CN117292266B - Method and device for detecting concrete cracks of main canal of irrigation area and storage medium - Google Patents
Method and device for detecting concrete cracks of main canal of irrigation area and storage medium Download PDFInfo
- Publication number
- CN117292266B CN117292266B CN202311576460.4A CN202311576460A CN117292266B CN 117292266 B CN117292266 B CN 117292266B CN 202311576460 A CN202311576460 A CN 202311576460A CN 117292266 B CN117292266 B CN 117292266B
- Authority
- CN
- China
- Prior art keywords
- image
- crack
- main canal
- area
- concrete
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000003973 irrigation Methods 0.000 title claims abstract description 63
- 230000002262 irrigation Effects 0.000 title claims abstract description 63
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000003860 storage Methods 0.000 title claims abstract description 11
- 238000001514 detection method Methods 0.000 claims abstract description 63
- 238000013136 deep learning model Methods 0.000 claims abstract description 32
- 230000006870 function Effects 0.000 claims description 31
- 238000012549 training Methods 0.000 claims description 19
- 238000007781 pre-processing Methods 0.000 claims description 17
- 238000013135 deep learning Methods 0.000 claims description 16
- 238000010586 diagram Methods 0.000 claims description 14
- 238000005381 potential energy Methods 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 12
- 238000000605 extraction Methods 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 10
- 238000005457 optimization Methods 0.000 claims description 7
- 230000009467 reduction Effects 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 2
- 238000004590 computer program Methods 0.000 description 7
- 230000011218 segmentation Effects 0.000 description 5
- 230000001276 controlling effect Effects 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000012271 agricultural production Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012806 monitoring device Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- 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
- G06V10/52—Scale-space analysis, e.g. wavelet analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06V10/7715—Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method, a device and a storage medium for detecting cracks of main canal concrete in a pouring area, wherein the method constructs a deep learning model for detecting the areas of the cracks, the model takes a main canal concrete image in the pouring area as input, a crack detection result as output, and the model adopts multi-path depth direction banded convolution to extract multi-scale characteristic information of an input image; inputting the real-time acquired main canal concrete image of the irrigation area into the deep learning model to obtain an irrigation area main canal concrete crack detection result; and then optimizing by adopting a random field to obtain a final crack detection result. The invention can accurately identify the concrete crack condition in the main channel and avoid the occurrence of false alarm or missing alarm.
Description
Technical Field
The invention relates to a method and a device for detecting concrete cracks of a main canal of a irrigated area and a storage medium, and belongs to the technical field of safety monitoring of facilities of the irrigated area.
Background
Irrigation area main channels are an important component of irrigation systems and play a vital role in agricultural irrigation. The importance of the main canal concrete crack detection in the irrigation area is reflected in the aspects of guaranteeing the reliability and stability of an irrigation system, the efficient utilization of water resources, disaster prevention, risk management and the like. By periodically detecting cracks and timely maintaining and repairing, the normal operation of an irrigation system can be ensured, the resource loss and risk are reduced to the greatest extent, and the benefits and sustainable development of agricultural production are improved. Conventional main channel concrete crack detection typically relies on visual detection by experienced staff, however this approach is labor intensive and inefficient. In recent years, with the rapid development of deep learning technology, how to implement automatic crack detection has become a current research hotspot.
Disclosure of Invention
The invention aims to provide a method, a device and a storage medium for detecting cracks of main canal concrete in a filling area, which adopt a deep convolutional neural network to perform image processing, effectively distinguish the crack area of the main canal concrete in the filling area from a normal area, and improve the detection accuracy.
In order to achieve the above purpose, the invention adopts the following technical scheme:
the first aspect of the invention provides a method for detecting concrete cracks of a main canal in a irrigated area, which comprises the following steps:
constructing a deep learning model for crack region detection; the model takes a preprocessed main canal concrete image of the irrigation area as input and takes a crack area detection result as output, and adopts multi-path depth direction banded convolution to extract multi-scale characteristic information of the input image; the preprocessing is to adopt an image denoising algorithm based on total variation to denoise the main canal concrete image of the irrigation area;
preprocessing a real-time acquired main canal concrete image of the irrigation area, and inputting the preprocessed main canal concrete image into the deep learning model to obtain a detection result of a crack area of the main canal concrete of the irrigation area;
and optimizing the detection result of the crack region of the main canal concrete in the irrigation area by adopting a random field to obtain a final detection result of the crack region.
