CN112766260A - Image identification method and system for positioning air reservoir for accelerating and relieving railway train - Google Patents
Image identification method and system for positioning air reservoir for accelerating and relieving railway train Download PDFInfo
- Publication number
- CN112766260A CN112766260A CN202110056516.8A CN202110056516A CN112766260A CN 112766260 A CN112766260 A CN 112766260A CN 202110056516 A CN202110056516 A CN 202110056516A CN 112766260 A CN112766260 A CN 112766260A
- Authority
- CN
- China
- Prior art keywords
- accelerated release
- release reservoir
- positioning
- frame
- reservoir
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000001514 detection method Methods 0.000 claims abstract description 23
- 238000011176 pooling Methods 0.000 claims abstract description 8
- 238000012549 training Methods 0.000 claims description 22
- 238000005070 sampling Methods 0.000 claims description 17
- 230000007246 mechanism Effects 0.000 claims description 12
- 230000003321 amplification Effects 0.000 claims description 9
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 9
- 238000012216 screening Methods 0.000 claims description 8
- 238000010586 diagram Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 230000001133 acceleration Effects 0.000 abstract description 17
- 230000000116 mitigating effect Effects 0.000 abstract description 12
- 238000012423 maintenance Methods 0.000 abstract description 5
- 230000006870 function Effects 0.000 description 15
- 230000008569 process Effects 0.000 description 7
- 238000013473 artificial intelligence Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 2
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 1
- 206010039203 Road traffic accident Diseases 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000003416 augmentation Effects 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 238000013519 translation 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
- G06V10/20—Image preprocessing
- G06V10/255—Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Multimedia (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Train Traffic Observation, Control, And Security (AREA)
Abstract
本发明公开了一种铁路列车加速缓解风缸定位的图像识别方法及系统，解决了现有人工检测效率低稳定性差的技术问题，属于图像识别技术领域，其中，该方法包括：构建加速缓解风缸的样本数据集；利用样本数据集对改进的SSD网络进行训练，得到加速缓解风缸定位模型，其中，改进的SSD模型的结构为：resnet50作为主干网络，删除顶层的BN层、池化层和分类输出层，额外增加两层卷积层，卷积核大小为3*3，步长为2；采集实际图像，输入至加速缓解风缸定位模型中，获得加速缓解风缸位置信息。该方法一方面相较于人工检修可提高检测效率和稳定性，另一方面可以提高检测的准确率。
The invention discloses an image recognition method and system for accelerating and relieving wind cylinder positioning of railway trains, which solves the technical problem of low efficiency and poor stability of existing manual detection, and belongs to the technical field of image recognition. The sample data set of the cylinder; using the sample data set to train the improved SSD network, the accelerated mitigation cylinder positioning model is obtained. The structure of the improved SSD model is: resnet50 is used as the backbone network, and the top BN layer and pooling layer are deleted. And the classification output layer, two additional convolution layers are added, the size of the convolution kernel is 3*3, and the step size is 2; the actual image is collected, and input to the acceleration and mitigation air cylinder positioning model to obtain the acceleration and mitigation air cylinder position information. On the one hand, the method can improve the detection efficiency and stability compared with manual maintenance, and on the other hand, it can improve the detection accuracy.
Description
Technical Field
The invention relates to the technical field of image recognition, in particular to an image recognition method and system for positioning an acceleration relieving air reservoir of a railway train.
Background
At present, the accelerated release reservoir is mainly positioned by manually checking images or checking on the spot, so that the mode has low efficiency, faults are easy to miss, and further, serious railway traffic accidents are easy to cause. Therefore, it is highly desirable to provide an automatic image recognition method for fault detection of the acceleration mitigation air cylinder, but since the acceleration mitigation air cylinders of different vehicle types are located at different positions on the vehicle body, an automatic image recognition method is required to first locate the position of the acceleration mitigation air cylinder, and then further perform subsequent fault detection.
Disclosure of Invention
The present invention is directed to solving, at least to some extent, one of the technical problems in the related art.
Therefore, the invention aims to provide an image recognition method for positioning an acceleration release reservoir of a railway train, which improves the detection efficiency, stability and accuracy.
Another object of the present invention is to provide an image recognition system for positioning an accelerated release reservoir of a railroad train.
In order to achieve the above object, an embodiment of the present invention provides an image recognition method for positioning an accelerated release reservoir of a railway train, including the following steps: step S1, constructing a sample data set of the accelerated release reservoir; step S2, training the improved SSD network by using the sample data set to obtain an accelerated release reservoir positioning model, wherein the improved SSD model has the structure that: resnet50 is used as a backbone network, a BN layer, a pooling layer and a classification output layer at the top layer are deleted, two convolution layers are additionally added, the size of a convolution kernel is 3 x 3, and the step length is 2; and step S3, acquiring an actual image, inputting the actual image into the accelerated release reservoir positioning model, and acquiring the position information of the accelerated release reservoir.
