CN112489031A - Mask-rcnn-based oil leakage detection method for snake-shaped-resistant shock absorber - Google Patents

Mask-rcnn-based oil leakage detection method for snake-shaped-resistant shock absorber Download PDF

Info

Publication number
CN112489031A
CN112489031A CN202011462317.9A CN202011462317A CN112489031A CN 112489031 A CN112489031 A CN 112489031A CN 202011462317 A CN202011462317 A CN 202011462317A CN 112489031 A CN112489031 A CN 112489031A
Authority
CN
China
Prior art keywords
mask
rcnn
network
snake
shock absorber
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
Application number
CN202011462317.9A
Other languages
Chinese (zh)
Inventor
张轶鑫
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.)
Harbin Kejia General Mechanical and Electrical Co Ltd
Original Assignee
Harbin Kejia General Mechanical and Electrical Co Ltd
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 Harbin Kejia General Mechanical and Electrical Co Ltd filed Critical Harbin Kejia General Mechanical and Electrical Co Ltd
Priority to CN202011462317.9A priority Critical patent/CN112489031A/en
Publication of CN112489031A publication Critical patent/CN112489031A/en
Pending legal-status Critical Current

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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/50—Image enhancement or restoration by the use of more than one image, e.g. averaging, subtraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
    • 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]
    • 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
    • G06T2207/20212—Image combination
    • G06T2207/20221—Image fusion; Image merging
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30204—Marker

Abstract

A method for detecting oil leakage of an anti-snake-shaped shock absorber based on mask-rcnn relates to the technical field of railway wagon detection. The invention aims to solve the problem that the existing method for detecting the oil leakage of the snake-shaped resistant shock absorber is low in accuracy. According to the method for detecting oil leakage of the snake-shaped resistant shock absorber based on the mask-rcnn, the high-definition imaging cameras are respectively erected at the two sides and the bottom of the railway wagon, and the wagon passes through the equipment to obtain images. And performing background segmentation, oil stain classification and regression positioning of the candidate region by adopting mask-rcnn network positioning twice, and improving the accuracy of detection. And uploading an image for resisting oil leakage of the snake-shaped shock absorber to give an alarm, and correspondingly processing the alarm part by a vehicle inspection personnel according to an artificial prior principle. Meanwhile, the structure of the traditional FPN feature extraction network is improved, the detection performance of the mask-rcnn network is optimized, and the accuracy of fault detection is further improved.

