CN108280440A - A kind of fruit-bearing forest recognition methods and system - Google Patents
A kind of fruit-bearing forest recognition methods and system Download PDFInfo
- Publication number
- CN108280440A CN108280440A CN201810133935.5A CN201810133935A CN108280440A CN 108280440 A CN108280440 A CN 108280440A CN 201810133935 A CN201810133935 A CN 201810133935A CN 108280440 A CN108280440 A CN 108280440A
- Authority
- CN
- China
- Prior art keywords
- fruit
- feature
- several
- sub
- target area
- 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
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Image Processing (AREA)
Abstract
A kind of fruit-bearing forest recognition methods of present invention offer and system.Wherein, method includes:Obtain the high-resolution remote sensing image and digital elevation model of target area;Target area is divided into several equal-sized sub-goal regions, the target signature in each sub-goal region is obtained according to the high-resolution remote sensing image of target area and digital elevation model;Target signature includes several spectral signatures, several textural characteristics and several featuress of terrain;The target signature in each sub-goal region is inputted into the random forest grader that builds in advance respectively, judges whether each sub-goal region is fruit-bearing forest, according to each sub-goal region whether be fruit-bearing forest judging result, the fruit-bearing forest in identification object region.System includes:Remote sensing module, characteristic extracting module and identification module.A kind of fruit-bearing forest recognition methods provided by the invention and system improve the accuracy rate of fruit-bearing forest identification.
Description
Technical field
The present invention relates to remote sensing technology fields, more particularly, to a kind of fruit-bearing forest recognition methods and system.
Background technology
With quickly and accurately extracting fruit-bearing forest space distribution information and area to stablizing orchard area, optimization space layout
And relevant departments formulate decision and are of great significance.The features such as macroscopical, quick, real-time possessed by remote sensing technology, to extract people
Work fruit tree woods information provides the technological means of efficient quick.
The areas such as southern region of China such as Hainan, Guangdong, Guangxi are important fruit-producing area, but these area ground species
Type is also complicated various, and vegetation is green throughout the year, and plot fragmentation degree is high and area is small, when single or more it is low, in point
Resolution image often cannot be satisfied the demand of high-precision fruit-bearing forest (such as mango woods) information extraction and variation monitoring.With aviation
Space technology, sensor are constantly progressive, and the application range of high-resolution remote sensing image is gradually expanded, and how to be based on high-resolution
The extracted with high accuracy that image carries out the artificial fruit tree woods information of complicated earth surface underlying surface needs to be studied.
In terms of the ground object information extraction research based on remote sensing image, tradition relies on merely spectral information based on single pixel
Information extracting method mainly utilize single image picture element spectral signature information extraction atural object method, low resolution in being suitable for
Rate is multispectral and Hyperspectral imaging;The information extracting method of expertise decision tree is the spectral signature for utilizing single pixel, sky
Between a kind of ground object information extraction method suitable for multi-source data for sorting out of relationship and other context relations, whole process knowledge
It obtains more complicated;Tradition be based on single picture dot pure spectrum and Knowledge Decision-making tree remote sensing information extracting method, merely with individually as
The spectral information of member, nicety of grading are very low.Often exist inside fruit-bearing forest and between fruit-bearing forest and other natural forests, economic forest
" same object different images ", " the different spectrum of jljl " phenomenon so that the existing skill of identification and the information extraction of fruit-bearing forest is carried out merely with spectral information
Art is not still high in accuracy rate.
Invention content
To overcome the shortcomings of that recognition accuracy of the existing technology is not high, the present invention provide a kind of fruit-bearing forest recognition methods and
System.
According to an aspect of the present invention, a kind of fruit-bearing forest recognition methods is provided, including:
S1, the high-resolution remote sensing image and digital elevation model for obtaining target area;
S2, the target area is divided into several equal-sized sub-goal regions, according to the target area
High-resolution remote sensing image and digital elevation model obtain the target signature in each sub-goal region;The target signature packet
Include several spectral signatures, several textural characteristics and several featuress of terrain;
S3, the target signature in each sub-goal region is inputted to the random forest grader built in advance respectively,
Judge whether each sub-goal region is fruit-bearing forest, according to each sub-goal region whether be fruit-bearing forest judging result,
Identify the fruit-bearing forest in the target area.
