CN114821165A - Track detection image acquisition and analysis method - Google Patents
Track detection image acquisition and analysis method Download PDFInfo
- Publication number
- CN114821165A CN114821165A CN202210408403.4A CN202210408403A CN114821165A CN 114821165 A CN114821165 A CN 114821165A CN 202210408403 A CN202210408403 A CN 202210408403A CN 114821165 A CN114821165 A CN 114821165A
- Authority
- CN
- China
- Prior art keywords
- image
- track
- detection image
- track detection
- analysis result
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 214
- 238000004458 analytical method Methods 0.000 title claims abstract description 109
- 238000010191 image analysis Methods 0.000 claims abstract description 33
- 238000012545 processing Methods 0.000 claims abstract description 19
- 238000003708 edge detection Methods 0.000 claims abstract description 18
- 238000012549 training Methods 0.000 claims description 37
- 238000000034 method Methods 0.000 claims description 31
- 230000002159 abnormal effect Effects 0.000 claims description 24
- 238000012360 testing method Methods 0.000 claims description 24
- 238000000605 extraction Methods 0.000 claims description 16
- 238000001914 filtration Methods 0.000 claims description 15
- 238000007689 inspection Methods 0.000 claims description 11
- 235000004522 Pentaglottis sempervirens Nutrition 0.000 claims description 10
- 230000000694 effects Effects 0.000 claims description 10
- 240000004050 Pentaglottis sempervirens Species 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 9
- 230000003190 augmentative effect Effects 0.000 claims description 8
- 238000002372 labelling Methods 0.000 claims description 8
- 239000006002 Pepper Substances 0.000 claims description 5
- 230000009466 transformation Effects 0.000 claims description 5
- 235000002566 Capsicum Nutrition 0.000 claims description 4
- 235000016761 Piper aduncum Nutrition 0.000 claims description 4
- 235000017804 Piper guineense Nutrition 0.000 claims description 4
- 235000008184 Piper nigrum Nutrition 0.000 claims description 4
- 238000000691 measurement method Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 150000003839 salts Chemical class 0.000 claims description 4
- 238000012790 confirmation Methods 0.000 claims description 3
- 244000203593 Piper nigrum Species 0.000 claims 1
- 230000009286 beneficial effect Effects 0.000 description 10
- 241000722363 Piper Species 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 230000003416 augmentation Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 229910000831 Steel Inorganic materials 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 230000003137 locomotive effect Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
- G06T2207/20032—Median filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Multimedia (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a track detection image acquisition and analysis method, which comprises the following steps: acquiring a plurality of track detection images based on a preset acquisition standard to obtain a track detection image data set; performing basic processing on the track detection image data set to generate a track detection image data set to be analyzed; an image analysis model is constructed based on an improved YOLO-v3 network structure and a Canny edge detection algorithm, and the image analysis model is utilized to analyze the rail detection image data set to be analyzed to obtain a rail working condition analysis result. According to the invention, an image analysis model is constructed by an improved YOLO-v3 network structure and a Canny edge detection algorithm, so that the rail detection image can be comprehensively and accurately analyzed, rail working condition abnormity is found, and the acquisition and analysis quality and efficiency of the rail detection image are improved.
Description
Technical Field
The invention relates to the technical field of rail detection, in particular to a rail detection image acquisition and analysis method.
Background
With the rapid development of deep learning and other related technologies, a convolutional neural network has been widely used for object detection and classification, and particularly, in the acquisition and analysis of a track detection image, a significant advantage is shown in the aspect of image feature extraction, but the requirement on real-time performance still cannot be well met in the aspect of detection speed, and the accuracy of analysis on the small size and the large aspect ratio of targets such as track lines in the track detection image is not enough. Therefore, a method for acquiring and analyzing an image for track inspection is needed to solve the above problems.
Disclosure of Invention
The invention provides an image acquisition and analysis method for rail detection, which is characterized in that an image analysis model is constructed through an improved YOLO-v3 network structure and a Canny edge detection algorithm, rail detection images can be comprehensively and accurately analyzed, rail working condition abnormity is found, the quality and the efficiency of rail detection image acquisition and analysis are improved, and the problems in the background technology are solved.
The invention provides a track detection image acquisition and analysis method, which comprises the following steps:
s1, acquiring a plurality of track detection images based on a preset acquisition standard to obtain a track detection image data set;
s2, performing basic processing on the track detection image data set to generate a track detection image data set to be analyzed;
s3, constructing an image analysis model based on the improved YOLO-v3 network structure and the Canny edge detection algorithm, and analyzing the rail detection image data set to be analyzed by using the image analysis model to obtain a rail working condition analysis result.
Further, the preset collection criteria in S1 include:
s101, selecting a collecting device, wherein the collecting device comprises one or more of an industrial camera, a high-definition camera and an infrared camera;
s102, setting a collection weather condition, wherein the collection weather condition comprises one or more of a normal weather condition, a rain, snow and fog weather and a night environment;
s103, determining the types of the collected track detection images, wherein the types of the track detection images comprise a linear track detection image, a curve track detection image and a shielding track detection image;
further, S2 includes:
s201, adding salt and pepper noise to the image in the track detection image dataset to obtain a track detection image added with noise, adding the track detection image added with noise into the track detection image dataset to generate an augmented track detection image dataset;
s202, filtering the augmented orbit detection image data set by adopting a median filtering method to obtain a filtered orbit detection image data set;
s203, performing region division on the image in the filtering track detection image data set according to a preset division ratio to obtain an interested region image;
and S204, performing inverse perspective transformation on the interesting region image to obtain a bird 'S-eye view of the track detection image, and summarizing the bird' S-eye view of the track detection image to obtain a track detection image data set to be analyzed.
Further, S3 includes:
s301, dividing a track detection image data set to be analyzed to obtain an image test set and an image training set, and performing target labeling on images in the image training set by using a target detection labeling tool;
s302, constructing a first image analysis sub-model based on an improved YOLO-v3 detection algorithm, and analyzing and predicting the image training set by using the first image analysis sub-model to obtain a basic analysis result;
s303, analyzing and predicting the track detection image data set to be analyzed based on a Canny edge detection algorithm to obtain an accurate analysis result;
s304, performing curve fitting operation on the accurate analysis result to obtain track line data of a track detection image, and inferring parameter information of a track line according to the track line data;
s305, comparing the parameter information of the track line with the preset standard parameter information of the track line to obtain a track working condition analysis result.
Further, S302 includes:
s3021, respectively performing uniform format preprocessing on the track detection images in the image training set and the image test set to obtain a preprocessed training set track detection image and a preprocessed test set track detection image;
s3022, inputting the preprocessed training set track detection image into a first image analysis sub-model, and performing image feature extraction by using a Darknet-53 network, wherein the image feature extraction comprises the following steps:
acquiring a bottom layer output characteristic quantity to generate a first image characteristic; carrying out a layer of convolution and one-time up-sampling processing on the first image characteristic to obtain a first processed image characteristic;
stacking the convolution layer output quantity and the first processing image characteristic to generate a second image characteristic; performing a layer of convolution and one-time up-sampling processing on the second image characteristic to obtain a second processed image characteristic;
stacking the intermediate layer output characteristic quantity and the second processed image characteristic to generate a third image characteristic;
s3023, inputting the first image feature, the second image feature and the third image feature into a YOLO layer respectively for training, stopping iteration until the number of training times is reached, and generating a weight model;
s3024, inputting the preprocessed test set track detection image into the Darknet-53 network for image feature extraction, and calling a weight model to perform recognition analysis on the preprocessed test set track detection image test set image to obtain a basic analysis result.
Further, S303 includes:
s3031, carrying out grid division on the image in the track detection image data set to be analyzed according to a preset division condition, wherein the division condition is that the longitudinal height value of a grid unit is 2 times of the transverse width value;
s3032, setting a threshold range of the confidence coefficient of the prediction frame, traversing the distribution of the image gray scale by adopting a maximum inter-class variance method, and calculating to obtain a proper threshold of the confidence coefficient of the prediction frame;
s3033, comparing the confidence coefficient threshold of the prediction frame with the confidence coefficient of the prediction frame, resetting the region when the confidence coefficient of the prediction frame is greater than the confidence coefficient threshold of the prediction frame, and turning to the step S3034; when the confidence of the prediction frame is greater than or equal to the threshold of the confidence of the quarter prediction frame and smaller than the threshold of the confidence of the prediction frame, resetting the region, and turning to the step S3035; the resetting area comprises expanding the coordinate pixels of the bounding box according to a preset pixel value;
s3034, performing edge detection based on a Canny operator, binarizing the region and determining a connected domain of the track;
s3035, setting the pixel value in the expanded region to 0;
s3036, combining step S3034 and step S3035, obtaining accurate analysis results.
Further, S304 includes:
s3041, fitting the randomly extracted pixels by using cubic Bessel splines to obtain a plurality of spline curves;
s3042, scoring the fitting effect to obtain a scoring result;
s3043, according to the grading result, taking a curve with high straightness and long length as a fitting line.
Further, S305 includes:
s3051, acquiring standard parameter information of the track line based on the big data information of the track line;
s3052, formulating the normal running condition of the working condition of the track according to the standard parameter information of the track line;
s3053, matching the parameter information of the track line with the condition, and when the parameter information meets the condition, indicating that the track working condition analysis result is normal; and when the parameter information does not meet the condition, indicating that the track working condition analysis result is abnormal.
Further, after step S3, the method includes: s4, carrying out abnormity confirmation on the rail working condition analysis result: the S4 includes:
s401, acquiring a track detection image corresponding to the abnormal track working condition analysis result, and acquiring an area where the track detection image is located;
s402, selecting a first acquisition image acquired by acquiring the area in a first acquisition mode; the first acquisition mode is normal weather condition acquisition; selecting a second acquisition image acquired by acquiring the area in a second acquisition mode; the second acquisition mode is rain, snow, fog or night environment;
s403, comparing the similarity of the first collected image and the second collected image by adopting a structural similarity measurement method; when the similarity value range is larger than a preset similarity threshold, determining that the track working condition analysis result is abnormal;
s404, when the similarity value range is smaller than a preset similarity threshold value, comparing the similarity of the first collected image and the track detection original image; when the similarity value range is smaller than a preset similarity threshold value, determining that the track working condition analysis result is abnormal; and the track detection original image is a track detection image corresponding to the area where the track is located when the track work detection result is normal.
Further, the S4 includes, after: s5, setting a solution for the abnormity of the analysis result of the rail working condition, wherein the S5 comprises:
s501, importing a track detection image corresponding to the abnormal track working condition analysis result into a preset blank database, and establishing a classification item by taking a track parameter as an index;
s502, establishing a corresponding fault type item according to the classification item; establishing a corresponding solution item according to the fault type item; establishing a fault type library according to the fault type item, and establishing a solution library according to the solution item;
s503, determining a corresponding solution according to the rail working condition analysis result abnormity, and perfecting and updating a solution library in time according to the solution result of the solution.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and drawings.
The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention. In the drawings:
FIG. 1 is a schematic diagram of an image acquisition and analysis method for track inspection according to the present invention;
FIG. 2 is a schematic diagram of the acquisition criteria preset in step S1 of the method for acquiring and analyzing an image for track inspection according to the present invention;
fig. 3 is a schematic diagram of a method of the track inspection image acquisition and analysis step S2 according to the present invention.
Detailed Description
The preferred embodiments of the present invention will be described in conjunction with the accompanying drawings, and it will be understood that they are described herein for the purpose of illustration and explanation and not limitation.
The invention provides a track detection image acquisition and analysis method, as shown in fig. 1, comprising the following steps:
s1, acquiring a plurality of track detection images based on a preset acquisition standard to obtain a track detection image data set;
s2, performing basic processing on the track detection image data set to generate a track detection image data set to be analyzed;
s3, constructing an image analysis model based on the improved YOLO-v3 network structure and the Canny edge detection algorithm, and analyzing the rail detection image data set to be analyzed by using the image analysis model to obtain a rail working condition analysis result.
The working principle of the technical scheme is as follows: the YOLO algorithm can quickly classify and locate objects and directly obtain the positions and categories of the objects in an output layer. YOLO-v1 is the first algorithm of the YOLO series, which increases the size of the input image to improve the detection capability of small objects, and the algorithm process is as follows: firstly, adjusting the size of an image, then carrying out corresponding convolution network operation on the image, and finally analyzing and adjusting a threshold value on a result through a target confidence coefficient obtained by model regression. YOLO-v2 adds an anchor box on the basis of YOLO-v1 and trains the bounding box with K-means to find better box sizes automatically, which makes predicting object positions easier and accurate. However, since information on small objects is less stored in the high-level feature map, the detection result is not satisfactory when there are too many neighboring objects; the YOLO-v3 adopted in the embodiment is composed of a DarkNet-53 feature extraction network and a YOLO layer; wherein the DarkNet-53 mainly comprises a convolution layer and a residual error layer and is used for extracting image characteristics; the YOLO layer predicts the target detection by using the features obtained by DarkNet-53, performs multi-scale prediction by adopting up-sampling and feature fusion of a pyramid structure similar to the features, extracts more effective features, generates three feature extraction results with different sizes and scales, and finally predicts the number of prior frames. The Canny algorithm can find edges of the input image and identify these edges in the output image; in the embodiment, on the basis of image analysis by using a YOLO-v3 construction model, the relocation recognition analysis is carried out on the track in the track detection image based on the Canny adaptive threshold detection algorithm; the method comprises the following specific steps: acquiring a plurality of track detection images based on a preset acquisition standard to obtain a track detection image data set; performing basic processing on the track detection image data set to generate a track detection image data set to be analyzed; an image analysis model is constructed based on an improved YOLO-v3 network structure and a Canny edge detection algorithm, and the image analysis model is utilized to analyze the rail detection image data set to be analyzed to obtain a rail working condition analysis result.
The beneficial effects of the above technical scheme are: by adopting the scheme provided by the embodiment, the image analysis model is constructed by the improved YOLO-v3 network structure and the Canny edge detection algorithm, the track detection image can be comprehensively and accurately analyzed, the track working condition abnormity is found, and the track detection image acquisition and analysis quality and efficiency are improved.
In one embodiment, as shown in fig. 2, the preset collection criteria in S1 include:
s101, selecting a collecting device, wherein the collecting device comprises one or more of an industrial camera, a high-definition camera and an infrared camera;
s102, setting a collection weather condition, wherein the collection weather condition comprises one or more of a normal weather condition, a rain, snow and fog weather and a night environment;
s103, determining the types of the collected track detection images, wherein the types of the track detection images comprise a linear track detection image, a curve track detection image and a shielding track detection image;
the working principle of the technical scheme is as follows: the data set of the track detection image is enriched by setting various acquisition conditions; in this embodiment, a collecting device is selected first, where the collecting device includes one or more of an industrial camera, a high-definition camera, and an infrared camera; secondly, setting a collection weather condition, wherein the collection weather condition comprises one or more of a normal weather condition, a rain, snow and fog weather and a night environment; finally, determining the types of the collected track detection images, wherein the types of the track detection images comprise a linear track detection image, a curve track detection image and a shielding track detection image;
the beneficial effects of the above technical scheme are: by adopting the scheme provided by the embodiment, the data representativeness and diversity of the track detection image can be ensured, and the analysis effect of the track detection image is improved.
In one embodiment, as shown in fig. 3, S2 includes:
s201, adding salt and pepper noise to the image in the track detection image dataset to obtain a track detection image added with noise, adding the track detection image added with noise into the track detection image dataset to generate an augmented track detection image dataset;
s202, filtering the augmented orbit detection image data set by adopting a median filtering method to obtain a filtered orbit detection image data set;
s203, performing region division on the image in the filtering track detection image data set according to a preset division ratio to obtain an interested region image;
and S204, performing inverse perspective transformation on the interesting region image to obtain a bird 'S-eye view of the track detection image, and summarizing the bird' S-eye view of the track detection image to obtain a track detection image data set to be analyzed.
The working principle of the technical scheme is as follows: the data set augmentation is the most common method for avoiding overfitting of model training in the deep learning and training process at present, and the augmentation of an original data set is to add certain noise on the original data set so that a generated image has certain difference with an original image; salt and pepper noise is a common noise of digital images; the embodiment obtains a noise-added track detection image by adding salt-pepper noise to an image in the track detection image dataset, and adds the noise-added track detection image to the track detection image dataset to generate an augmented track detection image dataset;
median filtering is a common non-linear filtering method; the method arranges the surrounding pixels according to a specific order and replaces the original pixels with intermediate values; the relevance among image pixel points is neglected by median filtering, and when the fine texture of a detected object is complex, the non-differentiated texture of the detected object can be damaged by the result of the median filtering; in this embodiment, a median filtering method is adopted to perform filtering processing on the augmented orbit detection image dataset to obtain a filtered orbit detection image dataset;
by dividing the region of interest to be analyzed, the calculation amount of the algorithm can be reduced, and the real-time performance of the system is improved;
the image of the region of interest can be mapped to a new plane by using an inverse perspective transformation mapping method, so that the track lines are displayed in a bird's-eye view in a parallel manner, and the track lines can be accurately divided; in the embodiment, the image of the region of interest is subjected to inverse perspective transformation to obtain a bird's-eye view of the track detection image, and the bird's-eye view of the track detection image is summarized to obtain a data set of the track detection image to be analyzed.