According to the image recognition method for positioning the accelerated release reservoir of the railway train, the position of the accelerated release reservoir is positioned in an automatic image recognition mode, on one hand, the detection efficiency and stability can be improved compared with manual maintenance, on the other hand, the detection accuracy can be improved along with the continuous maturity of deep learning and artificial intelligence technology, and meanwhile, through improving the SSD algorithm matching strategy, each real frame ground channel box can obtain equal number of predicted frames anchor, so that the problem of unbalanced positive and negative samples is solved, the convergence of loss functions in the training process is accelerated, and the accuracy of the algorithm for positioning smaller accelerated release reservoirs is also improved.
In addition, the image recognition method for positioning the accelerated release reservoir of the railway train according to the embodiment of the invention can also have the following additional technical characteristics:
further, in an embodiment of the present invention, step S1 specifically includes: acquiring coarse positioning images of a plurality of accelerated release air cylinders as an original image set; marking the area where the accelerated release reservoir is located in the original image set as a marking information set, wherein the original image set corresponds to the marking information set one by one; and amplifying the original image set and the marked information set by using an amplification algorithm to obtain a sample data set.
Further, in an embodiment of the present invention, step S2 specifically includes: inputting the sample data set into an improved SSD network to obtain a 5-layer characteristic diagram; respectively constructing two convolution kernels with the size of 3 x 3 on the 5 layers of feature graphs as a regression predictor and a multi-classification predictor; predicting the anchor of each position on a preset feature map by using a regression predictor and a multi-classification predictor, generating a default frame for each feature map by using a default frame generation rule to obtain a fixed number of output predictions, and screening the output predictions by using a non-maximum value to inhibit NMS and a top _ k threshold value to obtain an output frame; sampling the output frame by a random sampling or difficult sampling method to calculate the loss function, calculating the gradient value of the loss function, updating the parameters according to the gradient value, and performing iterative training until the maximum iteration times is reached to obtain the accelerated release reservoir positioning model.
Further, in an embodiment of the present invention, the IOUs corresponding to each real frame ground channel box are sorted in a descending order, then predicted frames anchor corresponding to the first M IOUs are taken as samples, each predicted frame anchor has a chance of matching with the real frame ground channel only once, M is a positive integer, and the IOU is an intersection ratio of the predicted frame and the real frame, wherein the input of the matching mechanism of the improved SSD network is: each predicted frame anchor and the IOU set corresponding to the real frame, and a parameter N; the matching mechanism output of the improved SSD network is: each predicted frame anchor matches the index of the resulting real frame.
Further, in an embodiment of the present invention, step S3 specifically includes: collecting an actual image; and zooming the actual image into an image with the size of 1600 × 256, inputting the zoomed image into an accelerated release reservoir positioning model for positioning detection of an accelerated release reservoir, judging a threshold value of an output result, and if the length-width ratio of a position frame is greater than 0.8 and less than 1.2, determining that the positioned reservoir is the accelerated release reservoir, and further outputting the position information of the accelerated release reservoir.
In order to achieve the above object, an embodiment of another aspect of the present invention provides an image recognition system for positioning an acceleration release reservoir of a railway train, including: the construction module is used for constructing a sample data set for accelerating the relieving of the reservoir; the training module is used for training the improved SSD network by utilizing the sample data set to obtain an accelerated release reservoir positioning model, wherein the improved SSD model has the structure that: resnet50 is used as a backbone network, a BN layer, a pooling layer and a classification output layer at the top layer are deleted, two convolution layers are additionally added, the size of a convolution kernel is 3 x 3, and the step length is 2; and the acquisition module is used for acquiring an actual image, inputting the actual image into the accelerated release reservoir positioning model and acquiring the accelerated release reservoir position information.
According to the image recognition system for positioning the accelerated release reservoir of the railway train, the position of the accelerated release reservoir is positioned in an automatic image recognition mode, on one hand, the detection efficiency and stability can be improved compared with manual maintenance, on the other hand, the detection accuracy can be improved along with the continuous maturity of deep learning and artificial intelligence technology, and meanwhile, through improving the SSD algorithm matching strategy, each real frame ground channel box can obtain equal number of predicted frames anchor, so that the problem of unbalanced positive and negative samples is solved, the convergence of loss functions in the training process is accelerated, and the accuracy of the algorithm for positioning smaller accelerated release reservoirs is also improved.