Preferably, the specific steps for obtaining the random forest grader built include:
The training sample that S01, extraction obtain in advance concentrates the feature of the first quantity of each training sample;First number
The feature of amount includes several spectral signatures, several textural characteristics and several featuress of terrain;
S02, there is randomly select preset first ratio training sample with putting back to from training sample concentration;
S03, the feature that the second quantity is randomly selected from the feature of first quantity use pumping according to splitting rule
The feature of second quantity of the training sample taken obtains the split point of decision tree, and decision tree is generated according to the split point;
Second quantity is less than first quantity;
S04, step S02 and S03 are repeated, until the number extracted reaches preset times;
S05, the whole decision tree according to generation, generate random forest grader to be verified;
S06, using the verification sample set obtained in advance, the classification for obtaining the random forest grader to be verified is accurate
Rate, when the classification accuracy is more than accuracy rate threshold value, using the random forest grader to be verified as the structure
Good random forest grader, using the feature of current first quantity as the target signature.
Preferably, the training sample set is obtained in advance and the specific steps for verifying sample set include:
Obtain the high-resolution remote sensing image and digital elevation model of several sample areas;
Each sample areas is divided into several equal-sized increment one's respective areas, the increment one's respective area it is big
It is small equal in magnitude with the target area;
To sample set all be formed as sample in the increment one's respective area;
The sample of preset second ratio is extracted from the sample set as training sample, forms the training sample
Collection;Using the sample not being extracted in the sample set as verification sample, the verification sample set is formed.
Preferably, the spectral signature in the feature of first quantity includes:
Blue wave band spectral information, green light band spectral information, red spectral band spectral information and near infrared band spectrum letter
Breath, normalized differential vegetation index, soil lightness index and soil adjust vegetation index;
Textural characteristics in the feature of first quantity include:It is the variances of gray level co-occurrence matrixes, correlation, non-similar
Property, homogeney, contrast, second moment and entropy;
Features of terrain in the feature of first quantity includes:Elevation and the gradient.
Preferably, the step S06 further includes:
When the classification accuracy is less than accuracy rate threshold value, obtains the training sample and concentrate each training sample several
The feature of several described new features and current first quantity is collectively formed the spy of the first new quantity by a new feature
Sign, and step S02 is executed, until the classification accuracy rate is more than accuracy rate threshold value.
Preferably, the new feature includes at least new spectral signature and/or new textural characteristics;
The new spectral signature includes at least:Enhancement mode meta file, ratio vegetation index, green normalization difference are planted
By one or more of index, triangle vegetation index, difference vegetation index and normalization difference aqua index;
The new textural characteristics include at least:The unfavourable balance of gray level co-occurrence matrixes divides square, intermediate value, covariance, contrast, symbiosis
With one or more of mean value, symbiosis and variance, symbiosis and entropy, symbiosis difference mean value, symbiosis difference variance and symbiosis difference entropy.
Preferably, further include between the step S05 and step S06:
Using the training sample not being extracted when generating every decision tree, current random forest grader to be verified is obtained
The outer data error of bag;
According to data error outside the bag, the importance of each feature in the feature of current first quantity is obtained;
According to the importance of each feature in the feature of current first quantity, if in the feature of current first quantity of reservation
Dry feature, as the feature of the first new quantity, and executes step S02, until generating new random forest to be verified point
Class device.
Preferably, the specific steps of the high-resolution remote sensing image for obtaining target area include:
The clear cloudless high-resolution satellite image for obtaining the target area specific period, defends the high-resolution
Star image is pre-processed, and the high-resolution remote sensing image of the target area is obtained;
To the high-resolution satellite image carry out pretreated method include radiation calibration, atmospheric correction, ortho-rectification,
Geometric accurate correction and projective transformation.
Preferably, described that each institute is obtained according to the high-resolution remote sensing image and digital elevation model of the target area
The specific steps for stating the target signature in sub-goal region include:
For each sub-goal region, which is obtained according to the high-resolution remote sensing image of the target area
Several spectral signatures and several textural characteristics in region, the specific item is obtained according to the digital elevation model of the target area
Mark several featuress of terrain in region.
According to another aspect of the present invention, a kind of fruit-bearing forest identifying system is provided, including:
Remote sensing module, high-resolution remote sensing image and digital elevation model for obtaining target area;
Characteristic extracting module, for the target area to be divided into several equal-sized sub-goal regions, according to
The high-resolution remote sensing image and digital elevation model of the target area obtain the target signature in each sub-goal region;
The target signature includes several spectral signatures, several textural characteristics and several featuress of terrain;
Identification module builds the input of the target signature in each sub-goal region for respectively random gloomy in advance
Woods grader judges whether each sub-goal region is fruit-bearing forest, whether is fruit-bearing forest according to each sub-goal region
Judging result identifies the fruit-bearing forest in the target area.
A kind of fruit-bearing forest recognition methods provided by the invention and system, it is special by extracting spectral signature, textural characteristics and landform
The feature as sub-goal region is levied, and is classified using random forest grader, the fruit-bearing forest in identification object region improves
The accuracy rate of fruit-bearing forest identification can meet the demands such as fruit-bearing forest normality monitoring, the accuracy rate with the identification of higher fruit-bearing forest.
Description of the drawings
Fig. 1 is a kind of flow chart of fruit-bearing forest recognition methods of the embodiment of the present invention;
Fig. 2 is a kind of functional block diagram of fruit-bearing forest identifying system of the embodiment of the present invention.
Specific implementation mode
With reference to the accompanying drawings and examples, the specific implementation mode of the present invention is described in further detail.Implement below
Example is not limited to the scope of the present invention for illustrating the present invention.
Fig. 1 is a kind of flow chart of fruit-bearing forest recognition methods of the embodiment of the present invention.As shown in Figure 1, a kind of fruit-bearing forest recognition methods
Including:Step S1, the high-resolution remote sensing image and digital elevation model of target area are obtained;Step S2, target area is drawn
It is divided into several equal-sized sub-goal regions, is obtained according to the high-resolution remote sensing image of target area and digital elevation model
Take the target signature in each sub-goal region;Target signature include several spectral signatures, several textural characteristics and several
Features of terrain;Step S3, the target signature in each sub-goal region is inputted into the random forest grader built in advance respectively,
Judge whether each sub-goal region is fruit-bearing forest, according to each sub-goal region whether be fruit-bearing forest judging result, identify target
Fruit-bearing forest in region.
It should be noted that fruit-bearing forest recognition methods provided by the invention and system are based on high-resolution remote sensing image, identification
Fruit-bearing forest in some target area obtains the distribution of fruit-bearing forest in the target area.Fruit-bearing forest recognition methods provided by the invention and it is
Identification of the system suitable for various fruit-bearing forests, such as mango woods, palm grove, are particularly suitable for the identification of artificial fruit-bearing forest.
The rapid development of information technology and sensor technology brings the rapid development of remote sensing technology.Satellite remote-sensing image
Resolution ratio (including spatial resolution, spectral resolution and temporal resolution etc.) be all greatly improved, from 1972
The order of magnitude is 1m grades of even 1m grades of high-resolution below by now for the earth observation satellite development that first resolution ratio of transmitting is 80m
Rate remote sensing satellite.1999, U.S.'s space was imaged first Remote Sensing Satellites with High Resolution IKONOS of company (Yi Ke Northeys)
Successful launch, started the new era of Remote Sensing Satellites with High Resolution.Currently, the High Resolution Remote Sensing Satellites of mainstream provide
High-resolution remote sensing image resolution ratio be all up 1m even 1m or less.
Specifically, step S1, when needing the fruit-bearing forest in identification object region, the high-resolution for obtaining the target area is distant
Feel image and digital elevation model.
Current main High Resolution Remote Sensing Satellites have the high score No.1 of China, a high score two, the IKONOS in the U.S.,
QuickBird, WorldView series, GeoEye-1, French SPOT series, Pleiades, EROS-A, the EROS- of Israel
B, Russian " resource DK " etc..
Above-mentioned High Resolution Remote Sensing Satellites can be utilized, the high-resolution remote sensing image of the target area is obtained.
Digital elevation model (Digital Elevation Model, vehicle economy M) is by limited terrain elevation data
Realize the digitized simulation (i.e. the digital expression of topographical surface form) to ground surface or terrain.It is with one group of orderly array of values
Form indicates a kind of actual ground model of ground elevation, is digital terrain model (Digital Terrain Model, abbreviation
DTM thus a branch), various other topographic index can derive from.
It is generally believed that DTM is various geomorphologic factors of the description including elevation, such as the gradient, slope aspect, change of slope
The spatial distribution of linear and nonlinear combination including the factor.Wherein, DEM is the simple individual event digital land value model model of zeroth order,
His such as gradient, slope aspect and change of slope landforms characteristic can derive from the basis of DEM.
