CN110245642A - A kind of radar spectrum recognition method and system based on deep learning - Google Patents
A kind of radar spectrum recognition method and system based on deep learning Download PDFInfo
- Publication number
- CN110245642A CN110245642A CN201910541303.7A CN201910541303A CN110245642A CN 110245642 A CN110245642 A CN 110245642A CN 201910541303 A CN201910541303 A CN 201910541303A CN 110245642 A CN110245642 A CN 110245642A
- Authority
- CN
- China
- Prior art keywords
- layer
- radar
- spectrum data
- deep learning
- radar spectrum
- 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
- 238000001228 spectrum Methods 0.000 title claims abstract description 118
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000013135 deep learning Methods 0.000 title claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 39
- 238000013136 deep learning model Methods 0.000 claims abstract description 29
- 201000010099 disease Diseases 0.000 claims abstract description 21
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims abstract description 21
- 238000013527 convolutional neural network Methods 0.000 claims description 39
- 238000012545 processing Methods 0.000 claims description 8
- 238000001514 detection method Methods 0.000 claims description 7
- 230000007423 decrease Effects 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 238000013434 data augmentation Methods 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 230000002087 whitening effect Effects 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 abstract description 9
- 238000004458 analytical method Methods 0.000 abstract description 8
- 238000011160 research Methods 0.000 abstract description 4
- 230000009466 transformation Effects 0.000 description 5
- 230000008569 process Effects 0.000 description 4
- 238000012546 transfer Methods 0.000 description 4
- 230000007547 defect Effects 0.000 description 3
- 238000013519 translation Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000000149 penetrating effect Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- 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
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration using local operators
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/176—Urban or other man-made structures
-
- 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
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10044—Radar 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
- G06T2207/20081—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]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The radar spectrum recognition method and system based on deep learning that the present invention relates to a kind of, this method comprises: collecting the radar spectrum data of preset quantity, the radar spectrum data includes no road disease spectrum data and comprising empty disease spectrum data；Using the radar spectrum data as training set training deep learning model；The radar spectrum data acquired in real time is input in the deep learning model, so that the deep learning model carries out identification classification to the radar spectrum data acquired in real time.Technical solution provided by the invention, by in the radar spectrum recognition problem that introduces depth learning technology in underground engineering, the deep learning model of radar atlas analysis identification is established in research, realize the automatic identification and difference of radar map, artificial radar spectrum recognition method compared to the prior art, technical solution provided by the invention, human input is few, high-efficient, accuracy is high.
Description
Technical field
The present invention relates to geology detecting technical fields, and in particular to a kind of radar spectrum recognition method based on deep learning
And system.
Background technique
With the development of economy and society, city underground engineering (pipe network, subway tunnel, track, pipe gallery engineering) scale
It sharply expands, causes the diseases such as loose underground, water damage, cavity gradually, wherein empty disease can seriously directly result in road surface and collapse,
Cause serious economic loss and social influence.It therefore, is to ensure that city is handed over for the timely detection of road foundation cavity disease
Lead to the key point of safety.
Currently, the empty Defect inspection of road foundation mainly passes through analysis ground penetrating radar image, discovery underground is empty
Hole disease obtains road foundation radar map by radar detection, then explain cavity in identification map using human expert
Disease matches spectrum data library directly using computer to determine that map whether there is empty disease.Due at present to city road
The Defect inspection on road is mainly the method for pinpointing radar and detecting, subsequent artefacts' radar atlas analysis, verify on the spot, depends on people unduly
It is discontinuous that information is extracted for experience, radar, causes that the detection and analysis speed of urban road radar map is relatively slow, testing result error
It is larger:
1, disease region erroneous judgement (futile drillhole validation, waste of manpower and material resources)；
2, defect information fails to judge (cannot early warning road hidden danger in time, cause to collapse accident frequently occur).
With the generation and development of computer depth learning technology, depth learning technology has been penetrated into every field,
Especially in computer picture visual field, depth learning technology is solving the problems, such as that many Shangrao are effective, but at present also not
See application study of the depth learning technology in radar atlas analysis identification problem.
Summary of the invention
In view of this, it is an object of the invention to overcome the deficiencies of the prior art and provide a kind of thunders based on deep learning
Up to spectrum recognition method and system, to solve in the prior art, artificial radar spectrum recognition human input is big, accuracy rate is not high
Problem.
