CN114119577B - High-speed rail tunnel leakage cable buckle detection method - Google Patents
High-speed rail tunnel leakage cable buckle detection method Download PDFInfo
- Publication number
- CN114119577B CN114119577B CN202111449917.6A CN202111449917A CN114119577B CN 114119577 B CN114119577 B CN 114119577B CN 202111449917 A CN202111449917 A CN 202111449917A CN 114119577 B CN114119577 B CN 114119577B
- Authority
- CN
- China
- Prior art keywords
- gradient
- pixel
- sampling
- gradient direction
- buckle
- 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.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 24
- 238000005070 sampling Methods 0.000 claims abstract description 56
- 238000000034 method Methods 0.000 claims abstract description 39
- 238000010586 diagram Methods 0.000 claims abstract description 23
- 238000000605 extraction Methods 0.000 claims abstract description 11
- 238000004364 calculation method Methods 0.000 claims description 15
- 238000003491 array Methods 0.000 claims description 4
- 238000005192 partition Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 238000002474 experimental method Methods 0.000 description 11
- 230000008569 process Effects 0.000 description 8
- 230000000694 effects Effects 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 230000004927 fusion Effects 0.000 description 6
- 238000003384 imaging method Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 3
- 238000005065 mining Methods 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 101000822805 Naja atra Cytotoxin A5 Proteins 0.000 description 2
- 101000822803 Naja kaouthia Cytotoxin homolog Proteins 0.000 description 2
- 101000783567 Naja naja Cytotoxin 1 Proteins 0.000 description 2
- 101000822819 Naja naja Cytotoxin-like basic protein Proteins 0.000 description 2
- 230000000903 blocking effect Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000011835 investigation Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 241001270131 Agaricus moelleri Species 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000013024 troubleshooting Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
-
- 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
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Algebra (AREA)
- Computational Mathematics (AREA)
- Quality & Reliability (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a high-speed railway tunnel cable leakage buckle detection method, and relates to the technical field of image detection methods. The detection method comprises the following steps: improving a CS-LBP algorithm and calculating gradient direction characteristics of a high-speed railway tunnel cable leakage buckle image sampling circle domain one by one to obtain a complete primary gradient direction characteristic diagram; performing two continuous downsampling on the feature map and extracting continuous gradient features of the two sampled images respectively; and (3) connecting the two groups of gradient features in series to serve as descriptors, and completing the classification work of the fault buckle images by using the SVM. The method can improve the feature extraction efficiency, reduce the latitude of the description sub-features and more accurately reflect the outline features of the leaky cable buckle.
Description
Technical Field
The invention relates to the technical field of image processing methods, in particular to a high-speed railway tunnel leakage cable buckle detection method.
Background
At present, railway transportation is still a main transportation mode for solving population transfer and material transportation in China, and particularly, along with continuous increase of carrying capacity, it is important to ensure normal operation of a railway communication system. The communication network of mountain tunnel segments is mainly realized by covering leaky cables, which are usually suspended from the tunnel walls and fixed with special snaps. The air pressure and the ability wave that produce when the high-speed train passes through the tunnel can cause the influence to the buckle, and moist geological environment also can accelerate the buckle to appear becoming flexible even break away from the phenomenon simultaneously. The main detection means of the leakage cable buckle is manual fixed point periodical investigation one by one, and the detection means is severely limited by environmental factors and human factors and has great potential safety hazard. With the increasing mileage of railway tunnels in recent years, the conventional investigation method has hardly satisfied practical needs, and the realization of automatic detection of the leaky cable buckle becomes a necessary trend of development. The leakage cable buckle detection technology combined with computer vision mainly comprises the following two links: firstly, a high-speed camera is required to be arranged at a window of a train, and a full-section picture shot in the running process of a tunnel is collected along with the train; and processing the leaky cable buckle data by using a computer, extracting the characteristics frame by frame, and finishing the checking work of the buckle. The current data acquisition part is realized, but the detection part still stays in the stage of manually playing back the video for troubleshooting. The depth mining algorithm and the multi-feature fusion algorithm based on the local binary pattern in the prior art also have the problems of weak descriptor pertinence and over-high feature dimension.
Disclosure of Invention
The technical problem to be solved by the invention is how to provide a high-speed railway tunnel leaky cable buckle detection method capable of improving characteristic extraction efficiency, reducing descriptive sub-characteristic latitude and more accurately reflecting profile characteristics of a leaky cable buckle.