Description

Mask-rcnn-based oil leakage detection method for snake-shaped-resistant shock absorber
Technical Field
The invention belongs to the technical field of railway wagon detection.
Background
Freight transportation is an important component of rail transportation, and vehicles for carrying freight on rails are collectively called rail wagons. The railway department needs to perform safety inspection on the railway freight cars frequently to ensure the safe and stable operation of the railway freight cars. When the railway wagon runs at a high speed, the bogie inevitably presents a snake-shaped trend, the running track of the wheels and the track are poor in stability, and the problem of poor lateral stability of the railway wagon is caused.
In order to solve the problems, a railway motor car fault detection method of manual map checking is generally adopted, but the method has the phenomena of missing detection and false detection caused by fatigue of vehicle inspection personnel. Compared with the traditional manual method, the automatic fault detection method based on deep learning can remarkably improve the detection efficiency, reduce the cost and simultaneously avoid the phenomena of missed detection and false detection caused by fatigue of car inspection personnel. However, the anti-snake-shaped shock absorber images are located at the bottom and the side of the train, the image background is relatively disordered, the interference of a background shadow area and the like on image detection is relatively large, and the accuracy degree of fault detection only by adopting a deep learning network is relatively low.
Disclosure of Invention
The invention aims to solve the problem that the existing method for detecting the oil leakage of the anti-snake-shaped shock absorber is low in accuracy, and provides a mask-rcnn-based method for detecting the oil leakage of the anti-snake-shaped shock absorber.
The method for detecting oil leakage of the snake-shaped resistant shock absorber based on mask-rcnn comprises the following specific processes: acquiring an image of a region where a snake-shaped damper of a detected vehicle is located as a detected image, performing background segmentation on the detected image by adopting a main network in a trained mask-rcnn network to obtain a background separation image, and performing oil stain region detection on the background separation image by adopting an RPN network in the trained mask-rcnn network to obtain a detection result;
the loss function of the mask-rcnn network is expressed as follows:
Loss＝Lcls+Lbox+Lmask，
wherein L isclsTo classify errors, LboxAs a bounding box regression error, LmaskIs the mask error.
Classification error LclsThe expression of (a) is as follows:
Figure BDA0002829741280000011
wherein i is the category serial number of the interested region in the background separation image, NclsIs the total number of classes, piThe probability of predicting a positive sample for the ith region of interest, when the ith region of interest is a positive sample,
Figure BDA0002829741280000012
when the ith region of interest is a negative example,
Figure BDA0002829741280000021
the region of interest is an oil stain region output by the RPN network.
Bounding box regression error LboxThe expression of (a) is as follows:
Figure BDA0002829741280000022
wherein i is the category serial number of the interested region in the background separation image, NboxIs the total number of interested areas, i.e. the oil stain areas output by the RPN network, piThe probability of predicting a positive sample for the ith region of interest, when the ith region of interest is a positive sample,
Figure BDA0002829741280000023
when the ith region of interest is a negative example,
Figure BDA0002829741280000024
tia panning scaling parameter for the positive sample region of interest to a prediction region, which is an ideal region that the mask-rcnn network is expected to output,
Figure BDA0002829741280000025
for the panning scaling parameter from the positive sample region of interest to the real label region, R () is smoothL1The function, its expression is as follows:
Figure BDA0002829741280000026
wherein x is a function variable.
Mask error LmaskThe expression of (a) is as follows:
Lmask＝[Km2]i
wherein, i is the category serial number of the interested region in the background separation image, the interested region is the oil stain region output by the RPN network, K is the number of the categories of the classified objects, m2A mask of m x m size is generated for each region of interest for the mask branch.
The mask-rcnn network comprises a backbone network and an RPN network, wherein the backbone network is a resnet101+ FPN network structure, and the detection steps of the network structure are as follows:
the method comprises the steps of up-sampling a high-level background separation image in a characteristic pyramid, and connecting the high-level background separation image to a next-level background separation image to strengthen the high-level background separation image; meanwhile, the resnet101+ FPN network structure connects the bottom background separation image in the pyramid to the previous background separation image, so that the information of all size background separation images can be utilized;
