CN116468958B - Communication tower safety detection method and system - Google Patents

Communication tower safety detection method and system Download PDF

Info

Publication number
CN116468958B
CN116468958B CN202310559241.9A CN202310559241A CN116468958B CN 116468958 B CN116468958 B CN 116468958B CN 202310559241 A CN202310559241 A CN 202310559241A CN 116468958 B CN116468958 B CN 116468958B
Authority
CN
China
Prior art keywords
image
features
pixel
noise
feature
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
Application number
CN202310559241.9A
Other languages
Chinese (zh)
Other versions
CN116468958A (en
Inventor
洪球
尹进
高朝
黄娟辉
刘方科
欧军
孟正彪
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhongweijian Communication Technology Service Co ltd
Original Assignee
Zhongweijian Communication Technology Service Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Zhongweijian Communication Technology Service Co ltd filed Critical Zhongweijian Communication Technology Service Co ltd
Priority to CN202310559241.9A priority Critical patent/CN116468958B/en
Publication of CN116468958A publication Critical patent/CN116468958A/en
Application granted granted Critical
Publication of CN116468958B publication Critical patent/CN116468958B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G06V10/765—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects using rules for classification or partitioning the feature space
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/40—Image enhancement or restoration using histogram techniques
    • 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
    • G06T7/40—Analysis of texture
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/90—Determination of colour characteristics
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/467—Encoded features or binary features, e.g. local binary patterns [LBP]
    • 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
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/761—Proximity, similarity or dissimilarity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Medical Informatics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Multimedia (AREA)
  • General Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses a communication tower safety detection method and a communication tower safety detection system, wherein the communication tower safety detection method comprises the following steps: acquiring a close-range image of the iron tower through unmanned aerial vehicle aerial photographing; performing histogram equalization on the image; enhancing the image; performing an improved weighted median filtering of the image; feature extraction, including extracting edge features, color features, PCA features, and texture features; and classifying the characteristics by using an improved random forest method, and outputting the fault type of the communication tower. According to the invention, noise is filtered by improving a median filtering method, wherein the size of a filtering window is dynamically adjusted according to the image sparseness, and the product of the geometric position similarity and the sparseness factor is used as the weight of a pixel, so that the noise can be removed in a self-adaptive manner; the feature splitting node method in the random forest algorithm is improved, excessive feature discrete values are reduced, splitting of the features of the decision tree is quickened, and the quality of decision tree generation is improved.

Description

Communication tower safety detection method and system
Technical Field
The invention belongs to the technical field of communication tower safety, and particularly relates to a communication tower safety detection method and system.
Background
The communication iron tower can appear material ageing, bolt looseness, steel corrosion, appear the damage phenomenon of different degrees such as crack in long-term service in-process, leads to the structure to weaken constantly. If the manual inspection mode is used, a great deal of manpower is required to be input for periodic inspection. With the development of unmanned aerial vehicle technology, unmanned aerial vehicle cost is lower and lower, and compared with the method of installing various sensors on communication tower and detecting the trouble, the safety inspection through unmanned aerial vehicle aerial acquisition communication tower's image is more and more necessary.
Digital images are subject to environmental conditions during acquisition and transmission to produce noise, typical of which include evenly distributed noise, gaussian noise, pretzel noise, and gamma noise. These noises reduce the sharpness of the image, and thus it is necessary to remove the noises and extract image features in the image processing. The median filtering algorithm is a nonlinear filtering method in image processing, and has a good inhibition effect on bursty signals such as salt and pepper noise; the operation is simple and convenient, and complex parameter setting is not needed; image details and texture information are effectively preserved. However, since the median filtering algorithm belongs to a nonlinear filtering method, the image edge is blurred or distorted, and lost information caused by noise cannot be recovered; fixed noise cannot be suppressed, and only random noise can be suppressed; in some cases, the median filtering algorithm may not be ideal for smoothing the outdoor communication tower rainy and snowy weather image, and the filtering effect may affect the accuracy of subsequent image feature extraction.
The machine learning classification algorithm can be used for fault type classification, wherein the random forest algorithm has a plurality of advantages compared with other classification algorithms, such as high accuracy, capability of processing a large amount of data, better robustness and the like, but the problem that the feature splitting of the decision tree is influenced due to too many feature discrete values exists.
Disclosure of Invention
In view of this, the invention provides a communication tower safety detection system and method, noise is filtered by improving a median filtering method, wherein a filtering window is dynamically adjusted according to the image sparseness, the product of geometric position similarity and a sparseness factor is taken as the weight of a pixel, the pixel at the center point of the window of the current image is selected as the object of noise filtering, the pixel points in the window with the size of M multiplied by M are ordered according to the weight value, the median of the weight is selected as the weight value of the current pixel point, and the median is multiplied by the gray value of the pixel to obtain the gray value of the filtered pixel. After extracting the multidimensional features of the image, carrying out fault classification by using an improved random forest algorithm, wherein feature splitting nodes of the random forest algorithm are critical to the generation of a subsequent random decision tree, and the method randomly selects a part of features X from all the features 1 Further screening the feature set X from the part of the features by using a lasso regression method 2 Calculating the CMA sum of the random decision trees corresponding to the features; repeating the above steps for N times to obtain N feature sets X 2，N Selecting a feature set X 2，N M feature sets X with highest sum of CMA 2，M ，M<And N, calculating the similarity among M feature sets, and selecting two feature sets with the lowest similarity as feature splitting nodes, so that excessive feature discrete values are reduced, splitting of the features of the decision tree is quickened, and the quality of the decision tree generation is improved.
The invention discloses a communication tower safety detection method in a first aspect, which comprises the following steps:
acquiring a close-range image of the iron tower through unmanned aerial vehicle aerial photographing;
performing histogram equalization on the image;
enhancing the image;
performing an improved weighted median filtering of the image;
feature extraction, including extracting edge features, color features, PCA features, and the like;
and classifying the characteristics by using an improved random forest method, and outputting the fault type of the communication tower.
Further, the step of performing histogram equalization on the image is as follows:
histogram of statistical image normalized to between [0,1 ]:
wherein H, W are the height and width of the image, n k Representing a gray value r k The number of pixels of (a);
calculating a mapping function:
s k for the converted gray value, T (r k ) For the mapping function, the calculation process uses a cumulative histogram;
and processing the image by using the mapping function.
Further, the image enhancement step is as follows:
the image is enhanced using a Gamma transformation, the Gamma transformation formula being as follows:
output＝(input+ε) γ ；
in the formula, output is an output gray value, input is an input gray value, epsilon is a compensation coefficient, the value is generally 0, gamma is a gamma coefficient, and gamma is a positive number.
Further, the step of performing an improved weighted median filtering of the image is as follows:
1) Image noise detection and noise removal: calculating the statistical characteristics of the image, determining the noise type matched with the original image through the statistical characteristics, and if the probability density function of the noise is subjected to Gaussian distribution, the noise type is a Gaussian noise model, otherwise, the noise type is spiced salt noise;
for Gaussian noise, denoising by using an algorithm based on pixel gray value distribution;
for salt-pepper noise, denoising by using an algorithm based on the number of pixel values;
2) Calculating sparsity factors and geometric position similarity
The sparsity and compression ratio values of the image are obtained through a compressed sensing algorithm, and then coefficient factors are further calculated;
the gaussian kernel function formula for the geometrical position similarity is as follows:
wherein sigma d Representing the geometric diffusion factor, d (ζ, x) representing the Euclidean distance between the center pixel ζ of x and other pixels;
3) Determining the size of a dynamic filtering window, if the sparse factor of the area is large, increasing the filtering window, otherwise, using a smaller filtering window; the calculation formula of the filter window size is as follows:
4) Taking the product of the similarity of the geometric positions and the sparse factor as the weight of the pixel, selecting the pixel of the center point of the window of the current image as the noise filtering object, and performing the filtering on W wide ×W wide The pixel points in the window with the size are ordered according to the size of the weight value, the weight median value is selected as the weight value of the current pixel point, and the weight median value is multiplied by the gray value of the pixel to obtain the gray value of the filtered pixel;
5) Mapping and converting the filtered pixel gray values to a 0-255 interval, so that the pixel gray values < = 255:
x ij to map the gray value of the previous pixel, x max To map the gray maximum of the pixel before x ij ' is the gray value of the mapped pixel.
Further, the sparsity percentage is defined as:
the compression ratio is the ratio between the original and compressed signal vector lengths, defined as follows:
n represents the length of the signal X, M is the dimension of the sensing matrix, i.e. the number of measurements M of the accurately reconstructed signal X.
Further, the sparsity factor is defined as follows:
wherein a, b, c are experimentally obtained values, and a>b>c，a 1 And b 1 Is a preset threshold.
Further, the image features include edge features, PCA features, and texture features;
the edge features are extracted by using a Sobel edge extraction operator;
PCA features are extracted by calling PCA functions in a deconvolution library of a machine learning tool package sklearn;
texture features are extracted through a local binary pattern operator LBP;
the color features are obtained by extracting color moment features, including first-order moment, second-order moment and third-order moment color moments.
Further, the steps of performing fault identification on the image by using the improved random forest method are as follows:
sampling by using a Bootstrap;
training the subset;
selecting a characteristic splitting node;
constructing a decision subtree set;
performing preliminary classification on the test sample set to obtain a preliminary classification set;
calculating the weight of the AUC and carrying out weighted voting;
and obtaining a fault classification result.
Further, the step of selecting the feature split node is as follows:
randomly selecting a portion of features X from all features 1 Further screening the feature set X from the part of the features by using a lasso regression method 2 Calculating the CMA sum of the random decision trees corresponding to the features; repeating the above steps for N times to obtain N feature sets X 2，N Selecting a feature set X 2，N M feature sets X with highest sum of CMA 2，M ，M<And N, calculating the similarity among M feature sets, and selecting two feature sets with the lowest similarity as feature splitting nodes.
The second aspect of the invention discloses a communication tower safety detection system, comprising:
and an image acquisition module: acquiring a close-range image of the iron tower through unmanned aerial vehicle aerial photographing;
histogram equalization module: performing histogram equalization on the image;
an image enhancement module: enhancing the image;
and a filtering module: performing an improved weighted median filtering of the image;
the feature extraction module is used for extracting edge features, color features, PCA features and texture features;
and a fault classification module: and classifying the characteristics by using an improved random forest method, and outputting the fault type of the communication tower.
The beneficial effects of the invention are as follows:
according to the invention, noise is filtered by improving a median filtering method, wherein the size of a filtering window is dynamically adjusted according to the image sparseness degree, and the product of the geometric position similarity and the sparseness factor is used as the weight of a pixel, so that the noise can be removed in a self-adaptive manner;
judging the noise type in the image, and using a corresponding denoising method according to the noise type;
the feature splitting node method in the random forest algorithm is improved, excessive feature discrete values are reduced, splitting of the features of the decision tree is quickened, and the quality of decision tree generation is improved.
Drawings
FIG. 1 is a flow chart of the security detection of the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings, without limiting the invention in any way, and any alterations or substitutions based on the teachings of the invention are intended to fall within the scope of the invention.
Referring to fig. 1, the communication tower safety detection method provided by the invention is as follows:
acquiring a close-range image of the iron tower through unmanned aerial vehicle aerial photographing;
performing histogram equalization on the image;
the image is enhanced and the image is then processed,
performing an improved weighted median filtering of the image;
feature extraction, including extracting edge features, color features, PCA features, and the like;
and classifying the characteristics by using an improved random forest method, and outputting the fault type of the communication tower.
The steps of histogram equalization of the image are as follows:
1) Histogram of statistical image normalized to between [0,1 ]:
wherein H, W are the height and width of the image, n k Representing a gray value r k Is a number of pixels of the display panel.
2) Calculating a mapping function
s k For the converted gray value, T (r k ) For the mapping function, the calculation process uses a cumulative histogram.
3) And processing the image by using the mapping function.
The image is enhanced as follows:
the image is enhanced using a Gamma transformation, the Gamma transformation formula being as follows:
output＝(input+ε) γ ；
in the formula, output is an output gray value, input is an input gray value, epsilon is a compensation coefficient, the value is generally 0, gamma is a gamma coefficient, and gamma is a positive number. And correcting the picture with too high or too low gray level by using Gamma transformation, so as to enhance the contrast.
The steps of improved weighted median filtering of the image are as follows:
s1: image noise detection, calculating statistical characteristics of the image, such as pixel mean and variance.
And determining the noise type matched with the original image through the statistical characteristics, wherein the noise type is a Gaussian noise model if a probability density function of the noise is subjected to Gaussian distribution, and otherwise, the noise type is pretzel noise.
For Gaussian noise, an algorithm based on pixel gray value distribution is used for denoising, namely, points with pixel points smaller than a preset gray value distribution value are taken as noise, and the noise is removed.
For salt-pepper noise, denoising is performed by using an algorithm based on the number of pixel values, namely, points with the number smaller than the preset pixel values are taken as noise to be removed.
2) Calculating sparsity factors and geometric position similarity
And obtaining the sparsity and compression ratio value of the image through a compressed sensing algorithm, and further calculating to obtain a coefficient factor. The method specifically comprises the following steps: dividing an image into a plurality of M multiplied by M subareas, compressing the subareas by a compressed sensing method, calculating the sparsity and the compression ratio of the subareas, calculating the sparsity factor of the subareas, and taking the sparsity factor as the sparsity factor of all pixel points of the subareas. In some embodiments, under the condition of enough calculation power, the m×m sub-regions may be subjected to sliding calculation, and the calculated sparse factor is used as the sparse factor of the central pixel point of the m×m sub-regions.
Sparsity: if the elements in a domain of the signal are mostly zero or very small values, the signal is sparse in that domain. The effect of sparsity is that these zero elements can be effectively discarded without losing the relevant signal information. If a sparse signal contains S non-zero terms, then the signal is considered to be S sparse. For a signal of length N, this means that (N-S) signal coefficients can be removed while maintaining important information of this signal. In this case, the sparsity percentage is defined as:
compression Ratio (CR): a measure of the reduction of the samples required to represent the signal X data. In the case of CS, it is the ratio between the original and compressed signal vector lengths. If N represents the length of the signal X, M is the dimension of the sensing matrix, i.e., the number of measurements M needed to accurately reconstruct the signal X, then CR is defined as:
the sparsity factor is defined as follows:
wherein a, b, c are experimentally obtained values, and a>b>c，a 1 And b 1 Is a preset threshold.
The gaussian kernel function formula for the geometrical position similarity is as follows:
wherein sigma d Representing the geometric diffusion factor, d (ζ, x) represents the Euclidean distance between the center pixel ζ of x and other pixels.
S3: determining the size of the dynamic filter window, if the sparseness factor of the region is large, indicating that the effective pixels in the region are sparse, the filter window can be increased, otherwise, a smaller filter window is used. The calculation formula of the filter window size is as follows:
4) Taking the product of the similarity of the geometric positions and the sparse factor as the weight of the pixel, selecting the pixel of the center point of the window of the current image as the noise filtering object, and performing the filtering on W wide ×W wide The pixel points in the window with the size are ordered according to the weight value, the weight median value is selected as the weight value of the current pixel point, and the weight median value is multiplied by the gray value of the pixel to obtain the gray value of the filtered pixel. Mapping and converting the filtered pixel gray value to a 0-255 interval so as to enable the pixel gray value<＝255。
Image features are extracted, including edge features, PCA features, and texture features.
The extraction method of the edge features comprises the following steps:
edges are extracted using a Sobel edge extraction operator:
the Sobel edge extraction operator has a certain smoothing function and further removes noise.
Extracting PCA characteristics: and extracting PCA features of the image by using a PCA algorithm, and mapping pixels in the image into a low-dimensional space to realize dimension reduction of the image features. The method mainly comprises the following steps: the method comprises the steps of zero-equalizing original data, solving a covariance matrix, and solving eigenvectors and eigenvalues of the covariance matrix, wherein the eigenvectors form a new eigenvalue. The specific implementation method is that PCA feature extraction is carried out by calling PCA functions in a composition library of an open-source machine learning tool package sklearn based on python language.
Extracting texture features: LBP (Local BinaryPattern ) is an operator used to describe local texture features of an image, with multiple resolution, unchanged gray scale, unchanged rotation, etc. The method is mainly used for texture extraction in feature extraction. The invention acquires LBP characteristics of an image, comprising: the LBP features of the original LBP mode, the equivalent LBP mode, the rotation-invariant LBP mode, and the equivalent rotation-invariant LBP mode of the image are extracted and displayed.
Extracting color characteristics: the color moment feature is a simple and effective color feature representation method, and has a first moment (mean), a second moment (variance), a third moment (skew) and the like, and since color information is mainly distributed in low-order moments, the color moment feature representation method uses the first moment, the second moment and the third moment to express color distribution of an image, and the color moment has proved to be effective in representing the color distribution in the image.
Aiming at the problem that key parameters of a random forest algorithm are difficult to determine and the quality of decision trees is uneven to influence the overall effect of a model, the method uses an improved random forest algorithm to carry out safety detection on the iron tower image.
The steps for fault identification of images using the modified random forest method are as follows:
sampling by using a Bootstrap;
training the subset;
selecting a feature splitting node:
1) Randomly selecting a portion of features X from all features 1 Further screening the feature set X from the part of the features by using a lasso regression method 2 The sum of CMAs of the random decision trees corresponding to these features is calculated. The lasso regression method is common knowledge in the art and can be used for variable selection, and the invention is not repeated.
2) Repeating the step 1) for N times to obtain N feature sets X 2，N Selecting a feature set X 2，N M feature sets X with highest sum of CMA 2，M (M<N), calculating the similarity between M feature sets, and selecting two feature sets with the lowest similarity as feature splitting nodes; preferably, the similarity may be calculated using a cosine similarity method.
Constructing a decision subtree set;
performing preliminary classification on the test sample set to obtain a preliminary classification set;
calculating the weight of the AUC and carrying out weighted voting;
and obtaining a fault classification result.
The invention mainly uses accuracy, precision, recall and F value (F-score) to evaluate and compare the results of the improved random forest algorithm processing balance sample set.
The calculation formulas of the accuracy, the precision, the recall and the F value are as follows:
accuracy＝(TP+TN)/(TP+FP+TN+FN)；
precision＝TP/(TP+FP)，
indicating that the correct prediction is positive and the proportion of the total predictions is positive.
recall＝TP/(TP+FN)，
Indicating the proportion of correctly predicted positive to all positive samples.
F-score＝2precision×recall/(precision+recall)；
Wherein TP represents that the sample is positive and the prediction result is positive; FP indicates that the sample is negative and the predicted result is positive; TN indicates that the sample is negative and the predicted result is negative; FN indicates that the sample is positive and the prediction result is negative.
Another embodiment of the present invention discloses a communication tower safety detection system, including:
and an image acquisition module: acquiring a close-range image of the iron tower through unmanned aerial vehicle aerial photographing;
histogram equalization module: performing histogram equalization on the image;
an image enhancement module: enhancing the image;
and a filtering module: performing an improved weighted median filtering of the image;
the feature extraction module is used for extracting edge features, color features, PCA features and texture features;
and a fault classification module: and classifying the characteristics by using an improved random forest method, and outputting the fault type of the communication tower.
The beneficial effects of the invention are as follows:
according to the invention, noise is filtered by improving a median filtering method, wherein the size of a filtering window is dynamically adjusted according to the image sparseness degree, and the product of the geometric position similarity and the sparseness factor is used as the weight of a pixel, so that the noise can be removed in a self-adaptive manner.
And judging the noise type in the image, and using a corresponding denoising method according to the noise type.
The feature splitting node method in the random forest algorithm is improved, excessive feature discrete values are reduced, splitting of the features of the decision tree is quickened, and the quality of decision tree generation is improved.
The word "preferred" is used herein to mean serving as an example, instance, or illustration. Any aspect or design described herein as "preferred" is not necessarily to be construed as advantageous over other aspects or designs. Rather, use of the word "preferred" is intended to present concepts in a concrete fashion. The term "or" as used in this application is intended to mean an inclusive "or" rather than an exclusive "or". That is, unless specified otherwise or clear from the context, "X uses a or B" is intended to naturally include any of the permutations. That is, if X uses A; x is B; or X uses both A and B, then "X uses A or B" is satisfied in any of the foregoing examples.
Moreover, although the disclosure has been shown and described with respect to one or more implementations, equivalent alterations and modifications will occur to others skilled in the art based upon a reading and understanding of this specification and the annexed drawings. The present disclosure includes all such modifications and alterations and is limited only by the scope of the following claims. In particular regard to the various functions performed by the above described components (e.g., elements, etc.), the terms used to describe such components are intended to correspond, unless otherwise indicated, to any component which performs the specified function of the described component (e.g., that is functionally equivalent), even though not structurally equivalent to the disclosed structure which performs the function in the herein illustrated exemplary implementations of the disclosure. Furthermore, while a particular feature of the disclosure may have been disclosed with respect to only one of several implementations, such feature may be combined with one or other features of the other implementations as may be desired and advantageous for a given or particular application. Moreover, to the extent that the terms "includes," has, "" contains, "or variants thereof are used in either the detailed description or the claims, such terms are intended to be inclusive in a manner similar to the term" comprising.
The functional units in the embodiment of the invention can be integrated in one processing module, or each unit can exist alone physically, or a plurality of or more than one unit can be integrated in one module. The integrated modules may be implemented in hardware or in software functional modules. The integrated modules may also be stored in a computer readable storage medium if implemented in the form of software functional modules and sold or used as a stand-alone product. The above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, or the like. The above-mentioned devices or systems may perform the storage methods in the corresponding method embodiments.
In summary, the foregoing embodiment is an implementation of the present invention, but the implementation of the present invention is not limited to the embodiment, and any other changes, modifications, substitutions, combinations, and simplifications made by the spirit and principles of the present invention should be equivalent to the substitution manner, and all the changes, modifications, substitutions, combinations, and simplifications are included in the protection scope of the present invention.

Claims (9)

1. The communication iron tower safety detection method is characterized by comprising the following steps of:
acquiring a close-range image of the iron tower through unmanned aerial vehicle aerial photographing;
performing histogram equalization on the image;
enhancing the image;
the image is subjected to improved weighted median filtering as follows:
1) Image noise detection and noise removal: calculating the statistical characteristics of the image, determining the noise type matched with the original image through the statistical characteristics, and if the probability density function of the noise is subjected to Gaussian distribution, the noise type is a Gaussian noise model, otherwise, the noise type is spiced salt noise;
for Gaussian noise, denoising by using an algorithm based on pixel gray value distribution;
for salt-pepper noise, denoising by using an algorithm based on the number of pixel values;
2) Calculating sparsity factors and geometric position similarity
The sparsity and compression ratio values of the image are obtained through a compressed sensing algorithm, and then a sparsity factor is further calculated;
the geometrical position similarity is obtained through a Gaussian kernel function formula, wherein the Gaussian kernel function formula is as follows:
wherein sigma d Representing the geometric diffusion factor, d (ζ, x) representing the Euclidean distance between the center pixel ζ of x and other pixels;
3) Determining the size of a dynamic filtering window, if the sparse factor of the image processing area is large, increasing the filtering window, otherwise, using a smaller filtering window; the calculation formula of the filter window size is as follows:
wherein gamma represents a sparse factor, a, b, c are values obtained according to experiments, and a > b > c;
4) Taking the product of the similarity of the geometric positions and the sparse factor as the weight of the pixel, selecting the pixel of the center point of the window of the current image as the noise filtering object, and performing the filtering on W wide ×W wide The pixel points in the window with the size are ordered according to the size of the weight value, the weight median value is selected as the weight value of the current pixel point, and the weight median value is multiplied by the gray value of the pixel to obtain the gray value of the filtered pixel;
5) Mapping and converting the filtered pixel gray values to a 0-255 interval, so that the pixel gray values < = 255:
x ij to map the gray value of the previous pixel, x max To map the gray maximum of the pixel before x ij ' is the gray value of the mapped pixel;
feature extraction, including extracting edge features, color features, PCA features, and texture features; and classifying the characteristics by using an improved random forest method, and outputting the fault type of the communication tower.
2. The communication tower safety inspection method according to claim 1, wherein the step of histogram equalization of the image is as follows:
histogram of statistical image normalized to between [0,1 ]:
wherein H, W are the height and width of the image, n k Representing a gray value r k The number of pixels of (a);
calculating a mapping function:
s k for the converted gray value, T (r k ) For the mapping function, the calculation process uses a cumulative histogram;
and processing the image by using the mapping function.
3. The communication tower safety detection method according to claim 1, wherein the step of enhancing the image is as follows:
the image is enhanced using a Gamma transformation, the Gamma transformation formula being as follows:
output＝(input+ε) γ ；
in the formula, output is an output gray value, input is an input gray value, epsilon is a compensation coefficient, the value is generally 0, gamma is a gamma coefficient, and gamma is a positive number.
4. The communication tower safety detection method according to claim 1, wherein the sparseness percentage is defined as:
the compression ratio is the ratio between the original and compressed signal vector lengths, defined as follows:
n represents the length of the signal X, M is the dimension of the sensing matrix, i.e. the number of measurements M of the accurately reconstructed signal X.
5. The communication tower safety detection method according to claim 4, wherein the sparseness factor is defined as follows:
wherein a, b, c are experimentally obtained values, and a>b>c，a 1 And b 1 Is a preset threshold.
6. The method for detecting safety of communication towers according to claim 1, wherein the image features comprise edge features, PCA features and texture features;
the edge features are extracted by using a Sobel edge extraction operator;
PCA features are extracted by calling PCA functions in a deconvolution library of a machine learning tool package sklearn;
texture features are extracted through a local binary pattern operator LBP;
the color features are obtained by extracting color moment features, including first-order moment, second-order moment and third-order moment color moments.
7. The communication tower safety inspection method according to claim 1, wherein the step of performing fault recognition on the image using the modified random forest method comprises the steps of:
sampling by using a Bootstrap;
training the subset;
selecting a characteristic splitting node;
constructing a decision subtree set;
performing preliminary classification on the test sample set to obtain a preliminary classification set;
calculating the weight of the AUC and carrying out weighted voting;
and obtaining a fault classification result.
8. The method for detecting the safety of a communication tower according to claim 7, wherein the step of selecting the feature splitting node is as follows:
randomly selecting a portion of features X from all features 1 Further screening the feature set X from the part of the features by using a lasso regression method 2 Calculate these feature pairsThe sum of CMAs of the corresponding random decision trees;
repeating the above steps for N times to obtain N feature sets X 2，N Selecting a feature set X 2，N M feature sets X with highest sum of CMA 2，M ，M<And N, calculating the similarity among M feature sets, and selecting two feature sets with the lowest similarity as feature splitting nodes.
9. A pylon safety detection system using the method of any one of claims 1 to 8, comprising:
and an image acquisition module: acquiring a close-range image of the iron tower through unmanned aerial vehicle aerial photographing;
histogram equalization module: performing histogram equalization on the image;
an image enhancement module: enhancing the image;
and a filtering module: performing an improved weighted median filtering of the image;
the feature extraction module is used for extracting edge features, color features, PCA features and texture features;
and a fault classification module: and classifying the characteristics by using an improved random forest method, and outputting the fault type of the communication tower.
CN202310559241.9A 2023-05-18 2023-05-18 Communication tower safety detection method and system Active CN116468958B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310559241.9A CN116468958B (en) 2023-05-18 2023-05-18 Communication tower safety detection method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310559241.9A CN116468958B (en) 2023-05-18 2023-05-18 Communication tower safety detection method and system

Publications (2)

Publication Number Publication Date
CN116468958A CN116468958A (en) 2023-07-21
CN116468958B true CN116468958B (en) 2024-04-05

Family

ID=87179027

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310559241.9A Active CN116468958B (en) 2023-05-18 2023-05-18 Communication tower safety detection method and system

Country Status (1)

Country Link
CN (1) CN116468958B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5594807A (en) * 1994-12-22 1997-01-14 Siemens Medical Systems, Inc. System and method for adaptive filtering of images based on similarity between histograms
CN101179650A (en) * 2007-11-23 2008-05-14 西安交通大学 Video image contrast intensifying method and apparatus
CN105701835A (en) * 2016-02-26 2016-06-22 华北电力大学 Image edge detection method and system facing electric power facilities
CN107527333A (en) * 2017-07-31 2017-12-29 湖北工业大学 A kind of rapid image Enhancement Method based on gamma transformation
CN112949714A (en) * 2021-03-02 2021-06-11 北京城建设计发展集团股份有限公司 Fault possibility estimation method based on random forest

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9645377B2 (en) * 2015-02-06 2017-05-09 The Johns Hopkins University Compressive imaging systems and methods

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5594807A (en) * 1994-12-22 1997-01-14 Siemens Medical Systems, Inc. System and method for adaptive filtering of images based on similarity between histograms
CN101179650A (en) * 2007-11-23 2008-05-14 西安交通大学 Video image contrast intensifying method and apparatus
CN105701835A (en) * 2016-02-26 2016-06-22 华北电力大学 Image edge detection method and system facing electric power facilities
CN107527333A (en) * 2017-07-31 2017-12-29 湖北工业大学 A kind of rapid image Enhancement Method based on gamma transformation
CN112949714A (en) * 2021-03-02 2021-06-11 北京城建设计发展集团股份有限公司 Fault possibility estimation method based on random forest

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Michael Leigsnering ; Fauzia Ahmad ; Moeness G. Amin ; Abdelhak M. Zoubir.Compressive Sensing-Based Multipath Exploitation for Stationary and Moving Indoor Target Localization.《IEEE Journal of Selected Topics in Signal Processing 》.2015,1469 - 1483页. *
刘凯 ; 郑山红 ; 蒋权 ; 赵天傲.基于随机森林的自适应特征选择算法.《计算机技术与发展》.2018,第107-110页. *

Also Published As

Publication number Publication date
CN116468958A (en) 2023-07-21

Similar Documents

Publication Publication Date Title
Yousif et al. Improving urban change detection from multitemporal SAR images using PCA-NLM
Rajput et al. Comparative study of image enhancement techniques
Shen et al. A MAP-based algorithm for destriping and inpainting of remotely sensed images
Zheng et al. An advanced image fusion algorithm based on wavelet transform: incorporation with PCA and morphological processing
Kotwal et al. A novel approach to quantitative evaluation of hyperspectral image fusion techniques
CN109978848B (en) Method for detecting hard exudation in fundus image based on multi-light-source color constancy model
Ding et al. Image quality assessment using directional anisotropy structure measurement
CN111612741B (en) Accurate reference-free image quality evaluation method based on distortion recognition
CN111127387B (en) Quality evaluation method for reference-free image
KR101766536B1 (en) Method and apparatus of image denoising using multi-scale block region detection
WO2019049324A1 (en) Image processing device, image processing method and storage medium
CN111310771B (en) Road image extraction method, device and equipment of remote sensing image and storage medium
Morzelona Human visual system quality assessment in the images using the IQA model integrated with automated machine learning model
CN114255468A (en) Handwriting recognition method and related equipment thereof
Freitas et al. On the application LBP texture descriptors and its variants for no-reference image quality assessment
CN112633202A (en) Hyperspectral image classification algorithm based on dual denoising combined multi-scale superpixel dimension reduction
Lu et al. SAR image despeckling via structural sparse representation
CN116468958B (en) Communication tower safety detection method and system
CN110830043B (en) Image compressed sensing reconstruction method based on mixed weighted total variation and non-local low rank
Zehtabian et al. Adaptive restoration of multispectral datasets used for SVM classification
CN107085839B (en) SAR image speckle reduction method based on texture enhancement and sparse coding
Khmag et al. Natural image noise removal using non local means and hidden Markov models in stationary wavelet transform domain
CN113822361A (en) SAR image similarity measurement method and system based on Hamming distance
CN112766371B (en) High-resolution remote sensing image supervision and classification method based on few training samples
CN112541913B (en) Image local fuzzy detection and extraction method based on column rate spectral analysis and noise robustness

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