CN110930377A - Automatic detection method for drainage pipeline abnormal type based on multitask learning - Google Patents
Automatic detection method for drainage pipeline abnormal type based on multitask learning Download PDFInfo
- Publication number
- CN110930377A CN110930377A CN201911125639.1A CN201911125639A CN110930377A CN 110930377 A CN110930377 A CN 110930377A CN 201911125639 A CN201911125639 A CN 201911125639A CN 110930377 A CN110930377 A CN 110930377A
- Authority
- CN
- China
- Prior art keywords
- image
- pipeline
- depth feature
- residual error
- depth
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Abstract
The invention relates to a drainage pipeline abnormal type automatic detection method based on multitask learning. Firstly, classifying categories with overlapped characteristic spaces into the same group by analyzing the depth characteristic information of the pipeline defects; and then constructing a multi-task learning deep neural network based on grouping conditions, wherein the network divides a classification task into two stages of tasks, a high-level classification task tries to distinguish defect images of different groups, a low-level task has a plurality of subtasks which are respectively used for distinguishing defect types with similar feature spaces in the groups, and the final defect classification result comes from conditional probability. According to the method, the multi-task learning strategy is introduced, so that the model can reduce the precision loss caused by the overlapping of the characteristic space, and the detection precision of the model is improved, and therefore a more effective automatic detection system for the drainage pipeline defects is realized.
Description
Technical Field
The invention relates to a drainage pipeline abnormal type automatic detection method based on multitask learning.
Background
The existing automatic detection method for the abnormal type of the drainage pipeline based on deep learning has the publication number of CN108038850A, and the problems and the defects of the technology are as follows: the image inside the pipeline is different from a common image, and has the particularity that the pipeline image contains abundant details and has the characteristics of low brightness, single background, weak contrast, more noise and strong variability. At the same time, pipe images exhibit large changes in visual appearance due to differences in geometry, materials, defect properties, internal lining, camera specifications, etc. This puts high demands on the generalization performance of the classification model, and increases the difficulty of the classification task. Moreover, the pipeline image has a problem of overlapping depth feature spaces, which mainly appears in two aspects. Firstly, the image of the interior of the pipeline has the characteristic of single background. Due to the single background, there may be a great deal of similarity between different types of images, and only a few details are different for determining the type of defect. Secondly, in practical situations, defects may be associated, a plurality of defects may exist in a few images, and a plurality of defects occur simultaneously, which results in overlapping of classes in the feature space, and the classifier cannot make a correct judgment on the classes. In existing classification models, classes are mutually exclusive by definition, and classification errors occur when there is overlap of classes in the feature space.
Disclosure of Invention
The invention aims to overcome the problems and provides a drainage pipeline abnormal type automatic detection method based on multi-task learning, which introduces a multi-task learning strategy and a deep learning model into drainage pipeline defect detection, thereby improving the detection precision and generalization capability of the model.
In order to achieve the purpose, the technical scheme of the invention is as follows: a drainage pipeline abnormal type automatic detection method based on multitask learning comprises the following steps:
step S1, establishing a training data set: randomly extracting video frames from a historical detection report and a video shot by a drainage pipeline robot to form a pipeline image set, labeling the images according to a pipeline evaluation mode given in a town drainage pipeline detection and evaluation technical specification for each image in the pipeline image set to form a pipeline image set S and an image label set gamma, wherein for each image set X (n) in the pipeline image set S, the image label set gamma corresponding to the image label set gamma is (gamma (1), gamma (2) and … gamma (n)), and gamma (n) represents the pipeline abnormal type of the omega (n) th image; dividing the pipeline image set into a training set, a verification set and a test set according to the proportion of 70%, 10% and 20%; in the deep learning, in order to avoid overfitting, data enhancement and standardization processing are carried out on the image data in the training set;
step S2, training a deep residual error neural network by using the training set image obtained in the step S1 through a gradient descent algorithm;
s3, recognizing the training set image in the S1 by using the neural network trained in the S2, obtaining the feature vector of the penultimate layer of the neural network by the image sequentially passing through a convolutional layer and a pooling layer, constructing a depth feature space of the defect through the feature vector, and constructing a depth feature histogram based on the depth feature space;
step S4, similarity measurement is carried out by using the depth feature histograms of various categories obtained in the step S3, the similarity of the depth feature histograms of different categories is compared one by one, and defects are grouped according to the similarity;
s5, constructing a multitask depth residual error neural network by using the grouping result in the S4, and training the neural network by using the training set image obtained in the S1 through a gradient descent algorithm;
and S6, recognizing the image to be recognized by using the neural network trained in the step S4, and obtaining the type of the defect in the image after the image sequentially passes through the convolutional layer, the pooling layer and the activation layer.