The beneficial effects of the above technical scheme are: by adopting the scheme provided by the embodiment, the quality and the efficiency of the analysis of the track detection image can be improved by carrying out basic processing on the image in the track detection image data set.
In one embodiment, S3 includes:
s301, dividing a track detection image data set to be analyzed to obtain an image test set and an image training set, and performing target labeling on images in the image training set by using a target detection labeling tool;
s302, constructing a first image analysis sub-model based on an improved YOLO-v3 detection algorithm, and analyzing and predicting the image training set by using the first image analysis sub-model to obtain a basic analysis result;
s303, analyzing and predicting the track detection image data set to be analyzed based on a Canny edge detection algorithm to obtain an accurate analysis result;
s304, performing curve fitting operation on the accurate analysis result to obtain track line data of a track detection image, and inferring parameter information of a track line according to the track line data;
s305, comparing the parameter information of the track line with the preset standard parameter information of the track line to obtain a track working condition analysis result.
The working principle of the technical scheme is as follows: the image data set used in training needs to be marked, and the effect of model identification can be improved by using the complete marking frame and containing the track line target; then, analyzing the images through the two image analysis submodels to obtain an analysis result of the working condition of the track;
the analysis model constructed by the YOLO-v3 detection algorithm has a good analysis effect on the track detection image in a simple scene, and in order to improve the generalization capability of the image analysis model, the image analysis model can have a good analysis effect on the track detection image under complex conditions (different weather conditions and night environments), in the embodiment, on the basis of construction by the YOLO-v3 detection algorithm, a transfer learning method is adopted, the detected track is repositioned by the Canny-based adaptive threshold detection algorithm, the original unmarked image is used for retraining, and the identification and analysis effect of the image analysis model under the complex conditions can be realized; the embodiment comprises the following steps:
s301, dividing a track detection image data set to be analyzed to obtain an image test set and an image training set, and performing target labeling on images in the image training set by using a target detection labeling tool;
s302, constructing a first image analysis sub-model based on an improved YOLO-v3 detection algorithm, and analyzing and predicting the image training set by using the first image analysis sub-model to obtain a basic analysis result;
s303, analyzing and predicting the track detection image data set to be analyzed based on a Canny edge detection algorithm to obtain an accurate analysis result;
s304, performing curve fitting operation on the accurate analysis result to obtain track line data of a track detection image, and inferring parameter information of a track line according to the track line data;
s305, comparing the parameter information of the track line with the preset standard parameter information of the track line to obtain a track working condition analysis result.
The beneficial effects of the above technical scheme are: by adopting the scheme provided by the embodiment, the track detection image is analyzed by adopting the two sub-models, so that the analysis accuracy can be improved, and the track working condition can be effectively judged.
In one embodiment, S302 includes:
s3021, respectively performing uniform format preprocessing on the track detection images in the image training set and the image test set to obtain a preprocessed training set track detection image and a preprocessed test set track detection image;
s3022, inputting the preprocessed training set track detection image into a first image analysis sub-model, and performing image feature extraction by using a Darknet-53 network, wherein the image feature extraction comprises the following steps:
acquiring a bottom layer output characteristic quantity to generate a first image characteristic; carrying out a layer of convolution and one-time up-sampling processing on the first image characteristic to obtain a first processed image characteristic;
stacking the convolution layer output quantity and the first processing image characteristic to generate a second image characteristic; performing a layer of convolution and one-time up-sampling processing on the second image characteristic to obtain a second processed image characteristic;
stacking the intermediate layer output characteristic quantity and the second processed image characteristic to generate a third image characteristic;
s3023, inputting the first image feature, the second image feature and the third image feature into a YOLO layer respectively for training, stopping iteration until the number of training times is reached, and generating a weight model;
s3024, inputting the preprocessed test set track detection image into the Darknet-53 network for image feature extraction, and calling a weight model to perform recognition analysis on the preprocessed test set track detection image test set image to obtain a basic analysis result.
The working principle of the technical scheme is as follows: the DarkNet-53 is mainly composed of a series of convolution layers and residual layers, and can be used for extracting image features. The DarkNet-53 characteristic extraction roughly comprises the following operation steps: inputting an image, performing special convolution block convolution operation, then outputting three feature layers with different shapes through a series of residual error operations, putting the feature layers into a YOLO layer for decoding operation, and obtaining a basic analysis result through characteristic analysis; the embodiment comprises the following steps:
s3021, respectively performing uniform format preprocessing on the track detection images in the image training set and the image test set to obtain a preprocessed training set track detection image and a preprocessed test set track detection image;
s3022, inputting the preprocessed training set track detection image into a first image analysis sub-model, and performing image feature extraction by using a Darknet-53 network, wherein the image feature extraction comprises the following steps:
acquiring a bottom layer output characteristic quantity to generate a first image characteristic; carrying out a layer of convolution and one-time up-sampling processing on the first image characteristic to obtain a first processed image characteristic;
stacking the convolution layer output quantity and the first processed image characteristic to generate a second image characteristic; performing a layer of convolution and one-time up-sampling processing on the second image characteristic to obtain a second processed image characteristic;
stacking the intermediate layer output characteristic quantity and the second processed image characteristic to generate a third image characteristic;
s3023, inputting the first image feature, the second image feature and the third image feature into a YOLO layer respectively for training, stopping iteration after the number of training times is reached, and generating a weight model;
s3024, inputting the preprocessed test set track detection image into the Darknet-53 network for image feature extraction, and calling a weight model to perform recognition analysis on the preprocessed test set track detection image test set image to obtain a basic analysis result.
The beneficial effects of the above technical scheme are: by adopting the scheme provided by the embodiment, the image analysis sub-model is adopted for analysis, so that the effect of track detection image analysis can be improved.
