the three-dimensional point cloud pavement data space structure is not obvious; (2) unbalanced distribution of three-dimensional point cloud pavement data; (3) The three-dimensional point cloud pavement data has few trainable samples, is widely applied to a network architecture U-Net with a symmetrical encoding-decoding structure in the field of two-dimensional significance detection, and has proved to have excellent effects in various related application scenes and tasks. However, although the U-Net architecture is simple and powerful in processing the task related to the image, for the point cloud data with irregular structure and huge data volume, the simple U-Net architecture is difficult to obtain a satisfactory detection result, so the scheme is proposed to solve the early road crack detection.
Disclosure of Invention
The present invention aims to solve at least one of the technical problems existing in the prior art; therefore, the invention provides a pavement crack detection method based on data enhancement and multi-scale feature learning, which is used for solving the technical problems.
In order to achieve the above object, the present invention is solved by the following means: the method for detecting the pavement cracks based on data enhancement and multi-scale feature learning specifically comprises the following steps:
S1, carrying out data enhancement on original road point cloud data;
S2, constructing a convolutional neural network CrackU 2 Net based on multi-scale feature learning;
and S3, training a convolutional neural network based on multi-scale feature learning by utilizing the enhanced data, and detecting pavement crack points from the pavement point cloud.
As a further scheme of the invention, the specific steps of S1 are as follows:
S11, the original three-dimensional ground laser point cloud Dividing according to a road travelling method to obtain n blocks of road point cloud data, wherein the n blocks of road point cloud data are marked as C= { C 1,C2,…,Cn };
s12, dividing each point cloud in the C according to the sequence from left to right to obtain a left subset of the C And right subset/>For any i E [1, n ], satisfy
S13, dividing each point cloud in the C according to the sequence from top to bottom to obtain an upper subset of the CAnd lower subset/>For any i E [1, n ], satisfy
S14, splicing any point cloud mass in the left subset C l and any point cloud mass in the right subset C r in sequence from left to right to obtain a spliced point cloud setThe number of elements in C lr is n 2;
S15, splicing any point cloud mass in the upper subset C t and any point cloud mass in the lower subset C b in sequence from left to right to obtain a spliced point cloud set The number of elements in C tb is n 2;
s16, for any piece of splicing point cloud in C lr Calculation/>Normal vector/>Calculation/>Normal vector of (2)Adjustment/> according to normal vectorAnd/>Pose of (1) such that/>And/>The included angle between the two is minimum, and the adjusted/>And/>Respectively marked as/>And/>
S17, for any piece of splicing point cloud in C tb Calculation/>Normal vector/>Calculation/>Normal vector of (2)Adjustment/> according to normal vectorAnd/>Pose of (1) such that/>And/>The included angle between the two is minimum, and the adjusted/>And/>Respectively marked as/>And/>
S18, repeating the steps S14-S17 to obtain an adjusted splicing point cloud set AndThe enhanced dataset is noted as/>
S19, pairThe sample size is further expanded by random downsampling of each block of point cloud, and a fixed number of points are selected from the points cloud mass each time to serve as one input in a fixed point random sampling mode.
As a further aspect of the present invention, the specific cutting method in step S11 is as follows:
And scanning the experimental area road by adopting a mobile laser scanning system, dividing the experimental area road into n samples according to the road travelling direction, wherein each sample is of a fixed specification, and one sample is set as one road point cloud data to obtain a point cloud data set C.
As a further scheme of the invention, the specific operation steps of S2 are as follows:
CrackU 2 Net is a two-layer nested U-Net (downsampling-upsampling) structure network, the upper layer structure of the network is composed of 9 multi-scale feature learning units, each multi-scale feature learning unit comprises a multi-scale feature extraction component, and the specific structure of each learning unit is related to the layer number of the learning unit in the CrackU 2 Net network.
