AU2021106346A4 - Unsupervised coal flow anomaly detection method based on a generative adversarial learning - Google Patents

Unsupervised coal flow anomaly detection method based on a generative adversarial learning Download PDF

Info

Publication number
AU2021106346A4
AU2021106346A4 AU2021106346A AU2021106346A AU2021106346A4 AU 2021106346 A4 AU2021106346 A4 AU 2021106346A4 AU 2021106346 A AU2021106346 A AU 2021106346A AU 2021106346 A AU2021106346 A AU 2021106346A AU 2021106346 A4 AU2021106346 A4 AU 2021106346A4
Authority
AU
Australia
Prior art keywords
image
coal flow
set image
obtaining
test set
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.)
Ceased
Application number
AU2021106346A
Inventor
Zhipeng JIANG
Meng LEI
Yong Wang
Liang ZOU
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China University of Mining and Technology CUMT
Original Assignee
China University of Mining and Technology CUMT
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by China University of Mining and Technology CUMT filed Critical China University of Mining and Technology CUMT
Priority to AU2021106346A priority Critical patent/AU2021106346A4/en
Application granted granted Critical
Publication of AU2021106346A4 publication Critical patent/AU2021106346A4/en
Ceased legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B23/00—Testing or monitoring of control systems or parts thereof
    • G05B23/02—Electric testing or monitoring
    • G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
    • G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
    • G05B23/0243—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults model based detection method, e.g. first-principles knowledge model
    • G05B23/0254—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults model based detection method, e.g. first-principles knowledge model based on a quantitative model, e.g. mathematical relationships between inputs and outputs; functions: observer, Kalman filter, residual calculation, Neural Networks
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B23/00—Testing or monitoring of control systems or parts thereof
    • G05B23/02—Electric testing or monitoring
    • G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
    • G05B23/0259—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the response to fault detection
    • G05B23/0275—Fault isolation and identification, e.g. classify fault; estimate cause or root of failure
    • 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
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • E—FIXED CONSTRUCTIONS
    • E21—EARTH OR ROCK DRILLING; MINING
    • E21C—MINING OR QUARRYING
    • E21C41/00—Methods of underground or surface mining; Layouts therefor
    • E21C41/16—Methods of underground mining; Layouts therefor
    • E21C41/18—Methods of underground mining; Layouts therefor for brown or hard coal
    • E—FIXED CONSTRUCTIONS
    • E21—EARTH OR ROCK DRILLING; MINING
    • E21C—MINING OR QUARRYING
    • E21C41/00—Methods of underground or surface mining; Layouts therefor
    • E21C41/26—Methods of surface mining; Layouts therefor
    • E21C41/28—Methods of surface mining; Layouts therefor for brown or hard coal
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B23/00—Testing or monitoring of control systems or parts thereof
    • G05B23/02—Electric testing or monitoring
    • G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
    • G05B23/0259—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the response to fault detection
    • G05B23/0275—Fault isolation and identification, e.g. classify fault; estimate cause or root of failure
    • G05B23/0281—Quantitative, e.g. mathematical distance; Clustering; Neural networks; Statistical analysis
    • 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
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Automation & Control Theory (AREA)
  • Theoretical Computer Science (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

An unsupervised coal flow anomaly detection method based on a generative adversarial learning, comprising the following steps: obtaining a coal flow picture in a normal scene, and preprocessing the coal flow picture, and obtaining a first output image, the first output image containing a training set image and a test set image; taking the training set image as an input, training and establishing a detection model; taking the test set image as a model input, and obtaining a reconstructed second output image; calculating a first output result as an abnormal score for evaluating a reconstruction quality; and determining whether a coal flow is abnormal. The present invention collects images through a small instrument to solve the defects in the prior art that foreign object images are difficult to collect, and the neural network is used to achieve high-precision detection of abnormal coal flow scenes. -1/1 obtaining a coal flow picture in a normal scene, and preprocessesing the coal flow picture, and obtaining a first output image, the first output S100 image containing a training set image and a test set image taking the training set image as a input. training and establishing a detection model, a S101 according to the detection model taking the test set image as a model input and substituting the test set image into the detection S102 model, and then obtaining a reconstructed second output image calculating a distance L, between the test set image and the reconstructed second output image, calculating a vector distance L2 between the test set image and the reconstructed second output image in alatent space, performing a weighted addition S103 operation on the distance Li and the vector distance L2 to obtaining a first output result as an abnormal score for evaluating a reconstruction quality determining whether a coal Howis abnormal S104 FIG. 1