In one embodiment, S303 comprises:
s3031, carrying out grid division on the image in the track detection image data set to be analyzed according to a preset division condition, wherein the division condition is that the longitudinal height value of a grid unit is 2 times of the transverse width value;
s3032, setting a threshold range of the confidence coefficient of the prediction frame, traversing the distribution of the image gray scale by adopting a maximum inter-class variance method, and calculating to obtain a proper threshold of the confidence coefficient of the prediction frame;
s3033, comparing the confidence coefficient threshold of the prediction frame with the confidence coefficient of the prediction frame, resetting the region when the confidence coefficient of the prediction frame is greater than the confidence coefficient threshold of the prediction frame, and turning to the step S3034; when the confidence of the prediction frame is greater than or equal to the threshold of the confidence of the quarter prediction frame and smaller than the threshold of the confidence of the prediction frame, resetting the region, and turning to the step S3035; the resetting area comprises expanding the coordinate pixels of the bounding box according to a preset pixel value;
s3034, performing edge detection based on a Canny operator, binarizing the region and determining a connected domain of the track;
s3035, setting the pixel value in the expanded region to 0;
s3036, combining step S3034 and step S3035, obtaining accurate analysis results.
The working principle of the technical scheme is as follows: in the detection using YOLO-v3, the detector processes the image into SxS grids, and the candidate frames are distributed on the x and y axes with the same density; when the center of an object needing to be identified enters any grid, the grid is only responsible for predicting the target; b boundary frames can be predicted by any grid, and the frames not only feed back the position of the frames, but also predict the value of the confidence coefficient; the bounding box information contains five data values, x, y, w, h, c; wherein (x, y) represents the coordinates of the center position of the bounding box of the object obtained by current grid prediction, w and h represent the height and the corresponding width of the bounding box respectively, and c represents the confidence; however, in the bird's eye view of the orbit detection images, the orbits are sparsely distributed on the x-axis and continuously distributed on the y-axis, and the grid density of SxS cannot be used to detect small-sized and large-aspect-ratio targets such as orbit lines; in order to alleviate the influence of the aspect ratio on the object detection, the present embodiment divides the image into Sx2S grids, and improves the detector by increasing the longitudinal detection density so as to be more suitable for the track detection.
In the process of analyzing the track detection image, firstly predicting a selected candidate region meeting the conditions; and then, selecting a prediction result, obtaining prediction frames with high confidence, deleting frames with relatively low scores by assigning a threshold value because a prediction value with low confidence possibly is a non-orbital line after obtaining the confidence of each prediction frame, then processing the rest boundary frames by non-maximum inhibition to obtain a plurality of groups of high-score boundary frames, and finally obtaining the position parameters.
During model training, to speed up the training, the confidence value thresholds are adjusted, i.e., the area ranges of the track line bounding boxes are reset, and then the Canny-based adaptive edge detection algorithm is used to quickly reposition the track lines. The embodiment specifically includes:
s3031, carrying out grid division on the image in the track detection image data set to be analyzed according to a preset division condition, wherein the division condition is that the longitudinal height value of a grid unit is 2 times of the transverse width value;
s3032, setting a threshold range of the confidence coefficient of the prediction frame, traversing the distribution of the image gray scale by adopting a maximum inter-class variance method, and calculating to obtain a proper threshold of the confidence coefficient of the prediction frame;
s3033, comparing the confidence coefficient threshold of the prediction frame with the confidence coefficient of the prediction frame, resetting the region when the confidence coefficient of the prediction frame is greater than the confidence coefficient threshold of the prediction frame, and turning to the step S3034; when the confidence of the prediction frame is greater than or equal to the threshold of the confidence of the quarter prediction frame and smaller than the threshold of the confidence of the prediction frame, resetting the region, and turning to the step S3035; the resetting area comprises expanding the coordinate pixels of the bounding box according to a preset pixel value;
s3034, performing edge detection based on a Canny operator, binarizing the region and determining a connected domain of the track;
s3035, setting the pixel value in the expanded region to 0;
s3036, combining step S3034 and step S3035, obtaining accurate analysis results.
The beneficial effects of the above technical scheme are: by adopting the scheme provided by the embodiment, the edge detection analysis is carried out based on the improved Canny operator, so that the accuracy of the analysis of the track detection image can be improved, and the quality of the analysis is improved.
In one embodiment, S304 includes:
s3041, fitting the randomly extracted pixels by using cubic Bessel splines to obtain a plurality of spline curves;
s3042, scoring the fitting effect to obtain a scoring result;
s3043, according to the grading result, taking a curve with high straightness and long length as a fitting line.
The working principle of the technical scheme is as follows: in order to better understand and predict the environmental information around the rail when the steel mill locomotive is running, a curve fitting operation is performed on the detected rail line result. On the basis of a Bezier curve model, the method adjusts candidate points of a curve according to an actual curve to complete track fitting; the method comprises the following steps:
s3041, fitting the randomly extracted pixels by using cubic Bessel splines to obtain a plurality of spline curves;
s3042, scoring the fitting effect to obtain a scoring result;
s3043, according to the grading result, taking a curve with high straightness and long length as a fitting line.
Curve fitting is followed by establishing curve scoring criteria to evaluate the extent of curve fitting. In this embodiment, the longer the length is, the more nearly straight the shape is, the curve is used as the scoring standard, and the calculation formula is:
in the above formula, S represents the scoring result, p represents the pixel sum of the splines, and 1 and beta 2 A factor representing the regularization, α being the length of the spline; γ is the height of the image; delta 1 And delta 2 Representing the angle between lines of spline control points; and after grading, screening out a curve with high linearity and long length as a fitted line according to the grading numerical value.
The beneficial effects of the above technical scheme are: by adopting the scheme provided by the embodiment, the parameter information of the track line in the track detection image can be better generated by fitting the curve, and the analysis quality of the track detection image is improved.