In addition, the image recognition system for positioning the accelerated release reservoir of the railway train according to the embodiment of the invention can also have the following additional technical characteristics:
further, in one embodiment of the present invention, the building module comprises: the device comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring rough positioning images of a plurality of accelerated release air cylinders as an original image set; the marking unit is used for marking an area where the accelerated release reservoir is located in the original image set as a marking information set, wherein the original image set corresponds to the marking information set one by one; and the amplification unit is used for amplifying the original image set and the marked information set by using an amplification algorithm to obtain a sample data set.
Further, in one embodiment of the present invention, the training module comprises: the input unit is used for inputting the sample data set into an improved SSD network to obtain a 5-layer characteristic diagram; the construction unit is used for respectively constructing two convolution kernels with the size of 3 x 3 on the 5 layers of feature maps as a regression predictor and a multi-classification predictor; the prediction and screening unit is used for predicting the prediction frame anchor of each position on a preset feature map by using the regression predictor and the multi-classification predictor, generating a default frame for each feature map by using a default frame generation rule to obtain a fixed number of output predictions, and screening the output predictions by adopting a non-maximum value to inhibit NMS and a top _ k threshold value to obtain an output frame; and the iteration unit is used for sampling the output frame by a random sampling or difficult sampling method to calculate the loss function, calculating the gradient value of the loss function, updating the parameter according to the gradient value, and performing iterative training until the maximum iteration number is reached to obtain the accelerated release reservoir positioning model.
Further, in one embodiment of the present invention, the algorithm matching policy in the improved SSD network is: the IOU corresponding to each real frame ground truth box is subjected to descending order, then, the predicted frames anchor corresponding to the first M IOUs are taken as samples, each predicted frame anchor has the chance of being matched with the real frame ground truth box only once, M is a positive integer, and the IOU is the intersection and comparison of the predicted frames and the real frames, wherein the input of a matching mechanism of the improved SSD network is as follows: an IOU set and a parameter N corresponding to each predicted frame anchor and a real frame ground channel box; the matching mechanism output of the improved SSD network is: and the index of the real frame group truth box is obtained by matching each predicted frame anchor.
Further, in one embodiment of the present invention, the obtaining module includes: the acquisition unit is used for acquiring an actual image; and the positioning and output unit is used for zooming the actual image into an image with the size of 1600 × 256, inputting the zoomed image into the accelerated release reservoir positioning model for positioning detection of the accelerated release reservoir, judging a threshold value of an output result, and if the length-width ratio of the position frame is greater than 0.8 and smaller than 1.2, determining that the positioned reservoir is the accelerated release reservoir and further outputting the position information of the accelerated release reservoir.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The foregoing and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
FIG. 1 is a flow chart of an image recognition method for positioning a rail train accelerated release reservoir in accordance with one embodiment of the present invention;
FIG. 2 is a flow chart of a specific accelerated mitigation reservoir location identification of the present invention;
fig. 3 is a schematic structural diagram of an image recognition system for positioning an acceleration mitigation reservoir of a railway train according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention.
The following describes an image recognition method and system for positioning an acceleration release reservoir of a railway train according to an embodiment of the invention with reference to the accompanying drawings.
FIG. 1 is a flow chart of an image recognition method for positioning a rail train acceleration mitigation reservoir in accordance with an embodiment of the present invention.
As shown in fig. 1, the image recognition method for positioning the accelerated release reservoir of the railway train comprises the following steps:
in step S1, a sample data set of accelerated release reservoirs is constructed.
Specifically, as shown in fig. 2, a high-definition image of the train can be acquired through a high-definition imaging device built around the train track. It can be understood that, because the form and the position of the acceleration relieving air cylinder of different vehicle types are not fixed, the image of the middle part of the vehicle needs to be taken for positioning, and the positioned acceleration relieving air cylinder is further taken as a sub-image. Due to the fact that shooting conditions of different stations are different, images shot by the equipment are different, the camera can be influenced by natural conditions such as rainwater and mud stains, and the acquired accelerated release reservoir images are different. In order to ensure the diversity of the data set, images of acceleration mitigation air cylinders of different vehicle types, which are shot by each station under different conditions, need to be collected completely, that is, the rough positioning images of a plurality of acceleration mitigation air cylinders are used as an original image set.
After the original image set is obtained, the area where the accelerated release reservoir is located in the original image set is marked manually to serve as a marking information set, namely the marking information set is information of a rectangular frame of a target area. Wherein, the original image set and the mark information set are required to be in one-to-one correspondence.