In order to solve the technical problems, the invention adopts the following technical scheme: the method for detecting the high-speed railway tunnel leakage cable buckle is characterized by comprising the following steps of:
improving a CS-LBP algorithm and calculating gradient direction characteristics of a high-speed railway tunnel cable leakage buckle image sampling circle domain one by one to obtain a complete primary gradient direction characteristic diagram;
performing two continuous downsampling on the feature map and extracting continuous gradient features of the two sampled images respectively;
and (3) connecting the two groups of gradient features in series to serve as descriptors, and completing the classification work of the fault buckle images by using the SVM.
The further technical scheme is that the improved CS-LBP algorithm is as follows:
taking the sampling number P=8 in the image sampling circle domain, the sampling radius R=1 and the threshold T=10; firstly determining gray values of all sampling points in a circular neighborhood by using a bilinear interpolation method; sequentially calculating the gray difference value of each group of central symmetry sampling points; and finally, comparing the magnitude relation between the gray difference value and the threshold value T to determine the direction of the gradient.
The further technical proposal is that: if the absolute value of the gray difference value is larger than the threshold value T, marking the counting bit corresponding to the sampling point with large gray value as 1, wherein the counting bit is the gradient direction of the group of sampling pixel pairs; meanwhile, marking the counting bit corresponding to the sampling point with small gray value as 0; if the absolute value of the gray level difference value is not greater than the threshold value T, the number of the sampling points corresponding to the two sampling points is marked as 0, the sampling pixel pairs are not graded, the operation is repeated on the whole Gaussian filtered image, a complete preliminary gradient direction characteristic diagram is obtained, and the value of each pixel uniquely corresponds to a one-dimensional array formed by eight-bit binary digits 0 or 1.
The further technical proposal is that the calculation method of the key gradient characteristics is as follows:
the key gradient direction is adopted to represent the overall gradient trend of the pixel block, the whole pixel block is ordered according to a gradient direction diagram, numbers 1-8 distributed in the anticlockwise direction respectively correspond to 8 gradient directions, and after the binary pixel values of all pixels in the pixel block are accumulated one by one according to the same counting bit, the counting and the largest counting bit are the key gradient direction of the pixel block.
The further technical proposal is that: the Cell unit is a pixel block consisting of a, b, c, d four pixels, and the method for acquiring the key gradient direction characteristics of the Cell is as follows:
firstly, sequentially calculating pixel values of a one-dimensional array corresponding to the four pixels by using an improved CS-LBP algorithm; accumulating the four one-dimensional pixel values one by one according to the same counting bit, and taking the gradient direction with the largest counting direction as the key gradient direction; finally, obtaining the gradient direction of the Cell and the output coding value of the key gradient direction, wherein the calculation formula is defined as follows:
X i =[x i1 ,x i2 ,…x ij …,x im ]
wherein X is i Represents one-dimensional array, Y represents all one-dimensional arrays X in Cell structure i X is calculated by adding up, x ij E {0,1}, i e {1, n }, j e {1, m }, i and j are integers; the function max_id (Y) represents the position corresponding to the maximum value in the output one-dimensional array Y, namely the column coordinate j of the maximum value at the moment; repeating the steps on the whole preliminary gradient direction characteristic diagram, taking the step size stride as 2, and obtaining a gradient characteristic diagram after the first downsampling;
and carrying out the same iteration on the feature map after the first downsampling to obtain a second downsampled feature map.
The further technical proposal is that: the length and width of the Cell structure are set to be 2 multiplied by 2, the number of channels is 8, and the sliding step size is 2, so that the setting of relevant parameters in continuous gradient feature extraction can be facilitated, and meanwhile, the length and width of each layer of downsampling feature map are in a multiple relation with the size of the minimum partition sub-image obtained by the traditional LBP and a variant algorithm thereof.
The further technical proposal is that: when extracting continuous gradient features on the sampling feature map, defining a 3×3 window as a Block, wherein the number of channels is 8, and the sliding step length is 1.
The further technical proposal is that: and (3) carrying out annular sequencing in a Block according to the anticlockwise direction of the gradient azimuth graph, counting the maximum continuous sequence length of non-central pixels of all reference channels and adjacent channels, firstly determining the channel with the central pixel one by one and taking the channel as the reference, and if all the channels have no central pixel, outputting continuous gradient characteristics by the Block.
The further technical proposal is that: the calculation steps and the limiting conditions of the continuous gradient feature comprise the following three points:
firstly determining a channel with a central pixel in a Block of 8 channels, and if the central pixel does not exist, outputting no characteristics;
judging whether the maximum continuous sequence length in the neighborhood of the channel with the center pixel is not less than 3 or not by taking the channel with the center pixel as a reference, and if not, not outputting the characteristics;
and respectively counting the maximum continuous sequence length of the channels adjacent to the reference channel, wherein the coding form of continuous gradient characteristic output is the reference channel number and the maximum continuous sequence length of three corresponding channels, and the total number of codes of the strategy is 768.
