CN113139592A - Method, device and storage medium for identifying lunar meteorite crater based on depth residual error U-Net - Google Patents
Method, device and storage medium for identifying lunar meteorite crater based on depth residual error U-Net Download PDFInfo
- Publication number
- CN113139592A CN113139592A CN202110403032.6A CN202110403032A CN113139592A CN 113139592 A CN113139592 A CN 113139592A CN 202110403032 A CN202110403032 A CN 202110403032A CN 113139592 A CN113139592 A CN 113139592A
- Authority
- CN
- China
- Prior art keywords
- network model
- training
- data
- crater
- net
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000003860 storage Methods 0.000 title claims abstract description 7
- 238000012549 training Methods 0.000 claims abstract description 63
- 238000013507 mapping Methods 0.000 claims abstract description 9
- 238000012216 screening Methods 0.000 claims abstract description 7
- 238000004422 calculation algorithm Methods 0.000 claims description 19
- PEDCQBHIVMGVHV-UHFFFAOYSA-N Glycerine Chemical compound OCC(O)CO PEDCQBHIVMGVHV-UHFFFAOYSA-N 0.000 claims description 17
- 230000006870 function Effects 0.000 claims description 15
- 238000005070 sampling Methods 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 4
- 238000011176 pooling Methods 0.000 claims description 4
- 238000013506 data mapping Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 4
- 238000013527 convolutional neural network Methods 0.000 description 6
- 238000009826 distribution Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000015556 catabolic process Effects 0.000 description 2
- 238000006731 degradation reaction Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000014509 gene expression Effects 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Probability & Statistics with Applications (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a method, equipment and a storage medium for identifying lunar meteorite craters based on a depth residual U-Net, which comprises the steps of constructing a U-Net network model based on a residual unit, mapping and combining first historical data and second historical data to construct a training set, inputting the training set into the U-Net network model for training to obtain a first predicted value, screening and de-weighting the first predicted value to obtain a second predicted value, calculating error loss between the second predicted value and an actual value of the input data, updating weight parameters of the U-Net network model, and identifying the lunar meteorite crater image by using the trained U-Net network model with the error loss meeting a preset threshold. Different characteristics of the network model are combined by designing a U-shaped network based on a residual error unit, so that the identification effect and accuracy of the lunar meteorite crater are improved.
Description
Technical Field
The invention relates to the technical field of image recognition, in particular to a method, equipment and a storage medium for recognizing a lunar meteorite crater based on a depth residual error U-Net.
Background
At present, with the continuous development and progress of deep learning technology, Convolutional Neural Networks (CNNs) have been widely applied to the field of image processing, play an important role in the aspects of target detection, classification and identification, and the like, and in the field of remote sensing images, the appearance of the Convolutional Neural Networks also provides a new method for efficiently identifying meteor craters.
At present, the recognition of meteor craters is divided into visual recognition and automatic recognition algorithms, the detection of the meteor craters is realized through template matching, the meteor crater recognition is completed based on a Canny method and edge matching, and the like.
Some of the currently existing moon meteorite crater recognition algorithms are generally based on an image data set of a partial region, that is, there is a limitation in the data set, or analysis is performed by taking a typical impact crater as an example, although these algorithms can obtain a good effect in a training set region, model mobility is poor, and recognition accuracy of images outside a research region is low.
In a general CNN method, the deeper the hierarchy adopted by the network, the stronger the recognition capability of the network. However, the network with the deeper layer number is more difficult to train, the network training generally adopts a gradient descent and back propagation algorithm to perform parameter learning and optimization, and the too deep network cannot converge due to the fact that the too many layer numbers can cause gradient dispersion or explosion; when the neural network can ensure convergence, the model accuracy gradually saturates with the increase of the network depth and then decreases, namely the network degradation phenomenon; in the training process of the deep network, the parameters learned by the multilayer nonlinear network when the identity mapping is fitted have problems, and the model is degraded.
Disclosure of Invention
The invention aims to provide a method for identifying the lunar meteorite crater based on a depth residual error U-Net, which solves the problem of low identification precision of the traditional CNN meteorite crater identification method.
The invention provides a method for identifying a lunar meteorite crater based on a depth residual error U-Net, which comprises the following steps:
constructing a U-Net network model based on a residual error unit;
mapping and combining the first historical data and the second historical data to construct a training set;
inputting the training set into the U-Net network model for training to obtain a first predicted value;
screening and de-duplicating the first predicted value to obtain a second predicted value;
calculating error loss between the second predicted value and the real value of the input data, updating the weight parameter of the U-Net network model, and training the U-Net network model to be convergent by adopting a random gradient descent algorithm;
and identifying the lunar meteor crater image by using the trained U-Net network model with the error loss meeting the preset threshold.
Further, the residual unit includes:
two 3 x 3 convolution units and an identity map.
Further, the combining the first historical data with the second historical data map to construct the training set comprises:
the first historical data is a full DEM moon image;
the second historical data is meteorite crater recording data;
summarizing and unifying the full DEM moon image and meteor crater recorded data through a python script to obtain a corrected full DEM moon image and corrected meteor crater recorded data;
and marking the corrected meteorite pit recorded data on the corrected full DEM moon image through projection and data mapping.
Further, the inputting the training set into the U-Net network model for training includes:
utilizing a convolution layer and a pooling layer of a U-Net network model encoder to carry out down-sampling on the training set;
up-sampling the down-sampled data by a decoder;
and extracting a first predicted value from the up-sampled data by using a Sigmoid activation function.
Further, the calculating an error loss between the second predicted value and the real value of the input data, updating a weight parameter of the U-Net network model, and training the U-Net network model to converge by using a random gradient descent algorithm specifically includes:
calculating error loss between the second predicted value and the real value of the input data, updating the weight parameter of the U-Net network model through a parameter updating formula, and training the U-Net network model to be convergent by adopting a random gradient descent algorithm;
wherein, the parameter updating formula is as follows:
in the parameter updating formula, theta is weight, eta is learning rate, J (theta; x)i；yi) For the objective function to be optimized, xiFor the meteorite crater abscissa prediction result obtained by training the ith sample data through a U-Net network model, yiAnd obtaining the meteor crater ordinate prediction result of the ith sample through residual U-Net network model training.
Further, the screening and deduplication of the first predicted value comprises: reserving a predicted value which satisfies a first formula and a second formula simultaneously in the first predicted value as a second predicted value;
wherein the first formula is:
((xi-xj)2+(yi-yj)2)/min(ri，rj)2＜Dx，y；
the second formula is:
abs(ri-rj)/min(ri，rj)＜Dr
the first mentionedIn the first formula and the second formula, x is the abscissa of the meteorite crater to be predicted, y is the ordinate of the meteorite crater to be predicted, r is the radius of the meteorite crater to be predicted, i is the number of training samples, i is 0,1,2iAnd xjRespectively obtaining meteorite crater abscissa predicted values y of ith sample data and jth sample data through U-Net network model trainingiAnd yjRespectively obtaining meteorite crater ordinate predicted values r of ith sample data and jth sample data through residual error U-Net network model trainingiAnd rjRespectively obtaining meteorite crater radius predicted values D obtained by training the ith sample data and the jth sample data through a U-Net network modelx,yAnd DrIs an adjustable hyper-parameter.
Further, the calculating an error loss between the second predicted value and a true value of the input data includes:
calculating error loss between the second predicted value and the real value of the input data by adopting a pixel-by-pixel linear cross entropy function;
wherein the linear cross entropy function is:
l is the error loss value, N is the total number of training samples,
the true value of the data representing the ith training sample.
The invention also provides equipment for identifying the lunar meteorite crater, which comprises the following components: the device comprises a memory, a processor and an identification program which is stored on the memory and can run on the processor, wherein when the identification program is executed by the processor, the identification program realizes any one of the above-mentioned moon meteor crater identification methods based on the depth residual U-Net.
The invention also provides a storage medium, which is characterized in that the storage medium is stored with an identification program, and the identification program is executed by a processor to realize any one of the steps of the method for identifying the meteorite crater based on the depth residual error U-Net.
According to the method, a residual error unit-based U-Net is constructed, meteor crater recording data with different distributions, sizes and ranges are taken into consideration, the identification marking effect of a residual error network model on various sizes of meteor craters of images in different regions of the moon is improved, the existing CNN model for identification of the lunar meteor craters is improved and innovated integrally, and the final identification accuracy of the lunar meteor craters is improved.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description, serve to explain the principles of the invention.
FIG. 1 is a schematic diagram of a method for identifying the moon meteorite crater based on a depth residual U-Net model in the embodiment of the invention.
Fig. 2 is a detailed network structure diagram in the embodiment of the present invention.
Detailed Description
Various exemplary embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be noted that: the relative arrangement of the components and steps, the numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present invention unless specifically stated otherwise.
Meanwhile, it should be understood that the sizes of the respective portions shown in the drawings are not drawn in an actual proportional relationship for the convenience of description.
The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.
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.
Techniques, methods, and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail but are intended to be part of the specification where appropriate.
In all examples shown and discussed herein, any particular value should be construed as merely illustrative, and not limiting. Thus, other examples of the exemplary embodiments may have different values.
It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, further discussion thereof is not required in subsequent figures.
According to one aspect of the invention, as shown in FIG. 1, in one embodiment, a method for identifying a lunar meteorite crater based on a depth residual U-Net is disclosed, comprising the following steps:
the method comprises the following steps: firstly, constructing a large data set based on USGS moon image data and LRO WAC meteorite pit recorded data, namely forming an image set with a marked ring through coordinate projection and mapping, and introducing the image set into a pre-designed depth residual U-Net model for training, wherein the depth residual U-Net model is shown in figure 2, so that the image set has the capacity of marking and identifying moon meteorite pits with different ranges and sizes; the model is trained to be convergent by utilizing a stochastic gradient descent algorithm (SGD), and the stochastic gradient descent algorithm is used as a model optimization algorithm, because compared with the traditional batch gradient descent algorithm, the algorithm avoids the redundancy of repeated gradient calculation on similar samples before parameter updating every time in a mode of calculating one sample data every time, so that the algorithm has a faster optimization speed, namely the stochastic gradient descent algorithm trains one training sample xiAnd predicted value yiUpdating parameters, wherein the parameter updating formula is as follows:
where θ is the weight, η is the learning rate, J (θ; x)i；yi) For the objective function to be optimized, i is the number of training samples, xiIs predicted value of abscissa of ith sample data, yiFor the ith sample data sit verticallyMarking a predicted value; after each training, according to the input image picture (the training set for marking the ring), the network is subjected to feature training, so that the network has the capacity of identifying the meteor crater, the probability of the meteor crater can be judged according to the image information, whether the predicted meteor crater is marked by the ring is determined according to the predicted probability value, and the output meteor crater ring and the predicted probability value are obtained.
Step two: calculating the position coordinate of the meteor crater to be predicted, namely (x, y, r), of each input datum according to the depth residual U-Net model, wherein x and y are the horizontal and vertical coordinates of the meteor crater to be predicted, and r is the radius;
and screening and de-duplicating the calculated predicted value:
when the predicted value satisfies both equations (1) and (2), the predicted value is retained:
((xi-xj)2+(yi-yj)2)/min(ri，rj)2＜Dx，y (1)
abs(ri-rj)/min(ri，rj)<Dr (2)
wherein x is the abscissa of the meteorite crater to be predicted, y is the ordinate of the meteorite crater to be predicted, r is the radius of the meteorite crater to be predicted, i is the number of training samples, i is 0,1,2iAnd xjRespectively obtaining meteorite crater abscissa predicted values y of ith sample data and jth sample data through U-Net network model trainingiAnd yjRespectively obtaining meteorite crater ordinate predicted values r of ith sample data and jth sample data through residual error U-Net network model trainingiAnd rjRespectively obtaining meteorite crater radius predicted values D obtained by training the ith sample data and the jth sample data through a U-Net network modelx,yAnd DrAdjusting the recognition effect of the trained U-Net network model by continuously changing the value of the hyper-parameter value for adjustable hyper-parameter, continuously iterating the training process, ensuring the higher recognition precision of the U-Net network model, and reserving the maximum value of the model by selecting the optimal threshold rangeAnd (5) excellent predicted value.
Step three: obtaining a first predicted value of the model for the sample data set training, screening and de-duplicating the first predicted value, and adopting a pixel-by-pixel linear cross entropy function J (theta; x)i；yi) And calculating the error loss value of the real value of the training set sample and the predicted value after the filtering and the de-duplication by using the target function as a loss function, wherein the cross entropy loss function formula is as follows:
wherein L is the error loss value, N is the total number of training samples,
the real data value of the ith training sample is represented; and (3) calculating the accumulated error between the true value of the training sample and the predicted value obtained by the depth residual U-Net network model by using the loss function, selecting the trained depth residual U-Net network model with the accumulated error meeting the preset threshold, training the model to be convergent by the random gradient descent algorithm in the step one, finishing training of a model capable of accurately identifying the distribution of various lunar meteorite pits, and identifying the image distribution of the lunar meteorite pits by using the trained depth residual U-Net network model.