In order to achieve the above object, the present invention adopts the following technical scheme:
A kind of radar spectrum recognition method based on deep learning, comprising:
Step S1, the radar spectrum data of preset quantity is collected, the radar spectrum data includes no road disease map
Data and include empty disease spectrum data；
Step S2, using the radar spectrum data as training set training deep learning model；
Step S3, the radar spectrum data acquired in real time is input in the deep learning model, so that the depth
Learning model carries out identification classification to the radar spectrum data acquired in real time.
Preferably, the method also includes:
The radar spectrum data being collected into is pre-processed, using the pretreated radar spectrum data as training
Collect training deep learning model.
Preferably, the described pair of radar spectrum data being collected into pre-processes, comprising:
The radar spectrum data that will be collected into, be input to sliding filter carry out filter make an uproar；
Radar spectrum data after making an uproar to filter carries out data augmentation, to expand the quantity of the radar spectrum data.
Preferably, the radar spectrum data after described pair of filter is made an uproar carries out data augmentation, comprising:
Radar spectrum data after making an uproar to filter carries out whitening processing；
Radar spectrum data after making an uproar to filter carries out random left and right overturning；
The contrast of the random modified-image of radar spectrum data after making an uproar to filter.
Preferably, the deep learning model is convolutional neural networks model vgg16 model, described by the radar map
Data are as training set training deep learning model, comprising:
Construct multilayer convolutional neural networks vgg16 model；
Using the radar spectrum data being collected into, using ADAM algorithm to the error ladder of multilayer convolutional neural networks vgg16
Degree does steepest decline optimization, and off-line training constructs multilayer convolutional neural networks vgg16 model；
Multilayer convolutional neural networks vgg16 model utilizes Google open source deep learning system TensorFlow after constituting
It is developed, and deep learning algorithm is accelerated up to GPU using tall and handsome.
Preferably, the multilayer convolutional neural networks vgg16 model, comprising: 16 layers of weight layer, 5 layers of pond layer, 1 layer it is defeated
Enter layer and 1 layer of output layer；Wherein,
First layer is the input layer for the image block that size is S*S；Layers 2 and 3 is convolutional layer, and the size of convolution kernel is all
For 3*3, and every layer of convolution nuclear volume is all 64；4th layer is maximum value pond layer；5th layer and the 6th layer is convolutional layer,
The size of convolution kernel is 3*3, and every layer of convolution nuclear volume is all 128；7th layer is maximum value pond layer；8th, 9 and 10
Layer is convolutional layer, and the size of convolution kernel is all 3*3, and every layer of convolution nuclear volume is all 256；11th layer is maximum value pond
Layer；12nd, 13 and 14 layer is convolutional layer, and the size of convolution kernel is all 3*3, and every layer of convolution nuclear volume is all 512；15th
Layer is maximum value pond layer；16th, 17 and 18 layer is convolutional layer, and the size of convolution kernel is all 3*3, and every layer of convolution nucleus number
Amount is all 512；19th layer is maximum value pond layer；20th, 21 and 22 layer is full articulamentum, wherein the 20th, 21 full articulamentums
There are 4096 neurodes, the 22nd full articulamentum there are 1000 neurodes；23rd layer is softmax classification layer.
Preferably, the radar spectrum data for the preset quantity being collected into the step S1, comprising:
Multiple a roadbed cavity disease physical models are built, detect its radar from different perspectives using 250MHZ radar antenna
Map obtains multiple radar maps；
Carry out at least one site road cavity radar detection, extracts multiple undergrounds sky using 250MHZ radar antenna
Hole radar map；
The radar map buying multiple 250MHZ radar antennas and detecting is collected from relevant departments.
In addition, the invention also provides a kind of radar spectrum recognition system based on deep learning, comprising:
Collection module, for collecting the radar spectrum data of preset quantity；
Deep learning module, for using the radar spectrum data as training set training deep learning model；
Identification module, for carrying out identification classification to the radar spectrum data acquired in real time.
Preferably, the system also includes preprocessing modules, for being located in advance to the radar spectrum data being collected into
Reason, the preprocessing module includes image capture module and image processing module.
Preferably, the system also includes databases, and the database is for storing the radar spectrum data.