Further, the denoising processing of the irrigation area main canal concrete image by adopting an image denoising algorithm based on total variation comprises the following steps:
the following objective function is constructed:
；
wherein,for the purpose of +.>Representing the original image +.>And denoised image->Mean square error between>Representing regularization parameters, ++>Representing denoised image->Total variation of->Expressed as:，/>representing denoised image->In position->A gradient is provided at the point of the gradient,is a small positive number;
iterative optimization is carried out by adopting a gradient descent algorithm with the objective function of minimizing as a target, and a final denoised image is obtained。
Further, the constructing a deep learning model for crack region detection includes:
shooting crack images of the main canal concrete of the irrigation area under different angles, different distances and different light and shade conditions as a crack image data set;
preprocessing the image in the crack image dataset, and marking a crack region;
and constructing a deep learning network based on an encoder-decoder, taking the marked crack image data set as a training set, and training the deep learning network to obtain a trained deep learning model for crack region detection.
Further, the encoder-decoder based deep learning network includes:
the Stem module is used for performing dimension reduction processing on the input image;
the encoder is used for extracting characteristic diagrams of the input image on different receptive field scales; the encoder comprises four feature extraction modules, wherein each feature extraction module comprises two 1*1 convolution modules and a multipath depth direction banded convolution module; the multi-path depth direction banded convolution module comprises three depth-separable symmetrical convolutions, and convolution kernel sizes are 7, 11 and 21 respectively;
and the decoder is used for unifying the channel dimensions of the four feature images output by the encoder through the multi-layer perceptron, unifying the four feature images to be up-sampled to be the same size for splicing, adjusting the channel number of the spliced feature images to be 2, and outputting the prediction probability of pixels corresponding to the background area and the crack area.
Further, the Stem module downsamples with two identical overlapping convolutional layers, the convolution kernel size is 3, the step size is 2, and the zero padding is 1.
Further, in the training process of the deep learning network, an improved loss function is adopted, and the loss function is expressed as follows:
；
wherein,representing the probability of network prediction as the probability of positive class,/->Represents a factor for balancing the positive and negative sample weights,/->Representing a regulatory factor,/->Representing regularization coefficients, +.>Indicate->Network parameters.
Further, preprocessing a real-time obtained irrigation area main canal concrete image by adopting a mode with overlapped sliding windows, and inputting the preprocessed irrigation area main canal concrete image into the deep learning model;
the overlapping area of the sliding window is set to be 1/4 of the sliding window.
Further, optimizing the detection result of the crack region of the main canal concrete in the irrigation area by adopting a random field to obtain a final detection result of the crack region, wherein the method comprises the following steps:
constructing a random field according to the irrigation area main canal concrete image and the irrigation area main canal concrete crack area detection result;
the energy function of constructing the random field is as follows:
；
wherein,represents the potential energy of all pixels, +.>The label set representing all pixels is a detection result of a crack region of the main canal concrete of the irrigation area, and comprises background region pixels and crack region pixels, < +.>Indicate->Label of individual pixels>Indicate->Individual pixels +.>Potential energy of->Indicate->Individual pixels and->Potential energy between individual pixels;
expressed as: />Wherein->Indicate->Characteristics of individual pixels->Indicate->Individual pixels +.>Is the prediction result of the deep learning model;
expressed as: />Wherein->For weight parameter, ++>Parameters for controlling the similarity range;
and optimizing the label of each pixel by minimizing the energy function of the random field to obtain a final crack detection result.
The second aspect of the present invention provides a device for detecting cracks of concrete in a main canal of a irrigated area, for implementing the method for detecting cracks of concrete in a main canal of a irrigated area, the device comprising:
the model training module is used for constructing a deep learning model for crack region detection; the model takes a preprocessed main canal concrete image of the irrigation area as input and takes a crack area detection result as output, and adopts multi-path depth direction banded convolution to extract multi-scale characteristic information of the input image; the preprocessing is to adopt an image denoising algorithm based on total variation to denoise the main canal concrete image of the irrigation area;
the prediction module is used for preprocessing the real-time acquired main canal concrete image of the irrigation area and inputting the preprocessed main canal concrete image into the deep learning model to obtain a detection result of the crack area of the main canal concrete of the irrigation area;
and the optimization output module is used for optimizing the detection result of the crack region of the main canal concrete in the irrigation area by adopting a random field to obtain a final detection result of the crack region.
A third aspect of the invention provides a computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by a computing device, cause the computing device to perform any of the methods described above.
The beneficial effects of the invention are as follows:
(1) The method is based on a deep learning algorithm, can more accurately identify concrete cracks in the main channel, and avoids the situation of false report or missing report. The method has higher reliability, and can effectively detect the concrete crack condition in the main channel and repair the concrete crack in time;
(2) The method can automatically scan the shot main canal image of the irrigation area, reduces the workload of manual monitoring and improves the working efficiency.
Drawings
FIG. 1 is a flow chart of a method for detecting cracks of a main canal concrete in a filling area based on deep learning;
FIG. 2 is a structural diagram of a deep learning model for concrete crack detection provided by the invention;
FIG. 3 is a block diagram of a multi-path depth direction banded convolution module in a deep learning model for concrete crack detection provided by the invention;
FIG. 4 is a schematic diagram of a sliding window image input mode with overlapping in a deep learning model for concrete crack detection;
fig. 5 shows the image segmentation result of concrete cracks obtained in one embodiment of the present invention.
Detailed Description
The invention is further described below. The following examples are only for more clearly illustrating the technical aspects of the present invention, and are not intended to limit the scope of the present invention.
The crack areas of the main canal concrete in the irrigation area have obvious differences from the surrounding environment, the advantages of the deep convolutional neural network in terms of image processing and feature extraction can be fully utilized by the detection method based on the deep convolutional neural network, and the efficient and accurate identification of the crack areas of the main canal concrete in the irrigation area is realized by learning and classifying the features of different crack areas. In practical application, the method can realize accurate positioning and elimination of crack points through effective processing and analysis of the image information of the main channel, and ensure the safety and reliability of the main channel.
Based on the above inventive concept, the present invention provides a method for detecting cracks of main canal concrete in a irrigated area, referring to fig. 1, comprising:
constructing a deep learning model for crack region detection; the model takes the pretreated concrete image of the main canal of the irrigation area as input and takes the crack detection result as output; the model adopts multi-path depth direction banded convolution to extract multi-scale characteristic information of an input image; the preprocessing is to adopt an image denoising algorithm based on total variation to denoise the concrete image of the main canal of the irrigation area;
preprocessing a real-time acquired main canal concrete image of the irrigation area, and inputting the preprocessed main canal concrete image into the deep learning model to obtain a main canal concrete crack detection result of the irrigation area;
and optimizing the obtained crack detection result of the main canal concrete in the irrigated area by adopting a random field to obtain a final crack detection result.
In the invention, imaging equipment such as a mobile phone, a camera, a monitoring device and the like is adopted to shoot crack images under different angles, different distances and different light and shade conditions as a crack image data set.
In the invention, an image denoising algorithm based on total variation is adopted to denoise the concrete image of the main canal of the irrigation area, and the concrete implementation process comprises the following steps:
the total variance is the sum of the absolute values of the image gradients, which helps the image to remain smooth, thereby removing noise. In the imageCalculating total variation->Can be expressed as:
；
wherein,for denoised image->Total variation of->Representation of image->In position->Gradient at->Representing a small positive number for preventing the denominator from being zero.
The total variation-based image denoising algorithm removes noise by optimizing the following objective function:
；
wherein,representing an objective function +.>Representation of image->And denoised image->Mean square error between>Represents regularization parameters for balancing noise removal and total variation terms.
The image denoising algorithm based on the total variation comprises the following specific processes:
initializing: using noisy imagesInitializing clear image->；
Iterative optimization: updating images by iterative optimizationReduce the objective function->；
a. Gradient decrease: updating the image according to the gradient of the objective functionTo reduce the mean square error term;
b. total variation regularization term update: smoothing images using total variation regularization termThereby reducing noise;
c. parameter adjustment: by adjusting regularization parametersBalancing a mean square error term and a total variation term;
and (3) convergence judgment: judging whether the algorithm converges or not through a set threshold value of the change of the objective function, ending iteration after convergence, and outputting a final image。
In the invention, a deep learning model for crack region detection is constructed, and the specific implementation process is as follows:
s1, collecting an image data set of a main canal concrete crack in a irrigated area;
s2, marking a crack region after preprocessing the concrete crack image in the data set according to the mode;
and S3, constructing a deep learning network based on the encoder-decoder, taking the labeled concrete crack image data set as a training set, and training the constructed deep learning network to obtain a trained deep learning model for crack region detection.
Referring to fig. 2, in the present invention, a deep learning network includes:
and the Stem module is used for performing data dimension reduction and enhancing feature expression so as to better adapt to the subsequent network level. In this embodiment, the spatial resolution of the input image is reduced to 1/4 of the original; referring to fig. 2, in the present invention, the original input image is 512×512, and the Stem module image becomes: 128 x 128;
referring to fig. 2, in the present invention, the encoder includes four feature extraction modules, each including two standard 1*1 convolution modules and one multi-path depth-direction strip convolution module,
wherein a standard 1*1 convolution is used to adjust the number of signature channels to simulate the relationship between the different channels.
The multi-path depth direction banded convolution module is provided with three branches and is used for capturing multi-scale characteristic information; referring specifically to fig. 3, feature extraction is performed on three branches on different receptive field scales by depth-separable and symmetrical convolution, so as to obtain feature graphs of the three receptive field scales; the convolution kernel size of each branch is set to 7, 11 and 21, respectively, to obtain a multiscale receptive field and reduce model parameters and computational costs.
In fig. 2, the parameters of each feature extraction module of the encoder are specifically as follows:
table 1: parameters of each feature extraction module
| Module | Feature map size | Number of channels | Number of module repetition Ni |
| 1 | (128,128) | 32 | 3 |
| 2 | (64,64) | 64 | 3 |
| 3 | (32,32) | 160 | 5 |
| 4 | (16,16) | 256 | 2 |
Referring to fig. 2, in the present invention, a lightweight decoding mode is adopted by a decoder, which specifically includes:
unifying channel dimensions of the four feature maps output by the encoder through a multi-layer perceptron;
uniformly upsampling the four feature images to the same size and splicing the four feature images;
and finally, adjusting the channel number of the spliced feature map to 2, and respectively corresponding to the prediction probabilities of pixels in the background area and the crack area to realize pixel-level prediction classification.
Preferably, in the present invention, the Stem module uses two identical convolutions with overlapping convolutions to implement downsampling, the convolution kernel size is 3, the step size is 2, and the zero padding is 1. The middle layer uses a GELU activation function to enhance nonlinearity.
Preferably, the encoder in the present invention uses residual connection to solve the gradient vanishing/exploding and model degradation problems in deep neural networks.
Preferably, the loss function in the training process of the invention adopts improved focal loss, solves the problem of sample imbalance of a fracture data set, reduces the complexity of a model, and prevents overfitting, and is specifically defined as follows:
；
wherein:the probability representing model prediction is the probability of a positive category; />Representing a factor for balancing the positive and negative sample weights, taking 0.25; />Representing an adjusting factor which can be used for adjusting the weight of the difficult sample and taking 2.0; />Representing a regularization coefficient and controlling the influence degree of a regularization term; />Express model->And parameters.
Preferably, in the invention, the first-order moment and the second-order moment of the gradient are estimated by using the Adam optimizer in the training process to adapt to the update amplitude of different parameters, so that the optimal solution is searched more efficiently in a parameter space, and a parameter update formula is as follows:
；
；
；
；
；
wherein:expressed in time stepstModel parameters of (2); />Expressed in time stepstI.e. the gradient of the model parameters with respect to the loss function; />Is a moving average of the first moment estimate (mean) used to estimate the first moment of the gradient; />Is the moving average of the second moment estimates (unbiased variance) used to estimate the second moment of the gradient; />And->Super parameters for controlling the first moment and the second moment to estimate the attenuation rate are respectively 0.9 and 0.999; />And->The offset correction of the first moment and the second moment; />Is learning rate, controls the step length of each parameter update; />Representing a small constant, preventing the denominator from being zero.
In one embodiment of the present invention, the labeled fracture image dataset is divided into a training set and a test set, the training set is used for training the constructed deep learning model, the test set is used for testing the trained model, and parameters and structures of the model are further optimized.
In an embodiment of the present invention, the method of sliding windows with overlapping is adopted to pretreat the real-time obtained concrete image of the main canal of the irrigation area and input the preprocessed image into the deep learning model, and each sliding window overlaps with the previous window to ensure that the pixel information is not lost. See fig. 4 for a sliding window image input with overlap.
Preferably, the overlap area is set to 1/4 of the sliding window.
In the invention, based on the crack segmentation result, the relation among pixels is modeled by further adopting the random field model to improve the accuracy of the segmentation result. The relationship between pixels is represented using random fields, where each pixel corresponds to a random variable representing its semantic tag. The energy function of the random field can be defined as:
；
wherein,represents the potential energy of all pixels, +.>A label set representing all pixels,>indicate->Label of individual pixels>Is->Individual pixels +.>Energy of->Is->Individual pixels and->Potential energy between individual pixels (about label +.>And->Is set, is a power source of (a) energy).
Potential energy of single pixelTaking the confidence of prediction into consideration, using the prediction probability of the deep learning model as potential energy:
；
wherein,indicate->Characteristics of individual pixels->Indicate->Individual pixels +.>Is used for the prediction probability of (1).
Modeling potential between pixels using gaussian potential considering differences between pixel labels：
；
Wherein,for weight parameter, ++>To control parameters of the similarity range.
The specific implementation steps are as follows:
construction diagram: constructing a random field diagram according to the topological structure of the image, wherein each pixel is used as a node, and the edges represent the relation among the pixels;
defining a potential energy function: defining potential energy of a single pixel and a potential energy function between pixels according to the formula;
deducing the label: by minimizing the energy function of the random field, label inference is performed on each pixel to obtain a more consistent semantic segmentation result.
Fig. 5 is a graph of the result of detecting a concrete crack image by using the method for detecting a main canal concrete crack in an irrigation area according to an embodiment of the present invention, wherein two graphs a1 and a2 in fig. 5 are original crack images, and two graphs b1 and b2 are model segmentation results corresponding to a1 and a2, respectively. The model can realize accurate crack segmentation and can assist inspection personnel in crack detection.
Based on the above inventive concept, a second aspect of the present invention provides a device for detecting cracks of canal concrete in a irrigated area, for implementing the above method for detecting cracks of canal concrete in a irrigated area, the device comprising:
the model training module is used for constructing a deep learning model for crack region detection; the model takes a preprocessed main canal concrete image of the irrigation area as input and takes a crack detection result as output, and the model adopts multi-path depth direction banded convolution to extract multi-scale characteristic information of the input image; the preprocessing is to adopt an image denoising algorithm based on total variation to denoise the main canal concrete image of the irrigation area;
the prediction module is used for preprocessing the real-time acquired main canal concrete image of the irrigation area and inputting the preprocessed main canal concrete image into the deep learning model to obtain a crack detection result of the main canal concrete of the irrigation area;
and the optimization output module is used for optimizing the crack detection result of the main canal concrete in the irrigation area by adopting a random field to obtain a final crack detection result.
It should be noted that the embodiment of the apparatus corresponds to the embodiment of the method, and the implementation manner of the embodiment of the method is applicable to the embodiment of the apparatus and can achieve the same or similar technical effects, so that the description thereof is omitted herein.
A third aspect of the invention provides a computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by a computing device, cause the computing device to perform a method of irrigated area main canal concrete crack detection according to the foregoing.
A fourth aspect of the invention provides a computing device comprising,
one or more processors, memory, and one or more programs, wherein one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing the method of channel concrete crack detection according to the foregoing.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flowchart and/or block of the flowchart illustrations and/or block diagrams, and combinations of flowcharts and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the above embodiments are only for illustrating the technical aspects of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the above embodiments, it should be understood by those of ordinary skill in the art that: modifications and equivalents may be made to the specific embodiments of the invention without departing from the spirit and scope of the invention, which is intended to be covered by the claims.
Claims (7)
1. The method for detecting the concrete cracks of the main canal of the irrigation area is characterized by comprising the following steps:
constructing a deep learning model for crack region detection; the model takes a preprocessed main canal concrete image of the irrigation area as input and takes a crack area detection result as output, and adopts multi-path depth direction banded convolution to extract multi-scale characteristic information of the input image; the preprocessing is to adopt an image denoising algorithm based on total variation to denoise the main canal concrete image of the irrigation area;
the construction of the deep learning model for crack region detection comprises the following specific processes:
shooting crack images of the main canal concrete of the irrigation area under different angles, different distances and different light and shade conditions as a crack image data set;
preprocessing the image in the crack image dataset, and marking a crack region;
constructing an encoder-decoder based deep learning network, comprising:
the Stem module is used for performing dimension reduction processing on the input image;
the encoder is used for extracting characteristic diagrams of the input image on different receptive field scales; the encoder comprises four feature extraction modules, wherein each feature extraction module comprises two 1*1 convolution modules and a multipath depth direction banded convolution module; the multi-path depth direction banded convolution module comprises three depth-separable symmetrical convolutions, and convolution kernel sizes are 7, 11 and 21 respectively;
the decoder is used for unifying the channel dimensions of the four feature images output by the encoder through the multi-layer perceptron, unifying the four feature images to be up-sampled to be the same size for splicing, adjusting the channel number of the spliced feature images to be 2, and outputting the prediction probability of pixels corresponding to the background area and the crack area;
training the deep learning network by taking the marked crack image data set as a training set to obtain a trained deep learning model for crack region detection;
in the training process of the deep learning network, an improved loss function is adopted, and the loss function is expressed as follows:
where Loss represents a Loss function, p t Representing the network prediction probability, being the probability of a positive class, α t Represents a factor for balancing positive and negative sample weights, gamma represents an adjustment factor, lambda represents a regularization coefficient, θ i Representing an ith network parameter;
preprocessing a real-time acquired main canal concrete image of the irrigation area, and inputting the preprocessed main canal concrete image into the deep learning model to obtain a detection result of a crack area of the main canal concrete of the irrigation area;
and optimizing the detection result of the crack region of the main canal concrete in the irrigation area by adopting a random field to obtain a final detection result of the crack region.