The beneficial effects of adopting above-mentioned technical scheme to produce lie in: compared with the prior art, the method has the remarkable advantages that the dimension conversion of the outline features of the buckle is realized through the continuous downsampling feature map, the feature extraction efficiency is improved to a great extent, and the function of reducing the latitude of the description sub-features is also achieved. Meanwhile, the continuous gradient characteristic can also more accurately reflect the contour characteristic of the leaky cable buckle, particularly the gradient information of inflection points and side lines, and the method has a good detection effect in distinguishing fault buckle images in a high-speed railway tunnel scene.
Drawings
The invention will be described in further detail with reference to the drawings and the detailed description.
FIG. 1 is a diagram of LBP and CS-LBP sampling patterns in an embodiment of the present invention;
FIG. 2 is a graph of preliminary gradient direction calculated in an embodiment of the invention;
FIG. 3a is a graph of gradient azimuth in an embodiment of the invention;
FIG. 3b is a diagram of a Cell structure in an embodiment of the invention;
FIG. 4 is a graph of key gradients calculated in an embodiment of the invention;
FIG. 5 is a diagram of a downsampling process in an embodiment of the invention;
FIG. 6 is a graph of computing a continuous gradient profile in an embodiment of the invention;
FIG. 7 is a flow chart of an experiment of the method according to an embodiment of the present invention;
FIG. 8 is a graph of ROC in an embodiment of the invention;
FIG. 9 is a flow chart of a method according to an embodiment of the invention;
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways other than those described herein, and persons skilled in the art will readily appreciate that the present invention is not limited to the specific embodiments disclosed below.
As shown in fig. 9, generally, the embodiment of the invention discloses a method for detecting a high-speed rail tunnel leakage cable buckle with a multi-scale continuous gradient characteristic, which comprises the following steps:
firstly, improving a CS-LBP algorithm and calculating gradient direction characteristics of a sampling circle domain one by one to obtain a complete primary gradient direction characteristic diagram; then, carrying out continuous downsampling twice on the feature map and respectively extracting continuous gradient features of the two sampled images; finally, the two groups of gradient features are connected in series to serve as descriptors, and the SVM is used for completing the classification work of the fault buckle images.
The method can realize the scale transformation of the outline features of the buckle, effectively reduce the feature dimension of the descriptor, and improve the classification accuracy of the image by extracting continuous gradient features. The experiment compares the detection conditions of various variant algorithms such as CS-LBP, HOG and the like, and the result shows that the recall rate and the precision of the algorithm provided by the method reach 0.923 and 0.857 respectively, and the method has obvious advantages compared with the comparison algorithms.
The above method will be described in detail with reference to specific steps
Improved CS-LBP algorithm
Conventional LBP and CS-LBP:
the traditional LBP descriptor is originally proposed by Ojala et al, and has better rotation invariance and gray invariance because the traditional LBP descriptor can sharply capture gray level differences among fine textures, so far, the traditional LBP descriptor is still widely applied to the fields of face recognition and texture classification. The core idea of the algorithm is that in a circular neighborhood with R as a radius, the gray value between a central pixel and P sampling pixel points in the neighborhood is compared, a string of binary sequences only comprising 0 and 1 is calculated by using a Boolean function, and finally the binary sequences are converted into decimal LBP coding values according to a unified sequence, wherein the corresponding calculation formula is as follows:
n i gray value representing the ith sample point on the circular neighborhood, the size of which can be determined by bilinear interpolation, n c Representing the gray value of the center pixel. In order to enhance the representation capability of LBP descriptors in the spatial direction, heikkila et al propose CS-LBP descriptors, replacing the way of LBP about encoding the difference information of neighbor sampling points and center pixels, only aiming at the center symmetry of four groups of radial distributionThe sampling points are encoded, and the corresponding calculation formula is changed into:
the feature extraction effect of multiple groups of experiments needs to be compared to better evaluate the threshold T in the above formula (4). The method can be generally determined according to the global gray level condition of an input image or the gray level condition of a neighborhood sampling pixel, and the CS-LBP operator is selected as an improved basic model by the following three main reasons:
1) Under the condition of the same sampling number and sampling radius, as shown in fig. 1, the feature vector of CS-LBP coding is more compact, and the feature dimension of descriptors is far lower than LBP.
2) CS-LBP inherits good texture characteristics of LBP, and simultaneously expands gradient difference information of four groups of radial distribution.
3) CS-LBP can have stronger stability in the gentle region of the processed image, and many experiments on image matching also prove that CS-LBP operators have better robustness than gradient operators or magnitude operators alone.
Improving the directivity of CS-LBP:
the conventional CS-LBP only illustrates that there is a gradient difference in a certain diameter direction or a certain diameter direction within the sampling circle, but does not explicitly indicate which side the gradient direction is. Therefore, the method keeps the encoding mode of CS-LBP about comparing the gray level difference value of the central symmetry sampling points on the basis, but at the same time, the side with high gray level value of each group of central symmetry sampling points must be clearly identified so as to acquire more specific gradient direction information, namely, the binary encoding bit number is expanded to 8 bits. The primary gradient direction characteristic diagram output by the link is used for directly outputting the binary form of the coding mode without decimal conversion of the pixel value.
Fig. 2 shows a process of calculating a preliminary gradient direction by using a modified CS-LBP algorithm, where the number of samples p=8, the sampling radius r=1, and the threshold t=10 are taken in the circle. Firstly determining gray values of all sampling points in a circular neighborhood by using a bilinear interpolation method; sequentially calculating the gray difference value of each group of central symmetry sampling points; finally, comparing the magnitude relation between the difference value and the threshold value T to determine the direction of the gradient. If the absolute value is greater than the threshold value T, the counting bit corresponding to the sampling point with large gray value is marked as 1, and the counting bit is the gradient direction of the group of sampling pixel pairs. And meanwhile, marking the counting bit corresponding to the sampling point with small gray value as 0. If the absolute value is not greater than the threshold T, then the count bits corresponding to both sample points are marked 0, indicating that the set of sample pixel pairs has no gradient. The calculation shows that only three gradient directions of 45 degrees, 90 degrees and 315 degrees exist in the figure 2, so that only counting bits corresponding to the three directions are marked as 1, and the rest direction bits are marked as 0. Then, the encoding is performed in the counterclockwise direction from 0 °,10000110 is the binary encoding value. Repeating the operation on the whole Gaussian filtered image to obtain a complete primary gradient direction characteristic diagram, wherein the value of each pixel uniquely corresponds to a one-dimensional array formed by eight-bit binary digits 0 or 1.
Gradient direction feature map:
in the obtained primary gradient direction characteristic diagram, each one-dimensional array pixel value has 81 (3 4 ) If continuous gradient features are directly extracted from the feature map, the obtained description sub-feature dimension is too huge, so that the method provides an effective dimension reduction mode, on one hand, the total number of binary pixel value arrangement forms can be reduced, and on the other hand, the dimension conversion of the feature map can be realized.
The key gradient direction is adopted to represent the overall gradient trend of the pixel block, and the purpose is to convert 81 values of original single pixels into 8 effective values of the pixel block, so that continuous gradient characteristics are conveniently encoded. The sequence is carried out according to the gradient direction diagram of fig. 3a, the numbers 1-8 distributed in the anticlockwise direction respectively correspond to 8 gradient directions, and after the binary pixel values of all pixels in the pixel block are accumulated one by one according to the same counting bit, the counting and the largest counting bit are the key gradient directions of the pixel block. The Cell unit in fig. 3b is a pixel block composed of a, b, c, d four pixels, taking the key gradient direction feature of the Cell as an example, and fig. 4 is a calculation process of the key gradient. Firstly, sequentially calculating pixel values of a one-dimensional array corresponding to the four pixels by using an improved CS-LBP algorithm; then the four one-dimensional pixel values are accumulated one by one according to the same counting bit, so that the maximum counting in the 90-degree direction can be found, and four times of counting occur; finally, the gradient direction of the Cell is 90 degrees, namely the output coding value of the key gradient direction is 2, and the calculation formula is defined as follows:
X i =[x i1 ,x i2 ,…x ij …,x im ] (5)
wherein X is i Represents one-dimensional array, Y represents all one-dimensional arrays X in Cell structure i X is calculated by adding up, x ij E {0,1}, i e {1, n }, j e {1, m }, i and j are integers. The function max_id (Y) indicates the position corresponding to the maximum value in the output one-dimensional array Y, i.e., the column coordinate j of the maximum value at this time. Repeating the steps above for the whole preliminary gradient direction characteristic diagram, taking the step size stride as 2, and obtaining the gradient characteristic diagram after the first downsampling.
And carrying out the same iteration on the feature map after the first downsampling to obtain a second downsampled feature map. The length and width of the Cell structure are set to be 2 multiplied by 2, the number of channels is 8, and the sliding step size is 2, so that the setting of relevant parameters in continuous gradient feature extraction can be facilitated, and meanwhile, the length and width of each layer of downsampling feature map are in a multiple relation with the size of the minimum partition sub-image obtained by the traditional LBP and a variant algorithm thereof, and the comparison is facilitated. In addition, unlike maximum pooling in deep learning, here also value suppression is aimed at. Fig. 5 is a schematic diagram of the output result of the downsampling process, wherein the feature map of the other links is 8 channels except that the input image is a single-channel gray scale image, and each channel represents one gradient direction in fig. 3 a. After scaling, the typical gradient features at the target contour will be preserved, whereas some of the non-varying inflection points, edges, and noise will be removed during downsampling.
