CN115147664A - Fault image identification method for falling of anti-theft cover of brake valve of railway wagon - Google Patents
Fault image identification method for falling of anti-theft cover of brake valve of railway wagon Download PDFInfo
- Publication number
- CN115147664A CN115147664A CN202210900203.0A CN202210900203A CN115147664A CN 115147664 A CN115147664 A CN 115147664A CN 202210900203 A CN202210900203 A CN 202210900203A CN 115147664 A CN115147664 A CN 115147664A
- Authority
- CN
- China
- Prior art keywords
- network
- brake valve
- theft cover
- image
- convolution
- 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
- 238000000034 method Methods 0.000 title claims abstract description 49
- 238000001514 detection method Methods 0.000 claims abstract description 58
- 238000000605 extraction Methods 0.000 claims abstract description 17
- 239000011159 matrix material Substances 0.000 claims description 18
- 238000011176 pooling Methods 0.000 claims description 9
- 230000003014 reinforcing effect Effects 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 6
- 230000008054 signal transmission Effects 0.000 claims description 3
- 230000001629 suppression Effects 0.000 claims 1
- 238000013135 deep learning Methods 0.000 abstract description 8
- 238000007689 inspection Methods 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000005764 inhibitory process Effects 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000003321 amplification Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011179 visual inspection Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
- G06T7/74—Determining position or orientation of objects or cameras using feature-based methods involving reference images or patches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/776—Validation; Performance evaluation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/778—Active pattern-learning, e.g. online learning of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/09—Recognition of logos
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Valves And Accessory Devices For Braking Systems (AREA)
- Burglar Alarm Systems (AREA)
Abstract
A fault image identification method for falling of a railway wagon brake valve anti-theft cover relates to the technical field of railway wagon fault detection. The invention aims to solve the problem that the detection error rate is high when a brake valve anti-theft cover is identified and detected by a manual or deep learning detection network. The method is improved aiming at the characteristic extraction stage, the characteristic extraction is carried out by splitting the convolution kernel, the detection precision of the target can be improved, the hyper-parameters of the training model are adjusted in a random search mode, and the model training time and the model deduction time can be effectively reduced. The invention is suitable for detecting the anti-theft cover of the brake valve of the railway wagon.
Description
Technical Field
The invention belongs to the technical field of fault detection of rail wagons, and particularly relates to a deep learning network for fault detection.
Background
During operation of a railway wagon, parts of the wagon are generally inspected by means of manual inspection and manual visual inspection. The manual vehicle inspection workload is large, the time is long, the efficiency is low, the failure missing inspection probability is large, and the manual image inspection mode has overlarge image inspection amount and is easy to generate false inspection missing inspection. Therefore, in the prior art, a deep learning detection network is adopted to detect the pictures, so that the problem of false detection and missing detection caused by manual picture viewing is solved. However, in the fault image detection of the existing cascade-rcnn deep learning detection network for the brake valve anti-theft cover, the background of the railway wagon is complex, and the robustness of the model is poor. Therefore, the problems of high false detection rate, long detection time, low detection precision and the like still exist.
In conclusion, when the brake valve anti-theft cover is identified and detected by both manual work and deep learning detection networks, the problem of high detection error rate exists, and finally, the driving safety is seriously influenced.
Disclosure of Invention
The invention provides a fault image identification method for the falling-off of a brake valve anti-theft cover of a railway wagon, aiming at solving the problem that the detection error rate is high when the brake valve anti-theft cover is identified and detected by a manual or deep learning detection network.
A fault image recognition method for the falling of a brake valve anti-theft cover of a railway wagon comprises the steps of inputting a detected image into a trained cascade-rcnn network, obtaining a detection result, and judging whether the brake valve anti-theft cover falls off or not by utilizing the detection result; the method comprises the steps that a resnet convolution network is adopted in the cascade-rcnn network to conduct feature extraction on an image, the resnet convolution network comprises 12 groups of residual error units, each group of residual error units comprises 16 convolution branched chains which are connected in parallel, each convolution branched chain conducts 1 × 1 convolution operation and 3 × 3 convolution operation on input of the residual error unit, a filter is used for enabling feature matrixes output by the 16 convolution branched chains to be connected in series, then 1 × 1 convolution operation is conducted on the feature matrixes after the feature matrixes are connected in series, results obtained through a rule activation function after the convolution results are added with the input of the residual error units are input into the next group of residual error units, and the output results of the last group of residual error units are output of the res convolution network.