As a further aspect of the present invention, the 9 multi-scale feature learning units specifically are:
CrackU 2 Net contains five encoders (En 1, en2, en3, en4, en 5) and four decoders (De 4, de3, de2, de 1):
en1: the left side comprises 1 full connection layer (32), 4 local feature aggregation and random sampling layers (32-32-32-32) and 1 multi-layer perceptron layer (32), and the right side comprises 4 up-sampling and multi-layer perceptron layers (32-32-32-32) and 2 full connection layers (32-32);
En2: the left side comprises 1 full connection layer (128), 3 local feature aggregation and random sampling layers (64-64-64) and 1 multi-layer perceptron (64), and the right side comprises 3 up-sampling and multi-layer perceptrons (64-64-64) and 2 full connection layers (64-128);
En3: the left side comprises 1 full connection layer (256), 2 local feature aggregation and random sampling layers (128-128) and 1 multi-layer perceptron (128), and the right side comprises 2 up-sampling and multi-layer perceptron (128-128) and 2 full connection layers (128-256);
En4: the left side comprises 1 full connection layer (512), 1 local feature aggregation and random sampling layer (256) and 1 multi-layer perceptron (256), and the right side comprises 1 up-sampling and multi-layer perceptron (256) and 2 full connection layers (256-512);
en5: only one fully connected layer (512) is included.
De4: the left side comprises 1 full connection layer (256), 1 local feature aggregation and random sampling layer (256) and 1 multi-layer perceptron (256), and the right side comprises 1 up-sampling and multi-layer perceptron (256) and 2 full connection layers (256-256);
De3: the left side comprises 1 full connection layer (128), 2 local feature aggregation and random sampling layers (128-128) and 1 multi-layer perceptron (128), and the right side comprises 2 up-sampling and multi-layer perceptron (128-128) and 2 full connection layers (128-128);
De2: the left side comprises 1 full connection layer (32), 3 local feature aggregation and random sampling layers (64-64-64) and 1 multi-layer perceptron (64), and the right side comprises 3 up-sampling and multi-layer perceptrons (64-64-64) and 2 full connection layers (64-32);
De1: the left side contains 1 fully connected layer (32), 4 local feature aggregation and random sampling layers (32-32-32-32) and 1 multi-layer perceptron (32), and the right side has 4 up-sampling and multi-layer perceptrons (32-32-32-32) and 2 fully connected layers (32-2).
As a further scheme of the invention, the specific operation steps of S3 are as follows:
And (3) taking the enhanced data obtained in the step (S1) as a training set, training a convolutional neural network CrackU 2 Net based on multi-scale feature learning, and detecting pavement crack points from the pavement point cloud.
As a further scheme of the invention, the CrackU 2 Net model is a real-time training model, and each time, a fixed point is randomly sampled from each inputted road surface point cloud to be used as a model input.
Compared with the prior art, the invention has the beneficial effects that:
the invention provides a novel data enhancement strategy, which effectively expands the data diversity of a small sample data set through decomposing, recombining, calibrating and sampling the road point cloud, and experiments prove that the data enhancement strategy provided by the invention can effectively improve the performance of various deep learning models;
The invention provides a novel multi-scale feature learning network, which is characterized in that a multi-scale feature extraction block is designed to form a multi-scale feature learning network CrackU 2 Net to help the network extract crack features with various scales, so that the space geometry structure of pavement cracks can be comprehensively analyzed.
Drawings
FIG. 1 is a schematic diagram of a data enhancement strategy of the present invention;
FIG. 2 is a graph comparing the results of data enhancement by random downsampling according to the present invention;
FIG. 3 is a diagram of a multi-scale feature learning network CrackU 2 Net of the present invention;
FIG. 4 is a diagram of the three-dimensional point cloud pavement detection result of the present invention;
Fig. 5 is a general block diagram of the model process of the present invention.
