CN108846341A - A kind of remote sensing images lake ice classifying identification method neural network based - Google Patents
A kind of remote sensing images lake ice classifying identification method neural network based Download PDFInfo
- Publication number
- CN108846341A CN108846341A CN201810568238.2A CN201810568238A CN108846341A CN 108846341 A CN108846341 A CN 108846341A CN 201810568238 A CN201810568238 A CN 201810568238A CN 108846341 A CN108846341 A CN 108846341A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- sensing images
- neural network
- network model
- lake ice
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of remote sensing images lake ice classifying identification methods neural network based, the remote sensing images in annual ice age in winter are first obtained from database, and carry out radiation calibration to the remote sensing images of acquisition and handle to obtain preliminary remote sensing images;Obtained radiation calibration treated remote sensing images carry out FLAASH atmospheric correction and obtain the remote sensing images after final process;Construct BP neural network model;The neural network model that building obtains in step (3) is trained, and test is carried out to the neural network model after training and judges whether to meet required precision, (5) are entered step if meeting, is returned if being unsatisfactory for and rebuilds neural network model;Remote sensing images after obtained final process are added in the neural network model built, and are classified to the lake ice in remote sensing images.The nicety of grading that the present invention obtains is higher.
Description
Technical field
The invention belongs to neural network classification applied technical field, in particular to a kind of remote sensing images neural network based
Lake ice classifying identification method.
Background technique
The common technology of remote sensing images lake ice Classification and Identification is based on traditional statistical analysis, and specific algorithm includes parallel
Hexahedron, minimum range, mahalanobis distance, maximum likelihood.Parallelepiped has " angle " to be easy to cause wrong classification;Minimum range
The classification correctness of algorithm depends greatly on the average value of model;If covariance matrix makes in mahalanobis distance
Be easy to causeing transition to classify and input the numerical requirements of wave band each time with biggish value must be Normal Distribution;Most
When wave band number has increased slightly when maximum-likelihood is classified, calculation amount and calculating time can all be greatly increased, and result in the reduction of efficiency
And whole process is to whether Normal Distribution is more demanding.In these conventional classification techniques, due to lacking for itself algorithm
It falls into, classification results are often inaccurate.
Neural network refers to the structure with computer simulation human brain, and the neuron of biology is simulated with many small processing units,
Identification, the memory, thinking processes that human brain is realized with algorithm can largely solve to calculate based on traditional statistical analysis
The problems such as misclassification of method.
Summary of the invention
Goal of the invention:In order to overcome the deficiencies in the prior art, the present invention, which provides, a kind of will be based on BP neural network
The Technology application of model is not high with the accuracy that solves the problems, such as to classify in the prior art to remote sensing images lake ice classification field
Remote sensing images lake ice classifying identification method neural network based.
Technical solution:To achieve the above object, the present invention provides a kind of remote sensing images lake ice classification neural network based
Recognition methods includes the following steps:
(1) remote sensing images in annual ice age in winter are obtained from database, and radiation calibration is carried out to the remote sensing images of acquisition
Processing obtains preliminary remote sensing images;
(2) treated that remote sensing images carry out FLAASH atmospheric correction obtains finally for the radiation calibration obtained to step (1)
Treated remote sensing images;
(3) BP neural network model is constructed;
(4) neural network model that building obtains in step (3) is trained, and to the neural network mould after training
Type carries out test and judges whether to meet required precision, enters step (5) if meeting, the return step if being unsatisfactory for
(3);
(5) remote sensing images after final process obtained in step (2) are added in the neural network model built,
And classify to the lake ice in remote sensing images.
Further, the BP neural network model constructed in the step (3) is the BP nerve net containing one layer of hidden layer
Network model.
Further, the required precision that test judges in the step (4) is 96%.
Further, when classifying in the step (5) to the lake ice in remote sensing images, three classes are first split into:Lake
Ice, waters, other.
Further, in the step (4) to the obtained neural network model of building be trained the specific steps are:
First every piece image in the remote sensing images after final process in step (2) is classified, is divided into three classes:Lake ice, waters,
Other, select training sample of 10000 pixels as model, and will be after all final process obtained in step (2)
Remote sensing images and the training sample of selection are brought into neural network model and are trained.
Beneficial effect:Compared with the prior art, the present invention has the following advantages:
BP network used in the present invention is very widely used in field of remote sensing image processing, this is because it is with powerful
Output mapping ability;Sample decision problem is converted nonlinear optimal problem by BP algorithm, and network connection weight passes through along ladder
Degree decline, interative computation are corrected.Algorithm includes the forward-propagating of information and the backpropagation of error.Forward-propagating passes through
The output processing result for receiving each neuron, the information processing of middle layer, hidden layer and output layer of input layer is completed primary
Learning process;It is triggering that the reverse transfer of error is not inconsistent in the output result of forward-propagating and expectation.Error is under error gradient
Drop mode corrective networks connection weight, and press the opposite direction path anti-pass of forward-propagating.This process is recycled, when error is reduced to
Tolerance interval or the practice number for reaching setting obtain the information of all kinds of targets finally with trained network class,
Complete classification.The present invention improves lake ice point using the neural network algorithm processing remote sensing images lake ice classification problem for having supervision
Class accuracy of identification.
Detailed description of the invention
Fig. 1 is overview flow chart of the invention.
Specific embodiment
The present invention will be further explained with reference to the accompanying drawing.
As shown in Figure 1, a kind of remote sensing images lake ice classifying identification method neural network based of the present invention, including
Following steps:
Step 1:Radiation calibration
The Nam Co ice ages in winter in 2015 for selecting a width to download Landsat-8 remote sensing images (map sheet size for
170*185, image spatial resolution 30, i.e., each pixel point areas are 30*30), using general under 5.3 software of ENVI
Calibration tool Radiometric Calibration does radiation calibration processing to selected image.It needs to be arranged during this
Data type required for FLAASH atmospheric correction, design parameter setting are as follows:Calibration type selective radiation rate data (that is,
Radiance), BIL or BIP is arranged in storage sequence (Interleave), and data type (Data Type) is Float type,
Radiance data unit regulation coefficient (Scale Factor) selection 0.1.Wave spectrum is checked when showing radiation calibration result images
Curve, the numerical value after calibration are concentrated mainly within the scope of 0-10, and unit is μ W/ (cm2*sr*nm).
Step 2:Atmospheric correction
Landsat-8 remotely-sensed data processed to step 1 carries out FLAASH atmospheric correction:Use 5.3 software of ENVI
In FLAASHAtmospheric Correction tool obtain the geographical coordinate of this Landsat image, determine that central point passes through
Latitude Scene Center Location;It selects sensor type Sensor Type for Landsat-8OLI, obtains its correspondence
Sensor height (height of Landsat-8OLI sensor be 705) and image data resolution ratio (30 resolution ratio);It obtains
Original image imaging time is taken, is recorded to relative parameters setting;Atmospheric models parameter selection (Atmospheric
Model it) is selected according to the rule of imaging time and latitude information;In order to reduce result storage space, default reflectivity multiply in
10000, output reflection rate range becomes 0-10000.
Step 3:Construct BP neural network model
BP network is the ANN of a kind of combination error backpropagation algorithm and prime multitiered network.Currently, BP network is distant
It is very widely used to feel field of image processing, this is because it has powerful output mapping ability, does not need to obey certain spies
Fixed rule (such as normal distribution), it thus can it is only necessary to be trained study to image network by known mode
It accesses an input pattern and is mapped to desired output mode, be finally reached correct classifying quality.BP network is as a kind of
The multilayer neural network being made of n neuron, it includes input layer, output layer, is also possible to have one layer or more in network
Layer hidden layer, may include multiple nodes in each layer.The node that same layer is belonged in this network be mutually not attached to and
Each node has and the node of only upper layer line.BP network carry out classification of remote-sensing images transmission process be:Input
Layer receives input signal (i.e. training sample), and hidden layer implies the more BP network moulds of the number of plies by distributed method stored knowledge
Type is more complicated, calculation amount is bigger, exports result through output layer after sample judgement.
The building of model is specific as follows:
(3.1) addition of activation primitive can make the ability to express for promoting BP neural network model.Therefore at each layer
An activation primitive Activation, i.e. Logistic function are all added in output below, export result by this activation primitive.
Other parameters are tentatively arranged as follows:Initial threshold Training Threshold Contribution is set as 0.8, training rate
Training Rate is set as 0.2, and training factor of momentum Training Momentum is set as 0.9, BP network global error
Training RMS Exit Criteria is set as 0.05, and the frequency of training of model is set as 500 training.
(3.2) number of plies of hidden layer is determined first.The present invention is provided with the different implicit numbers of plies (respectively in building model
Use 1 layer, 2 layers, 4 layers of implicit layer building model), the present invention selects the BP network model of 1 layer of hidden layer first.
Step 4:Training pattern carries out lake ice Classification and Identification
(4.1) by carrying out parameter initialization in step 3;
(4.2) Landsat-8 image processed to steps 1 and 2 does feature decision, interprets diagnostic method by visual observation for shadow
As being divided into three classes:Lake ice, waters, other, select 10000 pixels (to randomly choose and determine 10000 of type of ground objects
Pixel) training sample as model.The sample point of entire remote sensing image and selection is input to the BP that step 3 constructs
In neural network model;
(4.3) the model output category recognition result for utilizing step 3, followed by the precision evaluation of following steps 5,
For determining whether classifying quality required for reaching the present invention (the present invention claims overall classification accuracies to reach 96%).
Step 5:The evaluation of lake ice classification and recognition
The evaluation index that precision evaluation uses in the present invention be overall classification accuracy OverallAccuracy, i.e., all three
The ratio of correct the classified pixels summation and total sum of all pixels of class atural object.
Precision evaluation is to utilize the Confusion Matrix Using Ground in 5.3 software of ENVI in the present invention
Truth ROIs tool.
(5.1) firstly the need of the verifying sample for determining every a kind of atural object using the earth's surface image in Google Earth software
Point, same 10000 pixels of selection (pay attention to this time selection based entirely on Google Earth high precision image and cannot be complete
It is complete identical with test sample point in step (4.2));
(5.2) the input verifying sample point in 5.3 software of ENVI, opens the Confusion in 5.3 software of ENVI
Matrix Using Ground Truth ROIs tool selects the classification of step (4.3) to export result image;
(5.3) every one kind of sample point and classification results figure will be verified in Match Classes Parameters panel
Atural object corresponds, and selects the display styles of pixel and percentage, clicks ok output as a result, if reaching required precision, most
The BP network model of whole lake ice classification constructs the lake ice Classification and Identification that successfully can be used for the remote sensing image of other lake ice;If not
Reach required precision, then return step 3 modifies basic parameter or the implicit number of plies until reaching required precision;
(5.4) so far, a lake ice classification and precision evaluation are completed.By many experiments, BP that the present invention finally establishes
Network model is the network model comprising 1 layer of hidden layer (other parameters such as step 3 is given).This is because the implicit number of plies is more,
Network model is more complicated, and calculation amount is also bigger, this is not suitable for lake ice Classification and Identification of the invention instead, therefore the present invention selects
Select 1 layer of hidden layer.
Step 6:The BP network model built is applied in the remote sensing image lake ice Classification and Identification of other lake ice.
The above is only a preferred embodiment of the present invention, it should be pointed out that:For the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications
It should be regarded as protection scope of the present invention.
Claims (5)
1. a kind of remote sensing images lake ice classifying identification method neural network based, which is characterized in that include the following steps:
(1) remote sensing images in annual ice age in winter are obtained from database, and radiation calibration processing is carried out to the remote sensing images of acquisition
Obtain preliminary remote sensing images;
(2) treated that remote sensing images carry out FLAASH atmospheric correction obtains final process for the radiation calibration obtained to step (1)
Remote sensing images afterwards;
(3) BP neural network model is constructed;
(4) neural network model that building obtains in step (3) is trained, and the neural network model after training is carried out
It tests and judges whether to meet required precision, enter step (5) if meeting, return step (3) if being unsatisfactory for;
(5) remote sensing images after final process obtained in step (2) are added in the neural network model built, and to distant
Lake ice in sense image is classified.
2. a kind of remote sensing images lake ice classifying identification method neural network based according to claim 1, feature exist
In the BP neural network model constructed in the step (3) is the BP neural network model containing one layer of hidden layer.
3. a kind of remote sensing images lake ice classifying identification method neural network based according to claim 1, feature exist
In the required precision that test judges in the step (4) is 96%.
4. a kind of remote sensing images lake ice classifying identification method neural network based according to claim 1, feature exist
In being first split into three classes when classifying in the step (5) to the lake ice in remote sensing images:Lake ice, waters, other.
5. a kind of remote sensing images lake ice classifying identification method neural network based according to claim 1, feature exist
In, in the step (4) to the obtained neural network model of building be trained the specific steps are:First by step (2) most
Every piece image in treated remote sensing images is classified eventually, is divided into three classes:Lake ice, waters, other, select 10000
Training sample of the pixel as model, and by after all final process obtained in step (2) remote sensing images and selection
Training sample is brought into neural network model and is trained.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810568238.2A CN108846341A (en) | 2018-06-05 | 2018-06-05 | A kind of remote sensing images lake ice classifying identification method neural network based |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810568238.2A CN108846341A (en) | 2018-06-05 | 2018-06-05 | A kind of remote sensing images lake ice classifying identification method neural network based |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108846341A true CN108846341A (en) | 2018-11-20 |
Family
ID=64211455
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810568238.2A Pending CN108846341A (en) | 2018-06-05 | 2018-06-05 | A kind of remote sensing images lake ice classifying identification method neural network based |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108846341A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110502978A (en) * | 2019-07-11 | 2019-11-26 | 哈尔滨工业大学 | A kind of laser radar waveform Modulation recognition method based on BP neural network model |
CN111476111A (en) * | 2020-03-20 | 2020-07-31 | 航天信德智图(北京)科技有限公司 | Potato remote sensing identification method based on BP neural network |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103559500A (en) * | 2013-10-15 | 2014-02-05 | 北京航空航天大学 | Multispectral remote sensing image land feature classification method based on spectrum and textural features |
CN107688780A (en) * | 2017-08-22 | 2018-02-13 | 河海大学 | A kind of Hyperspectral Remote Sensing Imagery Classification method |
-
2018
- 2018-06-05 CN CN201810568238.2A patent/CN108846341A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103559500A (en) * | 2013-10-15 | 2014-02-05 | 北京航空航天大学 | Multispectral remote sensing image land feature classification method based on spectrum and textural features |
CN107688780A (en) * | 2017-08-22 | 2018-02-13 | 河海大学 | A kind of Hyperspectral Remote Sensing Imagery Classification method |
Non-Patent Citations (1)
Title |
---|
张辉: "基于BP神经网络的遥感影像分类研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110502978A (en) * | 2019-07-11 | 2019-11-26 | 哈尔滨工业大学 | A kind of laser radar waveform Modulation recognition method based on BP neural network model |
CN111476111A (en) * | 2020-03-20 | 2020-07-31 | 航天信德智图(北京)科技有限公司 | Potato remote sensing identification method based on BP neural network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109993220B (en) | Multi-source remote sensing image classification method based on double-path attention fusion neural network | |
CN105528638B (en) | The method that gray relative analysis method determines convolutional neural networks hidden layer characteristic pattern number | |
CN107230113A (en) | A kind of house property appraisal procedure of multi-model fusion | |
CN111091105A (en) | Remote sensing image target detection method based on new frame regression loss function | |
CN110135267A (en) | A kind of subtle object detection method of large scene SAR image | |
Winiwarter et al. | Classification of ALS point clouds using end-to-end deep learning | |
CN108564109A (en) | A kind of Remote Sensing Target detection method based on deep learning | |
CN110334765A (en) | Remote Image Classification based on the multiple dimensioned deep learning of attention mechanism | |
CN107229904A (en) | A kind of object detection and recognition method based on deep learning | |
CN112070078B (en) | Deep learning-based land utilization classification method and system | |
CN106600595A (en) | Human body characteristic dimension automatic measuring method based on artificial intelligence algorithm | |
CN110245678A (en) | A kind of isomery twinned region selection network and the image matching method based on the network | |
CN109817276A (en) | A kind of secondary protein structure prediction method based on deep neural network | |
CN106991666A (en) | A kind of disease geo-radar image recognition methods suitable for many size pictorial informations | |
CN108154094A (en) | The non-supervisory band selection method of high spectrum image divided based on subinterval | |
CN108447057A (en) | SAR image change detection based on conspicuousness and depth convolutional network | |
CN113743417B (en) | Semantic segmentation method and semantic segmentation device | |
Jiang et al. | Application of back propagation neural network in the classification of high resolution remote sensing image: take remote sensing image of Beijing for instance | |
CN113077891A (en) | Big data disease diagnosis system based on algorithm, block chain and medical image | |
CN108734717A (en) | The dark weak signal target extracting method of single frames star chart background based on deep learning | |
CN108846341A (en) | A kind of remote sensing images lake ice classifying identification method neural network based | |
CN115860269A (en) | Crop yield prediction method based on triple attention mechanism | |
CN111814804A (en) | Human body three-dimensional size information prediction method and device based on GA-BP-MC neural network | |
CN113256733B (en) | Camera spectral sensitivity reconstruction method based on confidence voting convolutional neural network | |
CN117636183A (en) | Small sample remote sensing image classification method based on self-supervision pre-training |
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: 20181120 |
|
RJ01 | Rejection of invention patent application after publication |