Further, the input of the first group of residual error units is obtained by the following method:
the method comprises the steps of enabling a measured image to sequentially pass through a 7 × 7 convolution layer, a BN layer, a relu activation function and a maximum pooling layer to obtain a feature matrix with the channel number of 64, then respectively inputting the feature matrix into two first convolution blocks of 1 × 1, enabling the output of one first convolution block to sequentially pass through a second convolution block of 3 × 3 and a third convolution block of 1 × 1 to obtain a feature matrix with the channel number of 256, and enabling the sum of the feature matrix with the channel number of 256 and the output of the other first convolution block to serve as the input of a first group of residual error units.
Further, the cascade-rcnn network comprises a basic detection network and three detectors which are connected in series, and the output result of the basic detection network sequentially passes through the three detectors to obtain the output of the cascade-rcnn network;
the basic detection network comprises: the system comprises a feature extraction network, a regional suggestion network and a target positioning network, wherein each detector comprises: a region of interest pooling layer, a region suggestion network, and an object location network.
Further, the feature extraction network: and (4) performing feature extraction on the input image by adopting a resnet convolution network to obtain a feature map.
Further, the area suggests a network: and intercepting the candidate box in the feature map by utilizing the RPN network.
Further, the target positioning network: classifying the candidate frames intercepted by the RPN to obtain the candidate frames containing the detected target, then returning the candidate frames containing the detected target to the original position of the detected image, and finally selecting the candidate frame with the highest score as the output result of the target positioning network according to a non-maximum inhibition algorithm.
Further, the region of interest pooling layer: and adjusting the output result of the target positioning network to the dimension of the feature map output by the resnet convolution network.
Further, the IOU thresholds of the area-proposed networks in the three detectors are set to 0.55, 0.65 and 0.7 in sequence according to the signal transmission order.
Further, the specific method for judging whether the brake valve anti-theft cover falls off by using the detection result comprises the following steps:
respectively judging whether the detection results meet the following three points:
1) And judging whether the brake valve exists or not,
2) Judging whether the number of the complete hanging brackets of the anti-theft cover of the brake valve is 2 or not,
3) Judging whether the number of the reinforcing bolts is 2 or not,
if the detection results all accord with the three points, the brake valve anti-theft cover does not fall off, otherwise, the brake valve anti-theft cover falls off.
Further, the method for obtaining the detected image comprises the following steps:
the method comprises the steps of collecting a middle image of a vehicle to be tested, positioning a brake valve anti-theft cover in the middle image, intercepting a part containing the brake valve anti-theft cover to obtain a sub-image, marking a reinforcing bolt, a hanger and a brake valve of the brake valve anti-theft cover by using rectangular frames in the sub-image, and taking the marked sub-image as a tested image.
The fault image identification method for the falling of the anti-theft cover of the brake valve of the railway wagon has the beneficial effects that:
1. the invention adopts the deep learning algorithm to identify the image, can accurately position the position of the brake valve anti-theft cover at the bottom of the truck, and carries out fault detection on the brake valve anti-theft cover, thereby greatly improving the detection efficiency and accuracy, saving a lot of manpower and material resources and increasing a guarantee for the safety of railway operation.
2. The method is based on a Cascade-RCNN deep learning network, a Cascade-RCNN basic trunk network Resnet is improved, in each layer of convolution, a plurality of convolution kernels are used for being connected in parallel, each branch is called a base number, the original plurality of convolution kernels in the convolution layer are split into a plurality of kernels with small number, and the convolution kernels with small number have the same structure. By adopting the improved feature extraction network, the target information is merged into the deep network layer by layer, so that the feature expression of the target can be enhanced, the small target and the semi-shielded target can be better detected, and the target detection precision is improved. And in the parameter adjusting process of model training, the hyper-parameters are adjusted by adopting a random search method, so that the detection time of the model is greatly reduced.