Continuous gradient features:
some depth mining operators of LBP and the fusion feature operators of the series basically only perform texture characteristic analysis aiming at gray level difference values of local sampling points, and obvious limitations exist. Therefore, the method extracts continuous gradient features on the feature map after two downsampling respectively, and improves the quality of the descriptors through the gradient change relation of the target contour at the inflection point and the edge line. In order to distinguish from Cell units in the downsampling process, a 3×3 window is defined as a Block when continuous gradient features are extracted on a sampling feature map, the number of channels is 8, and the sliding step is 1. The processing unit selects 3×3 because the 8 field sampling points are adjacent to the center sampling point, which can better respond to the key characteristic of continuous gradient and also accords with coding habit.
And (3) performing annular sequencing in the Block according to the anticlockwise direction of the gradient azimuth graph, and counting the maximum continuous sequence length of non-central pixels of all the reference channels and the adjacent channels. First, the channels with the center pixel are determined one by one and based on the channel, if all channels have no center pixel, the Block does not output continuous gradient characteristics. Only channels 2, 4 in fig. 6 have a central pixel, so feature extraction is only performed for both channels. The maximum continuous sequence length of the 8 neighborhood sampling pixels in the channel 2 is 4, the maximum continuous sequence length of the adjacent channel 1 is 1, the maximum continuous sequence length of the channel 3 is 2, thus obtaining the continuous gradient characteristic of the channel 2, and the coding result is 2412, which means that the maximum continuous sequence lengths of the channels 2, 1 and 3 are 4, 1 and 2 respectively based on the channel 2. Since the channel 4 itself has a maximum consecutive sequence length of less than 3, the channel also does not output a feature. The calculation steps and the limiting conditions of the continuous gradient characteristics are summarized into the following three points:
1) And determining a channel with a central pixel in the 8-channel Block, and if the central pixel is not present, outputting no characteristic.
2) And judging whether the maximum continuous sequence length in the neighborhood of the channel with the center pixel is not less than 3 based on the channel with the center pixel, and if the maximum continuous sequence length is not less than 3, not outputting the characteristics.
3) And respectively counting the maximum continuous sequence length of the channels adjacent to the reference channel, wherein the coding form of continuous gradient characteristic output is the reference channel number and the maximum continuous sequence length of three corresponding channels, and the total number of codes of the strategy is 768.
The experimental process comprises the following steps:
the CPU adopted in the experimental operation environment is Intel Xeon E5-2697 v2, the operation memory is 98GB, and the hard disk storage space is 8TB. The leaky cable buckle image data is shot by a double-line CMOS camera in a whole course, wherein 195 images of a fault buckle and 900 images of a normal buckle are shot, 70% of an input image data set is taken as a training set, and the rest 30% is taken as a test set. Fig. 7 illustrates a complete process for obtaining continuous gradient features from raw data and then completing a fault buckle detection task with the SVM. The size of the input image after the preprocessing becomes 256×256, the basic unit for extracting the preliminary gradient direction feature and the continuous gradient feature is 3×3, and the basic unit for the two downsampling is 2×2.
Calculating an adaptive threshold T:
in order to obtain a proper threshold T, 5-6 images with global gray average value distribution in different intervals are selected from a data set, and the corresponding relation between gray and the threshold is analyzed through an experimental method. The original input pictures are preprocessed by a 9 x 9 gaussian filter, and then the imaging effects of the preliminary gradient direction feature patterns under different threshold conditions are compared, and the threshold T is as large as possible on the premise of meeting the following two requirements.
1) It is necessary to ensure that the preliminary gradient direction feature pattern can completely and clearly display the profile of the leaky cable fastener.
2) Background noise of the tunnel wall can be filtered to a certain extent.
TABLE 1 analysis of imaging effects for different gray-thresholds
In order to conveniently compare imaging effects of different gray scales and thresholds, the global gray scale average value is divided into 8 intervals, and 8 threshold levels are respectively corresponding to the intervals. Table 1 lists the imaging effect of the preliminary gradient direction feature map of the partial buckle image under different threshold values, and the specific calculation formula is:
when the global gray average value is between 80 and 220, calculating a corresponding threshold value T by adopting a rounding function; the gray average value is smaller than 80 and is the lowest threshold level T=4; the gray average value is greater than 220, which is the highest threshold level t=11. And (3) calculating the primary gradient direction characteristics one by adopting an improved CS-LBP algorithm to the input image according to the formula (8) to obtain a corresponding primary gradient direction characteristic diagram.
Calculating continuous gradient characteristics:
some typical features such as inflection points or edges of the leaky cable clasp are not apparent in the third downsampled feature map, or a large number of consecutive gradient features are not expressed on the third and subsequent sampled feature maps, so only two downsamples are performed during the experiment. And extracting continuous gradient features from the feature map after the two downsampling steps and connecting the two groups of continuous gradient features in series to obtain a final fusion feature descriptor, wherein the feature dimension is 1536, and the image is not required to be subjected to blocking processing.
Experimental results and analysis:
the experiment adopts Recall ratio Recall, precision, characteristic dimension Bin and AUC values of descriptors as evaluation indexes of detection results. Wherein Recall and Precision refer to the proportion of the buckle correctly judged to be faulty to all faulty buckles, and the proportion of the buckle correctly judged to be faulty to all faulty buckles respectively, and the specific calculation formula is as follows:
in order to prove that the method has the advantages of extracting the profile characteristics of the leaky cable buckle by the algorithm, and comparing the profile characteristics with various variant algorithms such as CS-LBP, HOG and the like, the experimental results of the table 2 are obtained, and the parameter setting conditions of the comparison algorithms are as follows:
1) The LBP sampling radius R is 1, the sampling number P is 8, and the bilinear interpolation method is used for sampling. In the experiment, 4×4 non-overlapping division is needed for the image, and the characteristic dimension of the descriptor is 4096.
2)The (complete binary pattern) is sampled by bilinear interpolation, the sampling radius r=1, and the sampling number p=8. The local sign operator clbp_s and the local magnitude operator clbp_m within each sample circle are encoded using a rotation invariant uniform mode. In the experiment, the method carries out 4×4 non-overlapping division on the image, and the characteristic dimension of each partitioned sub-image is 2 (P+2) 2 The output descriptor feature dimension is 3200, and the method will be referred to as CLBP in the following tables and figures.
3) The HOG features are first scaled down to 128 x 128 in size using nearest neighbor interpolation, and then feature extraction is performed. Where Cell is 8×8, block is 16×16, stripe is 16, and the bin of the gradient component in each Block unit is 9, so the feature dimension of the output descriptor is 2304.
(4) TPLBP and ARCS-LBP are sampled by bilinear interpolation, the sampling radius is R=1, the sampling number is P=8, 4×4 non-overlapping division is needed to be carried out on the image in the experiment, and the characteristic dimension of the descriptors is 4096.
5) HOG_LBP and HOG_CLBP refer to that the HOG descriptors obtained previously are respectively connected in series with the LBP descriptors and the CLBP descriptors to obtain fusion characteristics of the two descriptors, and the characteristic dimensions of the corresponding descriptors are directly accumulated.
By comparing the detection results of table 2 with the ROC graph of fig. 8, it can be seen that the multi-scale continuous gradient feature algorithm MSG-LBP proposed by the method shows significant advantages on the detection results of the leaky cable buckle. Where Recall and Precision reached 0.923 and 0.857, respectively, the AUC values for MSG-LBP were also highest in the ROC curve. Even though the fusion feature of hog_lbp can bring Recall to 0.889, precision is only 0.615, which indicates that there are a large number of normal snaps that are misjudged as failed snaps. The MSG-LBP descriptor does not carry out blocking processing on the image, but the feature map of the first layer of downsampling is consistent with the size of the minimum partitioned sub-image of the image which is divided by 2 multiplied by 2, and the feature map of the second layer of downsampling is consistent with the size of the minimum partitioned sub-image of the image which is divided by 4 multiplied by 4, so that the calculation amount of feature extraction is greatly reduced. The twice down-sampling feature map in the experiment not only contains key gradient information of the buckle outline features, but also realizes scale transformation of the buckle outline features. The dimension of the multi-scale continuous gradient feature descriptors is lower than the latitude of most of the depth mining descriptors and fusion feature descriptors of the series, and the detection capability is obviously higher.
Table 2 comparison of test results
In summary, compared with the series of algorithms, the method has the remarkable advantages that the scale transformation of the buckle outline features is realized through the continuous downsampling feature graphs, the feature extraction efficiency is improved to a great extent, and the function of reducing the latitude of the description sub-features is also realized. Meanwhile, the continuous gradient characteristic can also more accurately reflect the contour characteristic of the leaky cable buckle, in particular to gradient information at the inflection point and the edge line.
Claims (7)
1. The method for detecting the high-speed railway tunnel leakage cable buckle is characterized by comprising the following steps of:
improving a CS-LBP algorithm and calculating gradient direction characteristics of a high-speed railway tunnel cable leakage buckle image sampling circle domain one by one to obtain a complete primary gradient direction characteristic diagram;
the improved CS-LBP algorithm comprises the following steps: taking the sampling number P=8 in the image sampling circle domain, the sampling radius R=1 and the threshold T=10; firstly determining gray values of all sampling points in a circular neighborhood by using a bilinear interpolation method; sequentially calculating the gray difference value of each group of central symmetry sampling points; finally, comparing the magnitude relation between the gray difference value and the threshold value T to determine the gradient direction;
if the absolute value of the gray difference value is larger than the threshold value T, marking the counting bit corresponding to the sampling point with large gray value as 1, wherein the counting bit is the gradient direction of the group of sampling pixel pairs; meanwhile, marking the counting bit corresponding to the sampling point with small gray value as 0; if the absolute value of the gray level difference value is not greater than the threshold value T, the number of the corresponding sampling points is marked as 0, the number of the sampling points is not gradient, the operation is repeated on the whole Gaussian filtered image, a complete preliminary gradient direction characteristic diagram is obtained, and the value of each pixel is uniquely corresponding to a one-dimensional array formed by eight-bit binary digits 0 or 1;
performing two continuous downsampling on the feature map and extracting continuous gradient features of the two sampled images respectively;
and (3) connecting the two groups of gradient features in series to serve as descriptors, and completing the classification work of the fault buckle images by using the SVM.
2. The high-speed railway tunnel leaky cable buckle detection method according to claim 1, wherein: the key gradient feature calculation method is as follows:
the key gradient direction is adopted to represent the overall gradient trend of the pixel block, the whole pixel block is ordered according to a gradient direction diagram, numbers 1-8 distributed in the anticlockwise direction respectively correspond to 8 gradient directions, and after the binary pixel values of all pixels in the pixel block are accumulated one by one according to the same counting bit, the counting and the largest counting bit are the key gradient direction of the pixel block.
3. The method for detecting a high-speed rail tunnel cable-leaking buckle according to claim 2, wherein the Cell unit is a pixel block composed of a, b, c, d four pixels, and the method for obtaining the key gradient direction characteristics of the Cell is as follows:
firstly, sequentially calculating pixel values of a one-dimensional array corresponding to the four pixels by using an improved CS-LBP algorithm; accumulating the four one-dimensional pixel values one by one according to the same counting bit, and taking the gradient direction with the largest counting direction as the key gradient direction; finally, obtaining the gradient direction of the Cell and the output coding value of the key gradient direction, wherein the calculation formula is defined as follows:
wherein the method comprises the steps ofX i Representing a one-dimensional array of numbers,Yrepresenting all one-dimensional arrays within a Cell structureX i Is used for the sum of the (a),x ij ∈{0,1},i ∈{1,n},j ∈{1,m},iandjare integers; function ofRepresenting output one-dimensional arraysYThe position corresponding to the maximum value, i.e. the column coordinate of the maximum value at this timejThe method comprises the steps of carrying out a first treatment on the surface of the Repeating the steps on the whole preliminary gradient direction characteristic diagram, taking the step size stride as 2, and obtaining a gradient characteristic diagram after the first downsampling;
and carrying out the same iteration on the feature map after the first downsampling to obtain a second downsampled feature map.
4. The high-speed railway tunnel leaky cable buckle detection method according to claim 3, wherein: the length and width of the Cell structure are set to be 2 multiplied by 2, the number of channels is 8, and the sliding step size is 2, so that the setting of relevant parameters in continuous gradient feature extraction can be facilitated, and meanwhile, the length and width of each layer of downsampling feature map are in a multiple relation with the size of the minimum partition sub-image obtained by the traditional LBP and a variant algorithm thereof.
5. The high-speed railway tunnel leaky cable buckle detection method according to claim 1, wherein: when extracting continuous gradient features on the sampling feature map, defining a 3×3 window as a Block, wherein the number of channels is 8, and the sliding step length is 1.
6. The high-speed railway tunnel leaky cable buckle detection method according to claim 5, wherein: and (3) carrying out annular sequencing in a Block according to the anticlockwise direction of the gradient azimuth graph, counting the maximum continuous sequence length of non-central pixels of all reference channels and adjacent channels, firstly determining the channel with the central pixel one by one and taking the channel as the reference, and if all the channels have no central pixel, outputting continuous gradient characteristics by the Block.
7. The high-speed railway tunnel leaky cable buckle detection method according to claim 5, wherein: the calculation steps and the limiting conditions of the continuous gradient feature comprise the following three points:
firstly determining a channel with a central pixel in a Block of 8 channels, and if the central pixel does not exist, outputting no characteristics;
judging whether the maximum continuous sequence length in the neighborhood of the channel with the center pixel is not less than 3 or not by taking the channel with the center pixel as a reference, and if not, not outputting the characteristics;
and respectively counting the maximum continuous sequence length of the channels adjacent to the reference channel, wherein the coding form of continuous gradient characteristic output is the reference channel number and the maximum continuous sequence length of three corresponding channels, and the total number of codes of the strategy is 768.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111449917.6A CN114119577B (en) | 2021-11-30 | 2021-11-30 | High-speed rail tunnel leakage cable buckle detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111449917.6A CN114119577B (en) | 2021-11-30 | 2021-11-30 | High-speed rail tunnel leakage cable buckle detection method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114119577A CN114119577A (en) | 2022-03-01 |
CN114119577B true CN114119577B (en) | 2023-06-23 |
Family
ID=80368884
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111449917.6A Active CN114119577B (en) | 2021-11-30 | 2021-11-30 | High-speed rail tunnel leakage cable buckle detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114119577B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115100202B (en) * | 2022-08-25 | 2022-11-11 | 聊城市宏伟电器有限公司 | Cable quality detection method based on image processing |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103093226A (en) * | 2012-12-20 | 2013-05-08 | 华南理工大学 | Construction method of RATMIC descriptor for image feature processing |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102855634B (en) * | 2011-06-28 | 2017-03-22 | 中兴通讯股份有限公司 | Image detection method and image detection device |
CN102324030B (en) * | 2011-09-09 | 2013-11-06 | 广州灵视信息科技有限公司 | Target tracking method and system based on image block characteristics |
CN105404857A (en) * | 2015-11-04 | 2016-03-16 | 北京联合大学 | Infrared-based night intelligent vehicle front pedestrian detection method |
CN107315990B (en) * | 2017-05-04 | 2021-08-13 | 深圳市美好幸福生活安全***有限公司 | Pedestrian detection algorithm based on XCS-LBP characteristics |
-
2021
- 2021-11-30 CN CN202111449917.6A patent/CN114119577B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103093226A (en) * | 2012-12-20 | 2013-05-08 | 华南理工大学 | Construction method of RATMIC descriptor for image feature processing |
Also Published As
Publication number | Publication date |
---|---|
CN114119577A (en) | 2022-03-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114140462B (en) | Bearing wear degree assessment method based on image processing | |
CN110119728B (en) | Remote sensing image cloud detection method based on multi-scale fusion semantic segmentation network | |
CN108346154B (en) | Method for establishing lung nodule segmentation device based on Mask-RCNN neural network | |
CN102360421B (en) | Face identification method and system based on video streaming | |
CN110070091B (en) | Semantic segmentation method and system based on dynamic interpolation reconstruction and used for street view understanding | |
CN112633382B (en) | Method and system for classifying few sample images based on mutual neighbor | |
CN111080678B (en) | Multi-temporal SAR image change detection method based on deep learning | |
CN110751612A (en) | Single image rain removing method of multi-channel multi-scale convolution neural network | |
CN109635726B (en) | Landslide identification method based on combination of symmetric deep network and multi-scale pooling | |
CN111860823B (en) | Neural network training method, neural network image processing method, neural network training device, neural network image processing equipment and storage medium | |
CN111027377A (en) | Double-flow neural network time sequence action positioning method | |
CN116912257B (en) | Concrete pavement crack identification method based on deep learning and storage medium | |
CN112700418A (en) | Crack detection method based on improved coding and decoding network model | |
CN114119577B (en) | High-speed rail tunnel leakage cable buckle detection method | |
CN113988147A (en) | Multi-label classification method and device for remote sensing image scene based on graph network, and multi-label retrieval method and device | |
CN115033721A (en) | Image retrieval method based on big data | |
CN114612315A (en) | High-resolution image missing region reconstruction method based on multi-task learning | |
CN112614094B (en) | Insulator string abnormity positioning and identifying method based on sequence state coding | |
CN113378672A (en) | Multi-target detection method for defects of power transmission line based on improved YOLOv3 | |
CN115830514B (en) | Whole river reach surface flow velocity calculation method and system suitable for curved river channel | |
CN109815926B (en) | Improved LBP algorithm based on Euclidean distance-differential coding | |
CN113361407B (en) | PCANet-based spatial spectrum feature combined hyperspectral sea ice image classification method | |
CN112632601B (en) | Crowd counting method for subway carriage scene | |
CN104616302A (en) | Real-time object identification method | |
CN114140698A (en) | Water system information extraction algorithm based on FasterR-CNN |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |