CN114067245A - Method and system for identifying hidden danger of external environment of railway - Google Patents
Method and system for identifying hidden danger of external environment of railway Download PDFInfo
- Publication number
- CN114067245A CN114067245A CN202111357343.XA CN202111357343A CN114067245A CN 114067245 A CN114067245 A CN 114067245A CN 202111357343 A CN202111357343 A CN 202111357343A CN 114067245 A CN114067245 A CN 114067245A
- Authority
- CN
- China
- Prior art keywords
- hidden danger
- information
- railway
- external environment
- research object
- 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 47
- 238000011160 research Methods 0.000 claims abstract description 82
- 238000012545 processing Methods 0.000 claims description 23
- 230000008859 change Effects 0.000 claims description 20
- 238000007781 pre-processing Methods 0.000 claims description 13
- 238000012544 monitoring process Methods 0.000 claims description 11
- 238000006243 chemical reaction Methods 0.000 claims description 10
- 238000013135 deep learning Methods 0.000 claims description 8
- 238000011156 evaluation Methods 0.000 claims description 6
- 238000004458 analytical method Methods 0.000 claims description 5
- 238000013523 data management Methods 0.000 claims description 3
- 238000013500 data storage Methods 0.000 claims description 3
- 238000009826 distribution Methods 0.000 claims description 3
- 238000001514 detection method Methods 0.000 description 33
- 238000012549 training Methods 0.000 description 32
- 238000012360 testing method Methods 0.000 description 23
- 238000005516 engineering process Methods 0.000 description 9
- 230000008569 process Effects 0.000 description 9
- 230000000694 effects Effects 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 230000004927 fusion Effects 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000002372 labelling Methods 0.000 description 3
- 238000005065 mining Methods 0.000 description 3
- 230000000737 periodic effect Effects 0.000 description 3
- 241000251468 Actinopterygii Species 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000002360 preparation method Methods 0.000 description 2
- 238000012800 visualization Methods 0.000 description 2
- 238000006424 Flood reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 231100001267 hazard identification Toxicity 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 239000002689 soil Substances 0.000 description 1
- 230000033772 system development Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a method and a system for identifying hidden danger of external environment of a railway, wherein the method comprises the steps of obtaining high-resolution remote sensing image data and unmanned aerial vehicle aerial image data of the external environment of a target railway line region at the current stage, inputting the high-resolution remote sensing image data into a first image identification model to obtain a first research object identification result, and inputting the unmanned aerial vehicle aerial image data into a second image identification model to obtain a second research object identification result; and determining hidden danger information in the target railway line area at the current stage according to the first research object identification result and the second research object identification result. The method can determine the hidden danger information in the railway line area with high efficiency and high precision.
Description
Technical Field
The invention relates to the technical field of railway environment and image processing, in particular to a method and a system for identifying hidden dangers of a railway external environment.
Background
Railway traffic is typically linear traffic, which has the characteristics of long and narrow running route, traversing multiple regions, complex terrain and the like, and is often interfered by external environment. For example, in the periphery of a railway, the illegal occupation of the railway land, the illegal building, the operational mining, the soil taking, the sand digging, the ditch digging, the pipeline digging and burying, the fishpond and the natural disasters around the railway cause great difficulties in the safety early warning and the safety supervision of the running route.
For the problems existing around the railway, the problems of low checking efficiency, high cost, strong data subjectivity, low supervision precision and the like exist by taking manual on-site checking as a main means and taking video monitoring as an auxiliary tool.
Disclosure of Invention
The invention aims to provide a method and a system for identifying hidden dangers of an external environment of a railway, so as to achieve the purpose of identifying the hidden dangers of the external environment of the railway with high efficiency and high precision.
In order to achieve the purpose, the invention provides the following scheme:
a method for identifying hidden danger of external environment of a railway comprises the following steps:
acquiring high-resolution earth observation data of the external environment of the target railway along the area at the current stage; the high-resolution earth observation data comprises high-resolution remote sensing image data and unmanned aerial vehicle aerial image data;
inputting the high-resolution remote sensing image data of the external environment of the target railway line area at the current stage into a first image recognition model to obtain a first research object recognition result; the first subject identification result comprises a plurality of subjects and category information of each of the subjects, geographical location information of each of the subjects; wherein the research object is determined according to the ground feature characteristics of natural disasters in the area along the railway; the research object is a railway line or a hidden danger object;
inputting the unmanned aerial vehicle aerial image data of the external environment of the target railway line area at the current stage into a second image recognition model to obtain a second research object recognition result; the second subject identification result comprises a plurality of subjects and category information of each of the subjects, geographical location information of each of the subjects;
determining hidden danger information in a target railway line area at the current stage according to the first research object identification result and the second research object identification result; the hidden danger information is hidden danger-free object information or hidden danger object information; the information of the hidden danger objects comprises one or more hidden danger objects, category information of each hidden danger object and position information of each hidden danger object.
Optionally, the method further includes: and determining the change information of each hidden danger object according to the hidden danger information in the target railway edge region at the previous stage and the hidden danger information in the target railway edge region at the current stage.
Optionally, the inputting the high-resolution remote sensing image data of the external environment of the target railway line area at the current stage into the first image recognition model to obtain a first research object recognition result specifically includes:
preprocessing the high-resolution remote sensing image data of the external environment of the target railway line area at the current stage; the preprocessing comprises image enhancement processing and image format conversion processing;
and inputting the preprocessed high-resolution remote sensing image data of the external environment of the target railway line area at the current stage into the first image recognition model to obtain a first research object recognition result.
Optionally, the inputting the unmanned aerial vehicle aerial image data of the external environment of the target railway line area at the current stage into a second image recognition model to obtain a second research object recognition result specifically includes:
preprocessing the unmanned aerial vehicle aerial image data of the external environment of the target railway line area at the current stage; the preprocessing comprises image enhancement processing and image format conversion processing;
and inputting the preprocessed unmanned aerial vehicle aerial image data of the external environment of the target railway line area at the current stage into a second image recognition model to obtain a second research object recognition result.
Optionally, the first image recognition model is determined according to a first data set and a deep learning algorithm; the first data set comprises a plurality of preprocessed high-resolution remote sensing images and label information of each preprocessed high-resolution remote sensing image; the tag information comprises category information of each research object and geographical location information of each research object;
the second image recognition model is determined according to a second data set and a deep learning algorithm; the second data set comprises a plurality of preprocessed unmanned aerial vehicle aerial images and label information of each preprocessed unmanned aerial vehicle aerial image.
Optionally, the determining, according to the first research object identification result and the second research object identification result, hidden danger information in the target railway edge area at the current stage specifically includes:
merging the first research object identification result and the second research object identification result to obtain an external environment identification result of the target railway line area at the current stage;
and determining hidden danger information in the target railway line area at the current stage according to the category information of the research objects in the external environment identification result of the target railway line area at the current stage.
A railway external environment hazard identification system, comprising:
the data acquisition module is used for acquiring high-resolution earth observation data of the external environment of the target railway line area at the current stage; the high-resolution earth observation data comprises high-resolution remote sensing image data and unmanned aerial vehicle aerial image data;
the first research object identification result determining module is used for inputting the high-resolution remote sensing image data of the external environment of the target railway line area at the current stage into a first image identification model to obtain a first research object identification result; the first subject identification result comprises a plurality of subjects and category information of each of the subjects, geographical location information of each of the subjects; wherein the research object is determined according to the ground feature characteristics of natural disasters in the area along the railway; the research object is a railway line or a hidden danger object;
the second research object identification result determining module is used for inputting the unmanned aerial vehicle aerial image data of the external environment of the target railway line area at the current stage into a second image identification model to obtain a second research object identification result; the second subject identification result comprises a plurality of subjects and category information of each of the subjects, geographical location information of each of the subjects;
the hidden danger information determining module is used for determining hidden danger information in a target railway line area at the current stage according to the first research object identification result and the second research object identification result; the hidden danger information is hidden danger-free object information or hidden danger object information; the information of the hidden danger objects comprises one or more hidden danger objects, category information of each hidden danger object and position information of each hidden danger object.
Optionally, the method further includes: and the change information determining module is used for determining the change information of each hidden danger object according to the hidden danger information in the target railway edge area at the previous stage and the hidden danger information in the target railway edge area at the current stage.
Optionally, the method further includes: and the image data management module is used for acquiring the high-resolution earth observation data output by the data acquisition module, and performing image enhancement processing, image format conversion processing, image data storage and image data distribution on the high-resolution earth observation data.
Optionally, the method further includes: the railway environment hidden danger disaster early warning analysis module is used for carrying out railway environment hidden danger grade evaluation, disaster risk evaluation and thematic map output according to the change information of each hidden danger object output by the change information determination module; the thematic map comprises a landslide monitoring thematic subject, a debris flow monitoring thematic subject, a road collapse monitoring thematic subject and a violation building thematic subject; the thematic map is used for assisting in generating a corresponding emergency plan.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention provides a method and a system for identifying hidden danger of a railway external environment.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a schematic flow chart of a method for identifying hidden danger in external environment of a railway according to the present invention;
FIG. 2 is a flow chart of the construction of an image recognition model according to the present invention;
FIG. 3 is a diagram of image annotation recognition in accordance with the present invention;
FIG. 4 is a graph of the image recognition result of the present invention;
FIG. 5 is a graph of the prediction accuracy visualization result of the present invention;
fig. 6 is a schematic structural diagram of the railway external environment hidden danger identification system of the present invention.
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.
The invention provides a method and a system for identifying hidden danger of a railway external environment based on a high-resolution remote sensing image unmanned aerial vehicle periodic aerial survey technology, which take high-resolution earth observation data of different time phases as a data source and carry out preprocessing such as color enhancement, brightness enhancement, contrast enhancement and the like on image pixels; then, according to the feature of the natural disasters along the railway, detection, identification, classification, positioning and quantitative information extraction are carried out on the researched objects (such as flood disasters, illegal buildings, trees and the like), and the multi-temporal change detection technology of the researched objects is utilized to monitor the change conditions of scenes and areas which are mainly concerned by the natural disasters in the construction process of railway projects, accurately calibrate the change areas and the change types, and further develop the high-resolution remote sensing image natural disaster detection and early warning system along the railway.