Drawings
FIG. 1 is a diagram of a Cascade-Rcnn network structure;
fig. 2 is a structural diagram of a residual error unit in the resnet convolutional network.
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 should be noted that the embodiments and features of the embodiments may be combined with each other without conflict.
The first embodiment is as follows: according to the fault image identification method for the falling of the brake valve anti-theft cover of the railway wagon, the detected image is input into a trained cascade-rcnn network, a detection result is obtained, and whether the brake valve anti-theft cover falls or not is judged according to the detection result.
Specifically, in the present embodiment, a resnet convolution network is used in a cascade-rcnn network to perform feature extraction on an image, and as shown in fig. 1, the number of input channels, the size of a convolution kernel, and the number of output channels are taken as three parameter values corresponding to each layer. The resnet convolutional network comprises 12 groups of residual error units connected in series, and each group of residual error units comprises 16 convolutional branched chains connected in parallel, as shown in fig. 2. The input of each group of residual error units is 256 characteristic matrixes and is divided into two paths to operate. One path of input is respectively input into 16 convolution chains, each convolution branched chain firstly carries out 1 x 1 convolution operation on the input, so that an output result is changed into a characteristic matrix with 8 channel numbers, the dimensionality can be reduced, the calculation bottleneck can be reduced, the number of network layers can be increased, the expression capacity of the network can be improved, and the expression bottleneck of the network can be effectively avoided. And then performing convolution operation of 3 x 3, wherein the number of channels of the feature matrix is unchanged. And then, the filter is utilized to serially connect the characteristic matrixes output by the 16 convolution branched chains to form a characteristic matrix with 128 channels, and the aggregation in a lower dimensional space does not lose the expression capacity, better balances the width and the depth of the network and reduces the loss of information content. The feature matrix with 128 channels is convolved by 1 × 1 to make 256 channels. The feature matrix with 256 channels is added with the other input of the residual error unit, and the added sum obtains a new feature matrix with 256 channels through a rule activation function as the output of the residual error unit. The output of the previous group of residual error units is used as the input of the next group of residual error units, and the output result of the last group of residual error units is the output of the resnet convolution network.
The implementation mode is mainly improved aiming at the characteristic extraction stage, the original network adopts a ResNet network as a main network to extract the characteristics, and the core idea is to use a form of a plurality of residual error quick links. The Resnet convolution network extracts features of an image, and in each layer of convolution, a plurality of convolution kernels are used for being connected in parallel, attribute values of each branch in parallel are the same as those of other branches, namely each branch is called a base number to split original convolution kernels in a convolution layer into a plurality of kernels with small numbers, the convolution kernels with small numbers are identical in structure, and therefore the frequency of target detection accuracy reduction is reduced in the process of continuously deepening the network. The improved residual error calculation unit has fewer parameters than the traditional residual error calculation unit, and the use of memory and calculation resources is smaller. The method and the device can further improve the detection precision, carry out feature extraction by splitting the convolution kernel, improve the detection precision of the target, adjust the hyper-parameters of the training model by a random search mode, and effectively reduce the time of model training and the time of model inference.
In summary, the images of the anti-theft cover of the brake valve are detected by using the network, so that the hanger, the reinforcing bolt and the brake valve of the anti-theft cover of the brake valve can be detected under the complex background environment of the railway wagon, and the result is output.
The second embodiment is as follows: in the present embodiment, the method for identifying a fault image of a railway wagon brake valve anti-theft cover falling off according to the first embodiment is further described, in the present embodiment, the input of the first group of residual error units is obtained by the following method:
and (3) sequentially passing the detected image through a 7-by-7 convolutional layer, a BN layer, a relu activation function and a maximum pooling layer to obtain a feature matrix with 64 channels. And then inputting the feature matrix into two first convolution blocks of 1 × 1, wherein the output of one first convolution block passes through a second convolution block of 3 × 3 and a third convolution block of 1 × 1 in sequence to obtain a feature matrix with the channel number of 256, and the sum of the feature matrix with the channel number of 256 and the output of the other first convolution block is used as the input of a first group of residual error units.