In some embodiments, the full DEM moon image and two groups of large meteorite crater recorded data which are counted at home and abroad and are 5-20km and more than 20km are used, the two groups of meteorite crater recorded data are collected and unified in format through a python script, and all meteorite crater recorded data are marked on the full DEM moon image through projection and data mapping to form complete data sets with different sizes and distribution, so that the model has better generalization as a whole.
In other embodiments, the method for identifying a lunar meteor crater based on the depth residual U-Net further comprises: the depth residual U-Net network model extracts data characteristics through the residual U-Net network model according to each input data, the input training set data firstly passes through 4 residual modules of an encoder, each residual module comprises two convolution units of 3 multiplied by 3 and an identity mapping to complete characteristic extraction, down sampling is carried out through a pooling layer, and the size of a characteristic diagram is reduced; the decoder part also corresponds to 4 residual modules, training set data are subjected to upsampling after passing through a convolutional layer, and finally prediction results are extracted through the convolutional layer of a Sigmoid activation function.
In some embodiments, the residual module is composed of a convolution layer, a regularization layer, a ReLU activation layer and an identity mapping of input and output of a connection module, each sample picture is subjected to a series of convolution, pooling and connection through a coding layer, corresponding features are extracted, then a series of convolution and up-sampling are performed through a decoding layer, aggregation operation is performed on a feature graph of a module corresponding to the coding layer, residual connection is performed, and then a predicted value is obtained; the depth residual U-Net network model ensures that the low-level features of the network can be directly transmitted to the high level by adding the feature channel between the encoder and the decoder, ensures that the feature map of the image trained in the network not only contains high-level semantic information, but also integrates low-level detail information, ensures the integrity of the meteorite pit features under the collocation of the residual unit, and also avoids the overfitting phenomenon possibly caused by the over-deep network layer number.
In some embodiments, a combination of deep residual network and u-type network structure is used, based on ResNet50, the number of network layers is more, the parameter amount is larger, and the identification precision and the network model performance are also higher; in the design of a residual error unit of a network, the number of channels is reduced by convolution of 1 multiplied by 1, then convolution of 3 multiplied by 3 is carried out, the number of channels is kept unchanged, and then convolution of 1 multiplied by 1 is carried out to ensure that the number of output channels of a residual error bottleeck is equal to the number of input channels of the bottleeck, and the structure of the bottleeck effectively reduces the parameter number and the calculated amount of the convolution; after the final convolution layer of the network model, a 2048d convolution layer of 1 x 1 is added, the width of the network and the significance of information expression are increased, and the final recognition precision also reaches 93.62%.
The method has the advantages that the existing lunar meteorite pit data are counted and integrated to form a training set containing multiple types and sizes, so that the neural network can learn lunar meteorite pit picture characteristics with different distribution, sizes and ranges, and the identified meteorite pits are marked and output by adopting a proper marking algorithm, so that the problem of insufficient generalization of a model is solved to a certain extent.
The residual error unit of the depth residual error network is realized in a layer jump connection mode, the input and the output of the unit are added and then activated, when the input signal is transmitted in the forward direction, the input signal can be directly transmitted to a high layer from a low layer, so that the constant mapping is included, the network degradation problem is solved, and the identification performance of the model can be improved and the identification accuracy of the model is improved by combining the input signal with the characteristics of U-Net low-level detail information and high-level semantic information.
The above description is only exemplary of the present invention and should not be taken as limiting the invention, as any modification, equivalent replacement, or improvement made within the spirit and scope of the present invention should be included in the present invention.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in the process, method, article, or apparatus that comprises the element.
Claims (9)
1. A method for identifying a lunar meteorite crater based on a depth residual error U-Net is characterized by comprising the following steps:
constructing a U-Net network model based on a residual error unit;
mapping and combining the first historical data and the second historical data to construct a training set;
inputting the training set into the U-Net network model for training to obtain a first predicted value;
screening and de-duplicating the first predicted value to obtain a second predicted value;
calculating error loss between the second predicted value and the real value of the input data, updating the weight parameter of the U-Net network model, and training the U-Net network model to be convergent by adopting a random gradient descent algorithm;
and identifying the lunar meteor crater image by using the trained U-Net network model with the error loss meeting the preset threshold.
2. The method for identifying the lunar meteorite crater based on the depth residual U-Net according to claim 1, wherein the residual unit comprises:
two 3 x 3 convolution units and an identity map.
3. The method for identifying the lunar meteor crater based on the depth residual U-Net as claimed in claim 1, wherein the mapping and combining the first history data and the second history data to construct the training set comprises:
the first historical data is a full DEM moon image;
the second historical data is meteorite crater recording data;
summarizing and unifying the full DEM moon image and meteor crater recorded data through a python script to obtain a corrected full DEM moon image and corrected meteor crater recorded data;
and marking the corrected meteorite pit recorded data on the corrected full DEM moon image through projection and data mapping.
4. The method for identifying the lunar meteor crater based on the deep residual U-Net according to claim 1, wherein the inputting the training set into the U-Net network model for training comprises:
utilizing a convolution layer and a pooling layer of a U-Net network model encoder to carry out down-sampling on the training set;
up-sampling the down-sampled data by a decoder;
and extracting a first predicted value from the up-sampled data by using a Sigmoid activation function.
5. The method for identifying the lunar meteor crater based on the depth residual U-Net according to any one of claims 1 to 4, wherein the calculating of the error loss between the second predicted value and the true value of the input data, the updating of the weighting parameters of the U-Net network model, and the training of the U-Net network model to convergence by using the stochastic gradient descent algorithm specifically comprises:
calculating error loss between the second predicted value and the real value of the input data, updating the weight parameter of the U-Net network model through a parameter updating formula, and training the U-Net network model to be convergent by adopting a random gradient descent algorithm;
wherein, the parameter updating formula is as follows:
in the parameter updating formula, theta is weight, eta is learning rate, J (theta; x)i；yi) For the objective function to be optimized, xiFor the meteorite crater abscissa prediction result obtained by training the ith sample data through a U-Net network model, yiAnd obtaining the meteor crater ordinate prediction result of the ith sample through residual U-Net network model training.
6. The method for identifying the lunar meteorite crater based on the depth residual U-Net according to claim 1, wherein the step of screening and de-duplicating the first predicted value comprises the following steps:
reserving a predicted value which satisfies a first formula and a second formula simultaneously in the first predicted value as a second predicted value;
wherein the first formula is:
((xi-xj)2+(yi-yj)2)/min(ri，rj)2＜Dx，y；
the second formula is:
abs(ri-rj)/min(ri，rj)＜Dr
in the first formula and the second formula, x is an abscissa of the meteorite crater to be predicted, y is an ordinate of the meteorite crater to be predicted, r is the radius of the meteorite crater to be predicted, i is the number of training samples, i is 0,1,2iAnd xjRespectively obtaining meteorite crater abscissa predicted values y of ith sample data and jth sample data through U-Net network model trainingiAnd yjRespectively obtaining meteorite crater ordinate predicted values r of ith sample data and jth sample data through residual error U-Net network model trainingiAnd rjRespectively obtaining meteorite crater radius predicted values D obtained by training the ith sample data and the jth sample data through a U-Net network modelx,yAnd DrIs an adjustable hyper-parameter.
7. The method for identifying the lunar meteor crater based on the depth residual U-Net according to claim 1, wherein the calculating the error loss between the second predicted value and the true value of the input data comprises:
calculating error loss between the second predicted value and the real value of the input data by adopting a pixel-by-pixel linear cross entropy function;
wherein the linear cross entropy function is:
l is the error loss value, N is the total number of training samples,
the true value of the data representing the ith training sample.
8. A lunar meteorite crate identification device, the lunar meteorite crate identification device comprising: memory, processor and identification program stored on the memory and executable on the processor, the identification program when executed by the processor implementing a method of identifying a lunar meteorite crater based on a depth residual U-Net according to any one of claims 1 to 7.
9. A storage medium having stored thereon an identification program which, when executed by a processor, implements the steps of a method for identifying a moon meteorite crater based on a depth residual U-Net according to the method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110403032.6A CN113139592A (en) | 2021-04-14 | 2021-04-14 | Method, device and storage medium for identifying lunar meteorite crater based on depth residual error U-Net |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110403032.6A CN113139592A (en) | 2021-04-14 | 2021-04-14 | Method, device and storage medium for identifying lunar meteorite crater based on depth residual error U-Net |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113139592A true CN113139592A (en) | 2021-07-20 |
Family
ID=76812558
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110403032.6A Pending CN113139592A (en) | 2021-04-14 | 2021-04-14 | Method, device and storage medium for identifying lunar meteorite crater based on depth residual error U-Net |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113139592A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114332070A (en) * | 2022-01-05 | 2022-04-12 | 北京理工大学 | Meteor crater detection method based on intelligent learning network model compression |
| CN114519723A (en) * | 2021-12-24 | 2022-05-20 | 上海海洋大学 | Meteorite crater automatic extraction method based on pyramid image segmentation |
| CN114936327A (en) * | 2022-07-22 | 2022-08-23 | 腾讯科技（深圳）有限公司 | Element recognition model obtaining method and device, computer equipment and storage medium |
| CN114972720A (en) * | 2022-05-30 | 2022-08-30 | 西北工业大学 | High-precision unmanned positioning method based on autonomous image perception |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110334645A (en) * | 2019-07-02 | 2019-10-15 | 华东交通大学 | A kind of moon impact crater recognition methods based on deep learning |
-
2021
- 2021-04-14 CN CN202110403032.6A patent/CN113139592A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110334645A (en) * | 2019-07-02 | 2019-10-15 | 华东交通大学 | A kind of moon impact crater recognition methods based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| 刘玉良 等: "《深度学习》", 西安：西安电子科技大学出版社 * |
| 胡涛 等: "行星陨石坑检测算法研究综述", 《载人航天》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114519723A (en) * | 2021-12-24 | 2022-05-20 | 上海海洋大学 | Meteorite crater automatic extraction method based on pyramid image segmentation |
| CN114519723B (en) * | 2021-12-24 | 2024-05-28 | 上海海洋大学 | Pyramid image segmentation-based meteorite crater automatic extraction method |
| CN114332070A (en) * | 2022-01-05 | 2022-04-12 | 北京理工大学 | Meteor crater detection method based on intelligent learning network model compression |
| CN114332070B (en) * | 2022-01-05 | 2024-05-28 | 北京理工大学 | Meteorite detection method based on intelligent learning network model compression |
| CN114972720A (en) * | 2022-05-30 | 2022-08-30 | 西北工业大学 | High-precision unmanned positioning method based on autonomous image perception |
| CN114972720B (en) * | 2022-05-30 | 2024-02-20 | 西北工业大学 | Autonomous image perception-based high-precision unmanned positioning method |
| CN114936327A (en) * | 2022-07-22 | 2022-08-23 | 腾讯科技（深圳）有限公司 | Element recognition model obtaining method and device, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109934293B (en) | Image recognition method, device, medium and confusion perception convolutional neural network | |
| CN113139592A (en) | Method, device and storage medium for identifying lunar meteorite crater based on depth residual error U-Net | |
| CN110555399B (en) | Finger vein identification method and device, computer equipment and readable storage medium | |
| CN104866810A (en) | Face recognition method of deep convolutional neural network | |
| CN105279554B (en) | The training method and device of deep neural network based on Hash coding layer | |
| CN114119979A (en) | Fine-grained image classification method based on segmentation mask and self-attention neural network | |
| CN110232341A (en) | Based on convolution-stacking noise reduction codes network semi-supervised learning image-recognizing method | |
| CN110991257B (en) | Polarized SAR oil spill detection method based on feature fusion and SVM | |
| CN112927253B (en) | Rock core FIB-SEM image segmentation method based on convolutional neural network | |
| CN112950780B (en) | Intelligent network map generation method and system based on remote sensing image | |
| CN110599459A (en) | Underground pipe network risk assessment cloud system based on deep learning | |
| CN112819039A (en) | Texture recognition model establishing method based on multi-scale integrated feature coding and application | |
| CN108376257B (en) | Incomplete code word identification method for gas meter | |
| CN116824239A (en) | Image recognition method and system based on transfer learning and ResNet50 neural network | |
| CN114842343A (en) | ViT-based aerial image identification method | |
| CN115761735A (en) | Semi-supervised semantic segmentation method based on self-adaptive pseudo label correction | |
| CN112560948A (en) | Eye fundus map classification method and imaging method under data deviation | |
| CN116469561A (en) | Breast cancer survival prediction method based on deep learning | |
| CN112527959B (en) | News classification method based on pooling convolution embedding and attention distribution neural network | |
| CN116721291A (en) | Metal surface defect detection method based on improved YOLOv7 model | |
| CN116433909A (en) | Similarity weighted multi-teacher network model-based semi-supervised image semantic segmentation method | |
| CN114783072B (en) | Image identification method based on remote domain transfer learning | |
| CN117011219A (en) | Method, apparatus, device, storage medium and program product for detecting quality of article | |
| CN113420706B (en) | Vehicle detection method based on multi-layer feature fusion | |
| CN115239967A (en) | Image generation method and device for generating countermeasure network based on Trans-CSN |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210720 |
|
| RJ01 | Rejection of invention patent application after publication |