Description

-1/1
obtaining a coal flow picture in a normal scene, and preprocessesing the coal flow picture, and obtaining a first output image, the first output S100 image containing a training set image and a test set image
taking the training set image as a input. training and establishing a detection model, a S101 according to the detection model
taking the test set image as a model input and substituting the test set image into the detection S102 model, and then obtaining a reconstructed second output image
calculating a distance L, between the test set image and the reconstructed second output image, calculating a vector distance L 2 between the test set image and the reconstructed second output image in alatent space, performing a weighted addition S103 operation on the distance Li and the vector distance L2 to obtaining a first output result as an abnormal score for evaluating a reconstruction quality
determining whether a coal Howis abnormal S104
FIG. 1
UNSUPERVISED COAL FLOW ANOMALY DETECTION METHOD BASED ON A GENERATIVE ADVERSARIAL LEARNING TECHNICAL FIELD
[0001] The present invention relates to a field of computer vision technology in deep learning, especially relates to an unsupervised coal flow anomaly detection method based on a generative adversarial learning.
BACKGROUND
[0002] Existing detection algorithms such as the YOLO series often require a large number of labeled data sets during the training process. However, it is difficult to collect enough foreign material samples for training in actual mine production. Meanwhile, calibration of foreign objects is also a laborious task.
[0003] The ordinary generative model existing in the prior art often requires a long period of training to achieve a better reconstruction ability for normal samples. Secondly, during the training phase, the number of normal pictures is extremely large, and there are often some backgrounds that have nothing to do with coal flow in the pictures. After the generator is trained on such a data set, it may have a better reconstruction of anomalies in the testing phase. Therefore, the present invention proposes an unsupervised coal flow foreign body detection algorithm that can shorten the training period and improve the detection effect, using the auto-encoding model to replace the ordinary generative model and using the attention mechanism to guide the model to focus on the coal flow area, and then identifying the frame containing foreign matter in the coal flow video.
SUMMARY
[0004] The present invention collects images through a small instrument to solve the defects in the prior art that foreign object images are difficult to collect and have a high rate of misjudgment , and the neural network is used to achieve high-precision detection of abnormal coal flow scenes.
[0005] The unsupervised coal flow anomaly detection method based on the generative adversarial learning, comprising the following steps:
[0006] Si: obtaining a coal flow picture in a normal scene, and preprocessing the coal flow picture, and obtaining a first output image, the first output image containing a training set image and a test set image;
[0007] S2: taking the training set image as an input, training and establishing a detection model;
[0008] S3: taking the test set image as a model input and substituting the test set image into the detection model, and then obtaining a reconstructed second output image;
[0009] S4: calculating a distance Li between the test set image and the reconstructed second output image, calculating a vector distance L2 between the test set image and the reconstructed second output image in a latent space, performing a weighted addition operation on the distance Li and the vector distance L2 to obtaining a first output result as an abnormal score for evaluating a reconstruction quality; and
[0010] S5: determining whether a coal flow is abnormal.
[0011] The unsupervised coal flow anomaly detection method based on the generative adversarial learning according to the present invention, S2 can accelerating and optimizing a learning effect of the detection model with the help of a jump connection between a neural network channel and a guidance of an attention mechanism. Preferably, the attention mechanism can be introduced in the jump connection process, thereby reducing training time.
[0012] The unsupervised coal flow anomaly detection method based on the generative adversarial learning according to the present invention, in S2, in order to verify the abnormal detection performance of the detection model, it is preferable to use a data set containing normal and abnormal coal flow pictures to verify it.
[0013] The unsupervised coal flow anomaly detection method based on the generative adversarial learning according to the present invention, in S4, according to an abnormal score, preferably, selecting an appropriate threshold value with the help of a ROC curve.
[0014] The unsupervised coal flow anomaly detection method based on the generative adversarial learning according to the present invention, wherein, in S5, if a real-time image is larger than the threshold value, judging the real-time image as abnormal.
BRIEF DESCRIPTION OF DRAWINGS
[0015] In order to more clearly explain the technical solutions of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are of the present invention. For some of the embodiments of the invention, for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work.
[0016] FIG. 1 is a process diagram of the unsupervised coal flow anomaly detection method based on the generative adversarial learning.
DETAILED DESCRIPTION OF THE EMBODIMENTS
[0017] The method of the present invention will be described in further detail below in conjunction with the drawings and the embodiments of the present invention.
[0018] This embodiment takes the situation of the mine as an example, providing the unsupervised coal flow anomaly detection method based on the generative adversarial learning, comprising the following steps:
[0019] Si: obtaining a coal flow picture in a normal scene, and preprocessing the coal flow picture, and obtaining a first output image, the first output image containing a training set image and a test set image.
[0020] In one embodiment of the present invention, obtaining the coal flow picture in the normal scene, wherein, the coal flow pictures include normal pictures and abnormal pictures. Preferably, a camera is used to obtain the coal flow picture. Further preferably, the camera is a high-definition visible light camera with a parameter of 25 frames/s.
[0021] In one embodiment of the present invention, using manual inspection method to select the coal flow picture to obtain the first output image. Preferably, 10,000 to 20,000 normal scene pictures are selected as the training set images, and 600 to 800 pictures are used as the test set images. Further preferably, the test set includes 100 abnormal images.
[0022] S2: taking the training set image as an input, training and establishing a detection mode.
[0023] In one embodiment of the present invention, preferably, a jump connection is added between the encoding and decoding of the generative model to improve the generator's ability to learn samples and reduce training time. Further preferably, introducing an attention mechanism during the jump connection process to make the weight of the target area larger.
[0024] Preferably, using the feature map of the deep network to judge the importance of each element in the feature map of the shallow network and assign different weights, so as to guide the detection model to pay more attention to our target area and suppress the information of irrelevant areas such as background.
[0025] S3: taking the test set image as a model input and substituting the test set image into the detection model, and then obtaining a reconstructed second output image.
[0026] Preferably, the number of the test set images is the same as the number of the second output images.
[0027] S4: calculating a distance Li between the test set image and the reconstructed second output image, calculating a vector distance L2 between the test set image and the reconstructed second output image in a latent space, performing a weighted addition operation on the distance Li and the vector distance L2 to obtaining a first output result as an abnormal score for evaluating a reconstruction quality.
[0028] Calculating the distance Li and the vector distance L2: the calculation process is as following:
L2 = E[ log D(x)]+ E [log(1- D(x')) (2)
[0029] The x in the above formula is an input and x is an output.
[0030] Preferably, according to the first output result, an appropriate threshold value can be selected with the help of a ROC curve.
[0031] In order to better verify the present invention, a comparative experiment was made on whether there is a jump connection and whether the attention mechanism is introduced after the jump connection is added. The performance of the model is evaluated by the value of the area AUC under the ROC curve. The experimental results are shown in the following table: model ordinary adversarial models incorporating attention mechanism generative network jump connections guided model AUC 0.81 0.86 0.91
[0032] It can be seen from the table that the final model achieves a good detection effect for anomaly detection.
[0033] S5: determining whether a coal flow is abnormal.
[0034] If the inputted real-time image is larger than the threshold value, the real-time image can be judged as abnormal

Claims (5)

  1. What is claimed is: 1. An unsupervised coal flow anomaly detection method based on a generative adversarial learning, comprising the following steps: SI: obtaining a coal flow picture in a normal scene, and preprocessing the coal flow picture, and obtaining a first output image, the first output image containing a training set image and a test set image; S2: taking the training set image as an input, training and establishing a detection model, and according to the detection model; S3: taking the test set image as a model input and substituting the test set image into the detection model, and then obtaining a reconstructed second output image; S4: calculating a distance Li between the test set image and the reconstructed second output image, calculating a vector distance L2 between the test set image and the reconstructed second output image in a latent space, performing a weighted addition operation on the distance Li and the vector distance L2 to obtaining a first output result as an abnormal score for evaluating a reconstruction quality; and S5: determining whether a coal flow is abnormal.
  2. 2. The unsupervised coal flow anomaly detection method based on the generative adversarial learning according to claim 1, wherein, S2 can accelerating and optimizing a learning effect of the detection model with the help of a jump connection between a neural network channel and a guidance of an attention mechanism.
  3. 3. The unsupervised coal flow anomaly detection method based on the generative adversarial learning according to claim 1, wherein, in S2, in a case that a detection model parameter is not updated, preparing normal data and abnormal data set for a verification to verify a detection performance of the detection model against abnormalities.
  4. 4. The unsupervised coal flow anomaly detection method based on the generative adversarial learning according to claim 1, wherein, in S4, according to an abnormal score, selecting an appropriate threshold value with the help of a ROC curve.
  5. 5. The unsupervised coal flow anomaly detection method based on the generative adversarial learning according to claim 1, wherein, in S5, if a real-time image is larger than the threshold value, judging the real-time image as abnormal.
    -1/1-
    FIG. 1
AU2021106346A 2021-08-21 2021-08-21 Unsupervised coal flow anomaly detection method based on a generative adversarial learning Ceased AU2021106346A4 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
AU2021106346A AU2021106346A4 (en) 2021-08-21 2021-08-21 Unsupervised coal flow anomaly detection method based on a generative adversarial learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
AU2021106346A AU2021106346A4 (en) 2021-08-21 2021-08-21 Unsupervised coal flow anomaly detection method based on a generative adversarial learning