the RPN firstly generates a suggestion frame of a target for a background separation image output by a backbone network, and then screens all the generated suggestion frames to output an interested region.
The training process of the mask-rcnn network comprises the following steps:
the method comprises the following steps: a set of samples is established that is,
acquiring images of an area where the anti-snake-shaped shock absorber is located in different states of the railway wagon, and taking all the acquired images as a training sample set, wherein the image of the anti-snake-shaped shock absorber with oil stains is a positive sample, the image of the anti-snake-shaped shock absorber without oil stains is a negative sample, and the number of the positive sample and the number of the negative sample are consistent;
step two: the training of the segmentation network is carried out,
extracting a background separation image of each sample in a training sample set by using a backbone network in a mask-rcnn network, training a segmentation model of the mask-rcnn network on the background, generating a suggestion frame of a target by using an RPN network in the mask-rcnn network, screening all the generated suggestion frames to obtain an interested region, and training a positioning model of the mask-rcnn network on the oil stain position.
The mask-rcnn network against background segmentation model was used for:
marking the area of the object to be measured in the original image as class 0 to obtain the mask information of the object to be measured; and assigning the pixels of the background area outside the mask to be 255 and eliminating the area to segment the area to be measured and the background area, wherein the obtained area to be measured is the background separation image.
The positioning model of the mask-rcnn network on the oil stain positions is used for: and marking the oil stain area in the measured object area as class 0 to finish the positioning of the oil stain area.
Further, after the step one, the method further amplifies the samples in the training sample set, and the method for amplifying the samples in the training sample set includes: one or more of rotating, cropping, or adding noise.
And after the detection result is obtained, when oil stains are detected, the detection result is used as fault information to be uploaded to a vehicle inspection personnel side.
According to the method for detecting oil leakage of the snake-shaped resistant shock absorber based on the mask-rcnn, the high-definition imaging cameras are respectively erected at the two sides and the bottom of the railway wagon, and the wagon passes through the equipment to obtain images. And performing background segmentation, oil stain classification and regression positioning of the candidate region by adopting mask-rcnn network positioning twice, and improving the accuracy of detection. And uploading an image for resisting oil leakage of the snake-shaped shock absorber to give an alarm, and correspondingly processing the alarm part by a vehicle inspection personnel according to an artificial prior principle. Meanwhile, the structure of the traditional FPN feature extraction network is improved, the detection performance of the mask-rcnn network is optimized, and the accuracy of fault detection is further improved.
Drawings
FIG. 1 is a flow chart of the operation of a conventional resnet101+ FPN network;
FIG. 2 is a flow chart of the improved resnet101+ FPN network of the present invention;
FIG. 3 is a network training flow diagram;
FIG. 4 is a fault detection flow diagram;
FIG. 5 is an original image;
FIG. 6 is a background-separated image after background segmentation;
FIG. 7 is a graph showing the results of oil stain detection.
Detailed Description
It should be noted that, in the case of conflict, the features included in the embodiments or the embodiments disclosed in the present application may be combined with each other.
The first embodiment is as follows: specifically describing the embodiment with reference to fig. 4, the method for detecting oil leakage of an anti-snake-shaped shock absorber based on mask-rcnn in the embodiment specifically includes:
acquiring an image of the area of the snake-shaped resistant shock absorber of the detected vehicle as a detected image,
adopting a main network in a trained mask-rcnn network to carry out background segmentation on the detected image to obtain a background separation image,
and then, performing oil stain area detection on the background separation image by adopting the RPN in the trained mask-rcnn network to obtain a detection result.
The loss function expresses the difference degree between the predicted value and the real label, and the model is trained by reducing the loss function between the predicted value and the real label. The loss function of the mask-rcnn network is expressed as follows:
Loss＝Lcls+Lbox+Lmask，
wherein L isclsTo classify errors, LboxAs a bounding box regression error, LmaskIs the mask error.
The second embodiment is as follows: the embodiment further illustrates the mask-rcnn-based method for detecting oil leakage of the anti-snake-shaped shock absorber in the first specific embodimentMiddle, classification error LclsThe expression of (a) is as follows:
Figure BDA0002829741280000041
wherein i is the category serial number of an interested area in the background separation image, the interested area is an oil stain area output by the RPN, and N isclsIs the total number of classes, piThe probability of predicting a positive sample for the ith region of interest, when the ith region of interest is a positive sample,
Figure BDA0002829741280000042
when the ith region of interest is a negative example,
Figure BDA0002829741280000043
the third concrete implementation mode: the present embodiment is further described with reference to the mask-rcnn-based method for detecting oil leakage of an anti-snake-shaped shock absorber described in the first or second embodiments, where in the present embodiment, a regression error L of a bounding box is usedboxThe expression of (a) is as follows:
Figure BDA0002829741280000044
wherein i is the category serial number of an interested area in the background separation image, the interested area is an oil stain area output by the RPN, and N is the number of the interested area in the background separation imageboxIs the total number of regions of interest, piThe probability of predicting a positive sample for the ith region of interest, when the ith region of interest is a positive sample,
Figure BDA0002829741280000051
when the ith region of interest is a negative example,
Figure BDA0002829741280000052
tifor the translational scaling parameter from the interested region of the positive sample to the predicted region, the predicted region is the predicted mask-rcnn network energyThe ideal region of the output can be obtained,
Figure BDA0002829741280000053
for the panning scaling parameter from the positive sample region of interest to the real label region, R () is smoothL1The function, its expression is as follows:
Figure BDA0002829741280000054
wherein x is a function variable.
The fourth concrete implementation mode: in this embodiment, the method for detecting oil leakage of an anti-snake-shaped shock absorber based on mask-rcnn according to the first, second or third embodiment is further describedmaskThe expression of (a) is as follows:
Lmask＝[Km2]i
wherein i is the category serial number of an interested area in the background separation image, the interested area is an oil stain area output by the RPN, K is the number of categories of classified objects, m2A mask of m x m size is generated for each region of interest for the mask branch.
The fifth concrete implementation mode: the embodiment further illustrates the method for detecting oil leakage of the anti-snake-shaped shock absorber based on the mask-rcnn according to the first, second, third or fourth specific embodiments.
In the feature pyramid of the FPN, a pyramid level is defined by each stage, and the output of the last layer of each stage is used as a reference set of background separation images. The deepest layers at each stage should have the strongest features, and the features of the last residual structure at each stage can be used to activate the output. The outputs of these residual blocks are denoted C2, C3, C4, C5, corresponding one to each other at the outputs conv2, conv3, conv4 and conv5, while taking care that they have a step size of {4, 8, 16, 32} pixels with respect to the input image, and the upper layer background separation image is up-sampled and then connected laterally to the previous layer feature, by which process the upper layer feature is enhanced. The two layers of features that are connected laterally in the above process must be identical in spatial dimension. This is primarily to make use of the underlying positioning detail information. The convolution kernel used in this patent processes the already fused background separation image to generate the final background separation image that we need. The { C2, C3, C4 and C5} layers are in one-to-one correspondence with the { P2, P3, P4 and P5} layers of the fused feature layer, and the space sizes of the corresponding layers are communicated, as shown in FIG. 1.
The mask-rcnn FPN network described above causes two problems: one is that the feature mapping image of the highest layer is the same as the final output of the original feature extraction network structure, but the information of the large-size target needs to be mainly obtained from the feature mapping image of the highest layer, so that the accuracy rate of large target detection is similar to that of original network data sometimes even lower; secondly, analyzing the top-down path structure can know that each layer of a group of feature maps output by the FPN contains information of the layer and higher layers but not information of lower layers, and the RPN selects the optimal size feature map from the feature maps for inputting, so that the information of all size feature maps cannot be fully utilized, and the final detection accuracy is not a better value.
In order to solve the above problem, the present embodiment generates feature maps of different sizes for FPN, adds a bottom-up path, directly transmits the bottom layer information to the upper layer, and fully utilizes the information of feature maps of all sizes, and the improved network structure is shown in fig. 2.
Specifically, the mask-rcnn network includes a backbone network and an RPN network. The main network is a resnet101+ FPN network structure, the network structure can perform up-sampling on a high-level background separation image in a characteristic pyramid, and the high-level background separation image is connected to a next-level background separation image, so that the high-level background separation image is enhanced; the resnet101+ FPN network structure can also connect the bottom background separation image in the pyramid to the previous background separation image, so that the information of all size background separation images can be utilized. The RPN firstly generates a suggestion frame of a target for a background separation image output by a backbone network, and then screens all the generated suggestion frames to output an interested region.
The sixth specific implementation mode: specifically describing the present embodiment with reference to fig. 3, the present embodiment is further described with reference to the method for detecting oil leakage of a snake-shaped damper based on mask-rcnn in the first, second, third, fourth or fifth embodiment, in which the training process of the mask-rcnn network includes the following steps:
the method comprises the following steps: a set of samples is established that is,
high-definition imaging equipment is erected around a rail of a railway wagon, and after the wagon passes by, vehicle passing images in different states are obtained. And intercepting images of the area of the anti-snake-shaped shock absorber in the process image, and taking the images as a training sample set. The image of resisting snakelike bumper shock absorber and having the oil stain in the training sample set is positive sample, and the image of resisting snakelike bumper shock absorber no oil stain is the negative sample to guarantee that the number of positive sample keeps unanimous with the number of negative sample.
Step two: the training of the segmentation network is carried out,
extracting a background separation image of each sample in a training sample set by using a backbone network in a mask-rcnn network to train a segmentation model of the mask-rcnn network on the background,
and generating a suggestion frame of the target by using an RPN (resilient packet network) in the mask-rcnn network, screening all the generated suggestion frames to obtain an interested region, and training a positioning model of the mask-rcnn network on the oil stain position.
The seventh embodiment: in this embodiment, the method for detecting oil leakage of an anti-snake-shaped shock absorber based on mask-rcnn according to the first, second, third, fourth, fifth or sixth specific embodiments is further described, in this embodiment, a segmentation model of the mask-rcnn network for the background is used for:
marking the area of the object to be measured in the original image as class 0 to obtain the mask information of the object to be measured, wherein the original image is shown in FIG. 5;
the background area in the original image can be eliminated by assigning the pixels of the background area outside the mask to be 255, so that the area to be measured is divided from the background area, and the obtained area to be measured is the background separation image, as shown in fig. 6.
The specific implementation mode is eight: in this embodiment, a method for detecting oil leakage of an anti-snake-shaped shock absorber based on mask-rcnn according to a first, second, third, fourth, fifth, sixth or seventh specific embodiment is further described, in this embodiment, a model for positioning an oil spot position by a mask-rcnn network is used for:
and (3) training the image with the background removed as a training set again, namely marking the oil stain area in the measured object area as 0 class, and completing the positioning of the oil stain area, as shown in fig. 7.
The specific implementation method nine: in this embodiment, the method for detecting oil leakage of an anti-snake-shaped shock absorber based on mask-rcnn according to the first, second, third, fourth, fifth, sixth, seventh or eighth embodiment is further described, in which after the first step, the method for amplifying the samples in the training sample set further comprises: one or more of rotating, cropping, or adding noise.
The data amplification operation can enhance the generalization capability of the subsequent detection network and reduce the probability of network overfitting.
The detailed implementation mode is ten: the embodiment further describes the method for detecting oil leakage of the snake-shaped damper based on mask-rcnn according to the specific embodiment, namely, according to the embodiment, after a detection result is obtained, when oil stain is detected, the detection result is uploaded to a vehicle inspector as fault information, and the vehicle inspector carries out the next processing according to a fault message.

Claims (10)

1. The method for detecting oil leakage of the snake-shaped resistance shock absorber based on mask-rcnn is characterized by comprising the following steps of,
acquiring an image of the area of the snake-shaped resistant shock absorber of the detected vehicle as a detected image,
adopting a main network in a trained mask-rcnn network to carry out background segmentation on the detected image to obtain a background separation image,
then, an RPN network in the trained mask-rcnn network is adopted to detect the oil stain area of the background separation image, and a detection result is obtained;
the loss function of the mask-rcnn network is expressed as follows:
Loss＝Lcls+Lbox+Lmask，
wherein L isclsTo classify errors, LboxAs a bounding box regression error, LmaskIs the mask error.
2. The mask-rcnn-based anti-snake shock absorber oil leakage detection method according to claim 1, wherein the classification error L isclsThe expression of (a) is as follows:
Figure FDA0002829741270000011
wherein i is the category serial number of an interested area in the background separation image, the interested area is an oil stain area output by the RPN, and N is the number of the interested area in the background separation imageclsIs the total number of classes, piThe probability of predicting a positive sample for the ith region of interest, when the ith region of interest is a positive sample,
Figure FDA0002829741270000012
when the ith region of interest is a negative example,
Figure FDA0002829741270000013
3. the mask-rcnn-based anti-snake shock absorber oil leakage detection method according to claim 1, wherein the bounding box regression error LboxThe expression of (a) is as follows:
Figure FDA0002829741270000014
wherein i is the category serial number of the interested region in the background separation image, NboxIs the total number of interested areas, i.e. the oil stain areas output by the RPN network, piIs the ith interested areaThe probability that the field is predicted as a positive sample, when the ith region of interest is a positive sample,
Figure FDA0002829741270000015
when the ith region of interest is a negative example,
Figure FDA0002829741270000016
tia panning scaling parameter for the positive sample region of interest to a prediction region, which is an ideal region that the mask-rcnn network is expected to output,
Figure FDA0002829741270000017
for the panning scaling parameter from the positive sample region of interest to the real label region, R () is smoothL1The function, its expression is as follows:
Figure FDA0002829741270000021
wherein x is a function variable.
4. The mask-rcnn-based snake-like shock absorber oil leakage detection method as claimed in claim 1, wherein the mask error LmaskThe expression of (a) is as follows:
Lmask＝[Km2]i
wherein, i is the category serial number of the interested region in the background separation image, the interested region is the oil stain region output by the RPN network, K is the number of the categories of the classified objects, m2A mask of m x m size is generated for each region of interest for the mask branch.
5. The mask-rcnn-based serpentine damper oil leakage detection method according to claim 1, wherein the mask-rcnn network includes a backbone network and an RPN network,
the backbone network is a resnet101+ FPN network structure, and the detection steps of the network structure are as follows:
the method comprises the steps of up-sampling a high-level background separation image in a characteristic pyramid, and connecting the high-level background separation image to a next-level background separation image to strengthen the high-level background separation image;
meanwhile, the resnet101+ FPN network structure connects the bottom background separation image in the pyramid to the previous background separation image, so that the information of all size background separation images can be utilized;
the RPN firstly generates a suggestion frame of a target for a background separation image output by a backbone network, and then screens all the generated suggestion frames to output an interested region.
6. The mask-rcnn-based anti-serpentine shock absorber oil leakage detection method according to claim 1 or 5, wherein the training process of the mask-rcnn network comprises the following steps:
the method comprises the following steps: a set of samples is established that is,
acquiring images of an area where the anti-snake-shaped shock absorber is located in different states of the railway wagon, and taking all the acquired images as a training sample set, wherein the image of the anti-snake-shaped shock absorber with oil stains is a positive sample, the image of the anti-snake-shaped shock absorber without oil stains is a negative sample, and the number of the positive sample and the number of the negative sample are consistent;
step two: the training of the segmentation network is carried out,
extracting a background separation image of each sample in a training sample set by using a backbone network in a mask-rcnn network to train a segmentation model of the mask-rcnn network on the background,
and generating a suggestion frame of the target by using an RPN (resilient packet network) in the mask-rcnn network, screening all the generated suggestion frames to obtain an interested region, and training a positioning model of the mask-rcnn network on the oil stain position.
7. The mask-rcnn-based snake-like damper oil leakage detection method according to claim 6, wherein the segmentation model of the mask-rcnn network against the background is used for:
marking the area of the object to be measured in the original image as class 0 to obtain the mask information of the object to be measured;
and assigning the pixels of the background area outside the mask to be 255 and eliminating the area to segment the area to be measured and the background area, wherein the obtained area to be measured is the background separation image.
8. The mask-rcnn-based snake-shaped shock absorber oil leakage detection method as claimed in claim 7, wherein the mask-rcnn network positioning model of the oil stain position is used for:
and marking the oil stain area in the measured object area as class 0 to finish the positioning of the oil stain area.
9. The mask-rcnn-based snake damper oil leakage detection method according to claim 6, wherein after the first step, the samples in the training sample set are further amplified,
the method for amplifying the samples in the training sample set comprises the following steps: one or more of rotating, cropping, or adding noise.
10. The mask-rcnn-based oil leakage detection method for the serpentine-shaped shock absorber according to the claim 1, 2, 3, 4, 5, 7, 8 or 9, characterized in that after the detection result is obtained, when oil stain is detected, the detection result is uploaded to a vehicle inspector as fault information.
CN202011462317.9A 2020-12-10 2020-12-10 Mask-rcnn-based oil leakage detection method for snake-shaped-resistant shock absorber Pending CN112489031A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011462317.9A CN112489031A (en) 2020-12-10 2020-12-10 Mask-rcnn-based oil leakage detection method for snake-shaped-resistant shock absorber

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011462317.9A CN112489031A (en) 2020-12-10 2020-12-10 Mask-rcnn-based oil leakage detection method for snake-shaped-resistant shock absorber

Publications (1)

Publication Number Publication Date
CN112489031A true CN112489031A (en) 2021-03-12

Family

ID=74916880

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011462317.9A Pending CN112489031A (en) 2020-12-10 2020-12-10 Mask-rcnn-based oil leakage detection method for snake-shaped-resistant shock absorber

Country Status (1)

Country Link
CN (1) CN112489031A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115457297A (en) * 2022-08-23 2022-12-09 中国航空油料集团有限公司 Method and device for detecting oil leakage of aviation oil depot and aviation oil safety operation and maintenance system
CN116091870A (en) * 2023-03-01 2023-05-09 哈尔滨市科佳通用机电股份有限公司 Network training and detecting method, system and medium for identifying and detecting damage faults of slave plate seat

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110120047A (en) * 2019-04-04 2019-08-13 平安科技（深圳）有限公司 Image Segmentation Model training method, image partition method, device, equipment and medium
CN111639215A (en) * 2020-06-01 2020-09-08 南京索安电子有限公司 Vehicle bottom scanning system for image learning and image recognition of artificial intelligence and implementation method

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110120047A (en) * 2019-04-04 2019-08-13 平安科技（深圳）有限公司 Image Segmentation Model training method, image partition method, device, equipment and medium
CN111639215A (en) * 2020-06-01 2020-09-08 南京索安电子有限公司 Vehicle bottom scanning system for image learning and image recognition of artificial intelligence and implementation method

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
KAIMING HE 等: "Mask R-CNN", 《ARXIV》 *
SHU LIU 等: "Path Aggregation Network for Instance Segmentation", 《ARXIV》 *
付发 等: "基于卷积网络的遥感图像建筑物提取技术研究", 《软件工程》 *
周雯 等: "基于卷积神经网络的动车组行车安全图像缺陷检测与分割", 《铁道学报》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115457297A (en) * 2022-08-23 2022-12-09 中国航空油料集团有限公司 Method and device for detecting oil leakage of aviation oil depot and aviation oil safety operation and maintenance system
CN115457297B (en) * 2022-08-23 2023-09-26 中国航空油料集团有限公司 Oil leakage detection method and device for aviation oil depot and aviation oil safety operation and maintenance system
CN116091870A (en) * 2023-03-01 2023-05-09 哈尔滨市科佳通用机电股份有限公司 Network training and detecting method, system and medium for identifying and detecting damage faults of slave plate seat
CN116091870B (en) * 2023-03-01 2023-09-12 哈尔滨市科佳通用机电股份有限公司 Network training and detecting method, system and medium for identifying and detecting damage faults of slave plate seat

Similar Documents

Publication Publication Date Title
CN110261436B (en) Rail fault detection method and system based on infrared thermal imaging and computer vision
CN111652227B (en) Method for detecting damage fault of bottom floor of railway wagon
CN111079746B (en) Railway wagon axle box spring fault image identification method
CN112489031A (en) Mask-rcnn-based oil leakage detection method for snake-shaped-resistant shock absorber
CN115272334A (en) Method for detecting micro defects on surface of steel rail under complex background
CN111080621B (en) Method for identifying railway wagon floor damage fault image
CN111080612B (en) Truck bearing damage detection method
CN111080600A (en) Fault identification method for split pin on spring supporting plate of railway wagon
US11495022B2 (en) Method for recognizing an object of a mobile unit
CN111091551A (en) Method for detecting loss fault of brake beam strut opening pin of railway wagon
CN113111875A (en) Seamless steel rail weld defect identification device and method based on deep learning
CN108648210B (en) Rapid multi-target detection method and device under static complex scene
CN111079629A (en) Railway wagon center plate out-of-position fault image identification method
Yamazaki et al. Vehicle extraction and speed detection from digital aerial images
CN113221839B (en) Automatic truck image identification method and system
CN107247967A (en) A kind of vehicle window annual test mark detection method based on R CNN
CN111652228B (en) Railway wagon sleeper beam hole foreign matter detection method
CN112329858A (en) Image recognition method for breakage fault of anti-loosening iron wire of railway motor car
CN115424128A (en) Fault image detection method and system for lower link of freight car bogie
CN116486129A (en) Deep learning-based railway wagon cover plate fault identification method and device
CN114743166A (en) Method for detecting brake of railway wagon
CN113643258A (en) Method for detecting loss fault of skirt board at side part of train based on deep learning
CN114005097A (en) Train operation environment real-time detection method and system based on image semantic segmentation
CN112102293B (en) Rapid detection method for foreign matters in triangular holes of railway wagon
CN115170890B (en) Method for identifying breakage fault of connecting pull rod chain of railway wagon

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

Application publication date: 20210312

RJ01 Rejection of invention patent application after publication