In one embodiment, S305 includes:
s3051, obtaining standard parameter information of the track line based on big data information of the track line;
s3052, formulating the normal running condition of the working condition of the track according to the standard parameter information of the track line;
s3053, matching the parameter information of the track line with the condition, and when the parameter information meets the condition, indicating that the track working condition analysis result is normal; and when the parameter information does not meet the condition, indicating that the track working condition analysis result is abnormal.
The working principle of the technical scheme is as follows: by means of big data information, abnormal parameter information in the track detection image can be accurately compared, so that an abnormal state can be found; the embodiment comprises the following steps:
s3051, acquiring standard parameter information of the track line based on the big data information of the track line;
s3052, formulating the normal running condition of the working condition of the track according to the standard parameter information of the track line;
s3053, matching the parameter information of the track line with the condition, and when the parameter information meets the condition, indicating that the track working condition analysis result is normal; and when the parameter information does not meet the condition, indicating that the track working condition analysis result is abnormal.
The beneficial effects of the above technical scheme are: by adopting the scheme provided by the embodiment, the basis can be provided for the final analysis of the track detection image by means of the big data information, and the track working condition abnormal state can be conveniently and accurately found.
In one embodiment, after step S3, comprising: s4, carrying out abnormity confirmation on the rail working condition analysis result: the S4 includes:
s401, acquiring a track detection image corresponding to the abnormal track working condition analysis result, and acquiring an area where the track detection image is located;
s402, selecting a first acquisition image acquired by acquiring the area in a first acquisition mode; the first acquisition mode is normal weather condition acquisition; selecting a second acquisition image acquired by acquiring the area by adopting a second acquisition mode; the second acquisition mode is rain, snow, fog or night environment;
s403, comparing the similarity of the first collected image and the second collected image by adopting a structural similarity measurement method; when the similarity value range is larger than a preset similarity threshold, determining that the track working condition analysis result is abnormal;
s404, when the similarity value range is smaller than a preset similarity threshold value, comparing the similarity of the first collected image and the track detection original image; when the similarity value range is smaller than a preset similarity threshold, determining that the track working condition analysis result is abnormal; and the track detection original image is a track detection image corresponding to the area where the track is located when the track work detection result is normal.
The working principle of the technical scheme is as follows: after the track operating mode analysis result appears unusually, in order to differentiate the abnormal conditions better, avoid the emergence of erroneous judgement, need confirm track operating mode analysis result is unusual, and this embodiment includes:
s401, acquiring a track detection image corresponding to the abnormal track working condition analysis result, and acquiring an area where the track detection image is located;
s402, selecting a first acquisition image acquired by acquiring the area in a first acquisition mode; the first acquisition mode is normal weather condition acquisition; selecting a second acquisition image acquired by acquiring the area by adopting a second acquisition mode; the second acquisition mode is rain, snow, fog or night environment;
s403, comparing the similarity of the first collected image and the second collected image by adopting a structural similarity measurement method; when the similarity value range is larger than a preset similarity threshold, determining that the track working condition analysis result is abnormal;
s404, when the similarity value range is smaller than a preset similarity threshold value, comparing the similarity of the first collected image and the track detection original image; when the similarity value range is smaller than a preset similarity threshold value, determining that the track working condition analysis result is abnormal; and the track detection original image is a track detection image corresponding to the area where the track is located when the track work detection result is normal.
The beneficial effects of the above technical scheme are: by adopting the scheme provided by the embodiment, the track detection image analysis accuracy can be further ensured by confirming the abnormality of the track working condition analysis result.
In one embodiment, said S4 is followed by: s5, setting a solution for the abnormity of the analysis result of the rail working condition, wherein the S5 comprises:
s501, importing a track detection image corresponding to the abnormal track working condition analysis result into a preset blank database, and establishing a classification item by taking a track parameter as an index;
s502, establishing a corresponding fault type item according to the classification item; establishing a corresponding solution item according to the fault type item; establishing a fault type library according to the fault type item, and establishing a solution library according to the solution item;
s503, determining a corresponding solution according to the rail working condition analysis result abnormity, and perfecting and updating a solution library in time according to the solution result of the solution.
The working principle of the technical scheme is as follows: according to the abnormity of the rail working condition analysis result, in order to ensure that the fault type and the solution are found in time, the association relationship between the rail working condition analysis result and the fault type and the solution is necessary to be established, and the method is realized by establishing a subentry database; the embodiment comprises the following steps:
s501, importing a track detection image corresponding to the abnormal track working condition analysis result into a preset blank database, and establishing a classification item by taking a track parameter as an index;
s502, establishing a corresponding fault type item according to the classification item; establishing a corresponding solution item according to the fault type item; establishing a fault type library according to the fault type item, and establishing a solution library according to the solution item;
s503, determining a corresponding solution according to the rail working condition analysis result abnormity, and perfecting and updating a solution library in time according to the solution result of the solution.
The beneficial effects of the above technical scheme are: by adopting the scheme provided by the embodiment, the fault type can be timely distinguished and the solution can be found by establishing the incidence relation between the track working condition analysis result and the fault type and the solution, and the efficiency of track working condition maintenance is improved.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.
Claims (10)
1. A rail detection image acquisition and analysis method is characterized by comprising the following steps:
s1, acquiring a plurality of track detection images based on a preset acquisition standard to obtain a track detection image data set;
s2, performing basic processing on the track detection image data set to generate a track detection image data set to be analyzed;
s3, constructing an image analysis model based on the improved YOLO-v3 network structure and the Canny edge detection algorithm, and analyzing the rail detection image data set to be analyzed by using the image analysis model to obtain a rail working condition analysis result.
