CN112183187A - Liquid leakage detection method based on Selective Search and LeNet-5 - Google Patents
Liquid leakage detection method based on Selective Search and LeNet-5 Download PDFInfo
- Publication number
- CN112183187A CN112183187A CN202010822101.2A CN202010822101A CN112183187A CN 112183187 A CN112183187 A CN 112183187A CN 202010822101 A CN202010822101 A CN 202010822101A CN 112183187 A CN112183187 A CN 112183187A
- Authority
- CN
- China
- Prior art keywords
- liquid
- layer
- layers
- model
- lenet
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Examining Or Testing Airtightness (AREA)
Abstract
The invention relates to a liquid leakage detection method based on Selective Search and LeNet-5, which comprises the following steps: intercepting a monitoring video of the state of the pipeline from a monitoring system of a heating station, wherein the monitoring video comprises pictures of liquid leakage and other interference objects of the pipeline; extracting a candidate area of suspected leakage liquid from a static scene where a heating pipeline is located by each picture of the collected video through a Selective Search algorithm, wherein the candidate area comprises two pictures of the leaked liquid and other interference objects; dividing the data set of the candidate region into a training data set and a testing data set; setting parameters of a network model LeNet-5; and training the VGG-16 model and adjusting the hyperparameter.
Description
Technical Field
The invention relates to a liquid leakage detection method, belonging to the fusion of Selective Search, LeNet-5 technology and optical leakage detection field.
Background
The leakage of the pipeline directly causes environmental pollution and energy waste, even safety accidents, and causes great economic loss, so that the leakage detection and maintenance of the pipeline are an important link in pipeline engineering. Many researchers at home and abroad have been engaged in pipeline leakage diagnosis for many years and have researched various detection methods. In principle, the pipeline leakage detection method can be divided into two major categories, namely a hardware method and a software method. The hardware method comprises the steps of leak detection of an optical fiber temperature sensor, an acoustic measurement method and the like. The optical fiber temperature sensor leak detection is to detect the temperature of the surrounding environment of the pipeline so as to judge the running state and the leakage condition of the pipeline. However, the optical fiber temperature sensor has high detection error rate, large limitation and needs regular maintenance. The acoustic detection method is that when the pipeline leaks, whether the pipeline leaks or not is judged by analyzing high-frequency vibration noise generated by the pipeline wall. The method is susceptible to noise and environmental interference, resulting in misjudgment. The software methods include measuring flow rate and pressure changes, transient flow simulation, and the like. The method of measuring the flow rate and pressure changes is to determine if a leak has occurred by observing changes in the flow rate or pressure at the input and output of the tubing. The method can only detect large leakage and is only suitable for detecting the fluid leakage with low pressure and approximate to a static state. The transient flow simulation method needs to establish a real-time mathematical model of the pipeline, the model considers various variables to predict the state of the pipeline, and when the difference between an actual measured value and a calculated value of the model exceeds a threshold value, the existence of leakage is indicated. Although the method can determine the time of leakage and the size of leakage amount, the workload of modeling and calculation is quite large, and the maintenance cost is high. With the development of deep learning, image-based detection methods are becoming mainstream due to their characteristics of large detection range, high speed, high precision, strong anti-interference capability, and the like.
Disclosure of Invention
The invention aims to provide a novel non-contact liquid leakage detection method based on Selective Search and LeNet-5, which has high efficiency, strong robustness and generalization capability. The method uses image data with pipeline liquid leakage and other interference objects, adopts a scheme of region identification, extracts candidate regions (leaked liquid and other interference objects) with suspected liquid leakage in each picture by using a Selective Search algorithm, extracts corresponding features of each candidate region through a feature extraction layer of a network LeNet-5, performs secondary classification on the leaked liquid and other interference objects through a classification layer of the LeNet-5, and finally realizes detection on the pipeline leakage liquid. The technical scheme is as follows:
a liquid leakage detection method based on Selective Search and LeNet-5 comprises the following steps:
firstly, intercepting monitoring videos of the state of the pipeline from a monitoring system of a heating station, wherein the monitoring videos comprise pictures of liquid leakage and other interference objects of the pipeline.