Publications (1)

Publication Number Publication Date
AU2021106346A4 true AU2021106346A4 (en) 2021-11-04

Family

ID=78488375

Family Applications (1)

Application Number Title Priority Date Filing Date
AU2021106346A Ceased AU2021106346A4 (en) 2021-08-21 2021-08-21 Unsupervised coal flow anomaly detection method based on a generative adversarial learning

Country Status (1)

Country Link
AU (1) AU2021106346A4 (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114757177A (en) * 2022-03-11 2022-07-15 重庆邮电大学 Text summarization method for generating network based on BART fusion pointer
CN116993694A (en) * 2023-08-02 2023-11-03 江苏济远医疗科技有限公司 Non-supervision hysteroscope image anomaly detection method based on depth feature filling

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114757177A (en) * 2022-03-11 2022-07-15 重庆邮电大学 Text summarization method for generating network based on BART fusion pointer
CN116993694A (en) * 2023-08-02 2023-11-03 江苏济远医疗科技有限公司 Non-supervision hysteroscope image anomaly detection method based on depth feature filling
CN116993694B (en) * 2023-08-02 2024-05-14 江苏济远医疗科技有限公司 Non-supervision hysteroscope image anomaly detection method based on depth feature filling

Similar Documents

Publication Publication Date Title
CN113743514B (en) Knowledge distillation-based target detection method and target detection terminal
CN108428227B (en) No-reference image quality evaluation method based on full convolution neural network
AU2021106346A4 (en) Unsupervised coal flow anomaly detection method based on a generative adversarial learning
CN109671071B (en) Underground pipeline defect positioning and grade judging method based on deep learning
CN111028222B (en) Video detection method and device, computer storage medium and related equipment
US20230004811A1 (en) Learning processing device and learning processing method
US11974050B2 (en) Data simulation method and device for event camera
CN117237279A (en) Blind quality evaluation method and system for non-uniform distortion panoramic image
CN111797733A (en) Behavior recognition method, behavior recognition device, behavior recognition equipment and storage medium based on image
CN111768398A (en) Defect detection method and device for photovoltaic module, storage medium and computing equipment
CN113705349A (en) Attention power analysis method and system based on sight estimation neural network
CN110674925A (en) No-reference VR video quality evaluation method based on 3D convolutional neural network
CN112862838A (en) Natural image matting method based on real-time click interaction of user
CN114758199A (en) Training method, device, equipment and storage medium for detection model
CN115690100B (en) Semi-supervised signal point detection model training method, signal point detection method and device
CN114596244A (en) Infrared image identification method and system based on visual processing and multi-feature fusion
CN113034359A (en) Intelligent diagnosis method and device for generator set stator and rotor defects, computer equipment and storage medium
CN115620083B (en) Model training method, face image quality evaluation method, equipment and medium
CN114677670B (en) Method for automatically identifying and positioning identity card tampering
CN115931895A (en) Method and device for checking defects of power transmission line
CN113947567B (en) Defect detection method based on multitask learning
CN112233089B (en) No-reference stereo mixed distortion image quality evaluation method
CN113283429B (en) Liquid level meter reading method based on deep convolutional neural network
CN113315995B (en) Method and device for improving video quality, readable storage medium and electronic equipment
CN112052727B (en) Portrait recognition analysis system and method based on big data

Legal Events

Date Code Title Description
FGI Letters patent sealed or granted (innovation patent)
MK22 Patent ceased section 143a(d), or expired - non payment of renewal fee or expiry