In an embodiment of the present invention, in the step S1, the abnormal type of the pipeline is divided into deformation, deposition, dislocation, corrosion, crack, undulation, leakage, and root, and the corresponding γ values are 0, 1, 2.. 6, 7.
In an embodiment of the present invention, in the step S1, the process of performing data enhancement and normalization on the image data in the training set specifically includes the following steps:
step S11, performing data enhancement on the image data in the training set; the method for enhancing the data comprises the steps of horizontal turning, rotation, image color changing and noise adding;
s12, scaling the image set subjected to data enhancement in the step S11 to a fixed size in an equal proportion according to the short edge, wherein the range is 256-512;
step S13, randomly cutting out 224 × 224 sub-pictures from the picture reduced in step S12;
step S14, performing normalization processing on the subgraph generated in step S13, specifically using the following formula:
in the formula, xiRepresenting one of the pixel points in a graph; x is the number ofminRepresenting the minimum point, x, of all pixels in the diagrammaxRepresenting the maximum point of all pixels in the graph.
In an embodiment of the present invention, in step S2, the depth residual network structure includes 5 residual convolution blocks, the residual convolution blocks are connected by a short structure, each residual convolution block includes 3 convolution layers, and 3 convolution layers of 1 × 1, 3 × 3, and 1 × 1 are connected in series to form one residual convolution block.
In an embodiment of the present invention, in the step S3, the depth feature histogram is constructed as follows:
step S31, extracting depth features by using the training set classified in the step S1 and the neural network trained in the step S2, and taking 2048-dimensional vectors of the penultimate layer of the network in the step S2 as depth feature vectors of each image in the training set;
step S32, combining the depth feature vectors acquired in the step S31 together according to categories to form a depth feature matrix;
step S33, a depth feature histogram is plotted for the depth feature matrix of each category in step S32.
In an embodiment of the present invention, in step S4, the defects are grouped according to similarity as follows:
the Babbitt distance under normal distribution comparison is used as a basis for evaluating the overlapping degree of different types of depth feature spaces, the Babbitt distance of a depth feature histogram between every two types is sequentially calculated, and the types with the threshold value smaller than 0.1 are divided into the same group; the babbitt distance used to calculate histogram identity can be described as, for histograms H1 and H2:
in the formula, H1 and H2 denote histograms to be compared, and H1(I) and H2(I) denote the value of each pixel point in the histograms.
In an embodiment of the present invention, in the step S5, the multitask depth residual error neural network is constructed by using the grouping result in the step S4 as follows:
taking a residual error network as a backbone network, sharing the first 4 residual error volume blocks among different tasks, and forming a multi-task branch after the 4 th residual error volume block; each branch consists of a residual volume block, an average pooling layer, a full link layer and a softmax activation layer; the branch is functionally divided into two task modules according to grouping results: the high-level task module and the low-level task module, wherein the classification target of the high-level task module is to classify the defect images of different groups and finally output the probability of the image attribution groups; the low-level task module is provided with a plurality of subtasks, each subtask classifies different defect defects in the group, and finally, the probability of the specific category in the image attribution group is output; the defect classification results come from the conditional probabilities:
in the formula, P (a) represents the output probability of the high-level task module, P (B | a) represents the output probability of the low-level task module, and the grouped defects are definitely divided into corresponding groups, so that the conditional probability P (a | B) is constantly equal to 1, thereby obtaining the final classification result; the training is carried out by adopting a joint optimization training method, and the loss function of the joint optimization can be expressed as the following form:
in the formula (I), the compound is shown in the specification,
for the cross-entropy loss function of the high-level task,
the sum of the functions is lost for the subtasks of the low-level task.