In order to improve the stability and adaptation habit of the algorithm, the data set is augmented by utilizing an augmentation algorithm, which can include rotation, translation, scaling, mirror image, brightness adjustment and the like of the original image. And after amplification, the original image set and the marking information set form a sample data set for accelerating and relieving the air cylinder.
In step S2, the improved SSD network is trained using the sample data set to obtain an accelerated release reservoir positioning model, where the improved SSD model has a structure: resnet50 is used as a backbone network, the top BN layer, pooling layer and classification output layer are deleted, two convolution layers are added, the convolution kernel size is 3 × 3, and the step size is 2.
Specifically, the vehicle bottom image containing the accelerated release reservoir is input into the network, firstly, resnet50 is selected as a main network to replace an original VGG cascade network, a BN layer, a pooling layer and a classification output layer at the top layer are deleted, and then two convolution layers are additionally added, wherein the convolution kernel size is 3 x 3, and the step length is 2. The 3-layer profile from resnet plus the expanded two-layer profile form all profiles. The output of the 5-layer residual block or convolutional layer on top of the network is used as a feature map for the SSD model. Then, two convolution kernels with the size of 3 x 3 are respectively constructed on the 5 layers of feature maps to serve as a regression predictor and a multi-classification predictor. The two convolution kernels predict the prediction frame anchor of each position on a specific characteristic diagram so as to obtain a fixed number of predictions, and further adopt non-maximum suppression (NMS) and top _ k thresholds to screen the output predictions so as to obtain a final output frame of the target detection model. And then sampling an output frame by a random sampling or difficult sampling method for calculating a loss function, wherein the classification loss and the position loss need to be calculated respectively, calculating the gradient value of the loss function, updating the parameters according to the gradient value, and performing iterative training until the maximum iteration number is reached to obtain an accelerated release reservoir positioning model.
It should be noted that in the existing model matching process, firstly, each real frame group channel will select a predicted frame anchor with the largest IOU (intersection ratio between the predicted frame and the real frame) to match, so as to ensure that each real frame group channel is at least matched to one predicted frame anchor, then, in all the IOUs of the predicted frame anchor and the real frame group channel, a part with a threshold value greater than 0.5 is selected for a subsequent prediction link, and the predicted frame anchor is matched to an object corresponding to the largest IOU. However, for smaller or larger objects, the matching strategy has a very limited number of positive samples, which is much smaller than the number of samples matched by medium-sized objects. The excessive difference of the sample quantity is not beneficial to the convergence of the loss function in the training process, and has certain influence on the detection effect.
Therefore, the SSD algorithm is improved, specifically, because the accelerated release reservoir occupies a small area at the bottom of the whole vehicle body, in order to improve the detection effect of the SSD on small targets, a new algorithm matching strategy is selected: and performing descending order on the IOU corresponding to each real frame ground channel, and then taking the predicted frames anchor corresponding to the first M IOUs as samples, wherein each predicted frame anchor has a matching chance only once, and M is a positive integer. The matching mechanism inputs are: and an IOU set and a parameter N corresponding to each predicted frame anchor and the real frame ground channel box. The matching mechanism outputs the index of the real frame group truth box matched by each predicted frame anchor. The convergence of the loss function in the training process also improves the accuracy of the algorithm for positioning the small accelerated release reservoir. Through an improved algorithm matching strategy, each real frame ground truth box can obtain equal number of predicted frames anchor, the problem of imbalance of positive and negative samples is solved, convergence of loss functions in the training process is accelerated, and the accuracy of the algorithm on smaller accelerated release air cylinders is improved.
In step S3, an actual image is acquired and input to the accelerated release reservoir positioning model, and accelerated release reservoir position information is obtained.
Specifically, as shown in fig. 2, an actual image of a station passing through the vehicle is collected, the actual image is scaled to 1600 × 256, the actual image is input into a trained positioning network to perform positioning detection of the accelerated release reservoir, an output result is judged, if the aspect ratio of a position frame is greater than 0.8 and smaller than 1.2 and the type is the accelerated release reservoir, the positioned reservoir is considered to be the accelerated release reservoir, and further, position information of a partial sub-image of the accelerated release reservoir and the position information of the accelerated release reservoir at the bottom of the vehicle are output.
