CN112907585A - Multi-scale fusion steel rail bolt assembly fault detection method - Google Patents
Multi-scale fusion steel rail bolt assembly fault detection method Download PDFInfo
- Publication number
- CN112907585A CN112907585A CN202110341276.6A CN202110341276A CN112907585A CN 112907585 A CN112907585 A CN 112907585A CN 202110341276 A CN202110341276 A CN 202110341276A CN 112907585 A CN112907585 A CN 112907585A
- Authority
- CN
- China
- Prior art keywords
- bolt
- features
- bolt assembly
- steel rail
- network
- 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
- 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
- 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
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30164—Workpiece; Machine component
Abstract
The invention relates to the technical field of bolt image fault identification and detection, and discloses a multi-scale fusion steel rail bolt assembly fault detection method, which comprises the following steps: s1: performing multi-level feature extraction on the steel rail bolt assembly image, and extracting the features of the whole row of bolt assemblies at the steel rail joint; s2: pooling the characteristics of the whole row of bolt assemblies, and fusing the characteristics after pooling to obtain the characteristics of a single bolt assembly; s3: down-sampling the shallow features extracted in the step S1, up-sampling the deep features in the step S2, and fusing the sampled features to output three groups of bolt local features with different dimensions; s4: predicting the three groups of extracted features in the S3 and framing a bolt assembly and a state label in the original rail image according to the result; the method has high prediction precision and high detection speed; and a residual error edge is constructed in the backbone network, the learning capability of the network is enhanced, and the Msih activation function is adopted, so that the processed data is smoother, and the gradient descent processing is better.
Description
Technical Field
The invention relates to the technical field of bolt image fault identification and detection, in particular to a multi-scale fusion steel rail bolt assembly fault detection method.
Background
The rail bolt assembly refers to a combination of a bolt, a nut and a washer for fixing a rail clamping plate at a rail joint, and is a key for ensuring the safe connection of two sections of rails, various faults of the assembly occur due to the high-speed rolling of a heavy-duty train and the action of various uncertain factors, the typical faults of the assembly include the reduction of the torsional force, the fracture of a bolt, the crack or the loss of a nut, the fracture or the loss of a washer and the overall loss of the assembly, many train accidents are caused by the problems of the rail joint, the rail bolt assembly is positioned on the side surface of the rail, and one section of rail joint is arranged in the rail every 12.5 meters or 25 meters, so the detection and the maintenance of the bolt assembly are discontinuous, the bolt assembly is small, the existing rail bolt assembly identification and detection mostly depend on manpower, time and labor are wasted, and the intelligent detection method is adopted in the system due to the fact that the, The bolt fault characteristics are not obvious, the detection is not in place, the leakage rate is high, and the accuracy is low.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a multi-scale fusion steel rail bolt assembly fault detection method.
In order to achieve the above purpose, the invention provides the following technical scheme:
a multi-scale fusion steel rail bolt assembly fault detection method comprises the following steps:
s1: performing multi-level feature extraction on the steel rail bolt assembly image, and extracting the features of the whole row of bolt assemblies at the steel rail joint;
s2: pooling the characteristics of the whole row of bolt assemblies, and fusing the characteristics after pooling to obtain the characteristics of a single bolt assembly;
s3: down-sampling the shallow features extracted in the step S1, up-sampling the deep features in the step S2, and fusing the sampled features to output three groups of bolt local features with different dimensions;
s4: and predicting the three groups of extracted features in the S3 and framing the bolt assembly and the corresponding state label in the original steel rail image according to the result.
In the present invention, preferably, in step S1, a backbone network is used to perform multi-level feature extraction on the rail bolt assembly image, where the backbone network includes a two-dimensional convolution layer, a plurality of residual network blocks, and a convolution output layer, which are connected in sequence.
In the present invention, it is preferable that the two-dimensional volume base layer is connected to the convolution output layer through a residual edge, and the Msih activation function is provided in the two-dimensional volume base layer.
In the present invention, preferably, in step 2, the SPP network is used for pooling, and the SPP network includes pooling layers with pooling kernels of 13x13, 9x9, 5x5 and 1x1, and the pooling layers are used for pooling characteristics output by the convolution output layers.
In the present invention, it is preferable that the step S3 further includes the steps of:
s301: extracting shallow layer characteristics of a middle-layer network output bolt of the residual error network block and performing convolution operation to obtain a first group of characteristics;
s302: extracting and convolving the middle layer characteristics of the middle layer network output bolt of the residual error network block;
s303: up-sampling the deep features of the bolt output by the SPP network and fusing the deep features with the middle features of the bolt in S302;
s304: down-sampling the features output in the step S301, and fusing the features with the features obtained in the step S303 again to obtain a second group of features;
s305: and after downsampling the features obtained in the step S304, fusing the features with the deep features of the bolt output by the SPP network to obtain a third group of features.
In the present invention, it is preferable that the step S5 includes the steps of:
s401: predicting the characteristics to obtain a prediction frame;
s402: predicting the state of the extracted bolt features through a non-maximum suppression algorithm;
s403: generating a category information list and a final bolt prediction data frame;
s404: and calling a drawing function to draw the bolt prediction box.
In the present invention, it is preferable that the status tags in step S4 include (lose), (Bolt break), (Nut crack), (Missing Nut), (washber fraction), (Missing gap), (overhall), (NW dispear), and correspond to the failure of the Bolt assembly: reduced tightening force, bolt breakage, nut cracking, nut missing, washer breakage, washer missing, nut washer missing, and overall part missing.
Compared with the prior art, the invention has the beneficial effects that:
according to the method, a data enhancement method is adopted to expand the training set during training, the number of samples is increased, and through training of a large number of fault steel rail bolt component data sets, the detection of the network on the bolts and the bolt state prediction accuracy are high, and the detection speed is high; constructing a residual edge in a backbone network, enhancing the learning capability of the network, and adopting an Msih activation function to enable the processed data to be smoother, the gradient descent processing to be better, and avoiding missing detailed data of the bolt assembly in the subsequent processing; the bolt characteristic data is repeatedly sampled, processed, fused and spliced in a multi-level mode through the SPP network and the PANet network, and the characteristics are enhanced, so that the overall and local characteristics of the bolt on the steel rail can be well extracted, the accuracy is high, and the subsequent state prediction of the bolt assembly is facilitated; the bolt assembly state is obtained after prediction, and is marked in the steel rail bolt assembly image, so that the hidden danger of the steel rail can be well found by auxiliary personnel in time, the steel rail is maintained or replaced at the first time, and accidents are avoided.
Drawings
Fig. 1 is a schematic flow chart of a multi-scale fusion rail bolt assembly fault detection method according to the invention.
Fig. 2 is a partial structural schematic diagram of a multi-scale fusion rail bolt assembly fault detection method according to the invention.
Fig. 3 is a schematic diagram of a prediction result of the multi-scale fusion steel rail bolt assembly fault detection method.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It will be understood that when an element is referred to as being "secured to" another element, it can be directly on the other element or intervening elements may also be present. When a component is referred to as being "connected" to another component, it can be directly connected to the other component or intervening components may also be present. When a component is referred to as being "disposed on" another component, it can be directly on the other component or intervening components may also be present. The terms "vertical," "horizontal," "left," "right," and the like as used herein are for illustrative purposes only.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.
Referring to fig. 1 and 2, a preferred embodiment of the present invention provides a method for detecting a failure of a rail bolt assembly with multi-scale fusion, which is mainly used for detecting whether a bolt assembly for fixing at a rail joint has a failure, marking the failure, helping a person to find a problem early and troubleshoot hidden dangers, and includes the following steps: s1: performing multi-level feature extraction on the steel rail bolt assembly image, and extracting the features of the whole row of bolt assemblies at the steel rail joint; s2: pooling the characteristics of the whole row of bolt assemblies, and fusing the characteristics after pooling to obtain the characteristics of a single bolt assembly; s3: down-sampling the shallow features extracted in the step S1, up-sampling the deep features in the step S2, fusing the sampled features and outputting three groups of bolt local features with different dimensions; s4: predicting the three groups of extracted features in the S3, and framing bolt components and corresponding state labels in the original rail image according to the result; through from the shallow layer to the deep layer, from whole to local, the holistic bolt subassembly of junction is to single bolt subassembly local feature at last in order in the whole image extraction of rail to through sampling repeatedly and fusing, make the characteristic accuracy rate and the complete rate of extraction higher, thereby improve the recognition rate of bolt subassembly state.
Specifically, the network is mainly based on a YOLOV4 network, firstly, a whole network is trained by utilizing a plurality of manufactured problem steel rail bolt training sets, data expansion is carried out by adopting a samplepair enhancement method, two or more images are randomly extracted from the training sets to carry out basic data enhancement operation, pixels are averaged and then superposed to form a new sample, and therefore the number of the samples is increased, the training set based on bolt faults is enhanced, and the robustness and the detection effect of the YOLOV4 network are improved.
Furthermore, a plurality of Bolt training sets are self-made professional detection data sets of steel rail bolts, and each set comprises a twisting force reduction set, a Bolt fracture set, a Nut crack set, a Nut Missing set, a gasket fracture set, a gasket Missing set, a Nut and gasket Missing set and an integral component Missing set, each set corresponds to a label which is (lose), (Bolt break), (Nut crack), (Missing Nut), (Washer fracture), (Missing gap), (Overall Missing), (NW dispear), and an xml file is generated and stored after the labels are marked.
In this embodiment, in step S1, a backbone network is used to perform multi-level feature extraction on the steel rail bolt assembly image, the backbone network includes a two-dimensional convolution layer, a plurality of residual network blocks, and a convolution output layer, which are connected in sequence, wherein a two-dimensional convolution base layer is connected to the convolution output layer through a residual edge, and a Msih activation function is disposed in the two-dimensional convolution base layer.
Specifically, the trunk network adopts a CSPDarknet53 trunk network, the CSPDarknet53 trunk network is established based on a residual error neural network, an input layer of the CSPDarknet is connected with a two-dimensional convolutional layer, the Msih activation function is adopted, so that the two-dimensional convolutional layer has better accuracy and generalization capability, a good foundation is laid for subsequent feature extraction, the output of the two-dimensional convolutional layer is connected into a residual error network block, a plurality of residual error network blocks are sequentially connected, the image is subjected to convolution operation for many times, the residual error network block is input into the convolutional output layer and is subjected to DarknetConv2D _ BN _ Leaky convolution for three times, the output of the two-dimensional convolutional layer is introduced into the convolutional output layer through a residual error edge, the convolution operation is carried out together with the residual error network block extraction, and finally the bolt overall feature extracted from the overall image of the steel rail is output.
In this embodiment, in step 2, the feature fusion is performed by a convolution fusion block after the pooling output is performed by pooling an SPP network including pooling layers with pooling kernels of 13x13, 9x9, 5x5, and 1x1, respectively, and the pooling layers are used for pooling the features output by the convolution output layers, respectively.
Specifically, the overall bolt features extracted in step S1 are subjected to pooling treatment in different degrees through three different pooling layers, so that more bolt local features of original steel rail images are obtained, bolt features are obviously separated, and the accuracy of bolt feature extraction is ensured.
In the present embodiment, the step S3 is performed using a Path Aggregation Network (Path Aggregation Network), and includes the steps of:
s301: extracting shallow layer characteristics of the middle layer network output bolt of the residual error network block in the step S2 and performing convolution operation to obtain a first group of characteristic data; s302: extracting and convolving the middle layer characteristics of the middle layer network output bolt of the residual error network block; s303: up-sampling the deep features of the bolt output by the SPP network and fusing the deep features with the middle features of the bolt in S302; s304: down-sampling the features output in the step S301, and fusing the features with the features obtained in the step S303 again to obtain a second group of feature data; s305: and after downsampling the features obtained in the step S304, fusing the features with the deep features of the bolt output by the SPP network to obtain a third group of feature data.
Referring to fig. 3, further, since the pixels of the bolt assembly are small, in order to ensure that extraction of the characteristics of the bolt assembly is not omitted and accuracy is high, extracted shallow characteristic data is downsampled and deep characteristic data is upsampled, and characteristic values of different depths are repeatedly extracted and fused, so that the detailed part characteristics of the bolt are further extracted, and the subsequent bolt state judgment is facilitated.
In the present embodiment, step S4 includes the following steps:
s401: predicting the characteristics to obtain a prediction frame;
s402: predicting the state of the extracted bolt features through a non-maximum suppression (NMS) algorithm;
s403: generating a category information list and a final bolt prediction data frame;
s404: and calling a drawing function to draw the bolt prediction box.
Specifically, a yolo _ head algorithm is adopted to predict a characteristic value, a prediction frame is obtained, only bolt characteristics are in the prediction frame, then IOU calculation is carried out on the framed bolt characteristics and corresponding state characteristics through a non-maximum value suppression algorithm, if the IOU value is larger than a threshold value, the bolt characteristics and the corresponding state characteristics are judged to be the same target, the state characteristics of the bolt are obtained, the steps S401-S402 are repeated until all the bolts are predicted and compared, a corresponding information list and a bolt state prediction data frame are generated, then a drawing function is called to draw the bolt frame and a fault label corresponding to the fault bolt on a steel rail image, a person can quickly find the problem of the bolt assembly through the fault label, repair or replace the bolt assembly on the steel rail in batches, and early find out potential safety hazards, so that accidents are avoided.
The working principle is as follows:
firstly training a Yolov4 network, setting an attenuation coefficient to be 0.0005 and a learning rate to be 0.0001, reducing the learning rate to 10 percent of an initial learning rate when training iteration times reach 8000 and 14000 times, wherein an iteration parameter is 16000 and a batch processing parameter is 4, inputting a steel rail image to be detected after training is finished, sequentially performing multi-level feature extraction through a CSPDarknet53 backbone network, performing multi-time pooling through an SPP network, fusing the images through a convolution fusion block after pooling to form deep feature data, performing up-sampling on the deep feature data through a PANET network, performing multi-azimuth fusion on the deep feature data and shallow feature data of the CSPDarknet53 backbone network, finally obtaining bolt feature data, performing prediction comparison on the bolt feature data to obtain a bolt component state, and marking the bolt component state in a steel rail bolt component image, thereby assisting personnel to find hidden troubles of a steel rail in time, the maintenance or the replacement is carried out at the first time, so that accidents are avoided.
The above description is intended to describe in detail the preferred embodiments of the present invention, but the embodiments are not intended to limit the scope of the claims of the present invention, and all equivalent changes and modifications made within the technical spirit of the present invention should fall within the scope of the claims of the present invention.
Claims (7)
1. A multi-scale fusion rail bolt assembly fault detection method is characterized by comprising the following steps:
s1: performing multi-level feature extraction on the steel rail bolt assembly image, and extracting the features of the whole row of bolt assemblies at the steel rail joint;
s2: pooling the characteristics of the whole row of bolt assemblies, and fusing the characteristics after pooling to obtain the characteristics of a single bolt assembly;
s3: down-sampling the shallow features extracted in the step S1, up-sampling the deep features in the step S2, fusing the sampled features, and outputting three groups of bolt local features with different dimensions;
s4: and predicting the three groups of extracted features in the S3 and framing the bolt assembly and the corresponding state label in the original steel rail image according to the result.
2. The method for detecting the fault of the rail bolt assembly fused in the multi-scale mode according to the claim 1, wherein in the step S1, a backbone network is adopted to perform multi-level feature extraction on the rail bolt assembly image, and the backbone network comprises a two-dimensional convolution layer, a plurality of residual error network blocks and a convolution output layer which are connected in sequence.
3. The method for detecting the fault of the multi-scale fused steel rail bolt assembly according to claim 2, wherein the two-dimensional coil base layer is connected with the convolution output layer through a residual edge, and an Msih activation function is arranged in the two-dimensional coil base layer.
4. The method for detecting the fault of the multi-scale fused steel rail bolt assembly according to the claim 3, wherein in the step 2, the SPP network is used for pooling, the SPP network comprises pooling layers with pooling cores of 13x13, 9x9, 5x5 and 1x1 respectively, and the pooling layers are used for pooling characteristics output by the convolution output layers respectively.
5. The method for detecting the failure of the multi-scale fused steel rail bolt assembly according to claim 4, wherein in the step S3, the method further comprises the following steps:
s301: extracting shallow layer characteristics of a middle-layer network output bolt of the residual error network block and performing convolution operation to obtain a first group of characteristics;
s302: extracting and convolving the middle layer characteristics of the middle layer network output bolt of the residual error network block;
s303: up-sampling the deep features of the bolt output by the SPP network and fusing the deep features with the middle features of the bolt in S302;
s304: down-sampling the features output in the step S301, and fusing the features with the features obtained in the step S303 again to obtain a second group of features;
s305: and after downsampling the features obtained in the step S304, fusing the features with the deep features of the bolt output by the SPP network to obtain a third group of features.
6. The method for detecting the failure of the multi-scale fused steel rail bolt assembly according to the claim 5, wherein in the step S5, the method comprises the following steps:
s401: predicting the characteristics to obtain a prediction frame;
s402: predicting the state of the extracted bolt features through a non-maximum suppression algorithm;
s403: generating a category information list and a final bolt prediction data frame;
s404: and calling a drawing function to draw the bolt prediction box.
7. The method for detecting the failure of the rail Bolt assembly with multi-scale fusion according to claim 1, wherein the status labels in step S4 include (lose), (Bolt break), (Nut crack), (Missing Nut), (scrubber fraction), (Missing gate), (overhead Missing), (NW dispear), which respectively correspond to the failure of the Bolt assembly: reduced tightening force, bolt breakage, nut cracking, nut missing, washer breakage, washer missing, nut washer missing, and overall part missing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110341276.6A CN112907585A (en) | 2021-03-30 | 2021-03-30 | Multi-scale fusion steel rail bolt assembly fault detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110341276.6A CN112907585A (en) | 2021-03-30 | 2021-03-30 | Multi-scale fusion steel rail bolt assembly fault detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112907585A true CN112907585A (en) | 2021-06-04 |
Family
ID=76109733
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110341276.6A Pending CN112907585A (en) | 2021-03-30 | 2021-03-30 | Multi-scale fusion steel rail bolt assembly fault detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112907585A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111080603A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Method for detecting breakage fault of shaft end bolt of railway wagon |
| CN112288043A (en) * | 2020-12-23 | 2021-01-29 | 飞础科智慧科技(上海)有限公司 | Kiln surface defect detection method, system and medium |
| CN112418117A (en) * | 2020-11-27 | 2021-02-26 | 北京工商大学 | Small target detection method based on unmanned aerial vehicle image |
| CN112464846A (en) * | 2020-12-03 | 2021-03-09 | 武汉理工大学 | Automatic identification method for abnormal fault of freight train carriage at station |
-
2021
- 2021-03-30 CN CN202110341276.6A patent/CN112907585A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111080603A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Method for detecting breakage fault of shaft end bolt of railway wagon |
| CN112418117A (en) * | 2020-11-27 | 2021-02-26 | 北京工商大学 | Small target detection method based on unmanned aerial vehicle image |
| CN112464846A (en) * | 2020-12-03 | 2021-03-09 | 武汉理工大学 | Automatic identification method for abnormal fault of freight train carriage at station |
| CN112288043A (en) * | 2020-12-23 | 2021-01-29 | 飞础科智慧科技(上海)有限公司 | Kiln surface defect detection method, system and medium |
Non-Patent Citations (2)
| Title |
|---|
| 李彬，等: "改进YOLOv4 算法的航空发动机部件表面缺陷检测", 《激光与光电子学进展》 * |
| 王振，等: "基于YOLO v3的钢轨螺栓组件故障检测方法", 《机器人技术与应用》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109615019B (en) | Abnormal behavior detection method based on space-time automatic encoder | |
| CN106056079B (en) | A kind of occlusion detection method of image capture device and human face five-sense-organ | |
| CN111080620A (en) | Road disease detection method based on deep learning | |
| CN107967475A (en) | A kind of method for recognizing verification code based on window sliding and convolutional neural networks | |
| CN111797890A (en) | Method and system for detecting defects of power transmission line equipment | |
| Rizvi et al. | Crack detection in railway track using image processing | |
| CN110232380A (en) | Fire night scenes restored method based on Mask R-CNN neural network | |
| CN109063609A (en) | A kind of anomaly detection method based on Optical-flow Feature in conjunction with full convolution semantic segmentation feature | |
| WO2023082418A1 (en) | Power utility tunnel settlement crack identification method based on artificial intelligence technology | |
| CN111862029A (en) | Fault detection method for bolt part of vertical shock absorber of railway motor train unit | |
| CN110009622B (en) | Display panel appearance defect detection network and defect detection method thereof | |
| CN111401149B (en) | Lightweight video behavior identification method based on long-short-term time domain modeling algorithm | |
| CN116416497B (en) | Bearing fault diagnosis system and method | |
| CN113887654A (en) | Surface defect region segmentation method, device and equipment based on staggered feature fusion | |
| CN109902730B (en) | Power transmission line broken strand detection method based on deep learning | |
| CN103325123A (en) | Image edge detection method based on self-adaptive neural fuzzy inference systems | |
| CN115457304A (en) | Luggage damage analysis method and system based on target detection | |
| CN106355187A (en) | Application of visual information to electrical equipment monitoring | |
| CN108761263B (en) | Fault diagnosis system based on evidence theory | |
| CN112907585A (en) | Multi-scale fusion steel rail bolt assembly fault detection method | |
| Katsamenis et al. | A Few-Shot Attention Recurrent Residual U-Net for Crack Segmentation | |
| CN111310899B (en) | Power defect identification method based on symbiotic relation and small sample learning | |
| CN113159052A (en) | Flexural reinforced concrete simply supported beam failure mode identification method based on deep learning | |
| CN113343760A (en) | Human behavior recognition method based on multi-scale characteristic neural network | |
| CN113033624A (en) | Industrial image fault diagnosis method based on federal learning |
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: 20210604 |