Compared with the prior art, the invention has the following beneficial effects: the invention provides a drainage pipeline abnormal type automatic detection method based on multitask learning. And the problem of defect characteristic overlapping exists in the pipeline image, and the difficulty of a classification task is increased. The invention designs a depth residual error network based on multi-task learning for automatically classifying the defects of pipelines. Firstly, classifying the categories with overlapped feature spaces into the same group by analyzing the pipeline defect depth feature information. The classification task is then divided into two stages, the high-level classification task attempts to distinguish between different groups of defect images, and the low-level task has multiple subtasks, each for emphasizing a type of defect within a group having a similar feature space. The final defect classification results are from the conditional probabilities.
According to the automatic detection method for the drainage pipeline abnormal type based on the multitask learning, disclosed by the invention, the performance of the model is improved by introducing the multitask learning strategy through analyzing the characteristics of the pipeline image. Compared with the prior art, the invention has the following advantages:
(1) compared with the traditional manual characteristic extraction method, the method has better generalization performance.
(2) Compared with the prior deep learning method, the method has better detection precision.
Drawings
FIG. 1 is an overall flow chart of the present invention.
Fig. 2 is a schematic structural diagram of a depth residual neural network used in the present invention for extracting image depth features.
Fig. 3 is a Shortcut connection between residual volume blocks.
FIG. 4 is a flow chart for constructing a depth feature histogram.
FIG. 5 is a schematic diagram of an improved deep residual error neural network structure for classification task according to the present invention.
FIG. 6 is a diagram of a multitasking module according to the present invention.
Detailed Description
The technical scheme of the invention is specifically explained below with reference to the accompanying drawings.
As shown in FIG. 1, the invention provides a drainage pipeline abnormal type automatic detection method based on multitask learning, which comprises the following steps:
step S1, establishing a training data set: randomly extracting video frames from a historical detection report and a video shot by a drainage pipeline robot to form a pipeline image set, labeling the images according to a pipeline evaluation mode given in a town drainage pipeline detection and evaluation technical specification for each image in the pipeline image set to form a pipeline image set S and an image label set gamma, wherein for each image set X (n) in the pipeline image set S, the image label set gamma corresponding to the image label set gamma is (gamma (1), gamma (2) and … gamma (n)), and gamma (n) represents the pipeline abnormal type of the omega (n) th image; the abnormal type of the pipeline can be divided into deformation, deposition, stagger, corrosion, fracture, fluctuation, leakage and tree root, and the gamma values corresponding to the abnormal type of the pipeline are 0, 1, 2.. 6 and 7 in sequence; dividing the pipeline image set into a training set, a verification set and a test set according to the proportion of 70%, 10% and 20%; in the deep learning, in order to avoid overfitting, data enhancement and standardization processing are carried out on the image data in the training set;
step S2, training a deep residual error neural network by using the training set image obtained in the step S1 through a gradient descent algorithm;
s3, recognizing the training set image in the S1 by using the neural network trained in the S2, obtaining the feature vector of the penultimate layer of the neural network by the image sequentially passing through a convolutional layer and a pooling layer, constructing a depth feature space of the defect through the feature vector, and constructing a depth feature histogram based on the depth feature space;
step S4, similarity measurement is carried out by using the depth feature histograms of various categories obtained in the step S3, the similarity of the depth feature histograms of different categories is compared one by one, and defects are grouped according to the similarity;
s5, constructing a multitask depth residual error neural network by using the grouping result in the S4, and training the neural network by using the training set image obtained in the S1 through a gradient descent algorithm;
and S6, recognizing the image to be recognized by using the neural network trained in the step S4, and obtaining the type of the defect in the image after the image sequentially passes through the convolutional layer, the pooling layer and the activation layer.