According to the image recognition method for positioning the accelerated release reservoir of the railway train, provided by the embodiment of the invention, the position of the accelerated release reservoir is positioned in an automatic image recognition mode, so that on one hand, compared with manual maintenance, the detection efficiency and stability can be improved, on the other hand, along with the continuous maturity of deep learning and artificial intelligence technology, the detection accuracy can be improved, meanwhile, through improving the SSD algorithm matching strategy, each real frame group channel box can obtain equal number of predicted frames anchor, the problem of imbalance of positive and negative samples is solved, the convergence of loss functions in the training process is accelerated, and the accuracy of the algorithm for positioning smaller accelerated release reservoirs is also improved.
Next, an image recognition system for positioning an acceleration release reservoir of a railway train according to an embodiment of the present invention will be described with reference to the drawings.
Fig. 3 is a schematic structural diagram of an image recognition system for positioning an acceleration mitigation reservoir of a railway train according to an embodiment of the present invention.
As shown in fig. 3, the system 10 includes: a building module 100, a training module 200 and an acquisition module 300.
The building module 100 is configured to build a sample data set of the accelerated release reservoir. The training module 200 is configured to train the improved SSD network by using the sample data set, to obtain an accelerated release reservoir positioning model, where the improved SSD model has a structure: resnet50 is used as a backbone network, the top BN layer, pooling layer and classification output layer are deleted, two convolution layers are added, the convolution kernel size is 3 × 3, and the step size is 2. The obtaining module 300 is configured to collect an actual image, input the actual image into the accelerated release reservoir positioning model, and obtain the accelerated release reservoir position information.
Further, in one embodiment of the present invention, the building module 100 comprises: the acquiring unit 101 is configured to acquire coarse positioning images of multiple accelerated release air cylinders as an original image set; the marking unit 102 is configured to mark an area where the accelerated release reservoir is located in the original image set as a marking information set, where the original image set corresponds to the marking information set one to one; and the amplification unit 103 is configured to amplify the original image set and the labeled information set by using an amplification algorithm to obtain a sample data set.
Further, in one embodiment of the present invention, the training module 200 comprises: an input unit 201, configured to input the sample data set into an improved SSD network to obtain a 5-layer feature map; a constructing unit 202, configured to respectively construct two convolution kernels with a size of 3 × 3 on the 5-layer feature map as a regression predictor and a multi-class predictor; the prediction and screening unit 203 is used for predicting the prediction frame anchor of each position on a preset feature map by using a regression predictor and a multi-classification predictor, generating a default frame for each feature map by using a default frame generation rule to obtain a fixed number of output predictions, and screening the output predictions by using a non-maximum value to inhibit NMS and a top _ k threshold value to obtain an output frame; and the iteration unit 204 is configured to sample the output frame by using a random sampling or difficult sampling method to calculate the loss function, calculate a gradient value of the loss function, update parameters according to the gradient value, and perform iterative training until the maximum iteration number is reached, so as to obtain an accelerated release reservoir positioning model.
Further, in an embodiment of the present invention, the IOUs corresponding to each real frame ground channel box are sorted in a descending order, then predicted frames anchor corresponding to the first M IOUs are taken as samples, each predicted frame anchor has a chance of matching with the real frame ground channel only once, M is a positive integer, and the IOU is an intersection ratio of the predicted frame and the real frame, wherein the input of the matching mechanism of the improved SSD network is: an IOU set and a parameter N corresponding to each predicted frame anchor and a real frame ground channel box; the matching mechanism output of the improved SSD network is: and the index of the real frame group truth box is obtained by matching each predicted frame anchor.
Further, in an embodiment of the present invention, the obtaining module 300 includes: an acquisition unit 301 for acquiring an actual image; and the positioning and output unit 302 is configured to scale the actual image into an image with a size of 1600 × 256, input the scaled image into an accelerated release reservoir positioning model to perform positioning detection on the accelerated release reservoir, perform threshold judgment on an output result, and if the aspect ratio of the position frame is greater than 0.8 and smaller than 1.2, determine that the positioned reservoir is an accelerated release reservoir, and further output position information of the accelerated release reservoir.
It should be noted that the explanation of the foregoing embodiment of the image identification method for positioning the accelerated release reservoir of the railway train is also applicable to the system, and is not repeated here.
According to the image recognition system for positioning the accelerated release reservoir of the railway train, provided by the embodiment of the invention, the position of the accelerated release reservoir is positioned in an automatic image recognition mode, so that the detection efficiency and stability can be improved compared with manual maintenance on one hand, and the detection accuracy can be improved along with the continuous maturity of deep learning and artificial intelligence technology on the other hand.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "a plurality" means at least two, e.g., two, three, etc., unless specifically limited otherwise.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made to the above embodiments by those of ordinary skill in the art within the scope of the present invention.