Secondly, extracting a candidate area of suspected leakage liquid from a static scene where a heating pipeline is located by each picture of the collected video through an algorithm Selective Search, wherein the candidate area comprises two pictures of the leaked liquid and other interference objects;
thirdly, using the detection result of the Selective Search algorithm as a data set of candidate regions to be classified;
fourthly, dividing the data set of the candidate region into a training data set and a testing data set;
fifthly, setting parameters of a network model LeNet-5;
LeNet-5 has 7 layers in total, including 3 convolution layers, two pooling layers and two full-connection layers; the 7-layer network model structure comprises a convolution layer, a pooling layer, a convolution layer, a full-connection layer and a full-connection layer; local features of the leakage liquid or other interference object regions are extracted by the convolution layers and the pooling layers of the first five layers, the local features are integrated into global features by the fully-connected layers of the last two layers, and the richer global features of the leakage liquid or other interference object regions are extracted; an output layer, namely a classification layer, is arranged after the seventh full connecting layer, and the output is set to be 2, namely the second classification of the leaked liquid and other interference objects;
sixthly, training the VGG-16 model and adjusting the hyperparameter;
1) selecting a cross entropy function as a loss function, using an optimization algorithm Adam and a back propagation adjustment model structure and a hyper-parameter, and continuing training;
2) drawing a verification set accuracy and loss function curve of the model;
seventhly, testing and storing the model;
1) testing the classification accuracy of the model by using the test set;
2) and storing the model with the highest accuracy as a final model of the liquid leakage detection system.
Drawings
Fig. 1 is a schematic view of a leak condition in a pipe. (a) No liquid (water) leakage state of valve (b) liquid (water) leakage state of valve
FIG. 2LeNet-5 framework
FIG. 3 Algorithm flow chart of the present invention
Detailed Description
The invention is further illustrated and described below with reference to the accompanying drawings and specific examples.
Referring to fig. 1, in (a), a state where no liquid (water) leaks from the valve, and (b), a state where liquid (water) leaks from the valve.
See fig. 2, LeNet-5 framework.
Referring to fig. 3, the algorithm of the present invention is a flow chart. The corresponding steps are briefly described as follows:
1. and intercepting a monitoring video of the state of the pipeline from a monitoring system of the heating station, wherein the monitoring video comprises the leakage of the liquid in the pipeline and other interference objects.
2. And extracting a candidate area of suspected leakage liquid (leaked liquid and other interference objects) from a static scene where the heating pipeline is located by using a Selective Search algorithm for each picture of the acquired video.
The following is a specific method of the Selective Search algorithm:
1) dividing each picture in the collected video into a plurality of sub-blocks, and generating a region set R by the sub-blocks;
2) calculating the similarity S of each adjacent region sub-block in the region set R, wherein the similarity S is { S1, S2, … }, and the similarity calculation method adopts color space (RGB, HSV);
3) finding out two areas with highest similarity, merging the two areas into a new set, and adding the new set into S1;
4) removing from R all subsets relevant in S1; continuing to merge R subsets according to the method until the R subsets are empty;
5) according to the method of subset combination in R, calculating the similarity of all the subsets in the new set S1, and continuing to combine and add to S2; and continue to merge S2-S3. The regions of the subset of Sn are detected as leaking liquid or other interfering objects.
3. The results of the Selective Search algorithm detection (leaking liquid and interfering objects) are taken as the data set of the candidate regions to be classified.
4. The data set of the candidate region (10000 sample pictures) was divided into a training data set (8000 samples) and a test data set (2000 samples) at a ratio of 8: 2.