In this example, in step S1, the process of performing data enhancement and normalization processing on the image data in the training set specifically includes the following steps:
step S11, performing data enhancement on the image data in the training set; the method for enhancing the data comprises the steps of horizontal turning, rotation, image color changing and noise adding;
s12, scaling the image set subjected to data enhancement in the step S11 to a fixed size in an equal proportion according to the short edge, wherein the range is 256-512;
step S13, randomly cutting out 224 × 224 sub-pictures from the picture reduced in step S12;
step S14, performing normalization processing on the subgraph generated in step S13, specifically using the following formula:
in the formula, xiRepresenting one of the pixel points in a graph; x is the number ofminRepresenting the minimum point, x, of all pixels in the diagrammaxRepresenting the maximum point of all pixels in the graph.
In this example, in the step S2, the depth residual network structure (shown in fig. 2) is composed of 5 residual convolution blocks, the residual convolution blocks are connected by a short structure (shown in fig. 3), each residual convolution block is composed of 3 convolution layers, and 3 convolution layers of 1 × 1, 3 × 3, and 1 × 1 are concatenated together to form one residual convolution block. The detailed parameter settings are shown in table 1.
TABLE 1
In this example, in step S3, the depth feature histogram is constructed as follows (as shown in fig. 4):
step S31, extracting depth features by using the training set classified in the step S1 and the neural network trained in the step S2, and taking 2048-dimensional vectors of the penultimate layer of the network in the step S2 as depth feature vectors of each image in the training set;
step S32, combining the depth feature vectors acquired in the step S31 together according to categories to form a depth feature matrix;
step S33, a depth feature histogram is plotted for the depth feature matrix of each category in step S32.
In this example, in step S4, the defects are grouped according to similarity as follows:
the Babbitt distance under normal distribution comparison is used as a basis for evaluating the overlapping degree of different types of depth feature spaces, the Babbitt distance of a depth feature histogram between every two types is sequentially calculated, and the types with the threshold value smaller than 0.1 are divided into the same group; the babbitt distance used to calculate histogram identity can be described as, for histograms H1 and H2:
in the formula, H1 and H2 denote histograms to be compared, and H1(I) and H2(I) denote the value of each pixel point in the histograms.
In this example, in the step S5, the way of constructing the multitask depth residual error neural network using the grouping result in the step S4 (as shown in fig. 5) is as follows:
the network takes a residual error network as a backbone network, the first 4 residual error volume blocks are shared among different tasks, and a multi-task branch is formed after the 4 th residual error volume block; each branch consists of a residual volume block, an average pooling layer, a full link layer and a softmax activation layer; the branch is functionally divided into two task modules according to the grouping result (as shown in fig. 6): the high-level task module and the low-level task module, wherein the classification target of the high-level task module is to classify the defect images of different groups and finally output the probability of the image attribution groups; the low-level task module is provided with a plurality of subtasks, each subtask classifies different defect defects in the group, and finally, the probability of the specific category in the image attribution group is output; the defect classification results come from the conditional probabilities:
in the formula, P (a) represents the output probability of the high-level task module, P (B | a) represents the output probability of the low-level task module, and the grouped defects are definitely divided into corresponding groups, so that the conditional probability P (a | B) is constantly equal to 1, thereby obtaining the final classification result; the training is carried out by adopting a joint optimization training method, and the loss function of the joint optimization can be expressed as the following form:
in the formula (I), the compound is shown in the specification,
for the cross-entropy loss function of the high-level task,
the sum of the functions is lost for the subtasks of the low-level task.
The above are preferred embodiments of the present invention, and all changes made according to the technical scheme of the present invention that produce functional effects do not exceed the scope of the technical scheme of the present invention belong to the protection scope of the present invention.