Claims (10)
1. An image identification method for positioning an accelerated release reservoir of a railway train is characterized by comprising the following steps:
step S1, constructing a sample data set of the accelerated release reservoir;
step S2, training an improved SSD network by using the sample data set to obtain an accelerated release reservoir positioning model, wherein the improved SSD model has the structure that: resnet50 is used as a backbone network, a BN layer, a pooling layer and a classification output layer at the top layer are deleted, two convolution layers are additionally added, the size of a convolution kernel is 3 x 3, and the step length is 2;
and step S3, acquiring an actual image, inputting the actual image into the accelerated release reservoir positioning model, and acquiring the position information of the accelerated release reservoir.
2. The image recognition method for positioning the accelerated release reservoir of a railroad train according to claim 1, wherein the step S1 specifically comprises:
acquiring coarse positioning images of a plurality of accelerated release air cylinders as an original image set;
marking the area where the accelerated release reservoir is located in the original image set as a marked information set, wherein the original image set corresponds to the marked information set one by one;
and amplifying the original image set and the marked information set by using an amplification algorithm to obtain the sample data set.
3. The image recognition method for positioning the accelerated release reservoir of a railroad train according to claim 1, wherein the step S2 specifically comprises:
inputting the sample data set into the improved SSD network to obtain a 5-layer characteristic diagram;
respectively constructing two convolution kernels with the size of 3 x 3 on the 5 layers of feature maps as a regression predictor and a multi-classification predictor;
predicting a prediction frame anchor of each position on a preset feature map by using the regression predictor and the multi-classification predictor, generating a default frame for each feature map by using a default frame generation rule to obtain a fixed number of output predictions, and screening the output predictions by using a non-maximum value to inhibit NMS and a top _ k threshold value to obtain an output frame;
sampling the output frame by a random sampling or difficult sampling method to calculate a loss function, calculating a gradient value of the loss function, updating parameters according to the gradient value, and performing iterative training until the maximum iteration number is reached to obtain the accelerated release reservoir positioning model.
4. The method of claim 3, wherein the algorithm matching strategy in the improved SSD network is: the IOU corresponding to each real frame ground channel box is subjected to descending order, then predicted frames anchor corresponding to the first M IOUs are taken as samples, each predicted frame anchor has the chance of being matched with the real frame ground channel only once, M is a positive integer, and the IOU is the intersection and comparison of the predicted frames and the real frames, wherein the input of a matching mechanism of the improved SSD network is as follows: an IOU set and a parameter N corresponding to each predicted frame anchor and a real frame ground channel box; the matching mechanism output of the improved SSD network is: and the index of the real frame group truth box is obtained by matching each predicted frame anchor.
5. The image recognition method for positioning the accelerated release reservoir of a railroad train according to claim 1, wherein the step S3 specifically comprises:
collecting the actual image;
and zooming the actual image into an image with the size of 1600 × 256, inputting the zoomed image into the accelerated release reservoir positioning model for positioning detection of the accelerated release reservoir, judging a threshold value of an output result, and if the length-width ratio of a position frame is larger than 0.8 and smaller than 1.2, determining that the positioned reservoir is the accelerated release reservoir, and further outputting the position information of the accelerated release reservoir.
6. An image recognition system for accelerated release reservoir positioning of a railroad train, comprising:
the construction module is used for constructing a sample data set for accelerating the relieving of the reservoir;
a training module, configured to train an improved SSD network by using the sample data set, to obtain an accelerated release reservoir location model, where the improved SSD model has a structure: resnet50 is used as a backbone network, a BN layer, a pooling layer and a classification output layer at the top layer are deleted, two convolution layers are additionally added, the size of a convolution kernel is 3 x 3, and the step length is 2;
and the acquisition module is used for acquiring an actual image, inputting the actual image into the accelerated release reservoir positioning model and acquiring the accelerated release reservoir position information.
7. The image recognition system for railroad train accelerated release reservoir location of claim 6, wherein the building module comprises:
the device comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring rough positioning images of a plurality of accelerated release air cylinders as an original image set;
the marking unit is used for marking the area where the accelerated release reservoir is located in the original image set as a marking information set, wherein the original image set corresponds to the marking information set in a one-to-one mode;
and the amplification unit is used for amplifying the original image set and the marked information set by using an amplification algorithm to obtain the sample data set.
8. The image recognition system for accelerated release reservoir positioning of a railroad train of claim 6, wherein the training module comprises:
an input unit, configured to input the sample data set into the improved SSD network, to obtain a 5-layer feature map;
the construction unit is used for respectively constructing two convolution kernels with the size of 3 x 3 on the 5 layers of feature maps as a regression predictor and a multi-classification predictor;
the prediction and screening unit is used for predicting a prediction frame anchor of each position on a preset feature map by using the regression predictor and the multi-classification predictor, generating a default frame for each feature map by using a default frame generation rule to obtain a fixed number of output predictions, and screening the output predictions by using a non-maximum value to inhibit NMS and a top _ k threshold value to obtain an output frame;
and the iteration unit is used for sampling the output frame by a random sampling or difficult sampling method to calculate a loss function, calculating a gradient value of the loss function, updating parameters according to the gradient value, performing iterative training until the maximum iteration number is reached, and obtaining the accelerated release reservoir positioning model.
9. The image recognition system for accelerated release reservoir location of a railroad train of claim 8, wherein the algorithm matching strategy in the modified SSD network is: the IOU corresponding to each real frame ground channel box is subjected to descending order, then predicted frames anchor corresponding to the first M IOUs are taken as samples, each predicted frame anchor has the chance of being matched with the real frame ground channel only once, M is a positive integer, and the IOU is the intersection and comparison of the predicted frames and the real frames, wherein the input of a matching mechanism of the improved SSD network is as follows: an IOU set and a parameter N corresponding to each predicted frame anchor and a real frame ground channel box; the matching mechanism output of the improved SSD network is: and the index of the real frame group truth box is obtained by matching each predicted frame anchor.
10. The image recognition system for accelerated release reservoir positioning of a railroad train of claim 1, wherein the acquisition module comprises:
the acquisition unit is used for acquiring the actual image;
and the positioning and output unit is used for zooming the actual image into an image with the size of 1600 × 256, inputting the zoomed image into the accelerated release reservoir positioning model for positioning detection of the accelerated release reservoir, judging a threshold value of an output result, and if the length-width ratio of the position frame is greater than 0.8 and smaller than 1.2, determining that the positioned reservoir is the accelerated release reservoir, and further outputting the position information of the accelerated release reservoir.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110056516.8A CN112766260B (en) | 2021-01-15 | 2021-01-15 | Image identification method and system for positioning air reservoir for accelerating and relieving railway train |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110056516.8A CN112766260B (en) | 2021-01-15 | 2021-01-15 | Image identification method and system for positioning air reservoir for accelerating and relieving railway train |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112766260A true CN112766260A (en) | 2021-05-07 |
| CN112766260B CN112766260B (en) | 2021-09-14 |
Family
ID=75701955
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110056516.8A Active CN112766260B (en) | 2021-01-15 | 2021-01-15 | Image identification method and system for positioning air reservoir for accelerating and relieving railway train |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112766260B (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109300114A (en) * | 2018-08-30 | 2019-02-01 | 西南交通大学 | Detection method for tightness and missing of extremely small target parts of high-speed rail catenary support device |
| CN110197113A (en) * | 2019-03-28 | 2019-09-03 | 杰创智能科技股份有限公司 | A kind of method for detecting human face of high-precision anchor point matching strategy |
| CN110796643A (en) * | 2019-10-18 | 2020-02-14 | 四川大学 | Rail fastener defect detection method and system |
| CN111027260A (en) * | 2019-12-24 | 2020-04-17 | 大连圣力来监测技术有限公司 | Reciprocating compressor fault diagnosis method based on neural network |
| US20200124581A1 (en) * | 2018-10-19 | 2020-04-23 | The Climate Corporation | Detecting infection of plant diseases by classifying plant photos |
| CN111079819A (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 judging state of coupler knuckle pin of railway wagon based on image recognition and deep learning |
| CN111079627A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | An image recognition method for the fracture fault of the brake beam of a railway freight car |
| CN111080608A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Image recognition method of automatic brake valve plug door handle closing fault of railway freight car derailment |
| CN111079746A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Image recognition method of railway freight car axle box spring fault |
| CN111091551A (en) * | 2019-12-12 | 2020-05-01 | 哈尔滨市科佳通用机电股份有限公司 | A fault detection method for the loss of the split pin of the brake beam strut of a railway freight car |
| CN111126472A (en) * | 2019-12-18 | 2020-05-08 | 南京信息工程大学 | Improved target detection method based on SSD |
| CN111860623A (en) * | 2020-07-03 | 2020-10-30 | 北京林业大学 | Method and system for counting trees based on improved SSD neural network |
-
2021
- 2021-01-15 CN CN202110056516.8A patent/CN112766260B/en active Active
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109300114A (en) * | 2018-08-30 | 2019-02-01 | 西南交通大学 | Detection method for tightness and missing of extremely small target parts of high-speed rail catenary support device |
| US20200393435A1 (en) * | 2018-10-19 | 2020-12-17 | The Climate Corporation | Detecting infection of plant diseases by classifying plant photos |
| US20200124581A1 (en) * | 2018-10-19 | 2020-04-23 | The Climate Corporation | Detecting infection of plant diseases by classifying plant photos |
| CN110197113A (en) * | 2019-03-28 | 2019-09-03 | 杰创智能科技股份有限公司 | A kind of method for detecting human face of high-precision anchor point matching strategy |
| CN110796643A (en) * | 2019-10-18 | 2020-02-14 | 四川大学 | Rail fastener defect detection method and system |
| CN111091551A (en) * | 2019-12-12 | 2020-05-01 | 哈尔滨市科佳通用机电股份有限公司 | A fault detection method for the loss of the split pin of the brake beam strut of a railway freight car |
| CN111079627A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | An image recognition method for the fracture fault of the brake beam of a railway freight car |
| CN111080608A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Image recognition method of automatic brake valve plug door handle closing fault of railway freight car derailment |
| CN111079746A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Image recognition method of railway freight car axle box spring fault |
| CN111079819A (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 judging state of coupler knuckle pin of railway wagon based on image recognition and deep learning |
| CN111126472A (en) * | 2019-12-18 | 2020-05-08 | 南京信息工程大学 | Improved target detection method based on SSD |
| CN111027260A (en) * | 2019-12-24 | 2020-04-17 | 大连圣力来监测技术有限公司 | Reciprocating compressor fault diagnosis method based on neural network |
| CN111860623A (en) * | 2020-07-03 | 2020-10-30 | 北京林业大学 | Method and system for counting trees based on improved SSD neural network |
Non-Patent Citations (3)
| Title |
|---|
| WEI LIU ET AL.: "SSD: Single Shot MultiBox Detector", 《HTTPS://ARXIV.ORG/ABS/1512.02325》 * |
| 李萍 等: "基于深度学习的铁路列车关键零部件图像故障检测", 《铁道科学与工程学报》 * |
| 王士凯: "《铁路货车站修技术问答》", 31 October 1998, 中国铁道出版社 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112766260B (en) | 2021-09-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110827251B (en) | Power transmission line locking pin defect detection method based on aerial image | |
| CN107609525B (en) | Remote sensing image target detection method for constructing convolutional neural network based on pruning strategy | |
| CN109559302A (en) | Pipe video defect inspection method based on convolutional neural networks | |
| Cai et al. | Treepedia 2.0: applying deep learning for large-scale quantification of urban tree cover | |
| CN113160062B (en) | Infrared image target detection method, device, equipment and storage medium | |
| CN104700099B (en) | The method and apparatus for recognizing traffic sign | |
| CN108921875A (en) | A kind of real-time traffic flow detection and method for tracing based on data of taking photo by plane | |
| CN109147254A (en) | A kind of video outdoor fire disaster smog real-time detection method based on convolutional neural networks | |
| CN109766769A (en) | A road target detection and recognition method based on monocular vision and deep learning | |
| CN111523616B (en) | Coarse-grained soil filler grading identification method based on convolutional neural network and application system | |
| CN109766746A (en) | A method for detecting foreign objects on track in aerial video of unmanned aerial vehicle | |
| CN110766058B (en) | Battlefield target detection method based on optimized RPN (resilient packet network) | |
| CN109214308A (en) | A kind of traffic abnormity image identification method based on focal loss function | |
| CN109840904B (en) | A detection method for large-scale difference parts of high-speed rail catenary | |
| CN113988222A (en) | Forest fire detection and identification method based on fast-RCNN | |
| CN111091551A (en) | A fault detection method for the loss of the split pin of the brake beam strut of a railway freight car | |
| CN108629279A (en) | A method of the vehicle target detection based on convolutional neural networks | |
| CN111652211A (en) | A fault detection method for foreign objects hanging on the mounting seat of the anti-snake shock absorber of a motor vehicle | |
| CN113869248B (en) | Target detection method, device, equipment and storage medium | |
| CN108648210B (en) | Rapid multi-target detection method and device under static complex scene | |
| CN113392726B (en) | Method, system, terminal and medium for identifying and detecting head of person in outdoor monitoring scene | |
| CN115393802A (en) | A method for identifying uncommon intrusion targets in railway scenes based on small sample learning | |
| CN112418334A (en) | Method for identifying deformation fault of skirtboard grating of railway bullet train | |
| CN103605960B (en) | A kind of method for identifying traffic status merged based on different focal video image | |
| CN112766260B (en) | Image identification method and system for positioning air reservoir for accelerating and relieving railway train |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |