CN115660647A - Maintenance method for building outer wall - Google Patents
Maintenance method for building outer wall Download PDFInfo
- Publication number
- CN115660647A CN115660647A CN202211380269.8A CN202211380269A CN115660647A CN 115660647 A CN115660647 A CN 115660647A CN 202211380269 A CN202211380269 A CN 202211380269A CN 115660647 A CN115660647 A CN 115660647A
- Authority
- CN
- China
- Prior art keywords
- image
- building wall
- building
- cluster
- wall
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000012423 maintenance Methods 0.000 title claims abstract description 24
- 238000012544 monitoring process Methods 0.000 claims abstract description 34
- 238000012545 processing Methods 0.000 claims abstract description 11
- 238000007689 inspection Methods 0.000 claims description 17
- 238000013527 convolutional neural network Methods 0.000 claims description 14
- 238000001514 detection method Methods 0.000 claims description 14
- 238000012163 sequencing technique Methods 0.000 claims description 12
- 238000001914 filtration Methods 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 9
- 238000012216 screening Methods 0.000 claims description 7
- 238000009826 distribution Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 230000009467 reduction Effects 0.000 claims description 5
- 238000012937 correction Methods 0.000 claims description 4
- 238000003708 edge detection Methods 0.000 claims description 4
- 238000012935 Averaging Methods 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000005286 illumination Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 2
- 230000036541 health Effects 0.000 abstract description 4
- 230000010354 integration Effects 0.000 abstract description 3
- 238000012549 training Methods 0.000 description 6
- 230000007704 transition Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 238000011176 pooling Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 210000004205 output neuron Anatomy 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 238000005034 decoration Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Landscapes
- Image Processing (AREA)
Abstract
The method of the embodiment of the invention provides a maintenance method of a building outer wall, wherein the building outer wall image and position data shot by an unmanned aerial vehicle are sent to a task allocation server; the task allocation server preprocesses the building wall image to generate an image to be identified, generates a monitoring task and distributes the monitoring task to the distributed computing cluster; the distributed computing cluster carries out risk identification on the monitoring task and sends an identification result to the management server; and the management server respectively feeds back the maintenance terminal, the property terminal and the supervision terminal. The maintenance method of the building outer wall can be used for simultaneously operating different communities, improves the maintenance efficiency of the building outer wall, realizes the integration of computing resources, optimizes the efficiency of processing a large amount of data, realizes the linkage of all ends, realizes the timely maintenance, carries out the on-site protection and the timely supervision, avoids the occurrence of accidents of hurting people due to falling objects, and ensures the life health and property safety of people.
Description
[ technical field ] A
The invention relates to the technical field of buildings, in particular to a maintenance method of a building outer wall.
[ background ] A method for producing a semiconductor device
Because the building outer wall heat-insulating layer is architectural decoration exposed for a long time, the falling accident of the outer wall heat-insulating layer can be caused under the repeated action of severe environments such as wind, sunshine, rain, cold and hot and the like, the falling accident of objects at high altitude is caused, the life health and property loss of people are influenced, and negative social influence is caused.
The existing safety inspection of the building outer wall generally adopts a manual regular inspection mode. The manual inspection has great subjectivity, time and labor consumption in inspection, high cost and low working efficiency, and moreover, part of tall buildings or buildings with complex structures are not beneficial to the visual observation of inspectors, so that the possibility of missed inspection and false inspection is easily caused.
[ summary of the invention ]
In view of this, the embodiment of the present invention provides a method for maintaining an exterior wall of a building. The method comprises the following steps:
s1, receiving an image shooting instruction sent by an inspection end by an unmanned aerial vehicle, and sending shot building wall images and position data to a task allocation server;
s2, preprocessing the building wall image by the task distribution server to generate an image to be identified, generating a monitoring task and distributing the monitoring task to the distributed computing cluster;
s3, the distributed computing cluster carries out risk identification on the monitoring task and sends an identification result to the management server;
and S4, the management server sends the building wall image and the position data of the identified risk target to the maintenance terminal, grades the risk target, screens out the risk target and sends the risk target to the property terminal and the supervision terminal.
As to the above-mentioned aspect and any possible implementation manner, there is further provided an implementation manner, where the S1 specifically includes:
the unmanned aerial vehicle receives an image shooting instruction sent by the inspection end to shoot images, acquires current position data through the GPS module, and names and stores the shot building wall images according to the position data;
comparing the named building wall surface image with the previous image, if the position data is the same, judging whether the repeatability R and the definition D meet the requirement, and if the repeatability R is more than or equal to the repeatability threshold R 0 Deleting the building wall image; if degree of repetition R<Threshold value of degree of repetition R 0 And the definition D is more than or equal to the definition threshold D 0 Then add the serial number in the naming according to the time sequence; if degree of repetition R<Threshold value of degree of repetition R 0 And definition D<Definition threshold D 0 If the building wall image is deleted and a rephotograph request is sent to the checking end, when the definition D of the obtained rephotograph building wall image meets the requirement, the rephotograph building wall image is named by the position data and serial numbers are added in the naming according to the time sequence;
if the position data are different, judging whether the definition D meets the requirement, and if the definition D does not meet the requirement, judging whether the definition D meets the requirement<Definition threshold D 0 Deleting the building wall image and sending a rephotograph request to the checking end, if the definition D is more than or equal to the definition threshold D 0 Then, it is determined whether the distance L between the current position data and the previous position data is less than the distance threshold L 0 If the current building wall image and the previous building wall image are not the same, dividing the current building wall image and the previous building wall image into the same image group, otherwise, dividing the current building wall image and the previous building wall image into different image groups;
and compressing the same image group image and sending the compressed image to a task allocation server.
The above-described aspect and any possible implementation manner further provide an implementation manner, and the method for obtaining the repetition degree R includes:
selecting a named building wall image and a previous image, extracting and matching feature points through a Surf feature point detection algorithm, filtering out partial matching pairs through Hamming distance, and solving the matched feature points by using a least square method to obtain a homography matrix;
calculating a point set after the vertex of the first image is converted according to the homography matrix;
calculating a polygon intersection set by the named vertex set of the building wall image and the conversion point set of the previous image;
restoring the intersection of the polygons by using the homography matrix to the original point set of the last image so as to identify a repeated region;
and calculating the ratio of the total area of the repeated region to the named building wall surface image to obtain the repetition degree R.
The above-described aspect and any possible implementation manner further provide an implementation manner, and the method for obtaining the definition D includes:
carrying out Gaussian fuzzy denoising processing on the building wall surface image to be calculated to obtain a denoised building wall surface image, and carrying out Gaussian filtering according to the following formula:
wherein, I (x, y) represents an input image, w (m, n) represents a template of Gaussian filtering, the size of the template is a x b, and f (x, y) represents an output image;
converting the output image into a gray scale image, and performing fuzzy detection by using a Laplace operator;
carrying out histogram equalization, and carrying out normalized mapping on the histogram to the range of 0-255 gray levels;
and obtaining the definition D by averaging.
The above aspect and any possible implementation manner further provide an implementation manner, where the preprocessing is performed on the building wall image by the task allocation server in S2 to generate an image to be recognized, and specifically includes:
and the task allocation server processes the building wall image by adopting an image normalization method, and then generates an image to be identified by sequentially adopting histogram equalization, illumination correction, noise reduction and sharpening.
As to the above-mentioned aspect and any possible implementation manner, an implementation manner is further provided, where generating a monitoring task in S2 and distributing the monitoring task to a distributed computing cluster specifically includes:
grouping images to be identified based on different unmanned aerial vehicles serving as image sources to generate different monitoring tasks, and sequencing the monitoring tasks according to the number of the images to be identified from large to small;
calculating the computing resource state S of all cluster servers of the distributed computing cluster at a preset time interval T, and when the computing resource state S is less than a lower limit threshold value S 1 If so, taking the corresponding cluster server as an idle cluster server;
and acquiring idle cluster servers, sequencing the idle cluster servers from small to large according to the computing resource state S, and distributing the monitoring tasks to the idle cluster servers one by one in a circular traversal mode.
The above-described aspects and any possible implementations further provide an implementation, and the method further includes:
calculating the computing resource state S of all cluster servers of the distributed computing cluster within a preset time interval T and when T/2 is reached, and when the computing resource state S is larger than or equal to an upper limit threshold value S 2 If so, taking the corresponding cluster server as an overload cluster server;
and acquiring idle cluster servers and overloaded cluster servers, sequencing the overloaded cluster servers from large to small according to the computing resource state S, sequencing the idle cluster servers from small to large according to the computing resource state S, and transferring the monitoring tasks which are not processed by the overloaded cluster servers to the idle cluster servers one by one in a circular traversal mode.
The above-mentioned aspect and any possible implementation manner further provide an implementation manner, and a calculation formula of the calculation resource state S is as follows:
V ij ＝U 1j +U 2j +U 3j ，
wherein S is i For the computing resource status of the ith cluster server,
is the average of the computing resources of all cluster servers, n is the number of cluster servers,
is the average value of the computing resources of the m virtual machines of the ith cluster server, V ij For the computing resource of the jth virtual machine in the ith cluster server, U 1j 、U 2j 、U 3j Respectively is the CPU utilization rate, the memory utilization rate and the bandwidth utilization rate of the jth virtual machine.
As to the above-mentioned aspect and any possible implementation manner, an implementation manner is further provided, where the performing, by the distributed computing cluster, risk identification on the image to be identified in S3 specifically includes:
converting an image to be identified into a YCbCr color space, and performing wall surface detection by using a pre-constructed wall surface color model;
if the wall surface to be confirmed is detected, counting the number of pixels of the wall surface to be confirmed;
calculating the proportion K of the wall surface to be confirmed in the image to be recognized, wherein the proportion K is calculated according to the following formula:
wherein, P 1 Indicating the number of pixels, P, of the wall to be checked 2 Representing the number of pixels of the wall surface which is not to be confirmed in the image to be identified,p represents the total pixel number of the image to be recognized;
when the ratio K is more than or equal to the ratio threshold K 0 Determining that the building wall surface exists;
obtaining an area image of a building wall surface from an image to be identified by using pixel segmentation;
converting the area image into a gray image, and carrying out crack identification through a trained convolutional neural network model;
and carrying out nonlinear screening on the identified cracks, wherein the screening result is a risk target.
In the above aspect and any possible implementation manner, an implementation manner is further provided, in S4, the ranking is performed on the risk targets, the risk targets are screened out, and the risk targets are sent to the property terminal and the monitoring terminal, specifically including:
counting the number of pixel points contained in the dangerous target in the image to be identified, and marking the image to be identified as an image containing the dangerous target when the number of the pixel points is greater than a preset threshold value;
detecting the edge of the image in the area by using a Canny edge detection operator, identifying a closed contour, and marking the image to be identified as an image containing a dangerous target if the closed contour is overlapped with the dangerous target;
and sending the image and the position data containing the dangerous target to a property terminal and a supervision terminal.
One of the above technical solutions has the following beneficial effects:
the method provided by the embodiment of the invention can be used for simultaneously operating different communities, improving the maintenance efficiency of the outer wall of the building, realizing the integration of computing resources, optimizing the efficiency of processing a large amount of data, realizing the linkage of all ends, realizing the timely maintenance, timely carrying out field protection and timely supervision, avoiding the occurrence of accidents of hurting people due to falling objects, and ensuring the life health and property safety of people.
[ description of the drawings ]
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive labor.
FIG. 1 is a schematic flow chart of a maintenance method for an exterior wall of a building according to an embodiment of the present invention;
FIG. 2 is a schematic flow chart of S1 provided in the embodiment of the present invention;
fig. 3 is a schematic flow chart of obtaining the repetition degree R according to the embodiment of the present invention;
FIG. 4 is a schematic flow chart of sharpness D acquisition provided by an embodiment of the present invention;
FIG. 5 is a schematic flow chart of a portion S2 provided in an embodiment of the present invention;
FIG. 6 is a schematic flow chart of a portion S3 provided in an embodiment of the present invention;
fig. 7 is a schematic flow chart of part S4 according to an embodiment of the present invention.
[ detailed description ] embodiments
In order to make the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to specific embodiments and corresponding drawings. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Please refer to fig. 1, which is a schematic flow chart illustrating a maintenance method for an exterior wall of a building according to an embodiment of the present invention. As shown in fig. 1, the method comprises the steps of:
s1, an unmanned aerial vehicle receives an image shooting instruction sent by an inspection end and sends shot building wall images and position data to a task distribution server;
s2, preprocessing the building wall image by the task distribution server to generate an image to be identified, generating a monitoring task and distributing the monitoring task to the distributed computing cluster;
s3, carrying out risk identification on the monitoring task by the distributed computing cluster, and sending an identification result to the management server;
and S4, the management server sends the building wall image and the position data of the identified risk target to the maintenance terminal, grades the risk target, screens out the risk target and sends the risk target to the property terminal and the supervision terminal.
The method provided by the embodiment of the invention has the advantages that the linkage of all the ends is realized, the unmanned aerial vehicle receives the building wall images shot by the image shooting instruction sent by the inspection end, the inspection end controls the unmanned aerial vehicle to shoot the images with low cost and safety, a plurality of unmanned aerial vehicles and a plurality of inspection ends can simultaneously operate in different cells, the maintenance efficiency of the building outer wall is improved, the unmanned aerial vehicle carries the shooting module and the positioning module, and the inspection end can be a mobile terminal such as a mobile phone or a computer for controlling the unmanned aerial vehicle. The task allocation server preprocesses the building wall images to generate images to be identified, generates monitoring tasks and distributes the monitoring tasks to the distributed computing clusters, the distributed computing clusters perform risk identification on the monitoring tasks and send identification results to the management server, the risk identification accuracy is improved through image preprocessing, the distributed computing clusters achieve integration of computing resources, and the efficiency of processing a large amount of data is optimized. The management server sends building wall images and position data for identifying the risk targets to the maintenance terminal, the risk targets are graded, the risk targets are screened out and sent to the property terminal and the supervision terminal, maintenance personnel can acquire messages and maintain in time by timely informing the maintenance terminal, on-site protection is timely carried out by linkage of the property terminal and the supervision terminal, the occurrence of an event that people are injured by falling objects is avoided, life health and property safety of people are guaranteed, and the maintenance personnel are timely urged to maintain.
In a preferred embodiment of the present invention, as shown in fig. 2, S1 specifically includes:
the unmanned aerial vehicle receives an image shooting instruction sent by the inspection end to shoot images, acquires current position data through the GPS module, and names and stores shot building wall images according to the position data;
comparing the named building wall surface image with the previous image, if the position data is the same, judging whether the repeatability R and the definition D meet the requirement, and if the repeatability R is more than or equal to the repeatability threshold R 0 Deleting the building wall image; if degree of repetition R<Threshold value of degree of repetition R 0 And the definition D is more than or equal to the definition threshold D 0 Then add the serial number in the naming according to the time sequence; if degree of repetition R<Threshold value of degree of repetition R 0 And definition D<Definition threshold D 0 If the definition D of the obtained rephotographed building wall image meets the requirement, naming the rephotographed building wall image by using the position data and adding a serial number in the naming according to the time sequence;
if the position data are different, judging whether the definition D meets the requirement, and if the definition D does not meet the requirement, judging whether the definition D meets the requirement<Definition threshold D 0 Deleting the building wall image and sending a rephotograph request to the checking end, if the definition D is more than or equal to the definition threshold D 0 Then, it is determined whether the distance L between the current position data and the previous position data is less than the distance threshold L 0 If the current building wall image and the previous building wall image are not the same, dividing the current building wall image and the previous building wall image into the same image group, otherwise, dividing the current building wall image and the previous building wall image into different image groups;
and compressing the same image group image and sending the compressed image to a task allocation server.
Specifically, as shown in fig. 3, the method for obtaining the repetition degree R is as follows:
selecting a named building wall image and a previous image, extracting and matching feature points through a Surf feature point detection algorithm, filtering out partial matching pairs through Hamming distance, and solving the matched feature points by using a least square method to obtain a homography matrix;
calculating a point set after the vertex of the first image is converted according to the homography matrix;
calculating a polygon intersection set by the named vertex set of the building wall image and the conversion point set of the previous image;
restoring the intersection of the polygons to the original point set of the last image by using the homography matrix so as to identify a repeated region;
and calculating the ratio of the total area of the repeated region to the named building wall surface image to obtain the repetition degree R.
Specifically, as shown in fig. 4, the method for obtaining the definition D is as follows:
carrying out Gaussian fuzzy denoising processing on the building wall surface image to be calculated to obtain a denoised building wall surface image, and carrying out Gaussian filtering according to the following formula:
wherein, I (x, y) represents an input image, w (m, n) represents a template of Gaussian filtering, the size of the template is a × b, and f (x, y) represents an output image;
converting the output image into a gray scale image, and performing fuzzy detection by using a Laplacian operator;
histogram equalization is carried out, and the histogram is normalized and mapped into the range of 0-255 gray levels;
and obtaining the definition D by averaging.
According to the invention, through the judgment of the repeatability R and the definition D, the feedback of the unmanned aerial vehicle to the inspection end when the unmanned aerial vehicle collects the images which do not meet the requirements is realized, so that the shot building wall surface images meet the requirements, and the guarantee is provided for the accurate risk detection of the later building wall surface. The images are named by the position data, so that the target building wall surface can be accurately traced. The building wall surfaces of different communities have uniformity, and the building wall surface images are processed in a grouping mode, so that corresponding relations between different groups of building wall surface images and different buildings of different communities are established, the accuracy of the same type of outer wall identification and the data processing speed are improved, and information searching and processing are facilitated in later maintenance and protection work.
In a preferred embodiment of the present invention, the task allocation server in S2 preprocesses the building wall image to generate an image to be recognized, and specifically includes:
the task allocation server processes the building wall images by adopting an image normalization method, and then generates the images to be identified by sequentially adopting histogram equalization, illumination correction, noise reduction and sharpening.
Among them, histogram equalization enhances the contrast of an image to show more details, light correction removes shadows or adjusts dark areas, noise reduction removes unnecessary pixels by using a smoothing filter to ensure that details are recognizable, and the contrast is further enhanced by sharpening.
In a preferred embodiment of the present invention, as shown in fig. 5, generating a monitoring task in S2 and distributing the monitoring task to a distributed computing cluster specifically includes:
grouping images to be identified based on different unmanned aerial vehicles serving as image sources to generate different monitoring tasks, and sequencing the monitoring tasks according to the number of the images to be identified from large to small;
calculating the computing resource states S of all cluster servers of the distributed computing cluster at preset time intervals T, and when the computing resource states S are smaller than a lower limit threshold S 1 If so, taking the corresponding cluster server as an idle cluster server;
and acquiring idle cluster servers, sequencing the idle cluster servers from small to large according to the computing resource state S, and distributing the monitoring tasks to the idle cluster servers one by one in a circular traversal mode.
In addition, S2 further includes:
calculating the computing resource state S of all cluster servers of the distributed computing cluster within a preset time interval T and when T/2 is reached, and when the computing resource state S is more than or equal to an upper limit threshold S 2 If so, taking the corresponding cluster server as an overload cluster server;
and acquiring idle cluster servers and overloaded cluster servers, sequencing the overloaded cluster servers from large to small according to the computing resource state S, sequencing the idle cluster servers from small to large according to the computing resource state S, and transferring the monitoring tasks which are not processed by the overloaded cluster servers to the idle cluster servers one by one in a circular traversal mode.
Specifically, the calculation formula for calculating the resource state S is as follows:
V ij ＝U 1j +U 2j +U 3j ，
wherein S is i The computing resource status for the ith cluster server,
is the average of the computing resources of all cluster servers, n is the number of cluster servers,
is the average value of the computing resources of the m virtual machines of the ith cluster server, V ij Is the computing resource of the jth virtual machine in the ith cluster server, U 1j 、U 2j 、U 3j Respectively is the CPU utilization rate, the memory utilization rate and the bandwidth utilization rate of the jth virtual machine.
By optimizing the task allocation and task transfer mechanism, the optimization processing of computing resources is realized, and the data processing efficiency is improved.
In a preferred embodiment of the present invention, as shown in fig. 6, the performing risk identification on the image to be identified by the distributed computing cluster in S3 specifically includes:
converting an image to be identified into a YCbCr color space, and performing wall surface detection by using a pre-constructed wall surface color model;
if the wall surface to be confirmed is detected, counting the number of pixels of the wall surface to be confirmed;
calculating the occupation ratio K of the wall surface to be confirmed in the image to be recognized, wherein the occupation ratio K is calculated according to the following formula:
wherein, P 1 Number of pixels, P, representing wall to be identified 2 Representing the number of pixels on the wall surface which is not to be confirmed in the image to be recognized, wherein P represents the total number of pixels of the image to be recognized;
when the ratio K is more than or equal to the ratio threshold K 0 Determining that the building wall surface exists;
obtaining an area image of a building wall surface from an image to be identified by using pixel segmentation;
converting the area image into a gray image, and identifying cracks through a trained convolutional neural network model;
and carrying out nonlinear screening on the identified cracks, wherein the screening result is a risk target.
The invention utilizes the wall color model which is constructed in advance to carry out wall detection, so that the wall detection method is efficient and rapid, the accuracy of wall identification is further ensured through wall proportion confirmation, the crack identification is carried out through the trained convolutional neural network model, the identification accuracy is realized, and the non-linear screening can eliminate the original line element interference of the wall.
The method for constructing the wall color model comprises the following steps: the method comprises the steps of collecting a large number of images containing the building wall surface, counting the probability that various YCbCr color values in the images respectively appear as the building wall surface and the probability of a non-building wall surface, calculating the probability value of the building wall surface corresponding to each YCbCr color value according to a Bayesian rule, and selecting a building wall surface threshold value; for each pixel in the image, if the building wall probability of the color value of the pixel point is greater than a building wall threshold value, judging that the pixel point is a building wall pixel, otherwise, judging that the pixel point is not a building wall pixel; selecting a building wall surface threshold value to enable the ratio of the positive detection rate to the false detection rate of the building wall surface to be maximum; and the building wall surface probability and the building wall surface threshold value corresponding to each color value in the YCbCr color space form a wall surface color model.
The training method of the convolutional neural network model comprises the following steps:
acquiring a large number of building wall photos with cracks and building wall photos without cracks, converting the building wall photos into gray images after pretreatment, and distributing different labels for the building wall photos and the building wall photos;
randomly extracting the photos and the labels to form a training set, and using the rest 10 percent of the photos and the labels to a test set;
constructing a convolutional neural network model, wherein the first layer of the convolutional neural network model is an input layer and is used for receiving input images for classification, and the sizes of the convolutional neural network model are unified to be 256 multiplied by 3; the second layer is a convolution layer, and feature extraction is carried out, convolution kernel is 3 x 3, step length is 1, and a 126 x 64 feature map is obtained; the third layer is a maximum pooling layer, dimension reduction processing is carried out on the feature map after convolution of the convolutional layer, feature extraction is carried out by adopting a self-adaptive pooling method, the rectangular area is 3 multiplied by 3, the step length is 2, and a feature map of 64 multiplied by 64 is obtained; the fourth layer is a first dense block, and the size of an output feature map is 64 multiplied by 256; the fifth layer is a first transition block, and the size of the output characteristic diagram is 32 multiplied by 128; the sixth layer is a second dense block, and the size of the output feature map is 32 multiplied by 512; the seventh layer is a second transition block, and the size of the output characteristic diagram is 16 multiplied by 256; the eighth layer is a third dense block, and the size of an output feature map is 16 multiplied by 1024; the ninth layer is a third transition block, and the size of the output characteristic diagram is 8 multiplied by 512; the tenth layer is a fourth dense block, the size of an output characteristic diagram is 8 multiplied by 1024, the eleventh layer is a maximal pooling layer and outputs a one-dimensional vector, and the length is 3072; the twelfth layer is a full connection layer, comprises 512 ResNet full connection neuron nodes and is connected with all the features output by the eleventh layer; the thirteenth layer is an output layer and comprises 2 output neuron nodes using a softmax function, the output neuron nodes are fully connected with neurons of the full connection layer, output results are classified in a second mode, and building wall photos with cracks and building wall photos without cracks are identified;
wherein the basic components in the dense blocks are composed of 1 × 1 convolutional layers and 3 × 3 convolutional layers, and the first, second, third and fourth dense blocks have 6, 12, 24 and 16 basic components, respectively; the first transition block, the second transition block and the third transition block are respectively composed of a 1 × 1 convolution layer and a 2 × 2 average pooling layer, and the step length is 2;
the activation function of each convolution layer is ReLU;
during training, putting training sample pictures of a training set into an input layer, putting label vectors into an output layer, completing training of a convolutional neural network model, and measuring the accuracy of the trained convolutional neural network model through a test set;
and converting the area image into a gray image, and carrying out crack identification through a trained convolutional neural network model.
The convolutional neural network model has high selection and classification capability and can accurately and quickly identify cracks.
It should be noted that, other convolutional neural network models capable of identifying cracks can also be adopted in the convolutional neural network model of the present invention, as required.
In a preferred embodiment of the present invention, as shown in fig. 7, in S4, the risk targets are ranked, screened out, and sent to the property terminal and the monitoring terminal, which specifically includes:
counting the number of pixel points contained in the dangerous target in the image to be identified, and marking the image to be identified as an image containing the dangerous target when the number of the pixel points is greater than a preset threshold value;
detecting the edge of the image in the area by using a Canny edge detection operator, identifying a closed contour, and marking the image to be identified as an image containing a dangerous target if the closed contour is overlapped with the dangerous target;
and sending the image and the position data containing the dangerous target to a property terminal and a supervision terminal.
When the number of pixel points is larger than a preset threshold value, long cracks or complex cracks exist, so that the possibility of falling off of the wall surface is high, and the image to be identified is marked as an image containing a dangerous target; and detecting the edge of the image in the region by using a Canny edge detection operator, identifying a closed contour, if the closed contour is overlapped with a risk target, doubly identifying to eliminate interference factors, indicating that a falling wall surface or a bulge exists, and a wall skin or a heat preservation layer has a falling risk at any time, so that the image to be identified is marked as an image containing the risk target.
The above description is only an example of the present invention, and is not intended to limit the present invention. Various modifications and alterations to this invention will become apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.
Claims (10)
1. A method of maintaining an exterior wall of a building, the method comprising:
s1, receiving an image shooting instruction sent by an inspection end by an unmanned aerial vehicle, and sending shot building wall images and position data to a task allocation server;
s2, preprocessing the building wall image by the task distribution server to generate an image to be identified, generating a monitoring task and distributing the monitoring task to the distributed computing cluster;
s3, carrying out risk identification on the monitoring task by the distributed computing cluster, and sending an identification result to the management server;
and S4, the management server sends the building wall image and the position data of the identified risk target to the maintenance terminal, grades the risk target, screens out the risk target and sends the risk target to the property terminal and the supervision terminal.
2. The method for maintaining the exterior wall of the building according to claim 1, wherein the S1 specifically comprises:
the unmanned aerial vehicle receives an image shooting instruction sent by the inspection end to shoot images, acquires current position data through the GPS module, and names and stores shot building wall images according to the position data;
comparing the named building wall surface image with the previous image, if the position data is the same, judging whether the repeatability R and the definition D meet the requirement, and if the repeatability R is more than or equal to the repeatability threshold R 0 Deleting the building wall image; if degree of repetition R<Threshold value of degree of repetition R 0 And the definition D is more than or equal to the definition threshold D 0 Then add the serial number in the naming according to the time sequence; if degree of repetition R<Threshold value of degree of repetition R 0 And definition D<Definition threshold D 0 If the definition D of the obtained rephotographed building wall image meets the requirement, naming the rephotographed building wall image by using the position data and adding a serial number in the naming according to the time sequence;
if the position data are different, judging whether the definition D meets the requirement, if so, judging whether the definition D meets the requirement<Definition threshold D 0 Deleting the building wall image and sending a rephotograph request to the checking end, if the definition D is more than or equal to the definition threshold D 0 Then, it is determined whether the distance L between the current position data and the previous position data is less than the distance threshold L 0 If the current building wall image and the previous building wall image are not the same, dividing the current building wall image and the previous building wall image into the same image group, otherwise, dividing the current building wall image and the previous building wall image into different image groups;
and compressing the same image group image and sending the compressed image to a task distribution server.
3. The method for maintaining the exterior wall of the building as claimed in claim 2, wherein the repetition degree R is obtained by:
selecting a named building wall image and a previous image, extracting and matching feature points through a Surf feature point detection algorithm, filtering out partial matching pairs through Hamming distance, and solving the matched feature points by using a least square method to obtain a homography matrix;
calculating a point set after the vertex of the last image is converted according to the homography matrix;
calculating a polygon intersection by using the named vertex set of the building wall image and the conversion point set of the previous image;
restoring the intersection of the polygons to the original point set of the last image by using the homography matrix so as to identify a repeated region;
and calculating the ratio of the total area of the repeated region to the named building wall surface image to obtain the repetition degree R.
4. The method for maintaining the exterior wall of the building according to claim 2, wherein the definition D is obtained by the following steps:
carrying out Gaussian fuzzy denoising processing on the building wall surface image to be calculated to obtain a denoised building wall surface image, and carrying out Gaussian filtering according to the following formula:
wherein, I (x, y) represents an input image, w (m, n) represents a template of Gaussian filtering, the size of the template is a x b, and f (x, y) represents an output image;
converting the output image into a gray scale image, and performing fuzzy detection by using a Laplace operator;
carrying out histogram equalization, and carrying out normalized mapping on the histogram to the range of 0-255 gray levels;
and obtaining the definition D by averaging.
5. The maintenance method of the building outer wall according to claim 1, wherein the step S2 of preprocessing the building wall image by the task distribution server to generate an image to be recognized specifically comprises:
the task allocation server processes the building wall images by adopting an image normalization method, and then generates the images to be identified by sequentially adopting histogram equalization, illumination correction, noise reduction and sharpening.
6. The method for maintaining the external wall of the building according to claim 5, wherein the generating of the monitoring task in the S2 and the distributing to the distributed computing cluster specifically include:
grouping images to be identified based on different unmanned aerial vehicles serving as image sources to generate different monitoring tasks, and sequencing the monitoring tasks according to the number of the images to be identified from large to small;
calculating the computing resource state S of all cluster servers of the distributed computing cluster at a preset time interval T, and when the computing resource state S is less than a lower limit threshold value S 1 If so, taking the corresponding cluster server as an idle cluster server;
and acquiring idle cluster servers, sequencing the idle cluster servers from small to large according to the computing resource state S, and distributing the monitoring tasks to the idle cluster servers one by one in a circular traversal mode.
7. The method of maintaining an exterior wall of a building of claim 6, further comprising:
calculating the computing resource state S of all cluster servers of the distributed computing cluster within a preset time interval T and when T/2 is reached, and when the computing resource state S is larger than or equal to an upper limit threshold value S 2 If so, taking the corresponding cluster server as an overload cluster server;
and acquiring idle cluster servers and overloaded cluster servers, sequencing the overloaded cluster servers from large to small according to the computing resource state S, sequencing the idle cluster servers from small to large according to the computing resource state S, and transferring the monitoring tasks which are not processed by the overloaded cluster servers to the idle cluster servers one by one in a circular traversal mode.
8. The method for maintaining the exterior wall of the building according to claim 7, wherein the calculation formula of the calculation resource state S is as follows:
V ij ＝U 1j +U 2j +U 3j ，
wherein S is i The computing resource status for the ith cluster server,
is the average of the computing resources of all cluster servers, n is the number of cluster servers,
is the average value of the computing resources of the m virtual machines of the ith cluster server, V ij For the computing resource of the jth virtual machine in the ith cluster server, U 1j 、U 2j 、U 3j Respectively is the CPU utilization rate, the memory utilization rate and the bandwidth utilization rate of the jth virtual machine.
9. The maintenance method for the external wall of the building according to any one of claims 1 or 8, wherein the risk identification is performed on the image to be identified by the distributed computing cluster in S3, and specifically comprises:
converting an image to be identified into a YCbCr color space, and performing wall surface detection by using a pre-constructed wall surface color model;
if the wall surface to be confirmed is detected, counting the number of pixels of the wall surface to be confirmed;
calculating the occupation ratio K of the wall surface to be confirmed in the image to be recognized, wherein the occupation ratio K is calculated according to the following formula:
wherein, P 1 Number of pixels, P, representing wall to be identified 2 Representing the number of pixels of a non-to-be-confirmed wall surface in the image to be recognized, wherein P represents the total number of pixels of the image to be recognized;
when the ratio K is more than or equal to the ratio threshold K 0 Determining that the building wall surface exists;
obtaining an area image of a building wall surface from an image to be identified by using pixel segmentation;
converting the area image into a gray image, and carrying out crack identification through a trained convolutional neural network model;
and carrying out nonlinear screening on the identified cracks, wherein the screening result is a risk target.
10. The maintenance method of the building outer wall according to claim 9, wherein in S4, the risk targets are ranked, screened out and sent to a property terminal and a supervision terminal, and specifically comprises:
counting the number of pixel points contained in the dangerous target in the image to be identified, and marking the image to be identified as an image containing the dangerous target when the number of the pixel points is greater than a preset threshold value;
detecting the edge of the image in the area by using a Canny edge detection operator, identifying a closed contour, and marking the image to be identified as an image containing a dangerous target if the closed contour is overlapped with the dangerous target;
and sending the image and the position data containing the dangerous target to a property terminal and a supervision terminal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211380269.8A CN115660647A (en) | 2022-11-05 | 2022-11-05 | Maintenance method for building outer wall |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211380269.8A CN115660647A (en) | 2022-11-05 | 2022-11-05 | Maintenance method for building outer wall |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115660647A true CN115660647A (en) | 2023-01-31 |
Family
ID=85016902
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211380269.8A Pending CN115660647A (en) | 2022-11-05 | 2022-11-05 | Maintenance method for building outer wall |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115660647A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116661479A (en) * | 2023-07-28 | 2023-08-29 | 深圳市城市公共安全技术研究院有限公司 | Building inspection path planning method, equipment and readable storage medium |
| CN117495356A (en) * | 2023-12-18 | 2024-02-02 | 成都秦川物联网科技股份有限公司 | Intelligent gas pipeline maintenance method based on man-machine linkage and Internet of things system |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109670515A (en) * | 2018-12-13 | 2019-04-23 | 南京工业大学 | A kind of detection method and system changed for building in unmanned plane image |
| CN110322442A (en) * | 2019-07-11 | 2019-10-11 | 福州大学 | A kind of building surface crack detecting method based on SegNet |
| CN211347977U (en) * | 2019-10-30 | 2020-08-25 | 常州市建筑科学研究院集团股份有限公司 | Device for detecting quality of functional layer of building outer wall in unmanned aerial vehicle manner |
| KR102237096B1 (en) * | 2020-04-27 | 2021-04-08 | 주식회사 에이치에스씨앤디 | Building crack maintenance method using unmanned flying device and flight plan |
| CN112734739A (en) * | 2021-01-18 | 2021-04-30 | 福州大学 | Visual building crack identification method based on attention mechanism and ResNet fusion |
| CN114235814A (en) * | 2021-12-02 | 2022-03-25 | 福州市建筑科学研究院有限公司 | Crack identification method for building glass curtain wall |
| CN114705689A (en) * | 2022-04-14 | 2022-07-05 | 西安建筑科技大学 | Unmanned aerial vehicle-based method and system for detecting cracks of outer vertical face of building |
| CN115002113A (en) * | 2022-05-26 | 2022-09-02 | 南京邮电大学 | Mobile base station edge computing power resource scheduling method, system and electronic equipment |
-
2022
- 2022-11-05 CN CN202211380269.8A patent/CN115660647A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109670515A (en) * | 2018-12-13 | 2019-04-23 | 南京工业大学 | A kind of detection method and system changed for building in unmanned plane image |
| CN110322442A (en) * | 2019-07-11 | 2019-10-11 | 福州大学 | A kind of building surface crack detecting method based on SegNet |
| CN211347977U (en) * | 2019-10-30 | 2020-08-25 | 常州市建筑科学研究院集团股份有限公司 | Device for detecting quality of functional layer of building outer wall in unmanned aerial vehicle manner |
| KR102237096B1 (en) * | 2020-04-27 | 2021-04-08 | 주식회사 에이치에스씨앤디 | Building crack maintenance method using unmanned flying device and flight plan |
| CN112734739A (en) * | 2021-01-18 | 2021-04-30 | 福州大学 | Visual building crack identification method based on attention mechanism and ResNet fusion |
| CN114235814A (en) * | 2021-12-02 | 2022-03-25 | 福州市建筑科学研究院有限公司 | Crack identification method for building glass curtain wall |
| CN114705689A (en) * | 2022-04-14 | 2022-07-05 | 西安建筑科技大学 | Unmanned aerial vehicle-based method and system for detecting cracks of outer vertical face of building |
| CN115002113A (en) * | 2022-05-26 | 2022-09-02 | 南京邮电大学 | Mobile base station edge computing power resource scheduling method, system and electronic equipment |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116661479A (en) * | 2023-07-28 | 2023-08-29 | 深圳市城市公共安全技术研究院有限公司 | Building inspection path planning method, equipment and readable storage medium |
| CN116661479B (en) * | 2023-07-28 | 2023-11-07 | 深圳市城市公共安全技术研究院有限公司 | Building inspection path planning method, equipment and readable storage medium |
| CN117495356A (en) * | 2023-12-18 | 2024-02-02 | 成都秦川物联网科技股份有限公司 | Intelligent gas pipeline maintenance method based on man-machine linkage and Internet of things system |
| CN117495356B (en) * | 2023-12-18 | 2024-03-26 | 成都秦川物联网科技股份有限公司 | Intelligent gas pipeline maintenance method based on man-machine linkage and Internet of things system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111784685B (en) | Power transmission line defect image identification method based on cloud edge cooperative detection | |
| CN115660647A (en) | Maintenance method for building outer wall | |
| CN110148130B (en) | Method and device for detecting part defects | |
| CN108038846A (en) | Transmission line equipment image defect detection method and system based on multilayer convolutional neural networks | |
| CN114092389A (en) | Glass panel surface defect detection method based on small sample learning | |
| CN110136210B (en) | Intelligent early warning system for leakage of monitoring device | |
| CN112949704B (en) | Tobacco leaf maturity state identification method and device based on image analysis | |
| CN112367400B (en) | Intelligent inspection method and system for power internet of things with edge cloud coordination | |
| CN110503637B (en) | Road crack automatic detection method based on convolutional neural network | |
| CN115131283A (en) | Defect detection and model training method, device, equipment and medium for target object | |
| CN112241679A (en) | Automatic garbage classification method | |
| CN110852164A (en) | YOLOv 3-based method and system for automatically detecting illegal building | |
| CN116612098B (en) | Insulator RTV spraying quality evaluation method and device based on image processing | |
| CN111540203B (en) | Method for adjusting green light passing time based on fast-RCNN | |
| CN115147488B (en) | Workpiece pose estimation method and grabbing system based on dense prediction | |
| CN114998251A (en) | Air multi-vision platform ground anomaly detection method based on federal learning | |
| CN115311623A (en) | Equipment oil leakage detection method and system based on infrared thermal imaging | |
| CN114241338A (en) | Building measuring method, device, equipment and storage medium based on image recognition | |
| CN117078670B (en) | Production control system of cloud photo frame | |
| CN110765900B (en) | Automatic detection illegal building method and system based on DSSD | |
| CN115849202B (en) | Intelligent crane operation target identification method based on digital twin technology | |
| CN115479946A (en) | Pavement damage detection method, system, device and storage medium | |
| CN115660648A (en) | Building outer wall maintenance system | |
| CN115601684A (en) | Emergency early warning method and device, electronic equipment and storage medium | |
| CN113705442A (en) | Outdoor large-board advertising picture monitoring and identifying system and method |
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 |