CN110866872A - Pavement crack image preprocessing intelligent selection method and device and electronic equipment - Google Patents
Pavement crack image preprocessing intelligent selection method and device and electronic equipment Download PDFInfo
- Publication number
- CN110866872A CN110866872A CN201910959954.8A CN201910959954A CN110866872A CN 110866872 A CN110866872 A CN 110866872A CN 201910959954 A CN201910959954 A CN 201910959954A CN 110866872 A CN110866872 A CN 110866872A
- Authority
- CN
- China
- Prior art keywords
- neural network
- preprocessing
- picture
- pavement crack
- network model
- 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.)
- Granted
Links
- 238000007781 pre-processing Methods 0.000 title claims abstract description 142
- 238000010187 selection method Methods 0.000 title claims abstract description 24
- 238000003062 neural network model Methods 0.000 claims abstract description 93
- 230000009471 action Effects 0.000 claims abstract description 80
- 238000012549 training Methods 0.000 claims abstract description 47
- 238000012545 processing Methods 0.000 claims description 29
- 230000006870 function Effects 0.000 claims description 28
- 238000013528 artificial neural network Methods 0.000 claims description 25
- 238000000605 extraction Methods 0.000 claims description 18
- 238000001914 filtration Methods 0.000 claims description 13
- 230000002146 bilateral effect Effects 0.000 claims description 11
- 238000010586 diagram Methods 0.000 claims description 11
- 239000011159 matrix material Substances 0.000 claims description 11
- 230000000877 morphologic effect Effects 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 3
- 238000000034 method Methods 0.000 abstract description 22
- 230000002787 reinforcement Effects 0.000 abstract description 8
- 230000000875 corresponding effect Effects 0.000 description 14
- 230000000694 effects Effects 0.000 description 12
- 238000004891 communication Methods 0.000 description 10
- 230000008859 change Effects 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 230000007797 corrosion Effects 0.000 description 4
- 238000005260 corrosion Methods 0.000 description 4
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 230000003628 erosive effect Effects 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 230000007704 transition Effects 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000013529 biological neural network Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011217 control strategy Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000010339 dilation Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000006740 morphological transformation Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20036—Morphological image processing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30132—Masonry; Concrete
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an intelligent selection method and device for pavement crack image preprocessing and electronic equipment, wherein the pavement crack image preprocessing method comprises the following steps: acquiring a pavement crack picture to be processed; determining a preprocessing action according to the target neural network model aiming at the pavement crack picture to be processed; the target neural network model is obtained by training based on a training data set containing pavement crack pictures with different characteristics; and preprocessing the pavement crack picture to be processed by utilizing the preprocessing action. The intelligent selection method for pavement crack image preprocessing is based on a depth reinforcement learning algorithm, and an algorithm for automatically selecting an optimal preprocessing mode is realized aiming at pavement image data with different characteristics, so that the accuracy of pavement crack identification is further improved.
Description
Technical Field
The invention relates to the technical field of picture processing in artificial intelligence, in particular to an intelligent selection method and device for pavement crack picture preprocessing and electronic equipment.
Background
When identifying cracks in a road surface picture, the quality of the image directly affects the design of a subsequent algorithm and the precision of the effect thereof, and therefore, before identifying, the picture needs to be subjected to proper preprocessing operation. The image preprocessing is mainly used for eliminating some noise point information in the image, enhancing the readability of effective information, optimizing data information to a greater extent and providing high-quality conditions for the processing of subsequent images. The existing method often cannot select an optimal preprocessing mode aiming at road surface pictures with different characteristics, and accordingly cannot carry out targeted preprocessing operation on the road surface pictures.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide an intelligent selection method and an intelligent selection device for pavement crack image preprocessing and electronic equipment.
Based on the above purpose, the invention provides an intelligent selection method for preprocessing a pavement crack picture, which comprises the following steps:
acquiring a pavement crack picture to be processed;
determining a preprocessing action according to the target neural network model aiming at the pavement crack picture to be processed; the target neural network model is obtained by training based on a training data set containing pavement crack pictures with different characteristics;
and preprocessing the pavement crack picture to be processed by utilizing the preprocessing action.
In an embodiment of the present invention, the step of training the target neural network model based on road surface crack images with different characteristics includes:
setting circulation conditions;
randomly initializing a parameter theta of the neural network model and a parameter theta 'of the target neural network model to enable theta' to be theta;
processing the training data set by using the neural network model, and creating a memory bank P according to a processing result;
when the memory bank P meets the circulation condition, calculating a mean square error loss function according to data in the memory bank P, and updating the parameter theta of the neural network model according to the mean square error loss function;
and updating the parameter theta 'of the target neural network until the update frequency of the target neural network is met, so that the theta' is equal to the current parameter theta of the neural network model, and obtaining the target neural network model.
In an embodiment of the present invention, the neural network model is used to process the training data set, and the step of creating the memory library P according to the processing result includes:
randomly selecting a pavement crack picture from the training data set as an original picture, performing feature extraction to obtain an original picture feature picture as an initial state st；
Will be in the initial state stAs the input of the neural network model, obtaining the Q value output corresponding to each preprocessing action;
selecting the preprocessing action a that the original picture should take according to the epsilon-greedy policyt；
The original picture performs the selected pre-processing action atObtaining a processed picture, extracting the characteristics, and obtaining a processed picture characteristic diagram as a next state st+1；
Calculating the recognition accuracy rate acc 'of the processed picture, and comparing the recognition accuracy rate acc' with the recognition accuracy rate acc of the original picture to obtain the reward information rt；
Will(s)t,at,rt,st+1) To be stored in the memory bank P.
In an embodiment of the present invention, when the memory bank P satisfies the loop condition, a mean square error loss function is calculated according to data in the memory bank P, and the step of updating the parameter θ of the neural network model according to the mean square error loss function includes:
randomly selecting data from the memory bank P, and calculating the current Q value Q ═ rt+γ×max(Q'(st+1,at+1θ'), where γ is the attenuation parameter;
calculating a mean square error loss function according to the current Q value
Wherein m is the batch data volume;
and updating the parameter theta of the neural network model through back propagation of the neural network model according to the mean square error loss function.
In one embodiment of the present invention, the satisfying of the cycle condition is: current training round < ImaxWhile the current training round < EmaxWhile the current number of steps is < TmaxAnd the current round has not been finished yet; wherein, ImaxFor maximum training period, EmaxFor maximum number of exploration rounds, TmaxMaximum number of steps per round;
the step of updating the parameter theta 'of the target neural network until the update frequency of the target neural network is met, so that the theta' is equal to the current parameter theta of the neural network model, and obtaining the target neural network model comprises the following steps:
until the current number of rounds is EupdateThe integral multiple of the target neural network, updating a parameter theta 'of the target neural network to enable the theta' to be equal to the current parameter theta of the neural network model, and obtaining the target neural network model; wherein E isupdateThe frequency is updated for the target network.
In an embodiment of the present invention, the step of determining a preprocessing action according to the target neural network model for the road surface crack image to be processed includes:
performing feature extraction on the pavement crack picture to be processed to obtain a feature picture of the picture to be processed as an initial state st；
Will be in the initial state stThe Q value output corresponding to each preprocessing action is obtained as the input of the target neural network model;
determining a preprocessing action a to be taken by a pavement crack picture to be processed according to an epsilon-greedy strategyt。
In one embodiment of the present invention, the feature extraction is to obtain a 704 × 1088 matrix through a feature extraction network;
determining a preprocessing action a to be taken by the pavement crack picture to be processed according to an epsilon-greedy strategytComprises the following steps:
selecting the current optimal preprocessing action a corresponding to the maximum Q value by the probability 1-epsilontOr the pretreatment action a to be taken by randomly selecting the pavement crack picture to be treated by epsilont。
In one embodiment of the invention, the pre-processing action includes histogram equalization, bilateral filtering, morphological on-operation or direct output.
Based on the same inventive concept, the invention also provides an intelligent selection device for preprocessing the pavement crack picture, which comprises:
the acquisition module is used for acquiring a pavement crack picture to be processed;
the determination module is used for determining a preprocessing action according to the target neural network model aiming at the pavement crack picture to be processed; the target neural network model is obtained by training based on a training data set containing pavement crack pictures with different characteristics;
and the processing module is used for preprocessing the pavement crack picture to be processed by utilizing the preprocessing action.
Based on the same inventive concept, the invention further provides electronic equipment which 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 program to realize the intelligent selection method for preprocessing the pavement crack image.
Compared with the prior art, the invention has the following beneficial effects:
(1) by adopting the deep reinforcement learning algorithm, the perception capability of deep learning and the decision capability of reinforcement learning can be combined mutually, the advantages are complementary, the control strategy can be learned directly from high-dimensional original data, the method is an artificial intelligence method closer to the human thinking mode, and meanwhile, the intelligent selection of preprocessing can be better realized.
(2) The system adopts a mode based on end-to-end sensing and control, and has strong universality. The original image and the picture after the specific preprocessing operation method are used as the original image input, the neck support deep convolution neural network and the full-connection neural network output the state action function, and therefore end-to-end learning control is achieved.
(3) The convolutional neural network has greater advantages in image processing. The weight sharing network structure of the convolutional neural network is similar to that of a biological neural network, the convolutional neural network can reduce the complexity of a network model and the number of weights, and an image can be directly used as the input of the network when the image is processed, so that the complex processes of feature extraction and data reconstruction in the traditional recognition algorithm are avoided.
(4) With the empirical playback technique, it is possible to increase the efficiency of use between image data while reducing the correlation between data by repeatedly sampling the history data.
Drawings
FIG. 1 is a flow chart of an intelligent selection method for preprocessing a pavement crack picture according to an embodiment of the invention;
FIG. 2 is a schematic diagram of end-to-end based access and control according to an embodiment of the present invention;
FIG. 3 is a flowchart of the training steps of the target network model according to an embodiment of the present invention;
FIG. 4 is a flowchart of a method for creating a memory bank P according to an embodiment of the present invention;
FIG. 5 is a flowchart of a method for updating a parameter θ of a neural network model according to an embodiment of the present invention;
FIG. 6 is a basic flowchart illustrating the training steps of the target network model according to an embodiment of the present invention;
FIG. 7 is a schematic structural diagram of an intelligent selection device for preprocessing a pavement crack picture according to an embodiment of the invention;
fig. 8 is a schematic structural diagram of an electronic device according to an embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to specific embodiments and the accompanying drawings.
The preprocessing method adopted for the pavement crack picture in the prior art mainly comprises the following steps: histogram equalization, bilateral filtering, morphological opening operation, and the like. Each of the pretreatment modes will be described in detail below.
(1) Histogram equalization
An important application of histogram equalization gray scale transformation is generally used in image enhancement processing. The method utilizes the histogram to adjust the gray value of the image, enhances the global contrast of the image and enables the gray value to be more uniformly distributed on the histogram. The image processed by the histogram equalization is clearer in texture and richer in details, and meanwhile, the entropy of the image information is increased, so that an observer can obtain more useful information from the image.
Histogram equalization is to expand the gray area in a certain pixel point set in the original image and redistribute the pixel values, so that the gray histogram of the original image is changed from a certain gray interval in a comparative set to be uniformly distributed in the whole gray range, and the essence of the algorithm is to find a gray value mapping:
wherein A is0Representing image area (i.e. pixel summary), DmaxIs the maximum gray value of the original image, Da，DbBefore and after conversion, respectively, of gray values HiThe number of pixels is the i-th gray scale. And further acquiring a mapping relation from each different gray level of all the source images to each different gray level of the target image, and performing gray level conversion according to the mapping relation to finish histogram equalization operation.
(2) Bilateral filtering
Bilateral filtering has a very wide application in image processing, and belongs to a filter for edge preserving and denoising. Edge preserving filters refer to a class of special filters that effectively preserve edge information in an image during filtering. Bilateral filtering can preserve image edge information and the characteristics of the contour because this method can take into account both spatial and intensity differences of pixels. The bilateral filter evolves on the basis of gaussian filtering.
Wherein G isd(x, y) represents the distance Gaussian weight between pixels x, y, Gr(p (x), p (y)) represent the grey value gaussian weights between pixels x, y. From the formula, it can be seen that by the superposition of two gaussian weights, the overall appearance of bilateral filtering is: constructing weights by using different distances of pixels to blur noise, wherein the longer the distance is, the smaller the weight is; and (3) constructing a weight value by using the gray-scale values of the pixels to retain the edge information of the image, wherein the weight value is smaller when the gray-scale value difference is larger.
(3) Morphological transformation-on operation
The open operation is a filter based on geometric operations. The opening operation is a commonly used image processing method, and is mainly aimed at removing isolated points, burrs and the like, and simultaneously keeping the overall position and shape unchanged.
The two most basic operations in morphological transformation are erosion and dilation, which rely on a convolution kernel to slide over the original image to change the pixel values. In the erosion operation, if all the pixel values in the sliding convolution kernel are 255, the pixel value of the center element is kept unchanged, and this operation achieves boundary erosion of the foreground (white pixels). In the expansion operation, as long as one pixel in the sliding convolution kernel is 255, the pixel value of the central element is 255, and the operation realizes expansion of the foreground boundary (namely corrosion of the background black pixel). The opening operation is a group of combined operation of firstly carrying out corrosion operation and then carrying out expansion operation, white early points in the picture are eliminated in the corrosion operation process, information except white noise points which is not corroded is expanded in the expansion operation process, and the foreground information is restored, so that the identification accuracy is improved.
In the representation of mathematical morphology, the general technique is to use
Indicating corrosion by
The equation for the expansion, on operation can be expressed as follows:
the difference in the sizes of the structural elements of the on operation results in different filtering effects, and the selection of different structural elements results in different segmentations, i.e., different features are extracted.
In the algorithm of crack recognition, there are many factors influencing the recognition effect, wherein the picture preprocessing is a crucial link. Before the picture data enters the recognition algorithm, the picture data is often required to be cleaned or certain preprocessing operation is carried out, and the quality of the picture preprocessing selection is often directly related to the effect of subsequent recognition.
The inventor of the application finds that the prior method has the following defects: (1) the adopted pretreatment mode is single, and the pertinence is lacked; for data in a certain data set, only a certain preprocessing mode is often adopted for the selection of image preprocessing. However, when data is acquired, the picture quality is often uneven, and the picture is not subjected to targeted preprocessing operation, so that the characteristic quality of the picture is possibly reduced, and high-quality data information cannot be provided for subsequent crack identification operation; (2) horizontal hierarchy of application developers; before picture recognition and other operations, preprocessing operations performed on picture data are independently established by application developers, and currently, there is no specific description on which preprocessing mode different picture data should be processed most effectively. And the level of the application developer is uneven, the selected preprocessing mode is not suitable, and the problem of improper data processing exists.
The method is based on a deep reinforcement learning algorithm, and the algorithm of automatically selecting the optimal preprocessing mode is realized for the pavement picture data with different characteristics, so that the identification accuracy of the pavement cracks is further improved.
As shown in fig. 1, the present embodiment provides an intelligent selection method for preprocessing a pavement crack image, including:
optionally, a high-definition camera is used for collecting pavement crack images, and professional manual marking is performed.
in step 102, the road surface crack images with different characteristics are used as a training data set, a target neural network model is obtained through training, and when the road surface crack images to be processed are input into the target neural model, the target neural network model can automatically select an optimal preprocessing action aiming at the road surface crack images to be processed with different characteristics.
Optionally, the target neural network model is a Q network Based on deep reinforcement Learning (DQN), the DQN belongs to a Q-Learning algorithm, and is also a Value-Based algorithm, and instead of Learning a Policy directly, Learning criticic learns how to evaluate the current state, and further selects an optimal action according to the Q Value. Therefore, the neural network in the DQN can be regarded as a complex Q-function, the traditional reinforcement learning stores each state and the Q value owned by each behavior in the state in a table form, in the image field, the state information is difficult to store in the table form, the state and the action are used as the input of the neural network, namely the Q network, the Q value of the action is obtained after the neural network analysis, and therefore the Q value can be directly generated by using the neural network, and the optimal action is selected according to the Q value.
And 103, preprocessing the to-be-processed pavement crack picture by using the preprocessing action, namely executing the preprocessing action determined by the target neural network model on the to-be-processed pavement crack picture.
In step 103, the preprocessing action includes histogram equalization, bilateral filtering, morphological on operation or direct output (without any preprocessing action), but is not limited thereto.
The pavement crack image preprocessing intelligent selection method is developed based on a Python language and is realized by adopting a Pythrch frame.
The pavement crack image preprocessing intelligent selection method is based on a depth reinforcement learning algorithm, firstly, the identification effect of the current pavement crack is judged, a preprocessing mode is selected for an image with low identification accuracy according to a certain strategy, and then the image after preprocessing change is input into a crack identification network system again to identify the pavement crack, so that the pavement crack identification effect is improved. The key of this section requires the construction of several key elements: context, status, action, and reward functions.
1. Environment(s)
The characteristic space formed by all the road surface image data sets and the image data sets formed after the image data sets are subjected to certain preprocessing actions is used as the environment of the neural network model. In the environment, the neural network model distinguishes images with different characteristics through continuous learning and training, and which preprocessing mode is more beneficial to subsequent pavement crack identification.
2. Status of state
The state that can be observed by the neural network model in this environment is the feature map of the image. The original image feature map (i.e. the image without any pre-processing) is used as the initial state, and the image feature map after a certain pre-processing selection is used as the new state.
3. Movement of
The action of the neural network model is the transformation performed on the original image, and the transformation is the different preprocessing operations performed on the original image. These operations include: histogram equalization, morphological open operation, bilateral filtering, and direct output of the original picture (i.e., without any preprocessing action), for a total of 4 actions.
4. Return function
After the neural network model executes the action, some return information is received and is used as a basis for strategy learning. The return information is represented by whether or not the subsequent road surface crack recognition effect becomes better after the action of performing the preprocessing operation. And aiming at each input image, taking the confidence coefficient of correct recognition in the image recognition algorithm as a measure index. If the image is transferred to a new state after the action of the preprocessing operation is executed, the crack identification process is carried out again, the confidence coefficient of the real category is improved, the preprocessing intelligent selection system receives a certain positive return feedback, and otherwise, the preprocessing intelligent selection system receives a negative return feedback.
Wherein p istRepresenting the confidence at the moment of execution of the current action, CtrueRepresents a crack, raA reported value greater than 0.
As shown in fig. 2, the intelligent selection method for preprocessing the pavement crack image in the embodiment acquires target observation information from an environment based on end-to-end sensing and control, maps the current state to a corresponding action, provides state information in the current environment, judges the value of the action based on expected return, and has strong universality.
As shown in fig. 3, optionally, the step of training the target neural network model based on the road surface crack images with different characteristics includes:
In the present embodiment, the parameter θ of the neural network model is continuously updated, and the parameter θ' of the target neural network model is updated when the update frequency of the target neural network is satisfied.
As shown in fig. 4, optionally, in step 203, the step of processing the training data set by using the neural network model, and the step of creating a memory library P according to the processing result includes:
Extracting the characteristics of the original picture through a characteristic extraction network to obtain a matrix of 704 multiplied by 1088 as an initial state st(ii) a The adopted data is a single-channel gray-scale image, and the size of the single-channel gray-scale image is uniformly set to be 704 multiplied by 1088;
And taking the characteristics of the image as a state, wherein the original state is the characteristics of the original road image, the new state is the image characteristics after preprocessing, and the state transition is performed after each preprocessing operation.
The deep reinforcement learning is generally used for processing sequence state conversion, and here, the effect of a crack recognition algorithm is improved, intelligent selection operation is performed on preprocessing operation, and action and state transition change are different from general sequence state transition change.
The setting of the return function directly influences the quality of the whole pavement crack image preprocessing method, and the return function shows whether the following pavement crack identification effect becomes better or not after the preprocessing action is executed. The original data picture and the data after the specific preprocessing action are input into a crack identification network model, the identification accuracy rate acc 'of the processed picture and the identification accuracy rate acc of the original picture are respectively calculated, and if acc' is greater than acc, forward reward information r is obtainedtAfter the original image is subjected to the specific preprocessing action, the subsequent pavement crack identification effect becomes better, namely the original image is effective after the specific preprocessing action; if acc' is less than acc, obtaining reverse reward information rtAfter the original image is subjected to the specific preprocessing action, the subsequent pavement crack identification effect is not better, namely the original image is invalid after the specific preprocessing action;
As shown in fig. 5, optionally, in step 204, when the memory bank P satisfies the loop condition, the step of calculating a mean square error loss function according to data in the memory bank P, and updating the parameter θ of the neural network model according to the mean square error loss function includes:
and step 403, updating a parameter theta of the neural network model through back propagation of the neural network model according to the mean square error loss function.
For the neural network model, an original data picture and data subjected to specific preprocessing operation are firstly input into a crack recognition network, pavement cracks are recognized, and the accuracy of crack recognition is obtained. The neural network model is continuously learned to obtain a target neural network model, and the target neural network model can automatically select an optimized preprocessing mode aiming at pictures with different characteristics, so that the crack identification effect is improved.
Selecting preprocessing operation according to an epsilon-greedy strategy, executing the selected preprocessing operation on a road surface image, taking the change degree of the accuracy rate of a crack identification algorithm as a measurement standard of return information, feeding the return information back to an intelligent selection system, and updating a network by adopting gradient descent in the preprocessing intelligent selection algorithm.
As shown in fig. 6, optionally, the intelligent selection method for preprocessing the pavement crack image includes:
randomly selecting a road surface picture from the training data set, and extracting the characteristics of the road surface picture to obtain a matrix of 704 multiplied by 1088 as a state st；
The matrix is firstly input into a trained crack identification model to obtain the crack identification accuracy acc. (ii) a
Simultaneously, the matrix is input into a Q network, the Q network can select the preprocessing action to be taken by the original picture, and the pavement picture executes the preprocessing action at；
Extracting the characteristics of the processed picture to obtain a matrix of 704 multiplied by 1088 as a next state st+1；
Inputting the processed picture matrix into a trained crack recognition model, calculating the recognition accuracy acc 'of the processed picture, and comparing the recognition accuracy acc' of the processed picture with the recognition accuracy acc of the original picture to obtain reward information rt；
Will(s)t,at,rt,st+1) Storing the data into a memory bank P;
randomly selecting data from a memory library P to train the Q network.
In this embodiment, the training steps of the target neural network model are as follows:
the main training steps of the target neural network model will be illustrated below:
randomly selecting picture A from the database to obtain a matrix of 704 × 1088 as an initial state st；
In Q network, s istAs input, obtaining Q value output corresponding to all preprocessing actions (histogram equalization, morphological open operation, bilateral filtering or direct output of original picture respectively) of the Q network:
actions_value:[0.4076,-0.0076,0.2175,0.0901]；
selecting the preprocessing action a corresponding to the maximum Q value by the probability 1-epsilontOr randomly select action a with epsilont；
In this example, the preprocessing operation with the maximum Q value, namely the operation 1 histogram equalization preprocessing operation, is selected;
preprocessing the picture A by using histogram equalization to obtain a state s corresponding to the picture At+1；
Calculating the crack identification accuracy acc ' of the picture A ' through a crack identification model, and comparing the crack identification accuracy acc ' with the accuracy acc of the picture A;
acc' > acc, then get the forward reward value rt；
Will(s)t,at,rt,st+1) Storing the data into a memory bank P;
randomly selecting data from memory bank P
Calculating the current Q value Q ═ rt+γ×max(Q'(st+1,at+1,θ'))
Using a mean square error loss function
The parameter θ of the Q network is updated by back propagation through the neural network.
In this embodiment, optionally, in step 102, the step of determining a preprocessing action according to the target neural network model for the road surface crack image to be processed includes:
performing feature extraction on the pavement crack picture to be processed to obtain a feature picture of the picture to be processed as an initial state st(ii) a The characteristic extraction is to obtain a matrix of 704 multiplied by 1088 through a characteristic extraction network;
will be in the initial state stThe Q value output corresponding to each preprocessing operation is obtained as the input of the target neural network model;
determining a preprocessing action a to be taken by a pavement crack picture to be processed according to an epsilon-greedy strategytThe method specifically comprises the following steps: selecting the preprocessing action a corresponding to the maximum Q value by the probability 1-epsilontOr the pretreatment action a to be taken by randomly selecting the pavement crack picture to be treated by epsilont。
The pavement crack picture preprocessing device of the present invention will be described in detail below.
As shown in fig. 7, the present embodiment provides an intelligent selection device for preprocessing a pavement crack picture, which includes:
the acquisition module 11 is used for acquiring a road surface crack picture to be processed;
the determining module 12 is configured to determine a preprocessing action according to the target neural network model for the road surface crack image to be processed; the target neural network model is obtained by training based on a training data set containing pavement crack pictures with different characteristics;
and the processing module 13 is configured to perform preprocessing on the road surface crack image to be processed by using the preprocessing action.
The intelligent selection device for pavement crack image preprocessing carries out intelligent selection on the pavement image preprocessing operation, and can effectively select a more appropriate preprocessing mode and improve the accuracy of crack recognition compared with a pavement crack recognition algorithm adopting a single preprocessing mode.
In this embodiment, optionally, the determining module 12 is further configured to perform the following processing: setting circulation conditions;
randomly initializing a parameter theta of the neural network model and a parameter theta 'of the target neural network model to enable theta' to be theta;
processing the training data set by using the neural network model, and creating a memory bank P according to a processing result;
when the memory bank P meets the circulation condition, calculating a mean square error loss function according to data in the memory bank P, and updating the parameter theta of the neural network model according to the mean square error loss function;
and updating the parameter theta 'of the target neural network until the update frequency of the target neural network is met, so that the theta' is equal to the current parameter theta of the neural network model, and obtaining the target neural network model.
In this embodiment, optionally, the determining module 12 is further configured to perform the following processing:
randomly selecting a pavement crack picture from the training data set as an original picture, and performing feature extraction to obtain the original pictureCharacteristic diagram as initial state st；
Will be in the initial state stAs the input of the neural network model, obtaining the Q value output corresponding to each preprocessing action;
selecting the preprocessing action a that the original picture should take according to the epsilon-greedy policyt；
The original picture performs the selected pre-processing action atObtaining a processed picture, extracting the characteristics, and obtaining a processed picture characteristic diagram as a next state st+1；
Calculating the recognition accuracy rate acc 'of the processed picture, and comparing the recognition accuracy rate acc' with the recognition accuracy rate acc of the original picture to obtain the reward information rt；
Will(s)t,at,rt,st+1) To be stored in the memory bank P.
In this embodiment, optionally, the determining module 12 is further configured to perform the following processing:
randomly selecting data from the memory bank P, and calculating the current Q value Q ═ rt+γ×max(Q'(st+1,at+1θ'), where γ is the attenuation parameter;
calculating a mean square error loss function according to the current Q value
Wherein m is the batch data volume;
and updating the parameter theta of the neural network model through back propagation of the neural network model according to the mean square error loss function.
In this embodiment, optionally, the processing module 13 is further configured to perform the following processing:
the step of determining the preprocessing action according to the target neural network model aiming at the pavement crack picture to be processed comprises the following steps:
performing feature extraction on the pavement crack picture to be processed to obtain a feature picture of the picture to be processed as an initial state st(ii) a The characteristic extraction is to obtain a matrix of 704 multiplied by 1088 through a characteristic extraction network;
will be in the initial stateState stThe Q value output corresponding to each preprocessing action is obtained as the input of the target neural network model;
determining a preprocessing action a to be taken by a pavement crack picture to be processed according to an epsilon-greedy strategyt(ii) a The method specifically comprises the following steps: selecting the preprocessing action a corresponding to the maximum Q value by the probability 1-epsilontOr randomly selecting action a with probability epsilont。
Based on the same inventive concept, the embodiment provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and running on the processor, and when the processor executes the program, the intelligent selection method for preprocessing the road surface crack image according to any one of the above embodiments is implemented.
Fig. 8 is a schematic diagram illustrating a more specific hardware structure of an electronic device according to this embodiment, where the electronic device may include: a processor 1010, a memory 1020, an input/output interface 1030, a communication interface 1040, and a bus 1050. Wherein the processor 1010, memory 1020, input/output interface 1030, and communication interface 1040 are communicatively coupled to each other within the device via bus 1050.
The processor 1010 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an Application Specific Integrated Circuit (ASIC), or one or more Integrated circuits, and is configured to execute related programs to implement the technical solutions provided in the embodiments of the present disclosure.
The Memory 1020 may be implemented in the form of a ROM (Read Only Memory), a RAM (Random access Memory), a static storage device, a dynamic storage device, or the like. The memory 1020 may store an operating system and other application programs, and when the technical solution provided by the embodiments of the present specification is implemented by software or firmware, the relevant program codes are stored in the memory 1020 and called to be executed by the processor 1010.
The input/output interface 1030 is used for connecting an input/output module to input and output information. The i/o module may be configured as a component in a device (not shown) or may be external to the device to provide a corresponding function. The input devices may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output devices may include a display, a speaker, a vibrator, an indicator light, etc.
The communication interface 1040 is used for connecting a communication module (not shown in the drawings) to implement communication interaction between the present apparatus and other apparatuses. The communication module can realize communication in a wired mode (such as USB, network cable and the like) and also can realize communication in a wireless mode (such as mobile network, WIFI, Bluetooth and the like).
It should be noted that although the above-mentioned device only shows the processor 1010, the memory 1020, the input/output interface 1030, the communication interface 1040 and the bus 1050, in a specific implementation, the device may also include other components necessary for normal operation. In addition, those skilled in the art will appreciate that the above-described apparatus may also include only those components necessary to implement the embodiments of the present description, and not necessarily all of the components shown in the figures.
Those of ordinary skill in the art will understand that: the discussion of any embodiment above is meant to be exemplary only, and is not intended to intimate that the scope of the disclosure, including the claims, is limited to these examples; within the idea of the invention, also technical features in the above embodiments or in different embodiments may be combined and there are many other variations of the different aspects of the invention as described above, which are not provided in detail for the sake of brevity. Therefore, any omissions, modifications, substitutions, improvements and the like that may be made without departing from the spirit and principles of the invention are intended to be included within the scope of the invention.
In addition, well known power/ground connections to Integrated Circuit (IC) chips and other components may or may not be shown within the provided figures for simplicity of illustration and discussion, and so as not to obscure the invention. Furthermore, devices may be shown in block diagram form in order to avoid obscuring the invention, and also in view of the fact that specifics with respect to implementation of such block diagram devices are highly dependent upon the platform within which the present invention is to be implemented (i.e., specifics should be well within purview of one skilled in the art). Where specific details (e.g., circuits) are set forth in order to describe example embodiments of the invention, it should be apparent to one skilled in the art that the invention can be practiced without, or with variation of, these specific details. Accordingly, the description is to be regarded as illustrative instead of restrictive.
While the present invention has been described in conjunction with specific embodiments thereof, many alternatives, modifications, and variations of these embodiments will be apparent to those of ordinary skill in the art in light of the foregoing description. For example, other memory architectures (e.g., dynamic ram (dram)) may use the discussed embodiments.
The embodiments of the invention are intended to embrace all such alternatives, modifications and variances that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, substitutions, improvements and the like that may be made without departing from the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (10)
1. The intelligent selection method for preprocessing the pavement crack picture is characterized by comprising the following steps of:
acquiring a pavement crack picture to be processed;
determining a preprocessing action according to the target neural network model aiming at the pavement crack picture to be processed; the target neural network model is obtained by training based on a training data set containing pavement crack pictures with different characteristics;
and preprocessing the pavement crack picture to be processed by utilizing the preprocessing action.
2. The pavement crack picture preprocessing intelligent selection method as claimed in claim 1, wherein the step of training the target neural network model based on the training data set containing pavement crack pictures with different characteristics comprises:
setting circulation conditions;
randomly initializing a parameter theta of the neural network model and a parameter theta 'of the target neural network model to enable theta' to be theta;
processing the training data set by using the neural network model, and creating a memory bank P according to a processing result;
when the memory bank P meets the circulation condition, calculating a mean square error loss function according to data in the memory bank P, and updating the parameter theta of the neural network model according to the mean square error loss function;
and updating the parameter theta 'of the target neural network until the update frequency of the target neural network is met, so that the theta' is equal to the current parameter theta of the neural network model, and obtaining the target neural network model.
3. The intelligent selection method for preprocessing the pavement crack picture according to claim 2, wherein the neural network model is used for processing a training data set, and the step of creating a memory bank P according to the processing result comprises the following steps:
randomly selecting a pavement crack picture from the training data set as an original picture, performing feature extraction to obtain an original picture feature picture as an initial state st；
Will be in the initial state stAs the input of the neural network model, obtaining the Q value output corresponding to each preprocessing action;
selecting the preprocessing action a that the original picture should take according to the epsilon-greedy policyt；
The original picture performs the selected pre-processing action atObtaining a processed picture, extracting the characteristics, and obtaining a processed picture characteristic diagram as a next state st+1；
Calculating the recognition accuracy rate acc 'of the processed picture, and comparing the recognition accuracy rate acc' with the recognition accuracy rate acc of the original picture to obtain the reward information rt；
Will(s)t,at,rt,st+1) To be stored in the memory bank P.
4. The pavement crack image preprocessing intelligent selection method as claimed in claim 3, wherein when the memory bank P meets the circulation condition, the step of calculating a mean square error loss function according to data in the memory bank P, and updating the parameter θ of the neural network model according to the mean square error loss function comprises:
randomly selecting data from the memory bank P, and calculating the current Q value Q ═ rt+γ×max(Q'(st+1,at+1θ'), where γ is the attenuation parameter;
calculating a mean square error loss function according to the current Q value
Wherein m is the batch data volume;
and updating the parameter theta of the neural network model through back propagation of the neural network model according to the mean square error loss function.
5. The pavement crack picture preprocessing intelligent selection method according to claim 2, wherein the satisfying of the cycle condition is: current training round < ImaxWhile the current training round < EmaxWhile the current number of steps is < TmaxAnd the current round has not been finished yet; wherein, ImaxFor maximum training period, EmaxFor maximum number of exploration rounds, TmaxMaximum number of steps per round;
the step of updating the parameter theta 'of the target neural network until the update frequency of the target neural network is met, so that the theta' is equal to the current parameter theta of the neural network model, and obtaining the target neural network model comprises the following steps:
until the current number of rounds is EupdateThe integral multiple of the target neural network, updating a parameter theta 'of the target neural network to enable the theta' to be equal to the current parameter theta of the neural network model, and obtaining the target neural network model; wherein E isupdateThe frequency is updated for the target network.
6. The pavement crack picture preprocessing intelligent selection method as claimed in claim 1, wherein the step of determining a preprocessing action according to the target neural network model for the pavement crack picture to be processed comprises:
performing feature extraction on the pavement crack picture to be processed to obtain a feature picture of the picture to be processed as an initial state st；
Will be in the initial state stThe Q value output corresponding to each preprocessing action is obtained as the input of the target neural network model;
determining a preprocessing action a to be taken by a pavement crack picture to be processed according to an epsilon-greedy strategyt。
7. The intelligent selection method for preprocessing the pavement crack picture according to claim 6, wherein the feature extraction is to obtain a 704 x 1088 matrix through a feature extraction network;
determining a preprocessing action a to be taken by the pavement crack picture to be processed according to an epsilon-greedy strategytComprises the following steps:
selecting the preprocessing action a corresponding to the maximum Q value by the probability 1-epsilontOr the pretreatment action a to be taken by randomly selecting the pavement crack picture to be treated by epsilont。
8. The intelligent selection method for preprocessing the pavement crack picture according to claim 1, wherein the preprocessing action comprises histogram equalization, bilateral filtering, morphological open operation or direct output.
9. The utility model provides a road surface crack picture preliminary treatment intelligence selecting arrangement which characterized in that includes:
the acquisition module is used for acquiring a pavement crack picture to be processed;
the determination module is used for determining a preprocessing action according to the target neural network model aiming at the pavement crack picture to be processed; the target neural network model is obtained by training based on a training data set containing pavement crack pictures with different characteristics;
and the processing module is used for preprocessing the pavement crack picture to be processed by utilizing the preprocessing action.
10. An electronic device, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the intelligent selection method for pavement crack picture preprocessing according to any one of claims 1 to 8 when executing the program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910959954.8A CN110866872B (en) | 2019-10-10 | 2019-10-10 | Pavement crack image preprocessing intelligent selection method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910959954.8A CN110866872B (en) | 2019-10-10 | 2019-10-10 | Pavement crack image preprocessing intelligent selection method and device and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110866872A true CN110866872A (en) | 2020-03-06 |
| CN110866872B CN110866872B (en) | 2022-07-29 |
Family
ID=69652527
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910959954.8A Active CN110866872B (en) | 2019-10-10 | 2019-10-10 | Pavement crack image preprocessing intelligent selection method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110866872B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111489306A (en) * | 2020-03-31 | 2020-08-04 | 天津大学 | Image denoising method based on reinforcement learning |
| CN111489305A (en) * | 2020-03-31 | 2020-08-04 | 天津大学 | Image enhancement method based on reinforcement learning |
| CN111639766A (en) * | 2020-05-26 | 2020-09-08 | 上海极链网络科技有限公司 | Sample data generation method and device |
| CN114758114A (en) * | 2022-04-08 | 2022-07-15 | 北京百度网讯科技有限公司 | Model updating method, image processing method, device, electronic device and medium |
| CN115511836A (en) * | 2022-09-28 | 2022-12-23 | 丽水市市政设施管理中心(丽水市节约用水管理中心) | Bridge crack grade evaluation method and system based on reinforcement learning algorithm |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104182770A (en) * | 2013-05-24 | 2014-12-03 | 塔塔咨询服务有限公司 | Method and system for automatic selection of one or more image processing algorithm |
| CN108496200A (en) * | 2016-01-25 | 2018-09-04 | 皇家飞利浦有限公司 | Pre-processing image data |
| CN108550162A (en) * | 2018-03-27 | 2018-09-18 | 清华大学 | A kind of object detecting method based on deeply study |
| CN109190537A (en) * | 2018-08-23 | 2019-01-11 | 浙江工商大学 | A kind of more personage's Attitude estimation methods based on mask perceived depth intensified learning |
| CN110070487A (en) * | 2019-04-02 | 2019-07-30 | 清华大学 | Semantics Reconstruction face oversubscription method and device based on deeply study |
| CN110084245A (en) * | 2019-04-04 | 2019-08-02 | 中国科学院自动化研究所 | The Weakly supervised image detecting method of view-based access control model attention mechanism intensified learning, system |
-
2019
- 2019-10-10 CN CN201910959954.8A patent/CN110866872B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104182770A (en) * | 2013-05-24 | 2014-12-03 | 塔塔咨询服务有限公司 | Method and system for automatic selection of one or more image processing algorithm |
| CN108496200A (en) * | 2016-01-25 | 2018-09-04 | 皇家飞利浦有限公司 | Pre-processing image data |
| CN108550162A (en) * | 2018-03-27 | 2018-09-18 | 清华大学 | A kind of object detecting method based on deeply study |
| CN109190537A (en) * | 2018-08-23 | 2019-01-11 | 浙江工商大学 | A kind of more personage's Attitude estimation methods based on mask perceived depth intensified learning |
| CN110070487A (en) * | 2019-04-02 | 2019-07-30 | 清华大学 | Semantics Reconstruction face oversubscription method and device based on deeply study |
| CN110084245A (en) * | 2019-04-04 | 2019-08-02 | 中国科学院自动化研究所 | The Weakly supervised image detecting method of view-based access control model attention mechanism intensified learning, system |
Non-Patent Citations (5)
| Title |
|---|
| SHURA_R: "【强化学习】Deep Q Learning（DQN）算法详解", 《HTTPS://BLOG.CSDN.NET/QQ_30615903/ARTICLE/DETAILS/80744083?OPS_REQUEST_MISC=&REQUEST_ID=&BIZ_ID=102&UTM_TERM=DEEP%20QLEARNING&UTM_MEDIUM=DISTRIBUTE.PC_SEARCH_RESULT.NONE-TASK-BLOG-2~ALL~SOBAIDUWEB~DEFAULT-4-80744083.NONECASE&SPM=1018.2226.3001.4187》 * |
| 俞勇等: "《人工智能实践 动手做你自己的AI》", 30 August 2019, 上海科技教育出版社 * |
| 方卫华等: "《跨栏河建筑物安全状态感知、融合与预测》", 30 June 2019, 河海大学出版社 * |
| 许志刚: "《路面破损图像自动识别技术》", 30 September 2018, 西安电子科技大学出版社 * |
| 邓继忠等: "《数字图像处理技术》", 30 September 2005, 广东科技出版社 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111489306A (en) * | 2020-03-31 | 2020-08-04 | 天津大学 | Image denoising method based on reinforcement learning |
| CN111489305A (en) * | 2020-03-31 | 2020-08-04 | 天津大学 | Image enhancement method based on reinforcement learning |
| CN111489305B (en) * | 2020-03-31 | 2023-05-30 | 天津大学 | Image enhancement method based on reinforcement learning |
| CN111639766A (en) * | 2020-05-26 | 2020-09-08 | 上海极链网络科技有限公司 | Sample data generation method and device |
| CN111639766B (en) * | 2020-05-26 | 2023-09-12 | 山东瑞瀚网络科技有限公司 | Sample data generation method and device |
| CN114758114A (en) * | 2022-04-08 | 2022-07-15 | 北京百度网讯科技有限公司 | Model updating method, image processing method, device, electronic device and medium |
| CN115511836A (en) * | 2022-09-28 | 2022-12-23 | 丽水市市政设施管理中心(丽水市节约用水管理中心) | Bridge crack grade evaluation method and system based on reinforcement learning algorithm |
| CN115511836B (en) * | 2022-09-28 | 2023-10-27 | 丽水市市政设施管理中心(丽水市节约用水管理中心) | Bridge crack grade assessment method and system based on reinforcement learning algorithm |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110866872B (en) | 2022-07-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110866872B (en) | Pavement crack image preprocessing intelligent selection method and device and electronic equipment | |
| CN110399929B (en) | Fundus image classification method, fundus image classification apparatus, and computer-readable storage medium | |
| CN113298837B (en) | Image edge extraction method and device, storage medium and equipment | |
| CN110263819A (en) | A kind of object detection method and device for shellfish image | |
| CN108805016B (en) | Head and shoulder area detection method and device | |
| CN110930296B (en) | Image processing method, device, equipment and storage medium | |
| CN108229673B (en) | Convolutional neural network processing method and device and electronic equipment | |
| CN108229675B (en) | Neural network training method, object detection method, device and electronic equipment | |
| CN112836756B (en) | Image recognition model training method, system and computer equipment | |
| CN110910445B (en) | Object size detection method, device, detection equipment and storage medium | |
| CN116894985B (en) | Semi-supervised image classification method and semi-supervised image classification system | |
| CN111862040B (en) | Portrait picture quality evaluation method, device, equipment and storage medium | |
| CN116740362B (en) | Attention-based lightweight asymmetric scene semantic segmentation method and system | |
| CN111882555B (en) | Deep learning-based netting detection method, device, equipment and storage medium | |
| CN115358952B (en) | Image enhancement method, system, equipment and storage medium based on meta-learning | |
| CN108510478B (en) | Lung airway image segmentation method, terminal and storage medium | |
| CN112348808A (en) | Screen perspective detection method and device | |
| CN106023093A (en) | Non-local mean value image denoising method based on improved image black matching | |
| CN111985439B (en) | Face detection method, device, equipment and storage medium | |
| CN117253071B (en) | Semi-supervised target detection method and system based on multistage pseudo tag enhancement | |
| CN112270404A (en) | Detection structure and method for bulge defect of fastener product based on ResNet64 network | |
| CN109063749B (en) | Robust convolution kernel number adaptation method based on angular point radiation domain | |
| CN108447066B (en) | Biliary tract image segmentation method, terminal and storage medium | |
| CN115631108A (en) | RGBD-based image defogging method and related equipment | |
| CN104700416A (en) | Image segmentation threshold determination method based on visual analysis |
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 |