5. And setting parameters of the network model LeNet-5.
LeNet-5 has 7 layers in total, and the network model structures are respectively a convolution layer, a pooling layer, a convolution layer, a full-connection layer and a full-connection layer. The convolution layers and the pooling layers of the first five layers extract local characteristics of leaked liquid and other interference objects, and the fully-connected layers of the second two layers integrate the local characteristics into global characteristics to extract richer global characteristics of the leaked liquid and other interference objects. There is also an output layer after the seventh fully bonded layer, the output set to 2, i.e. a second classification of leaking liquid and other interfering objects.
The following are the parameter settings of the network model LeNet-5:
1) picture settings of incoming leaked liquid and other interfering objects.
Candidate regions detected by the Selective Search algorithm are different in size, and in order to facilitate data processing of leaked liquid and other interference objects in the later period, extracted candidate region pictures to be classified are unified from resize to 32 × 32.
2) And setting network model parameters.
The 7-layer model structure of the network is a convolution layer, a pooling layer, a convolution layer, a full-connection layer and a full-connection layer. The following is a specific set of 7 layers:
a first layer of convolutional layers: the convolution kernel size is 5 x 5, the convolution kernel type is 6, the input size of the leaked liquid and other interference objects is (32,32,1), and the output is (28,28, 6);
a second pooling layer: adopting a maximum pooling mode, wherein the pooling window is 2, the step length is 2, the input size is (28,28,6), and the output is (14,14, 6);
a third layer of convolutional layers: the convolution kernel size is 5 x 5, the convolution kernel type is 16, the input size is (14,14,6), and the output is (10,10, 16);
a fourth pooling layer: adopting a maximum pooling mode, wherein a pooling window is 2, the step length is 2, the input size is (10,10,16), and the output is (5,5, 16);
a fifth layer of convolutional layers: the convolution kernel size is 5 x 5, the convolution kernel type is 120, the input size is (5,5,16), and the output is (1, 120).
Sixth full tie layer: the input image size is (1 x 120), the output is 120;
a seventh fully-connected layer: input 120, output 84;
an output layer: the eigenvectors output by the seventh fully-connected layer pass through the output layer, and the output unit of the output layer is set to 2, i.e., two classifications of leaked liquid and other interfering objects.
6. Model training and adjusting hyper-parameters.
1) Selecting a cross entropy function as a loss function, using an optimization algorithm Adam and a back propagation adjustment model structure and a hyper-parameter, and continuing training;
2) and drawing a verification set accuracy and loss function curve of the model.
7. And testing and storing the model.
1) Testing the classification accuracy of the model by using the test set;
2) and storing the model with the highest accuracy as a final model of the liquid leakage detection system.
Liquid leak test evaluation based on Selective Search and LeNet-5:
the accuracy is as follows: 95.2 percent.
The invention has the following beneficial effects:
1) the invention combines the Selective Search algorithm, the LeNet-5 characteristic extraction and the SVM classification, provides a new idea of liquid leakage detection based on images, and realizes the detection of the leakage liquid which is difficult to be identified by human eyes.
2) The algorithm has strong robustness and generalization capability, and the experimental program can be used universally under various operating systems.
3) The whole design algorithm is simple in structure and high in speed. The method can detect in real time in the application of actual industrial environment and warn abnormal leakage in time.
Claims (2)
1. A liquid leakage detection method based on Selective Search and LeNet-5 comprises the following steps:
firstly, intercepting monitoring videos of the state of the pipeline from a monitoring system of a heating station, wherein the monitoring videos comprise pictures of liquid leakage and other interference objects of the pipeline.