2. The method for acquiring and analyzing track inspection images as claimed in claim 1, wherein the acquisition criteria preset in S1 include:
s101, selecting a collecting device, wherein the collecting device comprises one or more of an industrial camera, a high-definition camera and an infrared camera;
s102, setting a collection weather condition, wherein the collection weather condition comprises one or more of a normal weather condition, a rain, snow and fog weather and a night environment;
s103, determining the type of the acquired track detection image, wherein the type of the track detection image comprises a linear track detection image, a curve track detection image and a shielding track detection image.
3. The orbit detection image acquisition and analysis method of claim 1, wherein S2 comprises:
s201, adding salt and pepper noise to the image in the track detection image dataset to obtain a track detection image added with noise, adding the track detection image added with noise into the track detection image dataset to generate an augmented track detection image dataset;
s202, filtering the augmented orbit detection image data set by adopting a median filtering method to obtain a filtered orbit detection image data set;
s203, carrying out region division on the image in the filtering track detection image data set according to a preset division ratio to obtain an interested region image;
and S204, performing inverse perspective transformation on the interesting region image to obtain a bird 'S-eye view of the track detection image, and summarizing the bird' S-eye view of the track detection image to obtain a track detection image data set to be analyzed.
4. The orbit detection image acquisition and analysis method of claim 1, wherein S3 comprises:
s301, dividing a track detection image data set to be analyzed to obtain an image test set and an image training set, and performing target labeling on images in the image training set by using a target detection labeling tool;
s302, constructing a first image analysis sub-model based on an improved YOLO-v3 detection algorithm, and analyzing and predicting the image training set by using the first image analysis sub-model to obtain a basic analysis result;
s303, analyzing and predicting the track detection image data set to be analyzed based on a Canny edge detection algorithm to obtain an accurate analysis result;
s304, performing curve fitting operation on the accurate analysis result to obtain track line data of a track detection image, and inferring parameter information of a track line according to the track line data;
s305, comparing the parameter information of the track line with the preset standard parameter information of the track line to obtain a track working condition analysis result.
5. The track inspection image collection and analysis method of claim 4, wherein S302 comprises:
s3021, respectively performing uniform format preprocessing on the track detection images in the image training set and the image test set to obtain a preprocessed training set track detection image and a preprocessed test set track detection image;
s3022, inputting the preprocessed training set track detection image into a first image analysis sub-model, and performing image feature extraction by using a Darknet-53 network, wherein the image feature extraction comprises the following steps:
acquiring a bottom layer output characteristic quantity to generate a first image characteristic; carrying out a layer of convolution and one-time up-sampling processing on the first image characteristic to obtain a first processed image characteristic;
stacking the convolution layer output quantity and the first processing image characteristic to generate a second image characteristic; performing a layer of convolution and one-time up-sampling processing on the second image characteristic to obtain a second processed image characteristic;
stacking the intermediate layer output characteristic quantity and the second processed image characteristic to generate a third image characteristic;
s3023, inputting the first image feature, the second image feature and the third image feature into a YOLO layer respectively for training, stopping iteration until the number of training times is reached, and generating a weight model;
s3024, inputting the preprocessed test set track detection image into the Darknet-53 network for image feature extraction, and calling a weight model to perform recognition analysis on the preprocessed test set track detection image test set image to obtain a basic analysis result.
6. The method for acquiring and analyzing track inspection images as claimed in claim 4, wherein S303 comprises:
s3031, carrying out grid division on the image in the track detection image data set to be analyzed according to a preset division condition; the dividing condition is that the longitudinal height value of the grid unit cells is 2 times of the transverse width value;
s3032, setting a threshold range of the confidence coefficient of the prediction frame, traversing the distribution of the image gray scale by adopting a maximum inter-class variance method, and calculating to obtain a proper threshold of the confidence coefficient of the prediction frame;
s3033, comparing the confidence coefficient threshold of the prediction frame with the confidence coefficient of the prediction frame, resetting the region when the confidence coefficient of the prediction frame is greater than the confidence coefficient threshold of the prediction frame, and turning to the step S3034; when the confidence of the prediction frame is greater than or equal to the threshold of the confidence of the quarter prediction frame and smaller than the threshold of the confidence of the prediction frame, resetting the region, and turning to the step S3035; the resetting area comprises expanding the coordinate pixels of the bounding box according to a preset pixel value;
s3034, performing edge detection based on a Canny operator, binarizing the region and determining a connected domain of the track;
s3035, setting the pixel value in the expanded region to 0;
s3036, combining step S3034 and step S3035, obtaining accurate analysis results.
7. The method for acquiring and analyzing track inspection images as claimed in claim 4, wherein S304 comprises:
s3041, fitting the randomly extracted pixels by using cubic Bessel splines to obtain a plurality of spline curves;
s3042, scoring the fitting effect to obtain a scoring result;
s3043, according to the grading result, taking a curve with high straightness and long length as a fitting line.
8. The track inspection image collection and analysis method according to claim 4, wherein S305 comprises:
s3051, acquiring standard parameter information of the track line based on the big data information of the track line;
s3052, formulating the normal running condition of the working condition of the track according to the standard parameter information of the track line;
s3053, matching the parameter information of the track line with the condition, and when the parameter information meets the condition, indicating that the track working condition analysis result is normal; and when the parameter information does not meet the condition, indicating that the track working condition analysis result is abnormal.
9. The method for acquiring and analyzing track inspection images as claimed in claim 1, wherein after step S3, the method comprises: s4, carrying out abnormity confirmation on the rail working condition analysis result: the S4 includes:
s401, acquiring a track detection image corresponding to the abnormal track working condition analysis result, and acquiring an area where the track detection image is located;
s402, selecting a first acquisition image acquired by acquiring the area in a first acquisition mode; the first acquisition mode is normal weather condition acquisition; selecting a second acquisition image acquired by acquiring the area in a second acquisition mode; the second acquisition mode is rain, snow and fog weather or night environment;
s403, comparing the similarity of the first collected image and the second collected image by adopting a structural similarity measurement method; when the similarity value range is larger than a preset similarity threshold, determining that the track working condition analysis result is abnormal;
s404, when the similarity value range is smaller than a preset similarity threshold value, comparing the similarity of the first collected image and the track detection original image; when the similarity value range is smaller than a preset similarity threshold value, determining that the track working condition analysis result is abnormal; and the track detection original image is a track detection image corresponding to the area where the track is located when the track work detection result is normal.
