CN115880533A - Bridge apparent crack identification method based on adaptive subset search and deep learning - Google Patents
Bridge apparent crack identification method based on adaptive subset search and deep learning Download PDFInfo
- Publication number
- CN115880533A CN115880533A CN202211670986.4A CN202211670986A CN115880533A CN 115880533 A CN115880533 A CN 115880533A CN 202211670986 A CN202211670986 A CN 202211670986A CN 115880533 A CN115880533 A CN 115880533A
- Authority
- CN
- China
- Prior art keywords
- data set
- neural network
- network model
- deep neural
- subset
- 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
Images
Abstract
The application provides a bridge apparent crack identification method based on adaptive subset search and deep learning, which comprises the following steps: acquiring an apparent crack picture data set of a high-speed railway bridge, selecting a candidate data set from the picture data set, and selecting a current subset from the candidate data set; selecting a plurality of samples from the candidate data set as new training samples in a random sampling mode, marking the training samples, and adding the training samples into the training set; updating a deep neural network model for crack recognition by using the training set, and calculating the relative loss of the deep neural network model to the prediction result of the picture data set in the two adjacent iterative training processes; if the relative loss of the prediction result of the picture data set in continuous multiple iterations is smaller than an allowable value, judging that the deep neural network model is converged; and storing the deep neural network model, and identifying the apparent cracks of the high-speed rail bridge by using the deep neural network model. The method and the device can solve the technical problem of local optimal sampling in active learning of apparent crack identification of the high-speed railway bridge in the prior art, and improve the efficiency and the precision of crack identification.
Description
Technical Field
The invention relates to the technical field of deep learning, in particular to a bridge apparent crack identification method based on adaptive subset search and deep learning.
Background
The performance of a large-span high-speed railway (high-speed rail) bridge is degraded in the long-term service process, so that the safety performance of the bridge structure is reduced, and the life and property safety of people is seriously threatened. The method can effectively identify the performance degradation degree of the bridge and prevent disasters in advance by evaluating the state of the high-speed railway bridge, and is an important means for guaranteeing the safe operation of the high-speed railway bridge. The surface cracking is an important early indicator of the structural damage of the bridge, and the identification of the apparent cracks is an important task of the state evaluation of the high-speed railway bridge.
The traditional manual detection method is high in cost, potential safety hazards exist in manual high-altitude operation, and high-speed railway bridge detection in a large range is difficult to carry out efficiently. In recent years, computer vision technology based on deep learning is rapidly developed and successfully applied to the fields of structural apparent crack identification and the like, and the automatic implementation of the structural apparent crack identification of the high-speed railway bridge can be greatly promoted. However, deep learning models for computer vision identification of apparent cracks require a large number of labeled crack data sets to be trained to achieve sufficient identification accuracy. In actual engineering, it is generally difficult to establish a huge training data set, and the cost of labeling the data set is expensive. Compared with the traditional training method, the active learning method can train the deep learning model by using quite a few training data, and can obtain high enough recognition precision. However, the existing active learning method based on expression and uncertainty fails to balance exploration and utilization in the active learning process well, and the problem of local optimal sampling is easily caused, so that the efficiency of active learning is low, and the waste of a training data set is caused.
Disclosure of Invention
The invention aims to overcome the technical defects, provides a bridge apparent crack identification method based on adaptive subset search and deep learning, can solve the technical problem of local optimal sampling in active learning of high-speed railway bridge apparent crack identification in the prior art, and improves the efficiency and the precision of the high-speed railway bridge apparent crack identification.
In order to achieve the above technical purpose, in a first aspect, the technical solution of the present invention provides a bridge apparent crack identification method based on adaptive subset search and deep learning, including the following steps:
acquiring an apparent crack picture data set of a high-speed railway bridge, selecting a candidate data set from the picture data set, and selecting a current subset from the candidate data set;
selecting a plurality of samples from the candidate data set in a random sampling mode to serve as new training samples, marking the training samples, and adding the training samples into a training set;
updating a deep neural network model for crack recognition by using the training set, and calculating the relative loss of the deep neural network model to the image data set prediction result in the two adjacent iterative training processes;
if the relative loss of the image data set prediction result in a plurality of iterations is smaller than a first allowable value continuously, the deep neural network model converges;
and storing the deep neural network model, and identifying the apparent cracks of the high-speed rail bridge by using the deep neural network model.
Compared with the prior art, the invention has the beneficial effects that:
the invention provides a bridge apparent crack recognition method based on adaptive subset search and deep learning, which adopts a whole-to-local strategy to select subsets and training samples from a high-speed railway bridge apparent crack data set according to the uncertainty of the samples and train a deep neural network model for crack recognition; and calculating relative loss by adopting the model in the adjacent two iterations to the prediction result of the picture data set, thereby carrying out model convergence and subset convergence judgment. The method measures the global prediction precision of the deep neural network model according to subset convergence, adaptively reduces the sampling range, and balances exploration and utilization in the active learning process; and the model convergence is used as a termination condition of active learning, so that the efficiency and the precision of the apparent crack identification of the high-speed railway bridge can be effectively improved.
According to some embodiments of the present invention, if the deep neural network model does not satisfy the convergence condition, the prediction label of the current subset by the deep neural network model in the last iteration is used as a real label, and the prediction label of the current subset is predicted by the deep neural network model in the current iteration;
calculating the relative loss of the prediction result of the current subset in two iterations according to the loss functions of the real label, the prediction label and the network, and judging whether the current subset meets the convergence condition;
when the relative loss of the prediction result of the current subset in a plurality of continuous iterations is smaller than a second allowable value, the current subset meets the convergence condition, and the current subset is deleted from the candidate data set so as to narrow the search range of active learning;
and predicting samples of the candidate data set by adopting a deep neural network model, and calculating the uncertainty of the samples of the candidate data set based on the information entropy of the prediction result so as to update the uncertainty of the samples of the candidate data set.
According to some embodiments of the invention, after determining whether the current subset satisfies the convergence condition, the method comprises:
and if the current subset does not meet the convergence condition, calculating the uncertainty of the candidate data set samples based on the information entropy of the prediction result so as to update the uncertainty of the candidate data set samples.
According to some embodiments of the present invention, calculating the relative loss of the prediction result of the picture data set in the two adjacent iterative training processes of the deep neural network model comprises:
and adopting the prediction label of the deep neural network model on the picture data set in the last iteration as a real label, adopting the deep neural network model of the current iteration to predict the prediction label of the picture data set, and then calculating the relative loss of the data set in the two iterations according to the real label, the prediction label and the loss function of the deep neural network model.
According to some embodiments of the invention, the Loss function of the deep neural network model comprises two parts, namely a Dice Loss function and a cross entropy Loss function;
wherein the Dice Loss function is defined as:
in the formula, S i p And S i g Respectively classifying predicted and real pixels, wherein N is the total number of the pixels;
the cross entropy loss function is defined as:
x is input and output of the last layer of the deep neural network model, and class is a label index value of a pixel reality;
the loss function of the deep neural network model is as follows:
loss＝loss CE +loss Dice 。
in accordance with some embodiments of the present invention,
according to some embodiments of the invention, calculating the uncertainty of the candidate data set samples based on the information entropy of the prediction results comprises:
the uncertainty calculation formula based on the information entropy is as follows:
p(S i ) Is the probability that the pixel is predicted as i-class, N is the total number of classes, and N is the total number of pixels of the sample picture.
According to some embodiments of the invention, after updating the uncertainty of the samples of the candidate data set, comprising the steps of:
according to the uncertainty of the sample, selecting the top N with the minimum uncertainty from the candidate data set s Samples as a new current subset, where N s Is defined as:
N s ＝p s N c
p s is a proportionality coefficient, N c The number of remaining candidate dataset samples.
According to some embodiments of the present invention, if the relative loss of the picture data set prediction result in a plurality of iterations is smaller than a first allowable value, the deep neural network model converges, including the steps of:
if the relative loss of the picture data set prediction result in two iterations is less than the first allowable value, the first allowable value is 0.15, and the deep neural network model converges.
According to some embodiments of the present invention, when the relative loss of the prediction result of the current subset in a plurality of consecutive iterations is smaller than a second allowable value, the current subset satisfies a convergence condition, including the steps of:
when the relative loss of the prediction result of the current subset in the continuous iteration is less than a second allowable value, wherein the second allowable value is 0.3, the current subset meets the convergence condition.
In a second aspect, the present invention provides a bridge apparent crack identification system based on adaptive subset search and deep learning, including: a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the computer program implements the bridge apparent crack identification method based on adaptive subset search and deep learning according to any one of the first aspect.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The above and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which the abstract is to be fully consistent with one of the figures of the specification:
FIG. 1 is a flowchart of a bridge apparent crack identification method based on adaptive subset search and deep learning according to an embodiment of the present invention;
FIG. 2 is a flowchart of a bridge apparent crack identification method based on adaptive subset search and deep learning according to another embodiment of the present invention;
FIG. 3 is a flowchart of a bridge apparent crack identification method based on adaptive subset search and deep learning according to another embodiment of the present invention;
FIG. 4 is a schematic diagram of a computer vision model of a bridge apparent crack identification method based on adaptive subset search and deep learning according to another embodiment of the present invention;
fig. 5 is an original apparent crack image of a bridge apparent crack identification method based on adaptive subset search and deep learning according to another embodiment of the present invention;
fig. 6 is a predicted apparent crack image of a bridge apparent crack identification method based on adaptive subset search and deep learning according to another embodiment of the present invention;
fig. 7 is a fracture prediction D i ce coefficient convergence process of the bridge apparent fracture identification method based on adaptive subset search and deep learning according to another embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
It should be noted that although functional block divisions are provided in the system drawings and logical orders are shown in the flowcharts, in some cases, the steps shown and described may be performed in different orders than the block divisions in the systems or in the flowcharts. The terms first, second and the like in the description and in the claims, and the drawings described above, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order.
The invention provides a bridge apparent crack identification method based on self-adaptive subset search and deep learning, which adopts a sampling strategy from whole to local, selects subsets and training samples from a high-speed rail bridge apparent crack data set, and trains a deep neural network model for crack identification; and the efficiency and the precision of the identification of the apparent cracks of the high-speed railway bridge can be effectively improved by taking the model convergence as a termination condition of active learning. The method can effectively balance exploration and utilization in the active learning process, avoids the problem of local optimal sampling, and has higher sampling efficiency and crack identification precision.
The embodiments of the present invention will be further explained with reference to the drawings.
Referring to fig. 1 and fig. 2, fig. 1 is a flowchart of a bridge apparent crack identification method based on adaptive subset search and deep learning according to an embodiment of the present invention; FIG. 2 is a flowchart of a bridge apparent crack identification method based on adaptive subset search and deep learning according to another embodiment of the present invention; the bridge apparent crack identification method based on adaptive subset search and deep learning comprises the following steps:
step S110, acquiring an apparent crack picture data set of the high-speed railway bridge, selecting a candidate data set from the picture data set, and selecting a current subset from the candidate data set;
step S120, selecting a plurality of samples from the candidate data set as new training samples in a random sampling mode, marking the training samples, and adding the training samples into a training set;
step S130, updating the deep neural network model of crack recognition by using the training set, and calculating the relative loss of the deep neural network model to the image data set prediction result in the two adjacent iterative training processes;
step S140, if the relative loss of the prediction result of the picture data set in multiple iterations is less than a first allowable value, the deep neural network model is converged;
and S150, storing the deep neural network model, and identifying the apparent cracks of the high-speed rail bridge by using the deep neural network model.
In one embodiment, the bridge apparent crack identification method based on adaptive subset search and deep learning comprises the following steps: acquiring an apparent crack picture data set of a high-speed railway bridge, selecting a candidate data set from the picture data set, and selecting a current subset from the candidate data set; selecting a plurality of samples from the candidate data set as new training samples in a random sampling mode, marking the training samples, and adding the training samples into the training set; updating a deep neural network model for crack recognition by using the training set, and calculating the relative loss of the deep neural network model to the prediction result of the picture data set in the two adjacent iterative training processes; if the relative loss of the prediction result of the picture data set in multiple iterations is smaller than a first allowable value continuously, the deep neural network model is converged; and storing the deep neural network model, and identifying the apparent cracks of the high-speed rail bridge by using the deep neural network model.
The embodiment provides a bridge apparent crack identification method based on adaptive subset search and deep learning, which adopts an integral-to-local sampling strategy to select subsets and training samples from a high-speed railway bridge apparent crack data set and train a deep neural network model for crack identification; according to the uncertainty of the sample, a sampling strategy from whole to local is adopted, a subset and a training sample are selected from a high-speed railway bridge apparent crack data set, and a deep neural network model for crack identification is trained; and calculating relative loss by adopting the model in the two adjacent iterations to the prediction result of the data, thereby carrying out model convergence and subset convergence judgment. Measuring the global prediction precision of the deep neural network model according to subset convergence, reducing the sampling range in a self-adaptive manner, and balancing exploration and utilization in the active learning process; and the model convergence is used as a termination condition of active learning, so that the efficiency and the precision of the apparent crack identification of the high-speed railway bridge can be effectively improved. Compared with the existing method, the method can effectively balance exploration and utilization in the active learning process, avoids the problem of local optimal sampling, and has higher sampling efficiency and crack identification precision.
Referring to fig. 3, fig. 3 is a flowchart of a bridge apparent crack identification method based on adaptive subset search and deep learning according to another embodiment of the present invention; the bridge apparent crack identification method based on adaptive subset search and deep learning comprises the following steps:
step S210, if the deep neural network model does not meet the convergence condition, adopting the deep neural network model in the last iteration to predict the prediction label of the current subset as a real label, and adopting the deep neural network model of the current iteration to predict the prediction label of the current subset;
step S220, calculating the relative loss of the prediction result of the current subset in two iterations according to the loss function of the real label, the prediction label and the network, and judging whether the current subset meets the convergence condition;
step S230, when the relative loss of the prediction result of the current subset in a plurality of continuous iterations is less than a second allowable value, the current subset meets the convergence condition, and the current subset is deleted from the candidate data set so as to narrow the search range of active learning;
and S240, predicting the sample of the candidate data set by adopting the deep neural network model, and calculating the uncertainty of the sample of the candidate data set based on the information entropy of the prediction result so as to update the uncertainty of the sample of the candidate data set.
In one embodiment, the bridge apparent crack identification method based on adaptive subset search and deep learning comprises the following steps: acquiring an apparent crack picture data set of a high-speed railway bridge, selecting a candidate data set from the picture data set, and selecting a current subset from the candidate data set; selecting a plurality of samples from the candidate data set as new training samples in a random sampling mode, marking the training samples, and adding the training samples into the training set; updating a deep neural network model for crack recognition by using the training set, and calculating the relative loss of the deep neural network model to the prediction result of the picture data set in the two adjacent iterative training processes; if the relative loss of the prediction result of the picture data set in multiple iterations is smaller than a first allowable value, the deep neural network model is converged; and storing the deep neural network model, and identifying the apparent cracks of the high-speed rail bridge by using the deep neural network model.
If the deep neural network model does not meet the convergence condition, adopting the prediction label of the current subset by the deep neural network model in the last iteration as a real label, and adopting the deep neural network model of the current iteration to predict the prediction label of the current subset; calculating the relative loss of the prediction result of the current subset in the two iterations according to the real label, the prediction label and the loss function of the network, and judging whether the current subset meets the convergence condition; when the relative loss of the prediction result of the current subset in a plurality of continuous iterations is less than a second allowable value, the current subset meets the convergence condition, and the current subset is deleted from the candidate data set so as to narrow the search range of active learning; and predicting samples of the candidate data set by adopting the deep neural network model, and calculating the uncertainty of the samples of the candidate data set based on the information entropy of the prediction result so as to update the uncertainty of the samples of the candidate data set.
Referring to fig. 4 to 7, fig. 4 is a schematic diagram of a computer vision model of a bridge apparent crack identification method based on adaptive subset search and deep learning according to another embodiment of the present invention; fig. 5 is an original apparent crack image of a bridge apparent crack identification method based on adaptive subset search and deep learning according to another embodiment of the present invention; fig. 6 is a predicted apparent crack image of a bridge apparent crack identification method based on adaptive subset search and deep learning according to another embodiment of the present invention; fig. 7 is a fracture prediction D i ce coefficient convergence process of the bridge apparent fracture identification method based on adaptive subset search and deep learning according to another embodiment of the present invention.
(1) In one embodiment, the apparent crack identification method of the high-speed railway bridge based on adaptive subset search and deep learning is utilized for identifying the apparent crack of the high-speed railway bridge.
(2) Firstly, an apparent crack data set of the high-speed railway bridge is prepared. The original data set may be acquired by taking a photograph in the field. The data is then preprocessed by cropping, converting, etc., so that each picture sample is a 3-channel RGB picture, and all samples are adjusted to a resolution of 3 × 256 × 256. It is noted that samples in the data set do not need to be marked in advance, so that a sufficient number of samples can be acquired relatively easily. And finally, numbering the samples in the data set, so that the data can be conveniently called by a network.
(3) A candidate dataset is selected from the datasets. All data sets may generally be considered as selection candidate data sets;
(4) A subset is selected from the candidate data set. In the first iteration, an empty set can be selected as a current subset, in other iterations, a certain proportion of samples with small uncertainty are selected from a candidate data set as the current subset according to the uncertainty of the samples, and generally 10% of the candidate samples can be taken as the subset in each iteration;
(5) A new training sample is selected from the candidate data set. And selecting a small number of samples from the candidate data set as new training samples in a random sampling mode in each iteration. In the first iteration 50 samples can be taken, and 20 samples at a time. Marking a new sample, and adding the new sample into a training set;
(6) And updating the deep neural network model of the crack recognition by adopting the training set. The method is suitable for a plurality of deep neural network models in the current mainstream, such as U-Net deep neural network models. In the first iteration, a U-Net deep neural network model shown in figure 2 needs to be built, and in other iterations, the model of the previous iteration only needs to be introduced for retraining. The Loss function in the training process comprises two parts of a Dice pass and cross entropy, wherein the Dice pass is defined as
In the formula, S i p And S i g The predicted and true pixel classifications, respectively, and N is the total number of pixels. The cross entropy loss function is defined as
In the formula, x is the input and the output of the last layer of the deep neural network model, and class is the index value of the label of the pixel reality. After a network and a loss function are defined, a deep neural network model is trained in a small batch random gradient descent mode;
(7) And judging convergence of the deep neural network model. Adopting a prediction label of a deep neural network model in the last iteration on a data set as a real label
Predicting the label of the data set by adopting the deep neural network model of the current iteration>
Then, calculating the relative loss of the data set in two iterations according to the loss functions of the real label, the predicted label and the network d . When n is continuous data Loss in 2 iterations d Less than the allowable value loss dm If =0.15, the model convergence condition is considered to be satisfied, otherwise the model convergence condition is not satisfied;
(8) When the model convergence condition is not met, the next iterative calculation needs to be carried out again. First, a subset convergence determination is performed. Adopting the prediction label of the deep neural network model in the last iteration to the subset as a real label
And predicting the label of the subset by adopting the deep neural network model of the current iteration>
Then, calculating the relative loss of the subset prediction result in two iterations according to the loss functions of the real label, the prediction label and the network s . When n is continuous sub = loss in 1 iteration s Less than the allowable value loss sm If =0.3, the subset convergence condition is considered to be satisfied, otherwise the subset convergence condition is not satisfied;
(9) If the subset convergence condition is met, deleting the subset from the candidate data set, thereby reducing the search range of active learning;
(10) The uncertainty of the candidate data set samples is updated. And predicting the candidate data set samples by adopting a deep neural network model, and calculating the uncertainty of the candidate data set samples based on the information entropy of the prediction result. The uncertainty based on the entropy of information is defined as
In the formula, p (S) i ) The probability that the pixel is predicted to be in the i class is shown, N is the total number of the classes, and N is the total number of the pixels of the sample picture;
(11) According to the uncertainty of the sample, selecting the top N with the minimum uncertainty from the candidate data set s With individual samples as subsets. Wherein N is s Is composed of
N s ＝p s N c (4)
In the formula, p s For proportional coefficient, it can be 0.1; n is a radical of hydrogen c The number of remaining candidate dataset samples.
(12) And when the model convergence condition is not met, stopping the active learning iterative process and storing the trained deep neural network model.
(13) And (4) identifying the apparent cracks of the high-speed rail bridge by adopting the trained deep neural network model.
Fig. 5 and 6 are an original apparent fracture picture and a predicted apparent fracture picture, respectively. As can be seen from the figure, the method has higher prediction precision. FIG. 7 shows the convergence process of the Dice coefficient for crack prediction, which shows that the Dice coefficient of the invention converges faster, indicating that the invention has higher calculation efficiency.
In one embodiment, the bridge apparent crack identification method based on adaptive subset search and deep learning comprises the following steps: acquiring an apparent crack picture data set of a high-speed railway bridge, selecting a candidate data set from the picture data set, and selecting a current subset from the candidate data set; selecting a plurality of samples from the candidate data set as new training samples in a random sampling mode, marking the training samples, and adding the training samples into the training set; updating a deep neural network model for crack recognition by using the training set, and calculating the relative loss of the deep neural network model to the prediction result of the picture data set in the two adjacent iterative training processes; if the relative loss of the prediction result of the picture data set in multiple iterations is smaller than a first allowable value continuously, the deep neural network model is converged; and storing the deep neural network model, and identifying the apparent cracks of the high-speed rail bridge by using the deep neural network model.
If the deep neural network model does not meet the convergence condition, adopting the deep neural network model in the last iteration to predict the prediction label of the current subset as a real label, and adopting the deep neural network model of the current iteration to predict the prediction label of the current subset; calculating the relative loss of the prediction result of the current subset in the two iterations according to the real label, the prediction label and the loss function of the network, and judging whether the current subset meets the convergence condition; and if the current subset does not meet the convergence condition, calculating the uncertainty of the candidate data set samples based on the information entropy of the prediction result so as to update the uncertainty of the candidate data set samples.
In one embodiment, the bridge apparent crack identification method based on adaptive subset search and deep learning comprises the following steps: acquiring an apparent crack picture data set of a high-speed railway bridge, selecting a candidate data set from the picture data set, and selecting a current subset from the candidate data set; selecting a plurality of samples from the candidate data set as new training samples in a random sampling mode, marking the training samples, and adding the training samples into the training set; updating a deep neural network model for crack recognition by using the training set, and calculating the relative loss of the deep neural network model to the prediction result of the picture data set in the two adjacent iterative training processes; if the relative loss of the prediction result of the picture data set in multiple iterations is smaller than a first allowable value continuously, the deep neural network model is converged; and storing the deep neural network model, and identifying the apparent cracks of the high-speed rail bridge by using the deep neural network model.
Calculating the relative loss of the deep neural network model to the prediction result of the image data set in the two adjacent iterative training processes, and comprising the following steps of: and adopting the deep neural network model in the last iteration to predict the prediction label of the picture data set as a real label, adopting the deep neural network model of the current iteration to predict the prediction label of the picture data set, and then calculating the relative loss of the data set in the two iterations according to the loss functions of the real label, the prediction label and the deep neural network model.
In one embodiment, the bridge apparent crack identification method based on adaptive subset search and deep learning comprises the following steps: acquiring an apparent crack picture data set of a high-speed railway bridge, selecting a candidate data set from the picture data set, and selecting a current subset from the candidate data set; selecting a plurality of samples from the candidate data set as new training samples in a random sampling mode, marking the training samples, and adding the training samples into the training set; updating a deep neural network model for crack recognition by using the training set, and calculating the relative loss of the deep neural network model to the prediction result of the picture data set in the two adjacent iterative training processes; if the relative loss of the prediction result of the picture data set in multiple iterations is smaller than a first allowable value continuously, the deep neural network model is converged; and storing the deep neural network model, and identifying the apparent cracks of the high-speed rail bridge by using the deep neural network model.
Calculating the relative loss of the deep neural network model to the prediction result of the image data set in the two adjacent iterative training processes, and comprising the following steps of: and adopting the deep neural network model in the last iteration to predict the prediction label of the picture data set as a real label, adopting the deep neural network model of the current iteration to predict the prediction label of the picture data set, and then calculating the relative loss of the data set in the two iterations according to the loss functions of the real label, the prediction label and the deep neural network model. The Loss function of the deep neural network model comprises a Dice Loss function and a cross entropy Loss function;
wherein the D i ce Loss function is defined as:
in the formula (I), the compound is shown in the specification,
and &>
Respectively classifying the predicted and real pixels, wherein N is the total number of the pixels;
the cross entropy loss function is defined as:
x is input and output of the last layer of the deep neural network model, and class is a label index value of a pixel reality;
the loss function of the deep neural network model is:
loss＝loss CE +loss Dice 。
in one embodiment, the bridge apparent crack identification method based on adaptive subset search and deep learning comprises the following steps: acquiring an apparent crack picture data set of a high-speed railway bridge, selecting a candidate data set from the picture data set, and selecting a current subset from the candidate data set; selecting a plurality of samples from the candidate data set as new training samples in a random sampling mode, marking the training samples, and adding the training samples into the training set; updating a deep neural network model for crack recognition by using the training set, and calculating the relative loss of the deep neural network model to the prediction result of the picture data set in the two adjacent iterative training processes; if the relative loss of the prediction result of the picture data set in multiple iterations is smaller than a first allowable value continuously, the deep neural network model is converged; and storing the deep neural network model, and identifying the apparent cracks of the high-speed rail bridge by using the deep neural network model.
If the deep neural network model does not meet the convergence condition, adopting the prediction label of the current subset by the deep neural network model in the last iteration as a real label, and adopting the deep neural network model of the current iteration to predict the prediction label of the current subset; calculating the relative loss of the prediction result of the current subset in two iterations according to the loss functions of the real label, the prediction label and the network, and judging whether the current subset meets the convergence condition; when the relative loss of the prediction result of the current subset in a plurality of continuous iterations is less than a second allowable value, the current subset meets the convergence condition, and the current subset is deleted from the candidate data set so as to narrow the search range of active learning; and predicting samples of the candidate data set by adopting the deep neural network model, and calculating the uncertainty of the samples of the candidate data set based on the information entropy of the prediction result so as to update the uncertainty of the samples of the candidate data set.
Calculating the uncertainty of the candidate data set sample based on the information entropy of the prediction result, comprising:
the uncertainty calculation formula based on the information entropy is as follows:
p(S i ) Is the probability that a pixel is predicted as i class, N is the total number of classes, and N is the total number of pixels of the sample picture.
In one embodiment, the bridge apparent crack identification method based on adaptive subset search and deep learning comprises the following steps: acquiring an apparent crack picture data set of a high-speed railway bridge, selecting a candidate data set from the picture data set, and selecting a current subset from the candidate data set; selecting a plurality of samples from the candidate data set as new training samples in a random sampling mode, marking the training samples, and adding the training samples into the training set; updating a deep neural network model for crack recognition by using the training set, and calculating the relative loss of the deep neural network model to the prediction result of the picture data set in the two adjacent iterative training processes; if the relative loss of the prediction result of the picture data set in multiple iterations is smaller than a first allowable value continuously, the deep neural network model is converged; and storing the deep neural network model, and identifying the apparent cracks of the high-speed rail bridge by using the deep neural network model.
If the deep neural network model does not meet the convergence condition, adopting the prediction label of the current subset by the deep neural network model in the last iteration as a real label, and adopting the deep neural network model of the current iteration to predict the prediction label of the current subset; calculating the relative loss of the prediction result of the current subset in two iterations according to the loss functions of the real label, the prediction label and the network, and judging whether the current subset meets the convergence condition; when the relative loss of the prediction result of the current subset in a plurality of continuous iterations is less than a second allowable value, the current subset meets the convergence condition, and the current subset is deleted from the candidate data set so as to narrow the search range of active learning; and predicting samples of the candidate data set by using the deep neural network model, and calculating the uncertainty of the samples of the candidate data set based on the information entropy of the prediction result so as to update the uncertainty of the samples of the candidate data set.
According to the uncertainty of the sample, selecting the top N with the minimum uncertainty from the candidate data set s Samples as a new current subset, where N s Is defined as:
N s ＝p s N c
p s is a proportionality coefficient, N c The number of remaining candidate dataset samples.
In one embodiment, the bridge apparent crack identification method based on adaptive subset search and deep learning comprises the following steps: acquiring an apparent crack picture data set of a high-speed railway bridge, selecting a candidate data set from the picture data set, and selecting a current subset from the candidate data set; selecting a plurality of samples from the candidate data set as new training samples in a random sampling mode, marking the training samples, and adding the training samples into the training set; updating a deep neural network model for crack recognition by using the training set, and calculating the relative loss of the deep neural network model to the prediction result of the picture data set in the two adjacent iterative training processes; if the relative loss of the prediction result of the picture data set in two iterations is less than a first allowable value, the first allowable value is 0.15, and the deep neural network model is converged; and storing the deep neural network model, and identifying the apparent cracks of the high-speed rail bridge by using the deep neural network model.
In one embodiment, the bridge apparent crack identification method based on adaptive subset search and deep learning comprises the following steps: acquiring an apparent crack picture data set of a high-speed railway bridge, selecting a candidate data set from the picture data set, and selecting a current subset from the candidate data set; selecting a plurality of samples from the candidate data set as new training samples in a random sampling mode, marking the training samples, and adding the training samples into the training set; updating a deep neural network model for crack recognition by using the training set, and calculating the relative loss of the deep neural network model to the prediction result of the picture data set in the two adjacent iterative training processes; if the relative loss of the prediction result of the picture data set in multiple iterations is smaller than a first allowable value, the deep neural network model is converged; and storing the deep neural network model, and identifying the apparent cracks of the high-speed rail bridge by using the deep neural network model.
If the deep neural network model does not meet the convergence condition, adopting the deep neural network model in the last iteration to predict the prediction label of the current subset as a real label, and adopting the deep neural network model of the current iteration to predict the prediction label of the current subset; calculating the relative loss of the prediction result of the current subset in two iterations according to the loss functions of the real label, the prediction label and the network, and judging whether the current subset meets the convergence condition; when the relative loss of the prediction result of the current subset in a continuous iteration is smaller than a second allowable value, the second allowable value is 0.3, and the current subset meets the convergence condition, deleting the current subset from the candidate data set so as to narrow the search range of active learning; and predicting samples of the candidate data set by adopting the deep neural network model, and calculating the uncertainty of the samples of the candidate data set based on the information entropy of the prediction result so as to update the uncertainty of the samples of the candidate data set.
The invention also provides a bridge apparent crack identification system based on adaptive subset search and deep learning, which comprises the following steps: the system comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the computer program to realize the bridge apparent crack identification method based on adaptive subset search and deep learning.
The processor and memory may be connected by a bus or other means.
The memory, which is a non-transitory computer readable storage medium, may be used to store non-transitory software programs as well as non-transitory computer executable programs. Further, the memory may include high speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory optionally includes memory located remotely from the processor, and these remote memories may be connected to the processor through a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The above-described embodiments of the apparatus are merely illustrative, wherein the units illustrated as separate components may or may not be physically separate, i.e. may be located in one place, or may also be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment.
Furthermore, an embodiment of the present invention also provides a computer-readable storage medium, which stores computer-executable instructions, which are executed by a processor or a controller, for example, by a processor in the terminal embodiment, and can make the processor execute the bridge apparent crack identification method based on adaptive subset search and deep learning in the embodiment.
It will be understood by those of ordinary skill in the art that all or some of the steps, systems, and methods disclosed above may be implemented as software, firmware, hardware, or suitable combinations thereof. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on computer readable media, which may include computer storage media (or non-transitory media) and communication media (or transitory media). The term computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data, as is well known to those of ordinary skill in the art. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital Versatile Disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can accessed by a computer. In addition, communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media as known to those skilled in the art.
While the preferred embodiments of the present invention have been described in detail, it will be understood by those skilled in the art that the foregoing and various other changes, omissions and deviations in the form and detail thereof may be made without departing from the scope of this invention.
The above-described embodiments of the present invention should not be construed as limiting the scope of the present invention. Any other corresponding changes and modifications made according to the technical idea of the present invention should be included in the protection scope of the claims of the present invention.
Claims (10)
1. A bridge apparent crack identification method based on adaptive subset search and deep learning is characterized by comprising the following steps:
acquiring an apparent crack picture data set of a high-speed railway bridge, selecting a candidate data set from the picture data set, and selecting a current subset from the candidate data set;
selecting a plurality of samples from the candidate data set in a random sampling mode to serve as new training samples, marking the training samples, and adding the training samples into a training set;
updating a deep neural network model for crack recognition by using the training set, and calculating the relative loss of the deep neural network model to the prediction result of the picture data set in the two adjacent iterative training processes;
if the relative loss of the image data set prediction result in a plurality of iterations is smaller than a first allowable value continuously, the deep neural network model converges;
and storing the deep neural network model, and identifying the apparent cracks of the high-speed rail bridge by using the deep neural network model.
2. The bridge apparent crack identification method based on adaptive subset search and deep learning of claim 1, characterized in that if the deep neural network model does not satisfy the convergence condition, the predicted label of the current subset by the deep neural network model in the last iteration is used as a real label, and the predicted label of the current subset is predicted by the deep neural network model in the current iteration;
calculating the relative loss of the prediction result of the current subset in two iterations according to the loss functions of the real label, the prediction label and the network, and judging whether the current subset meets the convergence condition;
when the relative loss of the prediction result of the current subset in a plurality of continuous iterations is smaller than a second allowable value, the current subset meets the convergence condition, and the current subset is deleted from the candidate data set so as to narrow the search range of active learning;
and predicting samples of the candidate data set by adopting a deep neural network model, and calculating the uncertainty of the samples of the candidate data set based on the information entropy of the prediction result so as to update the uncertainty of the samples of the candidate data set.
3. The bridge apparent crack identification method based on adaptive subset search and deep learning of claim 2, characterized in that after judging whether the current subset meets the convergence condition, the method comprises the following steps:
and if the current subset does not meet the convergence condition, calculating the uncertainty of the candidate data set samples based on the information entropy of the prediction result so as to update the uncertainty of the candidate data set samples.
4. The bridge apparent crack identification method based on adaptive subset search and deep learning of claim 1, wherein the step of calculating the relative loss of the deep neural network model to the prediction result of the picture data set in the two adjacent iterative training processes comprises the following steps:
and adopting the prediction label of the deep neural network model on the picture data set in the last iteration as a real label, adopting the deep neural network model of the current iteration to predict the prediction label of the picture data set, and then calculating the relative loss of the data set in the two iterations according to the real label, the prediction label and the loss function of the deep neural network model.
5. The bridge apparent crack identification method based on adaptive subset search and deep learning of claim 4, wherein the Loss function of the deep neural network model comprises two parts, namely a Dice Loss function and a cross entropy Loss function;
wherein the Dice Loss function is defined as:
in the formula (I), the compound is shown in the specification,
and &>
Respectively classifying predicted and real pixels, wherein N is the total number of the pixels;
the cross entropy loss function is defined as:
x is input and output of the last layer of the deep neural network model, and class is a label index value of a pixel reality;
the loss function of the deep neural network model is as follows:
loss＝loss CE +loss Dice 。
6. the bridge apparent crack identification method based on adaptive subset search and deep learning according to claim 2 or 3, wherein the step of calculating the uncertainty of the candidate data set sample based on the information entropy of the prediction result comprises the following steps:
the uncertainty calculation formula based on the information entropy is as follows:
p(S i ) Is the probability that a pixel is predicted as i class, N is the total number of classes, and N is the total number of pixels of the sample picture.
7. The bridge apparent crack identification method based on adaptive subset search and deep learning of claim 2, characterized in that after updating the uncertainty of the samples of the candidate data set, comprising the steps of:
according to the uncertainty of the sample, selecting the top N with the minimum uncertainty from the candidate data set s Samples as a new current subset, where N s Is defined as:
N s ＝p s N c
p s is a proportionality coefficient, N c The number of remaining candidate dataset samples.
8. The bridge apparent crack identification method based on adaptive subset search and deep learning of claim 1, wherein if the relative loss of the prediction result of the picture data set in a plurality of iterations is less than a first allowable value, the deep neural network model converges, comprising the steps of:
if the relative loss of the picture data set prediction result in two iterations is less than the first allowable value, the first allowable value is 0.15, and the deep neural network model converges.
9. The method for identifying apparent cracks on a bridge based on adaptive subset search and deep learning of claim 1, wherein when the relative loss of the prediction result of the current subset in a plurality of continuous iterations is less than a second allowable value, the current subset satisfies a convergence condition, comprising the steps of:
when the relative loss of the prediction result of the current subset in a continuous iteration is smaller than a second allowable value, wherein the second allowable value is 0.3, the current subset meets the convergence condition.
10. A bridge apparent crack identification system based on adaptive subset search and deep learning is characterized by comprising the following steps: a memory, a processor and a computer program stored on the memory and executable on the processor, the processor when executing the computer program implementing the method for bridge apparent crack identification based on adaptive subset search and deep learning according to any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211670986.4A CN115880533A (en) | 2022-12-26 | 2022-12-26 | Bridge apparent crack identification method based on adaptive subset search and deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211670986.4A CN115880533A (en) | 2022-12-26 | 2022-12-26 | Bridge apparent crack identification method based on adaptive subset search and deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115880533A true CN115880533A (en) | 2023-03-31 |
Family
ID=85755519
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211670986.4A Pending CN115880533A (en) | 2022-12-26 | 2022-12-26 | Bridge apparent crack identification method based on adaptive subset search and deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115880533A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116975626A (en) * | 2023-06-09 | 2023-10-31 | 浙江大学 | Automatic updating method and device for supply chain data model |
-
2022
- 2022-12-26 CN CN202211670986.4A patent/CN115880533A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116975626A (en) * | 2023-06-09 | 2023-10-31 | 浙江大学 | Automatic updating method and device for supply chain data model |
| CN116975626B (en) * | 2023-06-09 | 2024-04-19 | 浙江大学 | Automatic updating method and device for supply chain data model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110176001B (en) | Grad-CAM algorithm-based high-speed rail contact net insulator damage accurate positioning method | |
| CN109189767B (en) | Data processing method and device, electronic equipment and storage medium | |
| CN110991311A (en) | Target detection method based on dense connection deep network | |
| CN106919957B (en) | Method and device for processing data | |
| CN110781818B (en) | Video classification method, model training method, device and equipment | |
| CN112633149A (en) | Domain-adaptive foggy-day image target detection method and device | |
| CN110909794A (en) | Target detection system suitable for embedded equipment | |
| CN112613569A (en) | Image recognition method, and training method and device of image classification model | |
| CN111723815A (en) | Model training method, image processing method, device, computer system, and medium | |
| CN115880533A (en) | Bridge apparent crack identification method based on adaptive subset search and deep learning | |
| CN114449343A (en) | Video processing method, device, equipment and storage medium | |
| CN116089883A (en) | Training method for improving classification degree of new and old categories in existing category increment learning | |
| CN113223011B (en) | Small sample image segmentation method based on guide network and full-connection conditional random field | |
| CN114565803A (en) | Method, device and mechanical equipment for extracting difficult sample | |
| CN111583321A (en) | Image processing apparatus, method and medium | |
| CN112508126A (en) | Deep learning model training method and device, electronic equipment and readable storage medium | |
| CN113052217A (en) | Prediction result identification and model training method and device thereof, and computer storage medium | |
| CN109543571B (en) | Intelligent identification and retrieval method for special-shaped processing characteristics of complex products | |
| CN115359322A (en) | Target detection model training method, device, equipment and storage medium | |
| CN113869317A (en) | License plate recognition method and device, electronic equipment and storage medium | |
| CN108985526B (en) | Transportation capacity prediction method and device, computer readable storage medium and terminal | |
| CN113192106A (en) | Livestock tracking method and device | |
| CN112507912A (en) | Method and device for identifying illegal picture | |
| CN111428770A (en) | Network model training and inductance defect identification method and device, and electronic equipment | |
| CN114463584B (en) | Image processing method, model training method, device, apparatus, storage medium, and program |
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 |