2. The method for detecting cracks of the main canal concrete in the irrigation area according to claim 1, wherein the denoising processing of the main canal concrete image in the irrigation area by adopting an image denoising algorithm based on total variation comprises the following steps:
the following objective function is constructed:
wherein E (I) represents an objective function,representing the mean square error between the original image B and the denoised image I, λ representing the regularization parameter, TV (I) being the total variation of the denoised image I, TV (I) being represented as:i (I, j) represents the gradient of the denoised image I at position (I, j), epsilon represents a small positive number;
and carrying out iterative optimization by adopting a gradient descent algorithm with the objective function of minimizing as a target to obtain a final denoised image I.
3. The method for detecting cracks of main canal concrete in a irrigated area according to claim 2, wherein the Stem module uses two identical convolution layers with overlapping for downsampling, the convolution kernel size is 3, the step size is 2, and zero padding is 1.
4. The method for detecting cracks of the main canal concrete in the irrigation area according to claim 1, wherein the main canal concrete image in the irrigation area, which is obtained in real time, is preprocessed by adopting a sliding window with overlapping and then is input into the deep learning model;
the overlapping area of the sliding window is set to be 1/4 of the sliding window.
5. The method for detecting cracks of the main canal concrete in the irrigated area according to claim 1, wherein the method for detecting the cracks of the main canal concrete in the irrigated area is characterized by optimizing the detection result of the main canal concrete in the irrigated area by adopting an random field to obtain the final detection result of the cracks, and comprises the following steps:
constructing a random field according to the irrigation area main canal concrete image and the irrigation area main canal concrete crack area detection result;
the energy function of constructing the random field is as follows:
wherein E (Y) represents potential energy of all pixels, Y represents label set of all pixels, and the label is detection result of crack region of the main canal concrete of the irrigation area, and comprises background region pixels and crack region pixels, Y i Label indicating ith pixel, ψ u (y i ) Representing the ith pixel with respect to label y i Potential energy of psi p (y i ,y j ) Representing potential energy between the i-th pixel and the j-th pixel;
ψ u (y i ) Expressed as: psi phi type u (y i )＝-logP(y i |x i ) Wherein x is i Features representing the ith pixel, P (y i |x i ) Representing the ith pixel with respect to label y i Is the prediction result of the deep learning model;
ψ p (y i ,y j ) Expressed as:wherein mu is a weight parameter, and sigma is a parameter for controlling the similarity range;
and optimizing the label of each pixel by minimizing the energy function of the random field to obtain a final crack detection result.
6. A device for detecting cracks in canal concrete in a irrigated area, characterized in that it is used for realizing the method for detecting cracks in canal concrete in a irrigated area according to any one of claims 1 to 5, said device comprising:
the model training module is used for constructing a deep learning model for crack region detection; the model takes a preprocessed main canal concrete image of the irrigation area as input and takes a crack area detection result as output, and adopts multi-path depth direction banded convolution to extract multi-scale characteristic information of the input image; the preprocessing is to adopt an image denoising algorithm based on total variation to denoise the main canal concrete image of the irrigation area; the method for constructing the deep learning model for crack region detection comprises the following specific steps:
shooting crack images of the main canal concrete of the irrigation area under different angles, different distances and different light and shade conditions as a crack image data set;
preprocessing the image in the crack image dataset, and marking a crack region;
constructing an encoder-decoder based deep learning network, comprising:
the Stem module is used for performing dimension reduction processing on the input image;
the encoder is used for extracting characteristic diagrams of the input image on different receptive field scales; the encoder comprises four feature extraction modules, wherein each feature extraction module comprises two 1*1 convolution modules and a multipath depth direction banded convolution module; the multi-path depth direction banded convolution module comprises three depth-separable symmetrical convolutions, and convolution kernel sizes are 7, 11 and 21 respectively;
the decoder is used for unifying the channel dimensions of the four feature images output by the encoder through the multi-layer perceptron, unifying the four feature images to be up-sampled to be the same size for splicing, adjusting the channel number of the spliced feature images to be 2, and outputting the prediction probability of pixels corresponding to the background area and the crack area;
training the deep learning network by taking the marked crack image data set as a training set to obtain a trained deep learning model for crack region detection;
in the training process of the deep learning network, an improved loss function is adopted, and the loss function is expressed as follows:
where Loss represents a Loss function, p t Representing the network prediction probability, being the probability of a positive class, α t Represents a factor for balancing positive and negative sample weights, gamma represents an adjustment factor, lambda represents a regularization coefficient, θ i Representing an ith network parameter;
the prediction module is used for preprocessing the real-time acquired main canal concrete image of the irrigation area and inputting the preprocessed main canal concrete image into the deep learning model to obtain a detection result of the crack area of the main canal concrete of the irrigation area;
and the optimization output module is used for optimizing the detection result of the crack region of the main canal concrete in the irrigation area by adopting a random field to obtain a final detection result of the crack region.
7. A computer readable storage medium storing one or more programs, wherein the one or more programs comprise instructions, which when executed by a computing device, cause the computing device to perform any of the methods of claims 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311576460.4A CN117292266B (en) | 2023-11-24 | 2023-11-24 | Method and device for detecting concrete cracks of main canal of irrigation area and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311576460.4A CN117292266B (en) | 2023-11-24 | 2023-11-24 | Method and device for detecting concrete cracks of main canal of irrigation area and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117292266A CN117292266A (en) | 2023-12-26 |
| CN117292266B true CN117292266B (en) | 2024-03-22 |
Family
ID=89253880
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311576460.4A Active CN117292266B (en) | 2023-11-24 | 2023-11-24 | Method and device for detecting concrete cracks of main canal of irrigation area and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117292266B (en) |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780493A (en) * | 2017-01-24 | 2017-05-31 | 广东省特种设备检测研究院珠海检测院 | A kind of circular flaw recognition methods based on weld seam X-ray digital image |
| CN108121225A (en) * | 2017-11-27 | 2018-06-05 | 中铁十二局集团有限公司 | Concrete continuous beam cantilever casting control system and control method |
| CN108898145A (en) * | 2018-06-15 | 2018-11-27 | 西南交通大学 | A kind of image well-marked target detection method of combination deep learning |
| CN110020652A (en) * | 2019-01-07 | 2019-07-16 | 新而锐电子科技(上海)有限公司 | The dividing method of Tunnel Lining Cracks image |
| CN110428432A (en) * | 2019-08-08 | 2019-11-08 | 梅礼晔 | The deep neural network algorithm of colon body of gland Image Automatic Segmentation |
| CN112948952A (en) * | 2021-04-08 | 2021-06-11 | 郑州航空工业管理学院 | Evolution prediction method for shield tunnel lining back cavity |
| CN113096088A (en) * | 2021-04-07 | 2021-07-09 | 浙江大学 | Concrete structure detection method based on deep learning |
| CN114283331A (en) * | 2021-12-02 | 2022-04-05 | 西安邮电大学 | Lightweight SAR image ship detection model and method based on strip pruning |
| CN114693615A (en) * | 2022-03-17 | 2022-07-01 | 常州工学院 | Deep learning concrete bridge crack real-time detection method based on domain adaptation |
| CN116563682A (en) * | 2023-05-11 | 2023-08-08 | 北京工业大学 | Attention scheme and strip convolution semantic line detection method based on depth Hough network |
| CN116778146A (en) * | 2023-06-02 | 2023-09-19 | 中国地质大学（武汉） | Road information extraction method and device based on multi-mode data |
| CN116823800A (en) * | 2023-07-17 | 2023-09-29 | 重庆交通大学 | Bridge concrete crack detection method based on deep learning under complex background |
| CN116975753A (en) * | 2023-03-01 | 2023-10-31 | 腾讯科技（深圳）有限公司 | Data category based prediction method, device, equipment and medium |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10699151B2 (en) * | 2016-06-03 | 2020-06-30 | Miovision Technologies Incorporated | System and method for performing saliency detection using deep active contours |
| CN114530224A (en) * | 2022-01-18 | 2022-05-24 | 深圳市智影医疗科技有限公司 | Auxiliary generation method and system for diagnosis report based on medical image |
-
2023
- 2023-11-24 CN CN202311576460.4A patent/CN117292266B/en active Active
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780493A (en) * | 2017-01-24 | 2017-05-31 | 广东省特种设备检测研究院珠海检测院 | A kind of circular flaw recognition methods based on weld seam X-ray digital image |
| CN108121225A (en) * | 2017-11-27 | 2018-06-05 | 中铁十二局集团有限公司 | Concrete continuous beam cantilever casting control system and control method |
| CN108898145A (en) * | 2018-06-15 | 2018-11-27 | 西南交通大学 | A kind of image well-marked target detection method of combination deep learning |
| CN110020652A (en) * | 2019-01-07 | 2019-07-16 | 新而锐电子科技(上海)有限公司 | The dividing method of Tunnel Lining Cracks image |
| CN110428432A (en) * | 2019-08-08 | 2019-11-08 | 梅礼晔 | The deep neural network algorithm of colon body of gland Image Automatic Segmentation |
| CN113096088A (en) * | 2021-04-07 | 2021-07-09 | 浙江大学 | Concrete structure detection method based on deep learning |
| CN112948952A (en) * | 2021-04-08 | 2021-06-11 | 郑州航空工业管理学院 | Evolution prediction method for shield tunnel lining back cavity |
| CN114283331A (en) * | 2021-12-02 | 2022-04-05 | 西安邮电大学 | Lightweight SAR image ship detection model and method based on strip pruning |
| CN114693615A (en) * | 2022-03-17 | 2022-07-01 | 常州工学院 | Deep learning concrete bridge crack real-time detection method based on domain adaptation |
| CN116975753A (en) * | 2023-03-01 | 2023-10-31 | 腾讯科技（深圳）有限公司 | Data category based prediction method, device, equipment and medium |
| CN116563682A (en) * | 2023-05-11 | 2023-08-08 | 北京工业大学 | Attention scheme and strip convolution semantic line detection method based on depth Hough network |
| CN116778146A (en) * | 2023-06-02 | 2023-09-19 | 中国地质大学（武汉） | Road information extraction method and device based on multi-mode data |
| CN116823800A (en) * | 2023-07-17 | 2023-09-29 | 重庆交通大学 | Bridge concrete crack detection method based on deep learning under complex background |
Non-Patent Citations (4)
| Title |
|---|
| Automatic concrete crack segmentation model based on transformer;Wenjun Wang等;《Automation in Construction》;第139卷(第2022期);第1-11页 * |
| Use of Parallel ResNet for High-Performance Pavement Crack Detection and Measurement;Zhun Fan等;《Sustainability》;第1-21页 * |
| 基于深度学习的灌区水工混凝土结构表面裂缝自动检测方法;曹国金 等;《吉林水利》;第2022卷(第12期);第14-17页 * |
| 改进YOLOX的药品泡罩铝箔表面缺陷检测方法;胡海涛 等;《图学学报》;第43卷(第5期);第803-814页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117292266A (en) | 2023-12-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111488789B (en) | Pedestrian detection method and device for monitoring based on image analysis | |
| CN111754498B (en) | Conveyor belt carrier roller detection method based on YOLOv3 | |
| CN111161315B (en) | Multi-target tracking method and system based on graph neural network | |
| CN107945210A (en) | Target tracking algorism based on deep learning and environment self-adaption | |
| CN113989257A (en) | Electric power comprehensive pipe gallery settlement crack identification method based on artificial intelligence technology | |
| Peng et al. | Non-uniform illumination image enhancement for surface damage detection of wind turbine blades | |
| CN110334775B (en) | Unmanned aerial vehicle line fault identification method and device based on width learning | |
| CN110795599B (en) | Video emergency monitoring method and system based on multi-scale graph | |
| CN115797846A (en) | Wind power generation blade block defect comparison method and device and electronic equipment | |
| CN116110042A (en) | Tomato detection method based on CBAM attention mechanism of YOLOv7 | |
| CN108549899B (en) | Image identification method and device | |
| CN117422695A (en) | CR-deep-based anomaly detection method | |
| CN117809164A (en) | Substation equipment fault detection method and system based on multi-mode fusion | |
| KR20230023263A (en) | Deep learning-based sewerage defect detection method and apparatus | |
| CN117523555A (en) | Aircraft part defect detection method based on self-encoder | |
| CN117292266B (en) | Method and device for detecting concrete cracks of main canal of irrigation area and storage medium | |
| McLeay et al. | Deep convolutional neural networks with transfer learning for waterline detection in mussel farms | |
| CN116680639A (en) | Deep-learning-based anomaly detection method for sensor data of deep-sea submersible | |
| CN115861223A (en) | Solar cell panel defect detection method and system | |
| CN114757941A (en) | Transformer substation equipment defect identification method and device, electronic equipment and storage medium | |
| CN112396126B (en) | Target detection method and system based on detection trunk and local feature optimization | |
| CN112488125B (en) | Reconstruction method and system based on high-speed visual diagnosis and BP neural network | |
| CN115359091A (en) | Armor plate detection tracking method for mobile robot | |
| CN113723210A (en) | Method and system for detecting leakage edge equipment of water supply pipe in intelligent park | |
| CN114387484A (en) | Improved mask wearing detection method and system based on yolov4 |
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 |