The third concrete implementation mode: the embodiment further illustrates a method for identifying a falling fault image of a railway wagon brake valve anti-theft cover, according to the first or second embodiment, in the embodiment, a cascade-rcnn network comprises a basic detection network and three detectors which are connected in series, and an output result of the basic detection network sequentially passes through the three detectors to obtain an output of the cascade-rcnn network;
the basic detection network comprises: the system comprises a feature extraction network, a regional suggestion network and a target positioning network, wherein each detector comprises: a region of interest pooling layer, a region suggestion network, and a target location network.
The fourth concrete implementation mode: the embodiment further describes a method for identifying a fault image of falling off of a railway wagon brake valve anti-theft cover, which is described in the third specific embodiment, and in the embodiment, the feature extraction network: and performing feature extraction on the input image by adopting a resnet convolution network to obtain a feature map.
The fifth concrete implementation mode: the embodiment further describes a method for identifying a fault image of falling off of a railway wagon brake valve anti-theft cover, which is described in the fourth specific embodiment, and in the embodiment, a regional suggestion network: and intercepting the candidate box in the feature map by using the RPN network.
In the embodiment, an anchor point mechanism is introduced into an RPN (resilient proxy Network), whether an anchor point is a background or a target is judged through an IOU threshold value, and then frame regression is utilized to correct the anchor point to obtain an accurate candidate frame.
The sixth specific implementation mode: the embodiment further describes a method for identifying a fault image of falling off of a railway wagon brake valve anti-theft cover in a fifth specific embodiment, and in the embodiment, a target positioning network: classifying the candidate frames intercepted by the RPN to obtain the candidate frames containing the detected target, then returning the candidate frames containing the detected target to the original position of the detected image, and finally selecting the candidate frame with the highest score as the output result of the target positioning network according to a non-maximum inhibition algorithm.
In the embodiment, the candidate frames generated by the RPN are classified, and meanwhile, the regression of the boundary frame is used again, so that the accurate position of the detection frame in the original detected image is finally obtained.
The seventh embodiment: in the present embodiment, a method for identifying a fault image of a railway wagon brake valve anti-theft cover falling off according to a sixth specific embodiment is further described, in the present embodiment, a pooling layer of an area of interest: and adjusting the output result of the target positioning network to the dimension of the characteristic diagram output by the resnet convolution network. I.e. the different sized profiles are adjusted to fixed dimensions according to the requirements of the fully connected layer.
The specific implementation mode is eight: in the embodiment, the fifth or seventh embodiment further describes a method for identifying a falling fault image of a railway wagon brake valve anti-theft cover, and in the embodiment, the IOU thresholds of the area recommendation network in the three detectors are sequentially set to be 0.55, 0.65 and 0.7 according to the signal transmission sequence.
In this embodiment, by continuously increasing the threshold of the IOU, the candidate box after fine adjustment in the upper layer can be defined as a new input and provided to the lower detector. And selecting a random search method to adjust the hyper-parameters in the model training process, promoting the quality of the candidate frame to be more ideal on the basis of continuous optimization, and finally improving the detection efficiency.
The specific implementation method nine: the embodiment further describes a method for identifying a fault image of falling of a railway wagon brake valve anti-theft cover, which is described in the specific embodiment, and in the embodiment, a specific method for judging whether the brake valve anti-theft cover falls off by using a detection result is as follows:
respectively judging whether the detection results meet the following three points:
1) And judging whether the brake valve exists or not,
2) Judging whether the number of the complete hanging brackets of the anti-theft cover of the brake valve is 2 or not,
3) Judging whether the number of the reinforcing bolts is 2 or not,
if the detection results all accord with the three points, the brake valve anti-theft cover does not fall off, otherwise, the brake valve anti-theft cover falls off.
In the embodiment, the image to be detected is input into the trained detection network, and the network outputs the detection result. And then, respectively judging which point does not meet the requirements, and outputting the coordinates of the point on the truck and carrying out fault alarm.
The detailed implementation mode is ten: in the present embodiment, a method for identifying a fault image of a brake valve anti-theft cover of a railway wagon falling off, which is one of the specific embodiments, will be further described,
and acquiring a high-definition image of the truck through high-definition imaging equipment built around the truck track. Because the shape and the position of the anti-theft cover of the brake valve of different vehicle types are not fixed, the middle image of the tested vehicle needs to be collected. And then positioning the brake valve anti-theft cover in the middle image according to the prior knowledge, and intercepting the part containing the brake valve anti-theft cover to obtain a sub-image. And marking the reinforcing bolt, the hanger and the brake valve of the anti-theft cover of the brake valve in the sub-image by using a rectangular frame, and taking the marked sub-image as a detected image.
In practical application, due to the fact that shooting conditions of different stations are different, images shot by equipment are different, a camera can be influenced by natural conditions such as rainwater and mud stains, and the images of the brake valve anti-theft cover obtained are different. In order to ensure the diversity of data sets, images of the brake valve anti-theft covers of different vehicle types, which are shot by all stations under different conditions, need to be collected completely.
Similarly, when training the cascade-rcnn network, a sample data set for detection needs to be constructed, where the sample data set includes: the original image set and the marker information set. The original image is the positioned sub-image. The marking information set is a rectangular frame containing a target area in the subimage, a reinforcing bolt of the anti-theft cover of the brake valve, a hanger for fixing the anti-theft cover and the brake valve in the anti-theft cover are arranged in the rectangular frame, and the mark can be obtained manually. The original image corresponds to the label information one to one. In order to improve the stability and the applicability of the network, data set amplification is also performed respectively, and mainly includes rotation, translation, scaling, mirror image, brightness adjustment and the like of an original image. Each operation is carried out randomly, and the diversity of the samples is ensured to the maximum extent.
The detection precision is obviously improved through the feature extraction network improved by the embodiment, the mode of adjusting the hyper-parameters in a random search mode is adopted, the time of model training is reduced while the detection precision is improved, and the optimal solution of the detection precision and the detection efficiency is realized.
Although the invention herein has been described with reference to particular embodiments, it is to be understood that these embodiments are merely illustrative of the principles and applications of the present invention. It is therefore to be understood that numerous modifications may be made to the illustrative embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims. It should be understood that features described in different dependent claims and herein may be combined in ways different from those described in the original claims. It is also to be understood that features described in connection with individual embodiments may be used in other described embodiments.
Claims (10)
1. A fault image recognition method for falling of a brake valve anti-theft cover of a railway wagon is characterized in that a detected image is input into a trained cascade-rcnn network to obtain a detection result, and whether the brake valve anti-theft cover falls off or not is judged by using the detection result;
the method is characterized in that a resnet convolution network is adopted in the cascade-rcnn network to extract the features of the image, the resnet convolution network comprises 12 groups of residual error units, each group of residual error units comprises 16 convolution branched chains which are connected in parallel, each convolution branched chain carries out 1 × 1 convolution operation and 3 × 3 convolution operation on the input of the residual error unit, a filter is used for serially connecting feature matrixes output by the 16 convolution branched chains, then 1 × 1 convolution operation is carried out on the feature matrixes which are serially connected, the convolution result and the input of the residual error units are added, then the result obtained through a rule activation function is input into the next group of residual error units, and the output result of the last group of residual error units is the output of the resnet convolution network.
2. The method for identifying the fault image of the falling of the anti-theft cover of the brake valve of the railway wagon as claimed in claim 1, wherein the input of the first group of residual error units is obtained by the following method:
the method comprises the steps of enabling a measured image to sequentially pass through a 7 × 7 convolution layer, a BN layer, a relu activation function and a maximum pooling layer to obtain a feature matrix with the channel number of 64, then respectively inputting the feature matrix into two first convolution blocks of 1 × 1, enabling the output of one first convolution block to sequentially pass through a second convolution block of 3 × 3 and a third convolution block of 1 × 1 to obtain a feature matrix with the channel number of 256, and enabling the sum of the feature matrix with the channel number of 256 and the output of the other first convolution block to serve as the input of a first group of residual error units.