Claims (7)
1. A drainage pipeline abnormal type automatic detection method based on multitask learning is characterized by comprising the following steps:
step S1, establishing a training data set: randomly extracting video frames from a historical detection report and a video shot by a drainage pipeline robot to form a pipeline image set, labeling the images according to a pipeline evaluation mode given in a town drainage pipeline detection and evaluation technical specification for each image in the pipeline image set to form a pipeline image set S and an image label set gamma, wherein for each image set X (n) in the pipeline image set S, the image label set gamma corresponding to the image label set gamma is (gamma (1), gamma (2) and … gamma (n)), and gamma (n) represents the pipeline abnormal type of the omega (n) th image; dividing the pipeline image set into a training set, a verification set and a test set according to the proportion of 70%, 10% and 20%; in the deep learning, in order to avoid overfitting, data enhancement and standardization processing are carried out on the image data in the training set;
step S2, training a deep residual error neural network by using the training set image obtained in the step S1 through a gradient descent algorithm;
s3, recognizing the training set image in the S1 by using the neural network trained in the S2, obtaining the feature vector of the penultimate layer of the neural network by the image sequentially passing through a convolutional layer and a pooling layer, constructing a depth feature space of the defect through the feature vector, and constructing a depth feature histogram based on the depth feature space;
step S4, similarity measurement is carried out by using the depth feature histograms of various categories obtained in the step S3, the similarity of the depth feature histograms of different categories is compared one by one, and defects are grouped according to the similarity;
s5, constructing a multitask depth residual error neural network by using the grouping result in the S4, and training the neural network by using the training set image obtained in the S1 through a gradient descent algorithm;
and S6, recognizing the image to be recognized by using the neural network trained in the step S4, and obtaining the type of the defect in the image after the image sequentially passes through the convolutional layer, the pooling layer and the activation layer.
2. The method as claimed in claim 1, wherein in step S1, the pipeline anomaly types are classified into deformation, deposition, dislocation, corrosion, cracking, fluctuation, leakage, and tree root, and the corresponding γ values are 0, 1, 2.
3. The method as claimed in claim 1, wherein in step S1, the process of performing data enhancement and normalization processing on the image data in the training set specifically includes the following steps:
step S11, performing data enhancement on the image data in the training set; the method for enhancing the data comprises the steps of horizontal turning, rotation, image color changing and noise adding;
s12, scaling the image set subjected to data enhancement in the step S11 to a fixed size in an equal proportion according to the short edge, wherein the range is 256-512;
step S13, randomly cutting out 224 × 224 sub-pictures from the picture reduced in step S12;
step S14, performing normalization processing on the subgraph generated in step S13, specifically using the following formula:
in the formula, xiRepresenting one of the pixel points in a graph; x is the number ofminRepresenting the minimum point, x, of all pixels in the diagrammaxRepresenting the maximum point of all pixels in the graph.
4. The method for automatically detecting the abnormal type of the drainage pipeline based on the multitask learning as claimed in claim 1, wherein in the step S2, the depth residual error network structure is composed of 5 residual error convolution blocks, the residual error convolution blocks are connected with each other through a short structure, each residual error convolution block is composed of 3 convolution layers, and 3 convolution layers of 1 × 1, 3 × 3 and 1 × 1 are connected in series to form one residual error convolution block.
5. The method for automatically detecting the abnormal type of the drainpipe pipeline based on the multitask learning as claimed in claim 1, wherein in the step S3, the depth feature histogram is constructed in the following way:
step S31, extracting depth features by using the training set classified in the step S1 and the neural network trained in the step S2, and taking 2048-dimensional vectors of the penultimate layer of the network in the step S2 as depth feature vectors of each image in the training set;
step S32, combining the depth feature vectors acquired in the step S31 together according to categories to form a depth feature matrix;
step S33, a depth feature histogram is plotted for the depth feature matrix of each category in step S32.
6. The method for automatically detecting the abnormal type of the drainpipe pipeline based on the multitask learning as claimed in the claim 1, wherein in the step S4, the defects are grouped according to the similarity as follows:
the Babbitt distance under normal distribution comparison is used as a basis for evaluating the overlapping degree of different types of depth feature spaces, the Babbitt distance of a depth feature histogram between every two types is sequentially calculated, and the types with the threshold value smaller than 0.1 are divided into the same group; the babbitt distance used to calculate histogram identity can be described as, for histograms H1 and H2:
in the formula, H1 and H2 denote histograms to be compared, and H1(I) and H2(I) denote the value of each pixel point in the histograms.
7. The method for automatically detecting the abnormal type of the drainpipe pipeline based on the multitask learning as claimed in claim 1, wherein in the step S5, the multitask depth residual error neural network is constructed by using the grouping result in the step S4 as follows:
taking a residual error network as a backbone network, sharing the first 4 residual error volume blocks among different tasks, and forming a multi-task branch after the 4 th residual error volume block; each branch consists of a residual volume block, an average pooling layer, a full link layer and a softmax activation layer; the branch is functionally divided into two task modules according to grouping results: the high-level task module and the low-level task module, wherein the classification target of the high-level task module is to classify the defect images of different groups and finally output the probability of the image attribution groups; the low-level task module is provided with a plurality of subtasks, each subtask classifies different defect defects in the group, and finally, the probability of the specific category in the image attribution group is output; the defect classification results come from the conditional probabilities:
in the formula, P (a) represents the output probability of the high-level task module, P (B | a) represents the output probability of the low-level task module, and the grouped defects are definitely divided into corresponding groups, so that the conditional probability P (a | B) is constantly equal to 1, thereby obtaining the final classification result; the training is carried out by adopting a joint optimization training method, and the loss function of the joint optimization can be expressed as the following form:
in the formula (I), the compound is shown in the specification,
for the cross-entropy loss function of the high-level task,
the sum of the functions is lost for the subtasks of the low-level task.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911125639.1A CN110930377B (en) | 2019-11-18 | 2019-11-18 | Drainage pipeline abnormal type automatic detection method based on multitask learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911125639.1A CN110930377B (en) | 2019-11-18 | 2019-11-18 | Drainage pipeline abnormal type automatic detection method based on multitask learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110930377A true CN110930377A (en) | 2020-03-27 |
| CN110930377B CN110930377B (en) | 2022-05-13 |
Family
ID=69854055
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911125639.1A Active CN110930377B (en) | 2019-11-18 | 2019-11-18 | Drainage pipeline abnormal type automatic detection method based on multitask learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110930377B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111598179A (en) * | 2020-05-21 | 2020-08-28 | 国网电力科学研究院有限公司 | Power monitoring system user abnormal behavior analysis method, storage medium and equipment |
| CN111814873A (en) * | 2020-07-07 | 2020-10-23 | 广州市运通水务有限公司 | Method for distinguishing drainage pipeline defect types and automatically identifying defect grades |
| CN112381165A (en) * | 2020-11-20 | 2021-02-19 | 河南爱比特科技有限公司 | Intelligent pipeline defect detection method based on RSP model |
| CN113221710A (en) * | 2021-04-30 | 2021-08-06 | 深圳市水务工程检测有限公司 | Neural network-based drainage pipeline defect identification method, device, equipment and medium |
| CN114120209A (en) * | 2022-01-27 | 2022-03-01 | 深圳市博铭维技术股份有限公司 | Pipeline defect detection method, system, equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060228027A1 (en) * | 2001-03-28 | 2006-10-12 | Canon Kabushiki Kaisha | Dynamically reconfigurable signal processing circuit, pattern recognition apparatus, and image processing apparatus |
| CN108038850A (en) * | 2017-12-08 | 2018-05-15 | 天津大学 | A kind of drainage pipeline Exception Type automatic testing method based on deep learning |
| CN109559302A (en) * | 2018-11-23 | 2019-04-02 | 北京市新技术应用研究所 | Pipe video defect inspection method based on convolutional neural networks |
| CN109783906A (en) * | 2018-12-29 | 2019-05-21 | 东北大学 | A kind of pipeline detection magnetic flux leakage data intelligent analysis system and method |
-
2019
- 2019-11-18 CN CN201911125639.1A patent/CN110930377B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060228027A1 (en) * | 2001-03-28 | 2006-10-12 | Canon Kabushiki Kaisha | Dynamically reconfigurable signal processing circuit, pattern recognition apparatus, and image processing apparatus |
| CN108038850A (en) * | 2017-12-08 | 2018-05-15 | 天津大学 | A kind of drainage pipeline Exception Type automatic testing method based on deep learning |
| CN109559302A (en) * | 2018-11-23 | 2019-04-02 | 北京市新技术应用研究所 | Pipe video defect inspection method based on convolutional neural networks |
| CN109783906A (en) * | 2018-12-29 | 2019-05-21 | 东北大学 | A kind of pipeline detection magnetic flux leakage data intelligent analysis system and method |
Non-Patent Citations (2)
| Title |
|---|
| JIAN FENG ET AL.: "Injurious or Noninjurious Defect Identification From MFL Images in Pipeline Inspection Using Convolutional Neural Network", 《IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT》 * |
| 户莹: "基于深度学习的地下排水管道缺陷智能检测技术研究", 《中国优秀博硕士学位论文全文数据库(硕士)工程科技Ⅱ辑》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111598179A (en) * | 2020-05-21 | 2020-08-28 | 国网电力科学研究院有限公司 | Power monitoring system user abnormal behavior analysis method, storage medium and equipment |
| CN111598179B (en) * | 2020-05-21 | 2022-10-04 | 国网电力科学研究院有限公司 | Power monitoring system user abnormal behavior analysis method, storage medium and equipment |
| CN111814873A (en) * | 2020-07-07 | 2020-10-23 | 广州市运通水务有限公司 | Method for distinguishing drainage pipeline defect types and automatically identifying defect grades |
| CN112381165A (en) * | 2020-11-20 | 2021-02-19 | 河南爱比特科技有限公司 | Intelligent pipeline defect detection method based on RSP model |
| CN112381165B (en) * | 2020-11-20 | 2022-12-20 | 河南爱比特科技有限公司 | Intelligent pipeline defect detection method based on RSP model |
| CN113221710A (en) * | 2021-04-30 | 2021-08-06 | 深圳市水务工程检测有限公司 | Neural network-based drainage pipeline defect identification method, device, equipment and medium |
| CN114120209A (en) * | 2022-01-27 | 2022-03-01 | 深圳市博铭维技术股份有限公司 | Pipeline defect detection method, system, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110930377B (en) | 2022-05-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110930377B (en) | Drainage pipeline abnormal type automatic detection method based on multitask learning | |
| CN110222604B (en) | Target identification method and device based on shared convolutional neural network | |
| CN112561910A (en) | Industrial surface defect detection method based on multi-scale feature fusion | |
| WO2021139197A1 (en) | Image processing method and apparatus | |
| CN110610475A (en) | Visual defect detection method of deep convolutional neural network | |
| US20230014823A1 (en) | Defect detection in image space | |
| CN112200045A (en) | Remote sensing image target detection model establishing method based on context enhancement and application | |
| CN113378829B (en) | Weak supervision target detection method based on positive and negative sample balance | |
| CN112766110A (en) | Training method of object defect recognition model, object defect recognition method and device | |
| CN111242955B (en) | Road surface crack image segmentation method based on full convolution neural network | |
| CN113256624A (en) | Continuous casting round billet defect detection method and device, electronic equipment and readable storage medium | |
| CN114863464B (en) | Second-order identification method for PID drawing picture information | |
| CN113362277A (en) | Workpiece surface defect detection and segmentation method based on deep learning | |
| CN114820625A (en) | Automobile top block defect detection method | |
| CN111104855B (en) | Workflow identification method based on time sequence behavior detection | |
| CN111862065A (en) | Power transmission line diagnosis method and system based on multitask deep convolutional neural network | |
| CN115170988A (en) | Power grid line fault identification method and system based on deep learning | |
| CN114565607A (en) | Fabric defect image segmentation method based on neural network | |
| CN111368865A (en) | Method and device for detecting remote sensing image oil storage tank, readable storage medium and equipment | |
| CN106952236B (en) | Fisheye lens shot image distortion correction method based on BP neural network | |
| CN111950357A (en) | Marine water surface garbage rapid identification method based on multi-feature YOLOV3 | |
| CN111539931A (en) | Appearance abnormity detection method based on convolutional neural network and boundary limit optimization | |
| CN115082449B (en) | Electronic component defect detection method | |
| CN111666822A (en) | Low-altitude unmanned aerial vehicle target detection method and system based on deep learning | |
| CN116740572A (en) | Marine vessel target detection method and system based on improved YOLOX |
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 |