The invention aims to solve the technical problems that:
(1) researching a railway disaster image classification technology based on spatial information;
(2) researching an image recognition technology of a railway external environment research object;
(3) researching early warning technologies of different time dimensions;
(4) the system development is performed based on a WebGIS (Web geographic information System) technology.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Example one
As shown in fig. 1, the method for identifying hidden dangers in the external environment of a railway based on the high-resolution remote sensing image unmanned aerial vehicle periodic aerial survey technology provided by the embodiment specifically includes the following steps.
Step 101: acquiring high-resolution earth observation data of the external environment of the target railway along the area at the current stage; the high-resolution earth observation data comprise high-resolution remote sensing image data and unmanned aerial vehicle aerial image data.
Step 102: inputting the high-resolution remote sensing image data of the external environment of the target railway line area at the current stage into a first image recognition model to obtain a first research object recognition result; the first subject identification result comprises a plurality of subjects and category information of each of the subjects, geographical location information of each of the subjects; wherein the research object is determined according to the ground feature characteristics of natural disasters in the area along the railway; the research object is a railway line or a hidden danger object; the method specifically comprises the following steps:
preprocessing the high-resolution remote sensing image data of the external environment of the target railway line area at the current stage; the preprocessing includes image enhancement processing and image format conversion processing.
And inputting the preprocessed high-resolution remote sensing image data of the external environment of the target railway line area at the current stage into the first image recognition model to obtain a first research object recognition result.
Step 103: inputting the unmanned aerial vehicle aerial image data of the external environment of the target railway line area at the current stage into a second image recognition model to obtain a second research object recognition result; the second subject identification result comprises a plurality of subjects and category information of each of the subjects, geographical location information of each of the subjects; the method specifically comprises the following steps:
preprocessing the unmanned aerial vehicle aerial image data of the external environment of the target railway line area at the current stage; the preprocessing includes image enhancement processing and image format conversion processing.
And inputting the preprocessed unmanned aerial vehicle aerial image data of the external environment of the target railway line area at the current stage into a second image recognition model to obtain a second research object recognition result.
Step 104: determining hidden danger information in a target railway line area at the current stage according to the first research object identification result and the second research object identification result; the hidden danger information is hidden danger-free object information or hidden danger object information; the information of the hidden danger objects comprises one or more hidden danger objects, the category information of each hidden danger object and the position information of each hidden danger object; the method specifically comprises the following steps:
and performing union processing on the first research object identification result and the second research object identification result to obtain an external environment identification result of the target railway line area at the current stage.
And determining hidden danger information in the target railway line area at the current stage according to the category information of the research objects in the external environment identification result of the target railway line area at the current stage.
Further, the method for identifying hidden danger in external environment of a railway provided by this embodiment further includes: and determining the change information of each hidden danger object according to the hidden danger information in the target railway edge region at the previous stage and the hidden danger information in the target railway edge region at the current stage.
The first image recognition model is determined according to a first data set and a deep learning algorithm; the first data set comprises a plurality of preprocessed high-resolution remote sensing images and label information of each preprocessed high-resolution remote sensing image; the tag information comprises category information of each research object and geographical location information of each research object;
the second image recognition model is determined according to a second data set and a deep learning algorithm; the second data set comprises a plurality of preprocessed unmanned aerial vehicle aerial images and label information of each preprocessed unmanned aerial vehicle aerial image.
Wherein each data set comprises a training set and a test set. The preparation process of the training set and the test set comprises the following steps:
the first step, data acquisition and management.
(1) High resolution remote sensing image data
The high-resolution remote sensing image data is provided by the Lanzhou office group of China, achieves the resolution effect of 0.6 m along the railway, is large in data volume, is stored in a server disk, is issued by a server and then called at any time, and can be updated and maintained at the later stage.
(2) Unmanned aerial vehicle aerial image data
The unmanned aerial vehicle aerial image data is provided by the Lanzhou office group, Inc. of China railway, and the Lanzhou office station section is provided with an unmanned aerial vehicle for remote sensing aerial photography of key lines, so that a large amount of unmanned aerial vehicle aerial image data is collected. The oblique photography data after aerial photography is converted into an OSGB format and then stored in a server disk, and the oblique photography data is issued by a server and then called at any time, and can be updated and maintained at a later stage.
And secondly, processing data.
(1) Image enhancement processing
In the process of acquiring the aerial image data and the high-resolution remote sensing image data of the unmanned aerial vehicle, the real radiation value of a ground object is influenced by factors such as a sensor, the sun direction, the landform and the atmospheric condition, so that before a research object (such as a building, a line and the like) in the image data is identified and extracted, the image data needs to be enhanced firstly, operations such as color enhancement, brightness enhancement, contrast enhancement and the like on image pixels are included, the processed image data is obtained, and the processed image data is still stored in the png format.
(2) Producing a data set
Because the data set required by the model training test is a json file in a coco format, the processed image data cannot be directly used for the model training and testing, and the format conversion of the processed image data is required. In this embodiment, a labelmme labeling tool is used to label each png format picture, after the labeling is completed, each picture correspondingly generates a labelmejson file, and then LabelMe2coco. The converted coco json files are used for model training and testing, and 80% of the json files are generally used as a test set, and the rest 20% are used as the test set.
Since the first image recognition model and the second image recognition model are constructed in substantially the same process, for the sake of clarity and brevity, only the first image recognition model will be described herein. It should be noted that the difference between the first image recognition model and the second image recognition model is that the training set and the test set are different, and the other sets are the same, that is, when the training set of the first image recognition model is the unmanned aerial vehicle aerial image data, the training set of the second image recognition model is the high-resolution remote sensing image data, and vice versa.
The construction process of the first image recognition model comprises the following steps: ,
(1) object classification
According to the unmanned aerial vehicle image data of taking photo by plane of gathering, can divide into the research object of railway external environment: railway lines, highway lines, colored-hillock tile houses, residential houses, illegal buildings, greenhouses, railway office buildings, tall buildings, farmlands, ponds (including fish ponds), mountainous regions, mining areas and the like. When the LabelMe marking tool is used for marking, each research object pair has a unique ID, namely the railway line 0; a high-speed line 1; the color post tile house 2; a residential house 3; a violation building 4; a greenhouse 5; a railway office building 6; a tall building 7; farmland 8, pond (including fish pond) 9, mountain land 10 and mining area 11.
(2) Model training
As shown in fig. 2, the training process of the first image recognition model includes the following 4 steps:
1) a training sample (training set) and a test sample (test set) are imported.
2) ResNet101 (deep residual neural network) was trained with the coco2017 dataset to obtain a pre-trained model. Since the coco2017 data set covers 9000 types including houses, lands, ponds, railway lines and the like to be detected in the embodiment, the pre-training model can also be used for target detection in the embodiment, but the pre-training model is slow in detection speed on target detection, relatively low in accuracy and inaccurate in positioning accuracy, and needs to retrain the pre-training model (namely the pre-training deep residual error neural network ResNet101) by means of the training set.
3) The method comprises the following specific processes of training a pre-training model through multiple iterations:
the training samples are input into a pre-training deep residual error neural network ResNet101 (comprising five convolution layers), one corresponding feature map is output by the ResNet101 after each convolution, and five feature maps are obtained after the convolution is completed. In general networks, a feature map output by the last convolutional layer is directly used, although the feature map has strong semantics, the position and the resolution are low, and small objects cannot be easily detected. Therefore, the embodiment further uses the FPN image feature pyramid mode to perform image feature enhancement on the obtained five feature maps, that is, the feature map of the upper layer is used to obtain the same length and width as those of the feature map of the next layer through upsampling and then added, so as to obtain a well-fused feature fusion map, and the feature fusion map not only has the high resolution and position information of the shallow layer, but also has the rich semantic information of the deep layer.
Then, 3 region of interest (ROI) candidates are set for each pixel point in each feature fusion map, and the size of the ROI candidates is generally determined by scale (area size of frame) and ratio (aspect ratio of frame). In general, the scale of feature map of each layer is kept constant, and only the ratio of each layer is changed. In the present model, the scale of each layer is (32, 64,128, 256, 512) and the ratio is (0.5, 1, 2), so that each pixel in each layer generates 3 candidate ROIs, and the pixel generates 15 candidate ROIs with different sizes.
These candidate ROIs are then fed into the RPN network (region-recommended network, based on convolutional neural network structure, to help the network recommend regions of interest). The specific implementation process comprises the following steps: firstly, generating a full-connection feature with the dimension of 512 dimensions on the convolution feature fusion graph of conv5-3 by using a 3-by-3 sliding window; then, using the generated 512-dimensional features, 2 fully-connected layer branches are generated by convolution with 1 × 11 × 1.
1) Regression layer (reg layer) with the effect of: and predicting coordinates x, y and width w, h of the candidate region ROI corresponding to the center anchor point of the candidate region (the coordinates are not absolute coordinates of the candidate ROI, but regression of deviation from actual coordinates is carried out, and meanwhile, in the regression process, a part of candidate ROI with larger deviation from the actual coordinates is filtered out).
2) A sorting layer (cls layer) for: it is determined whether the candidate region belongs to the foreground or the background, and the confidence level of the region (or the probability that the candidate region belongs to a certain category) is generated correspondingly.
The judgment is based on the following:
wherein: i denotes the number of boxes in batch;
pia probability value representing that the ith box belongs to the foreground;
when p isiWhen > 0.7, pi *1, the ith frame is taken as a target frame, and the regression loss of the ith frame is calculated;
when p isi<At 0.3, pi *When the frame is 0, the ith frame is taken as a background frame, and the regression loss of the frame is not calculated;
tiis the corresponding value after the frame scaling is predicted, ti *Is the corresponding value after the marker box prediction;
Nclstotal number of samples, NregThe size and the gamma weight coefficient of the characteristic diagram ensure that the classification loss and the regression loss are equal in weight;
and then ROIAlign operation is carried out on the remaining candidate ROIs, namely bilinear interpolation is adopted. For example, in building, rail road target detection, a 7 × 7 pooling operation is performed for each candidate ROI (candidate ROI has a mapping region on the feature map). Firstly, a mapping region is divided into 7 × 7 regions, each point in the region is supplemented into a 2 × 2 region by a bilinear interpolation method, and then maxporoling is performed after interpolation to obtain a final 7 × 7 ROI, i.e., roilign operation is completed.
These ROIs can be classified (12-class classification), BB regression, and Mask generation (i.e., contour generation of the detection object) by the fully connected convolutional neural network FCN.
4) And (4) repeating the step 3), training all samples to obtain a first image detection model which is finally optimized and adjusted, wherein the classification loss and the regression loss of the model reach the minimum, and the detection accuracy is the highest.
(3) Model testing
1) And testing the first image detection model by using the test sample to obtain a detection result of a new sample.
2) In order to ensure the recognition accuracy, a confidence threshold scale is set to be 0.5, when a test sample starts to call a first image detection model, the confidence of some recognition objects is lower than the threshold, and the test result is not satisfactory, the model training stage needs to be returned, the model and the training parameters are readjusted, and the model training is performed again; if the confidence of each recognition object recognition result is higher than the threshold, the test result meets the requirement, and the obtained first image detection model is also optimal (namely the classification loss and the regression loss are the same and the minimum, and the recognition and positioning accuracy is the highest), so that the first image recognition model is obtained. In addition, the detection sample identification result contains the category information, the position information and the confidence level of the research objects along the railway.
3) And processing high-resolution image data shot by the unmanned aerial vehicle at the past time and the present time for the same area, inputting the processed high-resolution image data into a final detection model, and comparing the high-resolution image data with the final detection model according to the type and position information output by the present image and the past image to obtain the change condition of the present area relative to the past area.
Taking the identification of houses and railway lines as an example, the identification process is as follows:
(1) data set production
The method comprises the steps of taking images shot by an unmanned aerial vehicle and images shot by a remote sensing satellite regularly as data sources, carrying out region segmentation on the images in the data sources, and dividing the images into a plurality of small image regions. Then, the segmented image is enhanced, and finally, the enhanced image is subjected to image annotation, as shown in fig. 3, buildings along the railway are divided into 7 types, which are respectively: 0, railway line; a high-speed line 1; the color post tile house 2; a residential house 3; a violation building 4; a greenhouse 5; a railway office building 6; a tall building 7. After image annotation was complete, the json data file for the respective LabelMe will be produced. And converting the json data file of the LabelMe into the json data file of the coco by using a self-programming program, and automatically dividing the data into a training set and a test set to finish the required data preparation in the early stage.
(2) Results and analysis of the experiments
1) Model training phase
Because the target detection model identification objects built above are 12 types in total and comprise buildings, railway lines and the like, the built target detection model can be used as a pre-training model, and in order to identify the buildings and the railway lines more accurately, the experiment trains the pre-training model by using the manufactured training set to obtain an optimized new target detection model (the model has small classification loss and regression loss and high identification and classification accuracy).
2) Stage of model testing
Inputting the test sample into the optimized new target detection model, and if the test result does not meet the requirement, indicating that the optimized new target detection model is not the optimal detection model, and returning to readjust the model and the training parameters; when the test result meets the requirement, the optimized new target detection model is the optimal detection model (the classification loss and the regression loss of the model are minimum, and the accuracy of identification and classification is highest), and the result of detection and identification is output and comprises position information, category information and confidence coefficient.
3) Test set identification results
All experiments are carried out on the same server by adopting GPU programming. The experimental parameters were set as follows: the learning rate is 0.01, the number of iterations is 2000, num _ works is 2, and the confidence threshold scale is 0.5. The test set identification results are shown in fig. 4.
The result of the prediction accuracy visualization is shown in fig. 5, and it can be seen from the graph that the recognition accuracy exceeds 90%.
The method combines the detection of the external environment of the railway with the deep learning, selects the images shot by the unmanned aerial vehicle and the remote sensing satellite as data sources on the basis of a Mask-RCNN target detection method, combines a ResNet101 network with a Mask-RCNN algorithm under a deep learning framework Detectron2 to carry out a multithreading iterative training model, and finally obtains an optimized railway and line surrounding environment detection model. The method effectively avoids the defects of traditional target detection, and does not need to manually set the characteristic value of the target, thereby improving the learning capability of the model. The reliability of the algorithm and the framework adopted by the research is proved through experiments, the experimental result achieves the effect of an expected model, the detection precision and the integral detection precision of the small buildings in the dense stacking area of the railway and the houses along the railway are improved, not only can the positioning frame for detecting the railway and the house targets along the railway be obtained, but also the binary Mask of the contour of the railway and the house along the railway can be obtained, the possibility is provided for further improving and obtaining the contour edge of the railway and the house along the railway, and compared with the traditional building detection method of aerial images, the method is quicker, more intelligent and more automatic.
Example two
On the basis of the above method, the embodiment provides a railway external environment hidden danger identification system of a high-resolution remote sensing image unmanned aerial vehicle periodic aerial survey technology, as shown in fig. 6, including:
the data acquisition module 601 is used for acquiring high-resolution earth observation data of the external environment of the target railway line area at the current stage; the high-resolution earth observation data comprise high-resolution remote sensing image data and unmanned aerial vehicle aerial image data.
A first research object identification result determining module 602, configured to input the high-resolution remote sensing image data of the external environment of the target railway area along the current stage into a first image identification model to obtain a first research object identification result; the first subject identification result comprises a plurality of subjects and category information of each of the subjects, geographical location information of each of the subjects; wherein the research object is determined according to the ground feature characteristics of natural disasters in the area along the railway; the research object is a railway line or a hidden danger object.
A second study object identification result determining module 603, configured to input the unmanned aerial vehicle aerial image data of the external environment of the target railway area along the current stage into a second image identification model to obtain a second study object identification result; the second subject identification result includes a plurality of subjects and category information for each of the subjects, geographic location information for each of the subjects.
A hidden danger information determining module 604, configured to determine hidden danger information in a target railway line area at the current stage according to the first research object identification result and the second research object identification result; the hidden danger information is hidden danger-free object information or hidden danger object information; the information of the hidden danger objects comprises one or more hidden danger objects, category information of each hidden danger object and position information of each hidden danger object.
Further, the system for identifying hidden danger in external environment of railway according to this embodiment further includes:
and the change information determining module is used for determining the change information of each hidden danger object according to the hidden danger information in the target railway edge area at the previous stage and the hidden danger information in the target railway edge area at the current stage.
The railway environment hidden danger disaster early warning analysis module is used for carrying out railway environment hidden danger grade evaluation, disaster risk evaluation and thematic map output according to the change information of each hidden danger object output by the change information determination module; the thematic map comprises a landslide monitoring thematic subject, a debris flow monitoring thematic subject, a road collapse monitoring thematic subject and a violation building thematic subject; the thematic map is used for assisting in generating a corresponding emergency plan, and traffic safety guarantee and emergency response capability are improved.
Further, the system for identifying hidden danger in external environment of railway according to this embodiment further includes: and the image data management module is used for acquiring the high-resolution earth observation data output by the data acquisition module, and performing image enhancement processing, image format conversion processing, image data storage and image data distribution on the high-resolution earth observation data. The module adopts a distributed object-relation database to manage unmanned aerial vehicle photographing data and high-resolution remote sensing image data.
Further, the system for identifying hidden danger in external environment of railway according to this embodiment further includes: an electronic map service module; the module adopts a basic electronic map which mainly comprises layers of a water system, a border and administrative region, a residential area, traffic, contour lines, a note and the like, integrates and stacks a railway road network map into the basic electronic map, and provides operation functions of zooming, roaming, ranging, labeling, navigation and the like.
The railway external environment hidden danger monitoring and early warning method and system based on the high-resolution remote sensing image unmanned aerial vehicle regular aerial survey technology have the following advantages:
(1) the railway line and the surrounding situation in the research area can be accurately positioned.
(2) The classification precision of disaster research objects such as railway lines, surrounding building buildings, landslides, debris flows, floods and the like is improved, and the accuracy rate of identifying the types of ground objects along the railway line reaches over 90 percent.
(3) Can combine three-dimensional geographic information system according to relevant data such as unmanned aerial vehicle aerial photograph and oblique photography and remote sensing image, can realize railway external environment hidden danger early warning analysis by the show that becomes more meticulous of key district.
(4) Compared with the traditional natural disaster auxiliary supervision means, the method provided by the research has high disaster checking efficiency and low cost, and greatly reduces the labor cost.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (10)
1. A method for identifying hidden danger of external environment of a railway is characterized by comprising the following steps:
acquiring high-resolution earth observation data of the external environment of the target railway along the area at the current stage; the high-resolution earth observation data comprises high-resolution remote sensing image data and unmanned aerial vehicle aerial image data;
inputting the high-resolution remote sensing image data of the external environment of the target railway line area at the current stage into a first image recognition model to obtain a first research object recognition result; the first subject identification result comprises a plurality of subjects and category information of each of the subjects, geographical location information of each of the subjects; wherein the research object is determined according to the ground feature characteristics of natural disasters in the area along the railway; the research object is a railway line or a hidden danger object;
inputting the unmanned aerial vehicle aerial image data of the external environment of the target railway line area at the current stage into a second image recognition model to obtain a second research object recognition result; the second subject identification result comprises a plurality of subjects and category information of each of the subjects, geographical location information of each of the subjects;
determining hidden danger information in a target railway line area at the current stage according to the first research object identification result and the second research object identification result; the hidden danger information is hidden danger-free object information or hidden danger object information; the information of the hidden danger objects comprises one or more hidden danger objects, category information of each hidden danger object and position information of each hidden danger object.
2. The method for identifying the hidden danger of the external environment of the railway according to claim 1, further comprising: and determining the change information of each hidden danger object according to the hidden danger information in the target railway edge region at the previous stage and the hidden danger information in the target railway edge region at the current stage.
3. The method for identifying the hidden danger in the external environment of the railway according to claim 1, wherein the step of inputting the high-resolution remote sensing image data of the external environment of the area along the target railway at the current stage into a first image identification model to obtain an identification result of a first research object specifically comprises the steps of:
preprocessing the high-resolution remote sensing image data of the external environment of the target railway line area at the current stage; the preprocessing comprises image enhancement processing and image format conversion processing;
and inputting the preprocessed high-resolution remote sensing image data of the external environment of the target railway line area at the current stage into the first image recognition model to obtain a first research object recognition result.
4. The method for identifying the hidden danger in the external environment of the railway according to claim 1, wherein the step of inputting the unmanned aerial vehicle aerial image data of the external environment of the area along the target railway at the current stage into a second image identification model to obtain a second research object identification result specifically comprises:
preprocessing the unmanned aerial vehicle aerial image data of the external environment of the target railway line area at the current stage; the preprocessing comprises image enhancement processing and image format conversion processing;
and inputting the preprocessed unmanned aerial vehicle aerial image data of the external environment of the target railway line area at the current stage into a second image recognition model to obtain a second research object recognition result.
5. The method for identifying the hidden danger of the external environment of the railway is characterized in that the first image identification model is determined according to a first data set and a deep learning algorithm; the first data set comprises a plurality of preprocessed high-resolution remote sensing images and label information of each preprocessed high-resolution remote sensing image; the tag information comprises category information of each research object and geographical location information of each research object;
the second image recognition model is determined according to a second data set and a deep learning algorithm; the second data set comprises a plurality of preprocessed unmanned aerial vehicle aerial images and label information of each preprocessed unmanned aerial vehicle aerial image.
6. The method for identifying hidden danger in external environment of railway according to claim 1, wherein the determining hidden danger information in the target railway edge area at the current stage according to the first research object identification result and the second research object identification result specifically comprises:
merging the first research object identification result and the second research object identification result to obtain an external environment identification result of the target railway line area at the current stage;
and determining hidden danger information in the target railway line area at the current stage according to the category information of the research objects in the external environment identification result of the target railway line area at the current stage.
7. A railway external environment hidden danger identification system is characterized by comprising:
the data acquisition module is used for acquiring high-resolution earth observation data of the external environment of the target railway line area at the current stage; the high-resolution earth observation data comprises high-resolution remote sensing image data and unmanned aerial vehicle aerial image data;
the first research object identification result determining module is used for inputting the high-resolution remote sensing image data of the external environment of the target railway line area at the current stage into a first image identification model to obtain a first research object identification result; the first subject identification result comprises a plurality of subjects and category information of each of the subjects, geographical location information of each of the subjects; wherein the research object is determined according to the ground feature characteristics of natural disasters in the area along the railway; the research object is a railway line or a hidden danger object;
the second research object identification result determining module is used for inputting the unmanned aerial vehicle aerial image data of the external environment of the target railway line area at the current stage into a second image identification model to obtain a second research object identification result; the second subject identification result comprises a plurality of subjects and category information of each of the subjects, geographical location information of each of the subjects;
the hidden danger information determining module is used for determining hidden danger information in a target railway line area at the current stage according to the first research object identification result and the second research object identification result; the hidden danger information is hidden danger-free object information or hidden danger object information; the information of the hidden danger objects comprises one or more hidden danger objects, category information of each hidden danger object and position information of each hidden danger object.
8. The system for identifying the hidden danger of the external environment of the railway according to claim 7, further comprising: and the change information determining module is used for determining the change information of each hidden danger object according to the hidden danger information in the target railway edge area at the previous stage and the hidden danger information in the target railway edge area at the current stage.
9. The system for identifying the hidden danger of the external environment of the railway according to claim 7, further comprising: and the image data management module is used for acquiring the high-resolution earth observation data output by the data acquisition module, and performing image enhancement processing, image format conversion processing, image data storage and image data distribution on the high-resolution earth observation data.
10. The system for identifying the hidden danger of the external environment of the railway according to claim 8, further comprising: the railway environment hidden danger disaster early warning analysis module is used for carrying out railway environment hidden danger grade evaluation, disaster risk evaluation and thematic map output according to the change information of each hidden danger object output by the change information determination module; the thematic map comprises a landslide monitoring thematic subject, a debris flow monitoring thematic subject, a road collapse monitoring thematic subject and a violation building thematic subject; the thematic map is used for assisting in generating a corresponding emergency plan.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111357343.XA CN114067245A (en) | 2021-11-16 | 2021-11-16 | Method and system for identifying hidden danger of external environment of railway |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111357343.XA CN114067245A (en) | 2021-11-16 | 2021-11-16 | Method and system for identifying hidden danger of external environment of railway |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114067245A true CN114067245A (en) | 2022-02-18 |
Family
ID=80273232
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111357343.XA Pending CN114067245A (en) | 2021-11-16 | 2021-11-16 | Method and system for identifying hidden danger of external environment of railway |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114067245A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114781823A (en) * | 2022-04-01 | 2022-07-22 | 中国科学院地理科学与资源研究所 | Urban ecological red line early warning method and device based on human activity feedback |
| CN116663912A (en) * | 2023-07-31 | 2023-08-29 | 中国铁路设计集团有限公司 | Railway external environment space safety risk assessment method |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109977921A (en) * | 2019-04-11 | 2019-07-05 | 广东电网有限责任公司 | A kind of transmission line of electricity perils detecting method |
| CN112419226A (en) * | 2020-10-12 | 2021-02-26 | 北京恒达时讯科技股份有限公司 | Railway external environment hidden danger monitoring system and method |
| WO2021051278A1 (en) * | 2019-09-17 | 2021-03-25 | 深圳市大疆创新科技有限公司 | Earth surface feature identification method and device, unmanned aerial vehicle, and computer readable storage medium |
| CN113033386A (en) * | 2021-03-23 | 2021-06-25 | 广东电网有限责任公司广州供电局 | High-resolution remote sensing image-based transmission line channel hidden danger identification method and system |
| CN113077424A (en) * | 2021-03-23 | 2021-07-06 | 广东电网有限责任公司广州供电局 | Power transmission line channel environment change detection method and system based on deep learning |
| WO2021184891A1 (en) * | 2020-03-20 | 2021-09-23 | 中国科学院深圳先进技术研究院 | Remotely-sensed image-based terrain classification method, and system |
-
2021
- 2021-11-16 CN CN202111357343.XA patent/CN114067245A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109977921A (en) * | 2019-04-11 | 2019-07-05 | 广东电网有限责任公司 | A kind of transmission line of electricity perils detecting method |
| WO2021051278A1 (en) * | 2019-09-17 | 2021-03-25 | 深圳市大疆创新科技有限公司 | Earth surface feature identification method and device, unmanned aerial vehicle, and computer readable storage medium |
| WO2021184891A1 (en) * | 2020-03-20 | 2021-09-23 | 中国科学院深圳先进技术研究院 | Remotely-sensed image-based terrain classification method, and system |
| CN112419226A (en) * | 2020-10-12 | 2021-02-26 | 北京恒达时讯科技股份有限公司 | Railway external environment hidden danger monitoring system and method |
| CN113033386A (en) * | 2021-03-23 | 2021-06-25 | 广东电网有限责任公司广州供电局 | High-resolution remote sensing image-based transmission line channel hidden danger identification method and system |
| CN113077424A (en) * | 2021-03-23 | 2021-07-06 | 广东电网有限责任公司广州供电局 | Power transmission line channel environment change detection method and system based on deep learning |
Non-Patent Citations (3)
| Title |
|---|
| 慎利;方灿明;王继成;戴延帅;: "场景解译框架下的高铁沿线建筑物自动识别", 遥感信息, no. 03, 14 May 2018 (2018-05-14), pages 33 - 34 * |
| 李爱农;南希;张正健;边金虎;: "特大山地灾害遥感应急响应调查方法与案例", 中国减灾, no. 19, 1 October 2018 (2018-10-01), pages 44 - 45 * |
| 樊泽明: "机器人学基础", 30 September 2021, 机械工业出版社, pages: 137 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114781823A (en) * | 2022-04-01 | 2022-07-22 | 中国科学院地理科学与资源研究所 | Urban ecological red line early warning method and device based on human activity feedback |
| CN116663912A (en) * | 2023-07-31 | 2023-08-29 | 中国铁路设计集团有限公司 | Railway external environment space safety risk assessment method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| AU2018360836A1 (en) | Methods and apparatus for automatically defining computer-aided design files using machine learning, image analytics, and/or computer vision | |
| Hurskainen et al. | Change detection of informal settlements using multi-temporal aerial photographs–the case of Voi, SE-Kenya | |
| KR100982237B1 (en) | Developing analysis methodology of temporal river changes using spatial imagery information and a gis and system of the same | |
| Dai et al. | Integrated edge detection and terrain analysis for agricultural terrace delineation from remote sensing images | |
| CN112712535A (en) | Mask-RCNN landslide segmentation method based on simulation difficult sample | |
| CN114067245A (en) | Method and system for identifying hidden danger of external environment of railway | |
| CN116539004A (en) | Communication line engineering investigation design method and system adopting unmanned aerial vehicle mapping | |
| CN113033390B (en) | Dam remote sensing intelligent detection method based on deep learning | |
| CN113823088B (en) | Urban road ponding depth prediction and early warning method based on visual recognition technology | |
| CN115471980B (en) | Mud-rock flow disaster early warning method | |
| CN113591617B (en) | Deep learning-based water surface small target detection and classification method | |
| CN115761513A (en) | Intelligent remote sensing identification method for mountain large landslide based on semi-supervised deep learning | |
| Dinis et al. | Hierarchical object-based classification of dense urban areas by integrating high spatial resolution satellite images and lidar elevation data | |
| Youn et al. | Extracting urban road networks from high-resolution true orthoimage and lidar | |
| Liu et al. | Architecture planning and geo-disasters assessment mapping of landslide by using airborne LiDAR data and UAV images | |
| CN116486282A (en) | Digital elevation model manufacturing method and system based on deep learning, electronic equipment and storage medium | |
| Zeybek et al. | Investigation of landslide-based road surface deformation in mountainous areas with single period UAV data | |
| CN113591668B (en) | Wide area unknown dam automatic detection method using deep learning and space analysis | |
| CN115638772A (en) | Method for extracting landslide disaster information in high mountain canyon region based on unmanned aerial vehicle image data | |
| Forghani et al. | Extracting terrain categories from multi-source satellite imagery | |
| Medhi et al. | Rural Road Extraction using Object Based Image Analysis (OBIA): A case study from Assam, India | |
| Sun et al. | Check dam extraction from remote sensing images using deep learning and geospatial analysis: A case study in the Yanhe River Basin of the Loess Plateau, China | |
| Forghani et al. | Object-based classification of multi-sensor optical imagery to generate terrain surface roughness information for input to wind risk simulation | |
| Meng | Remote Sensing of Urban Green Space | |
| CN114972991B (en) | Automatic recognition method and system for collapsing sentry |
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 |