3. The method for identifying the falling fault image of the anti-theft cover of the brake valve of the railway wagon according to the claim 1 or 2, wherein the cascade-rcnn network comprises a basic detection network and three detectors which are connected in series, and the output result of the basic detection network sequentially passes through the three detectors to obtain the output of the cascade-rcnn network;
the basic detection network comprises: the system comprises a feature extraction network, a regional suggestion network and a target positioning network, wherein each detector comprises: a region of interest pooling layer, a region suggestion network, and an object location network.
4. The method for identifying the fault image of the falling of the anti-theft cover of the brake valve of the railway wagon as claimed in claim 3, wherein the feature extraction network comprises the following steps: and performing feature extraction on the input image by adopting a resnet convolution network to obtain a feature map.
5. The method for identifying the fault image of the falling of the anti-theft cover of the brake valve of the railway wagon as claimed in claim 4, wherein the area recommendation network comprises the following steps: and intercepting the candidate box in the feature map by utilizing the RPN network.
6. The method for identifying the fault image of the falling of the anti-theft cover of the brake valve of the railway wagon as claimed in claim 5, wherein the target positioning network comprises the following steps: and classifying the candidate frames intercepted by the RPN to obtain the candidate frames containing the detected target, then returning the candidate frames containing the detected target to the original position of the detected image, and finally selecting the candidate frame with the highest score as the output result of the target positioning network according to a non-maximum suppression algorithm.
7. The method for identifying the fault image of the falling-off of the anti-theft cover of the brake valve of the railway wagon as claimed in claim 6, wherein the region-of-interest pooling layer comprises: and adjusting the output result of the target positioning network to the dimension of the feature map output by the resnet convolution network.
8. A fault image identification method for the falling-off of the anti-theft cover of the brake valve of the railway wagon as claimed in claim 5 or 7, wherein the IOU threshold values of the area recommendation network in the three detectors are set to be 0.55, 0.65 and 0.7 in sequence according to the signal transmission sequence.
9. The method for identifying the fault image of the falling of the anti-theft cover of the brake valve of the railway wagon according to claim 1, wherein the specific method for judging whether the anti-theft cover of the brake valve falls off by using the detection result comprises the following steps:
respectively judging whether the detection results meet the following three points:
1) And judging whether the brake valve exists or not,
2) Judging whether the number of the complete hanging brackets of the anti-theft cover of the brake valve is 2 or not,
3) Judging whether the number of the reinforcing bolts is 2 or not,
if the detection results all accord with the three points, the brake valve anti-theft cover does not fall off, otherwise, the brake valve anti-theft cover falls off.
10. The method for identifying the fault image of the falling of the anti-theft cover of the brake valve of the railway wagon as claimed in claim 1, wherein the method for obtaining the tested image comprises the following steps:
the method comprises the steps of collecting a middle image of a vehicle to be tested, positioning a brake valve anti-theft cover in the middle image, intercepting a part containing the brake valve anti-theft cover to obtain a sub-image, marking a reinforcing bolt, a hanger and a brake valve of the brake valve anti-theft cover in the sub-image by using a rectangular frame, and taking the marked sub-image as a tested image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210900203.0A CN115147664A (en) | 2022-07-28 | 2022-07-28 | Fault image identification method for falling of anti-theft cover of brake valve of railway wagon |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210900203.0A CN115147664A (en) | 2022-07-28 | 2022-07-28 | Fault image identification method for falling of anti-theft cover of brake valve of railway wagon |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115147664A true CN115147664A (en) | 2022-10-04 |
Family
ID=83414308
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210900203.0A Pending CN115147664A (en) | 2022-07-28 | 2022-07-28 | Fault image identification method for falling of anti-theft cover of brake valve of railway wagon |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115147664A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210056351A1 (en) * | 2018-06-04 | 2021-02-25 | Jiangnan University | Multi-scale aware pedestrian detection method based on improved full convolutional network |
| CN114049478A (en) * | 2022-01-13 | 2022-02-15 | 武汉理工大学 | Infrared ship image rapid identification method and system based on improved Cascade R-CNN |
| CN114332754A (en) * | 2021-12-14 | 2022-04-12 | 长讯通信服务有限公司 | Cascade R-CNN pedestrian detection method based on multi-metric detector |
-
2022
- 2022-07-28 CN CN202210900203.0A patent/CN115147664A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210056351A1 (en) * | 2018-06-04 | 2021-02-25 | Jiangnan University | Multi-scale aware pedestrian detection method based on improved full convolutional network |
| CN114332754A (en) * | 2021-12-14 | 2022-04-12 | 长讯通信服务有限公司 | Cascade R-CNN pedestrian detection method based on multi-metric detector |
| CN114049478A (en) * | 2022-01-13 | 2022-02-15 | 武汉理工大学 | Infrared ship image rapid identification method and system based on improved Cascade R-CNN |
Non-Patent Citations (4)
| Title |
|---|
| QQ_29631521: "《https://blog.csdn.net/qq_29631521/article/details/93136216》", 20 June 2019 * |
| SAINING XIE等: ""Aggregated Residual Transformations for Deep Neural Networks"", 《2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION(CVPR)》 * |
| 吉鹏飞: ""基于改进的Cascade－ RCNN 网络的人员检测算法"", 《智能计算机与应用》 * |
| 吉鹏飞: ""基于改进的Cascade-RCNN网络的人员检测算法"", 《智能计算机与应用》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114937055B (en) | Image self-adaptive segmentation method and system based on artificial intelligence | |
| CN111784633B (en) | Insulator defect automatic detection algorithm for electric power inspection video | |
| CN111652227B (en) | Method for detecting damage fault of bottom floor of railway wagon | |
| CN111797890A (en) | Method and system for detecting defects of power transmission line equipment | |
| CN110992349A (en) | Underground pipeline abnormity automatic positioning and identification method based on deep learning | |
| CN102013008A (en) | Smoke detection method based on support vector machine and device | |
| CN111080609B (en) | Brake shoe bolt loss detection method based on deep learning | |
| CN103093458B (en) | The detection method of key frame and device | |
| CN108648210B (en) | Rapid multi-target detection method and device under static complex scene | |
| CN112906534A (en) | Lock catch loss fault detection method based on improved Faster R-CNN network | |
| CN111079822A (en) | Method for identifying dislocation fault image of middle rubber and upper and lower plates of axle box rubber pad | |
| CN111091551A (en) | Method for detecting loss fault of brake beam strut opening pin of railway wagon | |
| CN113221839A (en) | Automatic truck image identification method and system | |
| CN113409267A (en) | Pavement crack detection and segmentation method based on deep learning | |
| CN117541534A (en) | Power transmission line inspection method based on unmanned plane and CNN-BiLSTM model | |
| CN115546223A (en) | Method and system for detecting loss of fastening bolt of equipment under train | |
| CN115527170A (en) | Method and system for identifying closing fault of door stopper handle of automatic freight car derailing brake device | |
| CN114549414A (en) | Abnormal change detection method and system for track data | |
| CN111652228B (en) | Railway wagon sleeper beam hole foreign matter detection method | |
| CN111179278B (en) | Image detection method, device, equipment and storage medium | |
| CN116958073A (en) | Small sample steel defect detection method based on attention feature pyramid mechanism | |
| CN115147664A (en) | Fault image identification method for falling of anti-theft cover of brake valve of railway wagon | |
| CN116309407A (en) | Method for detecting abnormal state of railway contact net bolt | |
| CN110163081A (en) | SSD-based real-time regional intrusion detection method, system and storage medium | |
| CN112712055B (en) | Double-path deformable CNN coal mine gateway belt conveying foreign matter monitoring method |
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: 20221004 |