The invention adopts the above technical scheme, at least have it is following the utility model has the advantages that
By the way that in the radar spectrum recognition problem that introduces depth learning technology in underground engineering, radar map is established in research
The deep learning model of spectrum analysis identification, to extract the TuPu method that can characterize different empty Damage Types in radar map,
And the radar spectrum data acquired in real time is input in the deep learning model, so that the deep learning module is according to institute
TuPu method is stated, identification classification is carried out to the radar spectrum data acquired in real time, to realize the automatic identification of radar map
And respectively, artificial radar spectrum recognition method compared to the prior art, technical solution provided by the invention, human input is few, effect
Rate is high, accuracy is high.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is the flow diagram for the radar spectrum recognition method based on deep learning that one embodiment of the invention provides；
Fig. 2 is the training process figure for the convolutional neural networks model that one embodiment of the invention provides；
Fig. 3 is the schematic block diagram for the radar spectrum recognition system based on deep learning that one embodiment of the invention provides.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, technical solution of the present invention will be carried out below
Detailed description.Obviously, described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Base
Embodiment in the present invention, those of ordinary skill in the art are obtained all without making creative work
Other embodiment belongs to the range that the present invention is protected.
Below by drawings and examples, technical scheme of the present invention will be described in further detail.
Referring to Fig. 1, a kind of radar spectrum recognition method based on deep learning of one embodiment of the invention offer, comprising:
Step S1, the radar spectrum data of preset quantity is collected, the radar spectrum data includes no road disease map
Data and include empty disease spectrum data；
Step S2, using the radar spectrum data as training set training deep learning model；
Step S3, the radar spectrum data acquired in real time is input in the deep learning model, so that the depth
Learning model carries out identification classification to the radar spectrum data acquired in real time.
It is understood that technical solution provided by the invention, by the way that depth learning technology is introduced into underground engineering
Radar spectrum recognition problem on, research establish radar atlas analysis identification deep learning model, to extract in radar map
The TuPu method of different empty Damage Types can be characterized, and the radar spectrum data acquired in real time is input to the depth
It practises in model, so that the deep learning module knows the radar spectrum data acquired in real time according to the TuPu method
Do not classify, to realize the automatic identification and difference of radar map, artificial radar spectrum recognition method compared to the prior art,
Technical solution provided by the invention, human input is few, high-efficient, accuracy is high.
Preferably, the method also includes:
The radar spectrum data being collected into is pre-processed, using the pretreated radar spectrum data as training
Collect training deep learning model.
Preferably, the described pair of radar spectrum data being collected into pre-processes, comprising:
The radar spectrum data that will be collected into, be input to sliding filter carry out filter make an uproar；
Radar spectrum data after making an uproar to filter carries out data augmentation, to expand the quantity of the radar spectrum data.
Preferably, the radar spectrum data after described pair of filter is made an uproar carries out data augmentation, comprising:
Radar spectrum data after making an uproar to filter carries out whitening processing；
Radar spectrum data after making an uproar to filter carries out random left and right overturning；
The contrast of the random modified-image of radar spectrum data after making an uproar to filter.
It is understood that allowing deep learning model generalization to obtain better best approach is instructed using more data
Practice.Since in practice, the data volume that we possess is very limited.A kind of method for solving this problem is creation false data
And it is added in training set.For some machine learning tasks, it is comparatively simple to create new false data.At present in the image done
Processing the inside is relatively more, common to have to data geometric transformation, gray proces etc..Common geometric transformation has: overturning, translation, rotation
Turn, scaling, similarity transformation and affine transformation etc..
Preferably, described that the radar spectrum data is input in existing convolutional neural networks model vgg16 model
It is trained, comprising:
Construct multilayer convolutional neural networks vgg16 model；
Using the radar spectrum data being collected into, using ADAM algorithm to the mistake of multilayer convolutional neural networks vgg16 model
Poor gradient does steepest decline optimization, and off-line training constructs multilayer convolutional neural networks vgg16 model；
Multilayer convolutional neural networks vgg16 is carried out after constituting using Google open source deep learning system TensorFlow
Exploitation, and deep learning algorithm is accelerated up to GPU using tall and handsome.
Preferably, referring to fig. 2, the multilayer convolutional neural networks vgg16 model, comprising: 16 layers of weight layer, 5 layers of pond
Layer, 1 layer of input layer and 1 layer of output layer；Wherein,
First layer is the input layer for the image block that size is S*S；Layers 2 and 3 is convolutional layer, and the size of convolution kernel is all
For 3*3, and every layer of convolution nuclear volume is all 64；4th layer is maximum value pond layer；5th layer and the 6th layer is convolutional layer,
The size of convolution kernel is 3*3, and every layer of convolution nuclear volume is all 128；7th layer is maximum value pond layer；8th, 9 and 10
Layer is convolutional layer, and the size of convolution kernel is all 3*3, and every layer of convolution nuclear volume is all 256；11th layer is maximum value pond
Layer；12nd, 13 and 14 layer is convolutional layer, and the size of convolution kernel is all 3*3, and every layer of convolution nuclear volume is all 512；15th
Layer is maximum value pond layer；16th, 17 and 18 layer is convolutional layer, and the size of convolution kernel is all 3*3, and every layer of convolution nucleus number
Amount is all 512；19th layer is maximum value pond layer；20th, 21 and 22 layer is full articulamentum, wherein the 20th, 21 full articulamentums
There are 4096 neurodes, the 22nd full articulamentum there are 1000 neurodes；23rd layer is softmax classification layer.
Preferably, the radar spectrum data for the preset quantity being collected into the step S1, comprising:
Multiple a roadbed cavity disease physical models are built, detect its radar from different perspectives using 250MHZ radar antenna
Map obtains multiple radar maps；
Carry out at least one site road cavity radar detection, extracts multiple undergrounds sky using 250MHZ radar antenna
Hole radar map；
The radar map buying multiple 250MHZ radar antennas and detecting is collected from relevant departments.
Such as: 3 roadbed cavity disease physical models are built, detect its thunder from different perspectives using 250MHZ radar antenna
Up to map, radar map totally 60 are obtained；
Carry out 1 site road cavity radar detection, extracts underground cavity radar map using 250MHZ radar antenna
Spectrum 5；
The radar map 10 that purchase 250MHZ radar antenna detects is collected from relevant departments to open.
The radar spectrum data trains CNN convolutional neural networks model as training set, and specific steps include:
(1) collect 10000 radar map spectrogram pieces, radar map spectrogram piece include no road disease map picture and comprising
Empty disease map picture cuts each map picture using picture crop tool, keeps map picture specification consistent.
(2) it is rotated, is translated using map picture of the method for OpenCV image procossing to deposit, feature scales and divides
Measure mean value pulverised.The rotation and translation of picture can increase the quantity of picture, to increase the quantity of training sample, enhance data
Source；The standardization of picture feature may be implemented in feature scaling and component mean value pulverised, improves CNN convolutional neural networks model.
(3) network parameter is set using Caffe frame, constructs CNN convolutional neural networks model.CNN convolutional neural networks
The training of model is the mode of learning for having supervision, need before training in network weight and bias carry out taking random number
First value operation, avoid because weight and bias are excessive so that the diverging of trained model, so as to cause failure to train.CNN
The propagated forward stage be similar to traditional neural network transfer mode from front to back, i.e., it is first that the sample of data set is random
It is divided into training set and test set, then training set is input in network, is exported by transformation step by step as a result, in a network
Input of upper one layer of the output as current layer, the input x of current layerl-1With output xlBetween relationship it is as follows: xl=f (Wlxl-1
+bl):
In formula: l represents the number of plies；W represents weight；B represents bias (placeholder)；F is a Relu activation primitive.Reversely
The process of propagation be exactly by error by way of back transfer successively to front transfer, make the neuron in one layer according to error
To automatically update itself weight and bias.The back-propagation algorithm of CNN passes through calculating by the way of based on gradient decline
The global error of network is adjusted come the direction for reducing the unit weights in network to error.Back-propagation phase is
The error of output layer is successively worth sample level, full articulamentum since CNN network is by convolutional layer to front transfer using inverse iteration
Composition, value sample level are obtained by carrying out simple scalability to convolutional layer, this to exist between convolutional layer and value sample level
Many-to-one mapping relations, the error of each neurode in value sample level correspond to a partial zones in convolutional layer
Domain.This just needs that the error of each neurode in value sample level is constituted one and part in convolutional layer in backpropagation
The corresponding error signal figure in region.Then by the partial derivative and errors signal graph of the Relu activation primitive in convolution algorithm
It is multiplied.Since weight is all β in value sample level, a β is multiplied on the basis of calculated result above, finally using reversed
Propagation algorithm hair calculates the gradient of weight.
CNN network shares 16 layers of weight layer, 5 layers of pond layer, 1 layer of input layer and 1 layer of output layer；Wherein,
First layer is the input layer for the image block that size is S*S, and 10000 radar map spectrogram pieces are input to input
Layer；Layers 2 and 3 is convolutional layer, and the size of convolution kernel is all 3*3, and every layer of convolution nuclear volume is all 64；4th
Layer is maximum value pond layer；5th layer and the 6th layer is convolutional layer, and the size of convolution kernel is 3*3, and every layer of convolution nuclear volume
It is all 128；7th layer is maximum value pond layer；8th, 9 and 10 layer is convolutional layer, and the size of convolution kernel is all 3*3, and every
Layer convolution nuclear volume is all 256；11th layer is maximum value pond layer；12nd, 13 and 14 layer is convolutional layer, the size of convolution kernel
It is all 3*3, and every layer of convolution nuclear volume is all 512；15th layer is maximum value pond layer；16th, 17 and 18 layer is convolution
Layer, the size of convolution kernel is all 3*3, and every layer of convolution nuclear volume is all 512；19th layer is maximum value pond layer；The
20,21 and 22 layers are full articulamentum, wherein the 20th, 21 full articulamentums have 4096 neurodes, and the 22nd full articulamentum has 1000
A neurode；23rd layer is softmax classification layer.
(4) setting learning rate base_lr is 0.01, and momentum parameter momentum is 0.9, weight attenuation coefficient weight_
Decay is 0.0005, and the optimisation strategy of gradient decline uses inv, maximum number of iterations 10000, and every iteration 5000 times saves one
It is secondary as a result, training hardware device selection GPU run
By above step, CNN convolutional neural networks model training is completed, and the radar spectrum data acquired in real time is inputted
Into trained CNN convolutional neural networks model, so that CNN convolutional neural networks model is to the radar map number acquired in real time
According to carrying out identification classification.
Radar spectrum recognition concrete operation step based on deep learning includes:
(1) IP camera is utilized, spectrum data is grabbed, spectrum data is transmitted by cloud and h.264 decompressed by computer
It obtains；
(2) under linux system, the spectrum data of deposit is cut, is rotated, is translated, feature scales and component is equal
Value zero can increase the quantity of samples pictures using the rotation and translation of picture, enhance data source；Feature scaling and component are equal
The standardization of picture feature may be implemented in value zero, improves CNN convolutional neural networks model；
(3) under VS2013 environment, neural network hyper parameter is configured using Python, learning rate base_lr is set
It is 0.01, momentum parameter momentum is 0.9, and weight attenuation coefficient weight_decay is 0.0005, the optimization of gradient decline
Strategy use inv, maximum number of iterations 10000, every iteration 5000 times save primary as a result, training hardware device selection GPU fortune
Row；
(4) trained deep learning model is added in program code such as Java, to collected every frame map
Compared with model, using model analysis map characteristic, the final tagsort for determining radar spectrum data.
In addition, referring to Fig. 3, the radar spectrum recognition system 100 based on deep learning that the invention also provides a kind of, packet
It includes:
Collection module 101, for collecting the radar spectrum data of preset quantity；
Deep learning module 102, for using the radar spectrum data as training set training deep learning model；
Identification module 103, for carrying out identification classification to the radar spectrum data acquired in real time.
It is understood that technical solution provided by the invention, by the way that depth learning technology is introduced into underground engineering
Radar spectrum recognition problem on, research establish radar atlas analysis identification deep learning model, to extract in radar map
The TuPu method of different empty Damage Types can be characterized, and the radar spectrum data acquired in real time is input to the depth
It practises in model, so that the deep learning module knows the radar spectrum data acquired in real time according to the TuPu method
Do not classify, to realize the automatic identification and difference of radar map, artificial radar spectrum recognition method compared to the prior art,
Technical solution provided by the invention, human input is few, high-efficient, accuracy is high.
Preferably, the system also includes preprocessing modules 104, pre- for carrying out to the radar spectrum data being collected into
Processing, preprocessing module 104 include image capture module and image processing module.
Preferably, the system also includes databases 105, and database 105 is for storing the radar spectrum data.
It is understood that database 105 with the training set data of storage depth learning algorithm and will can be adopted in real time
The radar spectrum data of collection is input in deep learning model, so that deep learning model is to the radar spectrum data acquired in real time
Carry out the result data that identification classification obtains.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Term " first ", " second " are used for descriptive purposes only and cannot be understood as indicating or suggesting relative importance.Term " multiple " refers to
Two or more, unless otherwise restricted clearly.
Claims (10)
1. a kind of radar spectrum recognition method based on deep learning characterized by comprising
Step S1, the radar spectrum data of preset quantity is collected, the radar spectrum data includes no road disease spectrum data
With include empty disease spectrum data；
Step S2, using the radar spectrum data as training set training deep learning model；
Step S3, the radar spectrum data acquired in real time is input in the deep learning model, so that the deep learning
Model carries out identification classification to the radar spectrum data acquired in real time.
2. the method according to claim 1, wherein further include:
The radar spectrum data being collected into is pre-processed, is assembled for training the pretreated radar spectrum data as training
Practice deep learning model.
3. according to the method described in claim 2, it is characterized in that, the described pair of radar spectrum data being collected into is located in advance
Reason, comprising:
The radar spectrum data that will be collected into, be input to sliding filter carry out filter make an uproar；
Radar spectrum data after making an uproar to filter carries out data augmentation, to expand the quantity of the radar spectrum data.
4. according to the method described in claim 3, it is characterized in that, the radar spectrum data after described pair of filter is made an uproar carries out data increasing
Extensively, comprising:
Radar spectrum data after making an uproar to filter carries out whitening processing；
Radar spectrum data after making an uproar to filter carries out random left and right overturning；
The contrast of the random modified-image of radar spectrum data after making an uproar to filter.
5. the method according to claim 1, wherein the deep learning model is convolutional neural networks model
Vgg16 model, it is described using the radar spectrum data as training set training deep learning model, comprising:
Construct multilayer convolutional neural networks vgg16 model；
Using the radar spectrum data being collected into, using ADAM algorithm to the error ladder of multilayer convolutional neural networks vgg16 model
Degree does steepest decline optimization, and off-line training constructs multilayer convolutional neural networks vgg16 model；
Multilayer convolutional neural networks vgg16 model is carried out after constituting using Google open source deep learning system TensorFlow
Exploitation, and deep learning algorithm is accelerated up to GPU using tall and handsome.
6. according to the method described in claim 5, it is characterized in that, the multilayer convolutional neural networks vgg16 model, comprising:
16 layers of weight layer, 5 layers of pond layer, 1 layer of input layer and 1 layer of output layer；Wherein,
First layer is the input layer for the image block that size is S*S；Layers 2 and 3 is convolutional layer, and the size of convolution kernel is all 3*
3, and every layer convolution nuclear volume is all 64；4th layer is maximum value pond layer；5th layer and the 6th layer is convolutional layer, convolution
The size of core is 3*3, and every layer of convolution nuclear volume is all 128；7th layer is maximum value pond layer；8th, 9 and 10 layer is
Convolutional layer, the size of convolution kernel is all 3*3, and every layer of convolution nuclear volume is all 256；11th layer is maximum value pond layer；
12nd, 13 and 14 layer is convolutional layer, and the size of convolution kernel is all 3*3, and every layer of convolution nuclear volume is all 512；15th layer
For maximum value pond layer；16th, 17 and 18 layer is convolutional layer, and the size of convolution kernel is all 3*3, and every layer of convolution nuclear volume
It is all 512；19th layer is maximum value pond layer；20th, 21 and 22 layer is full articulamentum, wherein the 20th, 21 full articulamentums have
4096 neurodes, the 22nd full articulamentum have 1000 neurodes；23rd layer is softmax classification layer.
7. the method according to claim 1, wherein the radar map for the preset quantity being collected into the step S1
Modal data, comprising:
Multiple a roadbed cavity disease physical models are built, detect its radar map from different perspectives using 250MHZ radar antenna
Spectrum, obtains multiple radar maps；
Carry out at least one site road cavity radar detection, extracts multiple underground cavity thunders using 250MHZ radar antenna
Up to map；
The radar map buying multiple 250MHZ radar antennas and detecting is collected from relevant departments.
8. a kind of radar spectrum recognition system based on deep learning characterized by comprising
Collection module, for collecting the radar spectrum data of preset quantity；
Deep learning module, for using the radar spectrum data as training set training deep learning model；
Identification module, for carrying out identification classification to the radar spectrum data acquired in real time.
9. system according to claim 8, which is characterized in that further include: preprocessing module, for the radar being collected into
Spectrum data is pre-processed, and the preprocessing module includes image capture module and image processing module.
10. system according to claim 8, which is characterized in that further include: database, the database is for storing institute
State radar spectrum data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910541303.7A CN110245642A (en) | 2019-06-21 | 2019-06-21 | A kind of radar spectrum recognition method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910541303.7A CN110245642A (en) | 2019-06-21 | 2019-06-21 | A kind of radar spectrum recognition method and system based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110245642A true CN110245642A (en) | 2019-09-17 |
Family
ID=67888631
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910541303.7A Pending CN110245642A (en) | 2019-06-21 | 2019-06-21 | A kind of radar spectrum recognition method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110245642A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110988839A (en) * | 2019-12-25 | 2020-04-10 | 中南大学 | Method for quickly identifying health condition of wall based on one-dimensional convolutional neural network |
| CN110988872A (en) * | 2019-12-25 | 2020-04-10 | 中南大学 | Method for rapidly identifying health state of wall body detected by unmanned aerial vehicle-mounted through-wall radar |
| CN110991507A (en) * | 2019-11-22 | 2020-04-10 | 北京中科蓝图科技有限公司 | Road underground cavity identification method, device and system based on classifier |
| CN111582284A (en) * | 2020-04-27 | 2020-08-25 | 中国科学院信息工程研究所 | Privacy protection method and device for image recognition and electronic equipment |
| CN111767874A (en) * | 2020-07-06 | 2020-10-13 | 中兴飞流信息科技有限公司 | Pavement disease detection method based on deep learning |
| CN111783784A (en) * | 2020-06-30 | 2020-10-16 | 创新奇智(合肥)科技有限公司 | Method and device for detecting building cavity, electronic equipment and storage medium |
| CN112036425A (en) * | 2020-05-09 | 2020-12-04 | 中铁四局集团有限公司 | Tunnel cavity state radar spectrum image recognition model construction method and tunnel cavity state radar spectrum image recognition method |
| WO2021082904A1 (en) * | 2019-11-01 | 2021-05-06 | 长安大学 | Ground penetrating radar map adaptive selection method for detecting water damage |
| WO2021083394A1 (en) * | 2019-11-01 | 2021-05-06 | 长安大学 | Method for detection of water damage to asphalt road surface based on adaptive selection of plot grayscales |
| CN113191391A (en) * | 2021-04-07 | 2021-07-30 | 浙江省交通运输科学研究院 | Road disease classification method aiming at three-dimensional ground penetrating radar map |
| CN115343685A (en) * | 2022-08-29 | 2022-11-15 | 北京国电经纬工程技术有限公司 | Multi-dimensional ground penetrating radar detection method, device and equipment applied to disease identification |
| CN115792919A (en) * | 2023-01-19 | 2023-03-14 | 合肥中科光博量子科技有限公司 | Method for identifying pollution hot spot area through horizontal scanning and monitoring of aerosol laser radar |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104729915A (en) * | 2015-01-26 | 2015-06-24 | 北京市市政工程研究院 | Device and method for searching looseness and emptiness of urban subway interval road in detail |
| CN107392896A (en) * | 2017-07-14 | 2017-11-24 | 佛山市南海区广工大数控装备协同创新研究院 | A kind of Wood Defects Testing method and system based on deep learning |
| CN108197655A (en) * | 2018-01-03 | 2018-06-22 | 重庆亲禾智千科技有限公司 | Road surface breakage disease geo-radar image sorting technique based on principal component analysis and neural network |
-
2019
- 2019-06-21 CN CN201910541303.7A patent/CN110245642A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104729915A (en) * | 2015-01-26 | 2015-06-24 | 北京市市政工程研究院 | Device and method for searching looseness and emptiness of urban subway interval road in detail |
| CN107392896A (en) * | 2017-07-14 | 2017-11-24 | 佛山市南海区广工大数控装备协同创新研究院 | A kind of Wood Defects Testing method and system based on deep learning |
| CN108197655A (en) * | 2018-01-03 | 2018-06-22 | 重庆亲禾智千科技有限公司 | Road surface breakage disease geo-radar image sorting technique based on principal component analysis and neural network |
Non-Patent Citations (2)
| Title |
|---|
| KAREN SIMONYAN ET AL.: "Very deep convolutional networks for large-scale image recognition", 《ARXIV》 * |
| 童峥: "基于深度学习和探地雷达技术的路面结构病害检测研究", 《中国优秀硕士学位论文全文数据库基础科学辑》 * |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021083394A1 (en) * | 2019-11-01 | 2021-05-06 | 长安大学 | Method for detection of water damage to asphalt road surface based on adaptive selection of plot grayscales |
| US11828840B2 (en) | 2019-11-01 | 2023-11-28 | Chang' an University | Method for detecting moisture damage on asphalt pavement based on adaptive selection of GPR image grayscale |
| WO2021082904A1 (en) * | 2019-11-01 | 2021-05-06 | 长安大学 | Ground penetrating radar map adaptive selection method for detecting water damage |
| CN110991507A (en) * | 2019-11-22 | 2020-04-10 | 北京中科蓝图科技有限公司 | Road underground cavity identification method, device and system based on classifier |
| CN110988872A (en) * | 2019-12-25 | 2020-04-10 | 中南大学 | Method for rapidly identifying health state of wall body detected by unmanned aerial vehicle-mounted through-wall radar |
| CN110988839B (en) * | 2019-12-25 | 2023-10-10 | 中南大学 | Quick identification method for wall health condition based on one-dimensional convolutional neural network |
| CN110988872B (en) * | 2019-12-25 | 2023-10-03 | 中南大学 | Rapid identification method for detecting wall health state by unmanned aerial vehicle through-wall radar |
| CN110988839A (en) * | 2019-12-25 | 2020-04-10 | 中南大学 | Method for quickly identifying health condition of wall based on one-dimensional convolutional neural network |
| CN111582284B (en) * | 2020-04-27 | 2023-04-07 | 中国科学院信息工程研究所 | Privacy protection method and device for image recognition and electronic equipment |
| CN111582284A (en) * | 2020-04-27 | 2020-08-25 | 中国科学院信息工程研究所 | Privacy protection method and device for image recognition and electronic equipment |
| CN112036425A (en) * | 2020-05-09 | 2020-12-04 | 中铁四局集团有限公司 | Tunnel cavity state radar spectrum image recognition model construction method and tunnel cavity state radar spectrum image recognition method |
| CN111783784A (en) * | 2020-06-30 | 2020-10-16 | 创新奇智(合肥)科技有限公司 | Method and device for detecting building cavity, electronic equipment and storage medium |
| CN111767874A (en) * | 2020-07-06 | 2020-10-13 | 中兴飞流信息科技有限公司 | Pavement disease detection method based on deep learning |
| CN111767874B (en) * | 2020-07-06 | 2024-02-13 | 中兴飞流信息科技有限公司 | Pavement disease detection method based on deep learning |
| CN113191391A (en) * | 2021-04-07 | 2021-07-30 | 浙江省交通运输科学研究院 | Road disease classification method aiming at three-dimensional ground penetrating radar map |
| CN115343685A (en) * | 2022-08-29 | 2022-11-15 | 北京国电经纬工程技术有限公司 | Multi-dimensional ground penetrating radar detection method, device and equipment applied to disease identification |
| CN115792919A (en) * | 2023-01-19 | 2023-03-14 | 合肥中科光博量子科技有限公司 | Method for identifying pollution hot spot area through horizontal scanning and monitoring of aerosol laser radar |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110245642A (en) | A kind of radar spectrum recognition method and system based on deep learning | |
| CN109345507A (en) | A kind of dam image crack detection method based on transfer learning | |
| Refsgaard et al. | Operational validation and intercomparison of different types of hydrological models | |
| CN104614766B (en) | Seismic horizon skeletonization | |
| CN110220909A (en) | A kind of Shield-bored tunnels Defect inspection method based on deep learning | |
| CN106980858A (en) | The language text detection of a kind of language text detection with alignment system and the application system and localization method | |
| CN113256562B (en) | Road underground hidden danger detection method and system based on radar images and artificial intelligence | |
| CN109325960A (en) | A kind of infrared cloud image cyclone analysis method and analysis system | |
| CN104166163B (en) | Tomography curved surface extraction method based on three-dimensional big data quantity seismic data cube | |
| CN110398320A (en) | A kind of gas leakage detection localization method and system being easy to Continuous optimization | |
| CN109740676A (en) | Object detection moving method based on similar purpose | |
| CN114154427B (en) | Volume fracturing fracture expansion prediction method and system based on deep learning | |
| CN110242865A (en) | A kind of gas leakage detection determination method and system being easy to Continuous optimization | |
| Zong et al. | A deep learning approach for urban underground objects detection from vehicle-borne ground penetrating radar data in real-time | |
| CN113836999A (en) | Tunnel construction risk intelligent identification method and system based on ground penetrating radar | |
| CN109712183A (en) | Electronic speckle interference intelligent information retrieval method based on deep learning | |
| CN109827081A (en) | A kind of buried drain pipe road plugging fault and branch pipe tee connection part diagnostic method based on acoustics active detecting | |
| CN107247927A (en) | A kind of remote sensing images coastline information extracting method and system based on K-T Transformation | |
| CN108288092A (en) | A method of obtaining tight sand permeability using nuclear magnetic resonance T 2 spectrum form | |
| CN113009553A (en) | Interpretable seismic motion parameter probability density distribution prediction method based on NGboost and SHAP values | |
| Yue et al. | Automatic recognition of defects behind railway tunnel linings in GPR images using transfer learning | |
| CN113326661B (en) | RS-XGboost-based tunnel advanced drilling quantitative interpretation method and device | |
| CN113656868A (en) | BIM technology-based hospital construction collaborative management platform | |
| Wu et al. | Post-flood disaster damaged houses classification based on dual-view image fusion and Concentration-Based Attention Module | |
| CN109766780A (en) | A kind of ship smog emission on-line checking and method for tracing based on deep learning |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190917 |