Detailed Description
The technical solutions of the present invention will be clearly and completely described in connection with the embodiments, and it is obvious that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1 to 5, the application provides a pavement crack detection method based on data enhancement and multi-scale feature learning, which specifically comprises the following steps:
Step one: extracting road point cloud in an experimental area:
Scanning the road of the experimental area by using a mobile laser scanning system to obtain a road surface crack point cloud data set, dividing the road of the experimental area into a plurality of samples, wherein each sample is of a fixed specification, in the embodiment, the length of each sample is 10 meters and the width of each sample is 8 meters, each point cloud mass comprises about 200000 points, and the points cloud mass are the samples;
Then, traversing the points cloud mass, marking all points in the points cloud mass as crack points and non-crack points, in this embodiment, manually marking, and after marking, the ratio of the crack points to the non-crack points contained in the point cloud in each sample is approximately 2:8, 8;
step two: splitting and splicing the point cloud mass:
In this embodiment, in order to expand the diversity of crack samples contained in the road surface point cloud data set, the data enhancement method is adopted by the invention to enhance the data of the original road surface point cloud, and the specific data enhancement method is as follows:
First, the original three-dimensional ground laser point cloud Dividing according to a road travelling method to obtain n blocks of road point cloud data, namely C= { C 1,C2,…,Cn }, namely dividing a road in an experimental area into n points cloud mass according to a road travelling direction, wherein n represents different points cloud mass;
dividing each point cloud in C according to the sequence from left to right to obtain a left subset of C And right subset/>And for any i E [1, n ], satisfyDividing each point cloud in C according to the sequence from top to bottom to obtain an upper subset/>, of CAnd lower subset/>For any i E [1, n ], satisfy
Any point cloud mass in the left subset C l and any point cloud mass in the right subset C r are spliced in sequence from left to right to obtain a spliced point cloud setThe number of elements in C lr is n 2; any point cloud mass in the upper subset C t and any point cloud mass in the lower subset C b are spliced in sequence from left to right to obtain a spliced point cloud setThe number of elements in C tb is n 2;
for example, c= { C1, C2, C3, C4}, which is obtained after division according to the left subset and the right subset is completed And/>After the right splicing is finished, a splicing point cloud set/> isobtained
Cloud for any block of C lr Calculation/>Normal vector/>Calculation/>Normal vector/>Adjustment/> according to normal vectorAnd/>Pose of (1) such that/>And/>The included angle between the two is minimum, and the adjusted/>And/>Respectively marked asAnd/>Cloud/>, for any piece of splice in C tb Calculation/>Normal vector/>Calculation/>Normal vector/>Adjustment/> according to normal vectorAnd/>Pose of (1) such that/>And/>The included angle between the two is minimum, and the adjusted/>And/>Respectively marked asAnd/>
For a pair ofMidpoint cloud mass random downsampling further expands the sample size, specifically, 40960 points are randomly taken from points cloud mass at a time as one input.
In this embodiment, the data enhancement strategy proposed by the present invention is shown in fig. 1, and the comparison result between the random sampling fixed point and other downsampling modes is shown in fig. 2;
step three: constructing a convolutional neural network CrackU 2 Net based on multi-scale feature learning:
CrackU 2 Net is a two-layer nested U-Net (downsampling-upsampling) structure network, the upper layer structure of the network is composed of 9 multi-scale feature learning units, each multi-scale feature learning unit comprises a multi-scale feature extraction component, and the specific structure of each learning unit is related to the layer number of the learning unit in the CrackU 2 Net network; in this embodiment, crackU 2 Net is shown in FIG. 3;
specifically, crackU 2 Net contains five encoders (En 1, en2, en3, en4, en 5) and four decoders (De 4, de3, de2, de 1):
en1: the left side comprises 1 full connection layer (32), 4 local feature aggregation and random sampling layers (32-32-32-32) and 1 multi-layer perceptron layer (32), and the right side comprises 4 up-sampling and multi-layer perceptron layers (32-32-32-32) and 2 full connection layers (32-32);
En2: the left side comprises 1 full connection layer (128), 3 local feature aggregation and random sampling layers (64-64-64) and 1 multi-layer perceptron (64), and the right side comprises 3 up-sampling and multi-layer perceptrons (64-64-64) and 2 full connection layers (64-128);
En3: the left side comprises 1 full connection layer (256), 2 local feature aggregation and random sampling layers (128-128) and 1 multi-layer perceptron (128), and the right side comprises 2 up-sampling and multi-layer perceptron (128-128) and 2 full connection layers (128-256);
En4: the left side comprises 1 full connection layer (512), 1 local feature aggregation and random sampling layer (256) and 1 multi-layer perceptron (256), and the right side comprises 1 up-sampling and multi-layer perceptron (256) and 2 full connection layers (256-512);
En5: comprises only one fully connected layer (512);
De4: the left side comprises 1 full connection layer (256), 1 local feature aggregation and random sampling layer (256) and 1 multi-layer perceptron (256), and the right side comprises 1 up-sampling and multi-layer perceptron (256) and 2 full connection layers (256-256);
De3: the left side comprises 1 full connection layer (128), 2 local feature aggregation and random sampling layers (128-128) and 1 multi-layer perceptron (128), and the right side comprises 2 up-sampling and multi-layer perceptron (128-128) and 2 full connection layers (128-128);
De2: the left side comprises 1 full connection layer (32), 3 local feature aggregation and random sampling layers (64-64-64) and 1 multi-layer perceptron (64), and the right side comprises 3 up-sampling and multi-layer perceptrons (64-64-64) and 2 full connection layers (64-32);
De1: the left side comprises 1 full connection layer (32), 4 local feature aggregation and random sampling layers (32-32-32-32) and 1 multi-layer perceptron (32), and the right side comprises 4 up-sampling and multi-layer perceptrons (32-32-32-32) and 2 full connection layers (32-2);
step four: training the data enhancement samples:
Taking the three-dimensional point cloud pavement data enhancement sample set obtained in the second step as a training set, training a model of a convolutional neural network CrackU 2 Net based on multi-scale feature learning in the third step, detecting pavement crack points from pavement point clouds, specifically taking 70% of samples in the enhancement data as the training set, 30% of samples as a test set, and training a CrackU 2 Net model;
It should be noted that, in the CrackU 2 Net model training process, no existing pre-training model is used, all training processes are started from the beginning, 40960 points are randomly sampled from each inputted road surface point cloud as a model input; in addition, the CrackU 2 Net model uses an ADAM optimizer, the initial learning rate of the model is set to be 0.01, attenuation is set to be 0.95 of an original value every 500 steps in the training process, and the network training is set to be 50 rounds; in this example, the road surface crack detection result is shown in fig. 4.
The partial data in the formula are all obtained by removing dimension and taking the numerical value for calculation, and the formula is a formula closest to the real situation obtained by simulating a large amount of collected data through software; the preset parameters and the preset threshold values in the formula are set by those skilled in the art according to actual conditions or are obtained through mass data simulation.
The above embodiments are only for illustrating the technical method of the present invention and not for limiting the same, and it should be understood by those skilled in the art that the technical method of the present invention may be modified or substituted without departing from the spirit and scope of the technical method of the present invention.
Claims (6)
1. The pavement crack detection method based on data enhancement and multi-scale feature learning is characterized by comprising the following steps of:
S1, carrying out data enhancement on original road point cloud data;
S2, constructing a convolutional neural network CrackU 2 Net based on multi-scale feature learning;
s3, training a convolutional neural network based on multi-scale feature learning by utilizing the enhanced data, and detecting pavement crack points from pavement point clouds;
The specific steps of S1 are as follows:
S11, the original three-dimensional ground laser point cloud Dividing according to a road travelling method to obtain n blocks of road point cloud data, wherein the n blocks of road point cloud data are marked as C= { C 1,C2,…,Cn };
s12, dividing each point cloud in the C according to the sequence from left to right to obtain a left subset of the C And right subset/>For any i E [1, n ], satisfy
S13, dividing each point cloud in the C according to the sequence from top to bottom to obtain an upper subset of the CAnd lower subset/>For any i E [1, n ], satisfy
S14, splicing any point cloud mass in the left subset C l and any point cloud mass in the right subset C r in sequence from left to right to obtain a spliced point cloud setThe number of elements in C lr is n 2;
S15, splicing any point cloud mass in the upper subset C t and any point cloud mass in the lower subset C b in sequence from left to right to obtain a spliced point cloud set The number of elements in C tb is n 2;
s16, for any piece of splicing point cloud in C lr Calculation/>Normal vector/>Calculation/>Normal vector/>Adjustment/> according to normal vectorAnd/>Pose of (1) such that/>And/>The included angle between the two is minimum, and the adjusted/>And/>Respectively marked as/>And/>
S17, for any piece of splicing point cloud in C tb Calculation/>Normal vector/>Calculation/>Normal vector/>Adjustment/> according to normal vectorAnd/>Pose of (1) such that/>And/>The included angle between the two is minimum, and the adjusted/>And/>Respectively marked asAnd/>
S18, repeating the steps S14-S17 to obtain an adjusted splicing point cloud setAndThe enhanced dataset is noted as/>
S19, pairThe sample size is further expanded by random downsampling of each block of point cloud, and a fixed number of points are selected from the points cloud mass each time to serve as one input in a fixed point random sampling mode.
2. The method for detecting the pavement crack based on the data enhancement and the multi-scale feature learning of claim 1, wherein the specific cutting method in the step S11 is as follows:
And scanning the experimental area road by adopting a mobile laser scanning system, dividing the experimental area road into n samples according to the road travelling direction, wherein each sample is of a fixed specification, and one sample is set as one road point cloud data to obtain a point cloud data set C.
3. The method for detecting the pavement cracks based on data enhancement and multi-scale feature learning according to claim 1, wherein the specific operation steps of S2 are as follows:
CrackU 2 Net is a two-layer nested U-Net structure network, the upper layer structure of the network is composed of 9 multi-scale feature learning units, each multi-scale feature learning unit comprises a multi-scale feature extraction component, and the specific structure of each learning unit is related to the layer number of the learning unit in the CrackU 2 Net network.
4. The method for detecting the pavement cracks based on data enhancement and multi-scale feature learning according to claim 3, wherein the 9 multi-scale feature learning units specifically comprise:
CrackU 2 Net contains five encoders and four decoders:
en1: the left side comprises 1 full-connection layer, 4 local feature aggregation and random sampling layers and 1 multi-layer perceptron layer, and the right side comprises 4 up-sampling and multi-layer perceptron layers and 2 full-connection layers;
En2: the left side comprises 1 full-connection layer, 3 local feature aggregation and random sampling layers and 1 multi-layer perceptron, and the right side comprises 3 up-sampling and multi-layer perceptron and 2 full-connection layers;
En3: the left side comprises 1 full-connection layer, 2 local feature aggregation and random sampling layers and 1 multi-layer perceptron, and the right side comprises 2 up-sampling and multi-layer perceptron and 2 full-connection layers;
En4: the left side comprises 1 full-connection layer, 1 local feature aggregation and random sampling layer and 1 multi-layer perceptron, and the right side comprises 1 up-sampling and multi-layer perceptron and 2 full-connection layers;
En5: only one full connection layer is included;
De4: the left side comprises 1 full-connection layer, 1 local feature aggregation and random sampling layer and 1 multi-layer perceptron, and the right side comprises 1 up-sampling and multi-layer perceptron and 2 full-connection layers;
De3: the left side comprises 1 full-connection layer, 2 local feature aggregation and random sampling layers and 1 multi-layer perceptron, and the right side comprises 2 up-sampling and multi-layer perceptron and 2 full-connection layers;
de2: the left side comprises 1 full-connection layer, 3 local feature aggregation and random sampling layers and 1 multi-layer perceptron, and the right side comprises 3 up-sampling and multi-layer perceptron and 2 full-connection layers;
De1: the left side comprises 1 full-connection layer, 4 local feature aggregation and random sampling layers and 1 multi-layer perceptron, and the right side comprises 4 up-sampling and multi-layer perceptron and 2 full-connection layers.
5. The method for detecting the pavement cracks based on data enhancement and multi-scale feature learning according to claim 1, wherein the specific operation steps of S3 are as follows:
And (3) taking the enhanced data obtained in the step (S1) as a training set, training a convolutional neural network CrackU 2 Net based on multi-scale feature learning, and detecting pavement crack points from the pavement point cloud.
6. The method for detecting the pavement cracks based on data enhancement and multi-scale feature learning according to claim 5, wherein CrackU 2 Net model is a real-time training model, and each time a fixed point is randomly sampled from each inputted pavement point cloud to be used as a model input.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311811088.0A CN117788409B (en) | 2023-12-27 | 2023-12-27 | Pavement crack detection method based on data enhancement and multi-scale feature learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311811088.0A CN117788409B (en) | 2023-12-27 | 2023-12-27 | Pavement crack detection method based on data enhancement and multi-scale feature learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117788409A CN117788409A (en) | 2024-03-29 |
| CN117788409B true CN117788409B (en) | 2024-06-25 |
Family
ID=90382774
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311811088.0A Active CN117788409B (en) | 2023-12-27 | 2023-12-27 | Pavement crack detection method based on data enhancement and multi-scale feature learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117788409B (en) |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113674403B (en) * | 2021-08-26 | 2023-12-26 | 上海交通大学 | Three-dimensional point cloud up-sampling method, system, equipment and medium |
| CN114638866A (en) * | 2022-03-25 | 2022-06-17 | 西安电子科技大学 | Point cloud registration method and system based on local feature learning |
| CN117078956A (en) * | 2023-09-04 | 2023-11-17 | 吉林大学 | Point cloud classification segmentation network based on point cloud multi-scale parallel feature extraction and attention mechanism |
-
2023
- 2023-12-27 CN CN202311811088.0A patent/CN117788409B/en active Active
Non-Patent Citations (2)
| Title |
|---|
| Scale-Space Data Augmentation for Deep Transfer Learning of Crack Damage from Small Sized Datasets;Tang, SM等;《JOURNAL OF NONDESTRUCTIVE EVALUATION》;20200930;全文 * |
| 基于多尺度卷积网络的路面图像裂缝分割方法;孙梦园;刘义;范文慧;;软件;20200515(05);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117788409A (en) | 2024-03-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110222701B (en) | Automatic bridge disease identification method | |
| Wei et al. | Instance-level recognition and quantification for concrete surface bughole based on deep learning | |
| CN108876780B (en) | Bridge crack image crack detection method under complex background | |
| Akagic et al. | Pothole detection: An efficient vision based method using rgb color space image segmentation | |
| CN108460764A (en) | The ultrasonoscopy intelligent scissor method enhanced based on automatic context and data | |
| CN108960135B (en) | Dense ship target accurate detection method based on high-resolution remote sensing image | |
| CN106980858A (en) | The language text detection of a kind of language text detection with alignment system and the application system and localization method | |
| CN111241994A (en) | Method for extracting remote sensing image rural highway desertification road section for deep learning | |
| CN113409314A (en) | Unmanned aerial vehicle visual detection and evaluation method and system for corrosion of high-altitude steel structure | |
| CN111008959A (en) | Grading ring defect detection method, device, medium and equipment based on aerial image | |
| CN110532914A (en) | Building analyte detection method based on fine-feature study | |
| Dong et al. | Innovative method for pavement multiple damages segmentation and measurement by the Road-Seg-CapsNet of feature fusion | |
| CN107462204A (en) | A kind of three-dimensional pavement nominal contour extracting method and system | |
| CN111597932A (en) | Road crack image identification method, device and system based on convolutional neural network | |
| CN114066808A (en) | Pavement defect detection method and system based on deep learning | |
| CN116777898B (en) | Method for realizing crack measurement in 3D printing retaining wall construction process based on AFFormer | |
| CN115984273B (en) | Road disease detection method, device, computer equipment and readable storage medium | |
| CN115223054A (en) | Remote sensing image change detection method based on partition clustering and convolution | |
| CN114998251A (en) | Air multi-vision platform ground anomaly detection method based on federal learning | |
| Fu et al. | Extended efficient convolutional neural network for concrete crack detection with illustrated merits | |
| CN115423743A (en) | Hot-rolled strip steel surface defect detection method based on improved YOLOv5 | |
| CN117788409B (en) | Pavement crack detection method based on data enhancement and multi-scale feature learning | |
| CN113643300A (en) | Pavement crack pixel level detection method based on Seg-CapsNet algorithm | |
| CN113744185A (en) | Concrete apparent crack segmentation method based on deep learning and image processing | |
| Li et al. | Automated bridge crack detection based on improving encoder–decoder network and strip pooling |
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 |