10. The method for acquiring and analyzing track inspection images as claimed in claim 9, wherein the step S4 is followed by the steps of: s5, setting a solution for the abnormity of the analysis result of the rail working condition, wherein the S5 comprises:
s501, importing a track detection image corresponding to the abnormal track working condition analysis result into a preset blank database, and establishing a classification item by taking a track parameter as an index;
s502, establishing a corresponding fault type item according to the classification item; establishing a corresponding solution item according to the fault type item; establishing a fault type library according to the fault type item, and establishing a solution library according to the solution item;
s503, determining a corresponding solution according to the rail working condition analysis result abnormity, and perfecting and updating a solution library in time according to the solution result of the solution.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210408403.4A CN114821165A (en) | 2022-04-19 | 2022-04-19 | Track detection image acquisition and analysis method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210408403.4A CN114821165A (en) | 2022-04-19 | 2022-04-19 | Track detection image acquisition and analysis method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114821165A true CN114821165A (en) | 2022-07-29 |
Family
ID=82506259
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210408403.4A Pending CN114821165A (en) | 2022-04-19 | 2022-04-19 | Track detection image acquisition and analysis method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114821165A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116664966A (en) * | 2023-03-27 | 2023-08-29 | 北京鹰之眼智能健康科技有限公司 | Infrared image processing system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108491758A (en) * | 2018-02-08 | 2018-09-04 | 深圳市睿灵创新科技开发有限公司 | A kind of track detection method and robot |
| CN109376605A (en) * | 2018-09-26 | 2019-02-22 | 福州大学 | A kind of electric inspection process image bird-resistant fault detection method |
| CN110532889A (en) * | 2019-08-02 | 2019-12-03 | 南京理工大学 | Track foreign matter detecting method based on rotor unmanned aircraft and YOLOv3 |
| CN113591973A (en) * | 2021-07-29 | 2021-11-02 | 中国铁路上海局集团有限公司科学技术研究所 | Intelligent comparison method for appearance state changes of track slabs |
-
2022
- 2022-04-19 CN CN202210408403.4A patent/CN114821165A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108491758A (en) * | 2018-02-08 | 2018-09-04 | 深圳市睿灵创新科技开发有限公司 | A kind of track detection method and robot |
| CN109376605A (en) * | 2018-09-26 | 2019-02-22 | 福州大学 | A kind of electric inspection process image bird-resistant fault detection method |
| CN110532889A (en) * | 2019-08-02 | 2019-12-03 | 南京理工大学 | Track foreign matter detecting method based on rotor unmanned aircraft and YOLOv3 |
| CN113591973A (en) * | 2021-07-29 | 2021-11-02 | 中国铁路上海局集团有限公司科学技术研究所 | Intelligent comparison method for appearance state changes of track slabs |
Non-Patent Citations (2)
| Title |
|---|
| 何静等: "基于Canny-YOLOv3的列车轮对踏面损伤检测", 《电子测量与仪器学报》 * |
| 孙科蒙: "一种基于深度学习的机车轨道检测方法", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116664966A (en) * | 2023-03-27 | 2023-08-29 | 北京鹰之眼智能健康科技有限公司 | Infrared image processing system |
| CN116664966B (en) * | 2023-03-27 | 2024-02-20 | 北京鹰之眼智能健康科技有限公司 | Infrared image processing system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108596166B (en) | Container number identification method based on convolutional neural network classification | |
| CN109767422B (en) | Pipeline detection and identification method based on deep learning, storage medium and robot | |
| Tan et al. | Automatic detection of sewer defects based on improved you only look once algorithm | |
| CN113378686B (en) | Two-stage remote sensing target detection method based on target center point estimation | |
| CN111598098B (en) | Water gauge water line detection and effectiveness identification method based on full convolution neural network | |
| CN110910350B (en) | Nut loosening detection method for wind power tower cylinder | |
| CN110992349A (en) | Underground pipeline abnormity automatic positioning and identification method based on deep learning | |
| CN112070135A (en) | Power equipment image detection method and device, power equipment and storage medium | |
| CN115239644B (en) | Concrete defect identification method, device, computer equipment and storage medium | |
| CN114926407A (en) | Steel surface defect detection system based on deep learning | |
| CN110751619A (en) | Insulator defect detection method | |
| CN113962960A (en) | Pavement disease detection method based on deep learning | |
| CN114723709A (en) | Tunnel disease detection method and device and electronic equipment | |
| CN111046950A (en) | Image processing method and device, storage medium and electronic device | |
| CN115830004A (en) | Surface defect detection method, device, computer equipment and storage medium | |
| CN111814852A (en) | Image detection method, image detection device, electronic equipment and computer-readable storage medium | |
| CN112801227A (en) | Typhoon identification model generation method, device, equipment and storage medium | |
| CN115439654A (en) | Method and system for finely dividing weakly supervised farmland plots under dynamic constraint | |
| CN116168246A (en) | Method, device, equipment and medium for identifying waste slag field for railway engineering | |
| Feng et al. | Improved SOLOv2 detection method for shield tunnel lining water leakages | |
| CN114821165A (en) | Track detection image acquisition and analysis method | |
| CN113205511B (en) | Electronic component batch information detection method and system based on deep neural network | |
| CN113158954B (en) | Automatic detection method for zebra crossing region based on AI technology in traffic offsite | |
| CN113592839A (en) | Distribution network line typical defect diagnosis method and system based on improved fast RCNN | |
| CN112686872A (en) | Wood counting method based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220729 |