Secondly, extracting a candidate area of suspected leakage liquid from a static scene where a heating pipeline is located by each picture of the collected video through an algorithm Selective Search, wherein the candidate area comprises two pictures of the leaked liquid and other interference objects;
thirdly, using the detection result of the Selective Search algorithm as a data set of candidate regions to be classified;
fourthly, dividing the data set of the candidate region into a training data set and a testing data set;
and fifthly, setting parameters of a network model LeNet-5:
LeNet-5 has 7 layers in total, including 3 convolution layers, two pooling layers and two full-connection layers; the 7-layer network model structure comprises a convolution layer, a pooling layer, a convolution layer, a full-connection layer and a full-connection layer; local features of the leakage liquid or other interference object regions are extracted by the convolution layers and the pooling layers of the first five layers, the local features are integrated into global features by the fully-connected layers of the last two layers, and the richer global features of the leakage liquid or other interference object regions are extracted; an output layer, namely a classification layer, is arranged after the seventh full connecting layer, and the output is set to be 2, namely the second classification of the leaked liquid and other interference objects;
sixthly, training the VGG-16 model, and adjusting the hyper-parameters:
1) selecting a cross entropy function as a loss function, using an optimization algorithm Adam and a back propagation adjustment model structure and a hyper-parameter, and continuing training;
2) drawing a verification set accuracy and loss function curve of the model;
seventhly, testing and storing the model;
1) testing the classification accuracy of the model by using the test set;
2) and storing the model with the highest accuracy as a final model of the liquid leakage detection system.
2. The method of claim 1, wherein the second step is performed by:
1) dividing the picture into a plurality of small areas through a region division algorithm;
2) by continuously aggregating adjacent small areas with the color space RGB or HSV as the similarity and the area size, the small areas are aggregated first, so that the small areas are prevented from being continuously aggregated by the large areas, and the incomplete hierarchical relationship is caused;
3) the area detected by the aggregation is the area containing the leaked liquid or other interfering objects.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010822101.2A CN112183187A (en) | 2020-08-15 | 2020-08-15 | Liquid leakage detection method based on Selective Search and LeNet-5 |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010822101.2A CN112183187A (en) | 2020-08-15 | 2020-08-15 | Liquid leakage detection method based on Selective Search and LeNet-5 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112183187A true CN112183187A (en) | 2021-01-05 |
Family
ID=73919937
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010822101.2A Pending CN112183187A (en) | 2020-08-15 | 2020-08-15 | Liquid leakage detection method based on Selective Search and LeNet-5 |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112183187A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180341859A1 (en) * | 2017-05-24 | 2018-11-29 | Southwest Research Institute | Detection of Hazardous Leaks from Pipelines Using Optical Imaging and Neural Network |
| CN109712140A (en) * | 2019-01-02 | 2019-05-03 | 中楹青创科技有限公司 | Method and device of the training for the full link sort network of evaporating, emitting, dripping or leaking of liquid or gas detection |
| CN110185939A (en) * | 2019-05-16 | 2019-08-30 | 西北工业大学 | Gas pipe leakage recognition methods based on convolutional neural networks |
| CN110243542A (en) * | 2019-05-31 | 2019-09-17 | 国网上海市电力公司 | A kind of SF6Gas leaks imaging detection method and detection device |
| CN110335294A (en) * | 2019-07-11 | 2019-10-15 | 中国矿业大学 | Mine water pump house leakage detection method based on frame difference method Yu 3D convolutional neural networks |
| CN110414478A (en) * | 2019-08-08 | 2019-11-05 | 东莞德福得精密五金制品有限公司 | The contingency liquefied gas leak supervision method of the non-Application inductor of artificial intelligence cloud computing |
-
2020
- 2020-08-15 CN CN202010822101.2A patent/CN112183187A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180341859A1 (en) * | 2017-05-24 | 2018-11-29 | Southwest Research Institute | Detection of Hazardous Leaks from Pipelines Using Optical Imaging and Neural Network |
| CN109712140A (en) * | 2019-01-02 | 2019-05-03 | 中楹青创科技有限公司 | Method and device of the training for the full link sort network of evaporating, emitting, dripping or leaking of liquid or gas detection |
| CN110185939A (en) * | 2019-05-16 | 2019-08-30 | 西北工业大学 | Gas pipe leakage recognition methods based on convolutional neural networks |
| CN110243542A (en) * | 2019-05-31 | 2019-09-17 | 国网上海市电力公司 | A kind of SF6Gas leaks imaging detection method and detection device |
| CN110335294A (en) * | 2019-07-11 | 2019-10-15 | 中国矿业大学 | Mine water pump house leakage detection method based on frame difference method Yu 3D convolutional neural networks |
| CN110414478A (en) * | 2019-08-08 | 2019-11-05 | 东莞德福得精密五金制品有限公司 | The contingency liquefied gas leak supervision method of the non-Application inductor of artificial intelligence cloud computing |
Non-Patent Citations (4)
| Title |
|---|
| |旧市拾荒|: "目标检测算法之Fast R-CNN和Faster R-CNN原理", 《博客园：HTTPS://WWW.CNBLOGS.COM/XIAOYH/P/11789625.HTML》 * |
| J. R. R. UIJLINGS，ET AL.: "Selective Search for Object Recognition", 《INT J COMPUT VIS》 * |
| JIHAO SHI，ET AL.: "Real-time leak detection using an infrared camera and Faster R-CNN technique", 《COMPUTERS AND CHEMICAL ENGINEERING》 * |
| 乐毅，王斌: "《深度学习-Caffe之经典模型详解与实战》", 31 December 2016, 电子工业出版社 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109555979B (en) | Water supply pipe network leakage monitoring method | |
| CN112258496A (en) | Underground drainage pipeline disease segmentation method based on full convolution neural network | |
| CN111915579B (en) | Liquid leakage detection method based on Gaussian mixture model and convolutional neural network | |
| CN100552668C (en) | Leakage detecting and locating method based on pressure and sound wave information fusion | |
| CN103968256B (en) | Piping for tank farm leakage detection method | |
| CN105389814A (en) | Air bubble detection method for air tightness test | |
| CN112016251B (en) | Nuclear power device fault diagnosis method and system | |
| Li et al. | An accurate leakage localization method for water supply network based on deep learning network | |
| CN112762362A (en) | Underwater pipeline leakage acoustic emission detection method based on convolutional neural network | |
| CN115751203A (en) | Natural gas pipeline leakage monitoring system based on thermal infrared imager | |
| CN111695465A (en) | Pipe network fault diagnosis and positioning method and system based on pressure wave mode identification | |
| CN112097125B (en) | Water supply pipe network pipe burst detection and positioning method based on self-adaptive checking | |
| CN113486950A (en) | Intelligent pipe network water leakage detection method and system | |
| CN112284619A (en) | Oil leakage real-time monitoring device and detection method | |
| Lučin et al. | Detailed leak localization in water distribution networks using random forest classifier and pipe segmentation | |
| CN112183185A (en) | Liquid leakage detection method based on optical flow method and CNN-SVM | |
| Leonzio et al. | Water Leak Detection and Localization using Convolutional Autoencoders | |
| CN112183187A (en) | Liquid leakage detection method based on Selective Search and LeNet-5 | |
| CN113970073B (en) | ResNet-based water supply network leakage accurate positioning method | |
| Han et al. | Novel Long Short-Term Memory Model Based on the Attention Mechanism for the Leakage Detection of Water Supply Processes | |
| CN113781513B (en) | Leakage detection method and system for water supply pipeline of power plant | |
| Wang et al. | Developing and Implementing an AI-Based Leak Detection System in a Long-Distance Gas Pipeline | |
| Ferreira et al. | Corrosion-like defect severity estimation in pipelines using convolutional neural networks | |
| CN113551157B (en) | Oil-gas data acquisition system and method based on sensor network | |
| Anqi et al. | Convolutional Neural Network-based Leakage Detection of Crude Oil Transmission Pipes |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210105 |