Digital elevation model can obtain by the following method:Photogrammetric, ground survey, have map digital,
Extraction etc. in the existing libraries DEM.
Target area can be divided into equal in magnitude by step S2 according to the size of the pixel of high-resolution remote sensing image
Several sub-goal regions.
Usually, the size in sub-goal region is equal to the size of the pixel of high-resolution remote sensing image, but not limited to this.
Pixel, also known as pixel or pixel point, i.e. image unit (picture element).Pixel is composition digitlization
The minimum unit of image.In remote sensing data acquiring, when such as scanning imagery, it is that sensor is scanned sampling to ground scenery
Minimum unit;In Digital Image Processing, it is sampled point when being scanned digitlization to analog image.Pixel is that composition is distant
The basic unit for feeling digital picture, is the sampled point during remotely sensed image.
Pixel is the important symbol for reflecting image feature.Pixel is the data for having simultaneously space characteristics and Spectral Characteristic
Member.The geometric meaning of pixel is that its data value determines representative floor area.The physical significance of pixel is its wave spectrum variable generation
In the table pixel in a certain specific band spectral response intensity, i.e., the atural object in same pixel, only there are one common gray scales
Value.Pixel size determines the image resolution and information content of digitized video.Pixel is smaller, and image resolution is higher, information content
It is bigger;Conversely, image resolution is lower, information content is smaller.
The resolution ratio of high-resolution remote sensing image obtains the High Resolution Remote Sensing Satellites of the high-resolution remote sensing image
Resolution ratio.
When the resolution ratio of high-resolution remote sensing image is 1 km, a pixel represents the face of the km of 1 km of ground × 1
Product, i.e. 1 sq-km;When the resolution ratio of high-resolution remote sensing image is 30 meters, a pixel represents 30 meters × 30 meters of ground
Area;When the resolution ratio of high-resolution remote sensing image is 1 meter, that is to say, that a pixel on image is equivalent to ground 1
Rice × 1 meter of area, i.e., 1 square metre.
Since the high-resolution remote sensing image and digital elevation model of acquisition are all based on same target area, by mesh
After region division is marked into several equal-sized sub-goal regions, high-resolution remote sensing image and digital elevation mould can be obtained
In type, the corresponding part in each sub-goal region.
It, can be according in the high-resolution remote sensing image and digital elevation model of target area for each sub-goal region
The corresponding part in sub-goal region obtains several spectral signatures in the sub-goal region, several textural characteristics and several
A features of terrain, the target signature as the sub-goal region.
Spectral signature refers to spectral characteristic of ground.Spectral characteristic of ground is that any atural object all has its own in nature
Electromagnetic radiation laws such as have reflection, absorb the characteristic of certain wave bands of external ultraviolet light, visible light, infrared ray and microwave,
They all have the characteristic for emitting certain infrared rays, microwave again;A small number of atural objects also characteristic with transmitted electromagnetic wave.This characteristic
Referred to as spectral characteristic of ground.
Texture is a kind of visual phenomenon of generally existing.Texture is a kind of visual signature reflecting homogeneity phenomenon in image,
What it embodied body surface has slowly varying or periodically variable surface textural alignment attribute.The texture of image
Feature is mainly shown as:Part spatial variations order constantly repeated in the region of bigger, sequence be by basic element it is non-with
Machine arranges and Anywhere has same structure size in composition and texture region.
Texture is showed by pixel and its intensity profile of surrounding space neighborhood, i.e. local grain information.In addition, part
The repeatability of texture information in varying degrees is exactly global texture information.While textural characteristics embody the property of global characteristics,
It also illustrates the surface nature of scenery corresponding to image or image-region.
Textural characteristics describe method and generally include statistical method, geometric method, modelling, signal processing method and structured analysis method
Deng.
Wherein, most common is statistical method.The Typical Representative of statistical method is that one kind being referred to as gray level co-occurrence matrixes
(GLCM) texture analysis method.Gray level co-occurrence matrixes are built upon on the basis of the second order hybrid conditional probability density of estimation image
A kind of method.This method utilizes the various statistical properties of gray level co-occurrence matrixes, describes textural characteristics.
Step S3, after the target signature for obtaining each sub-goal region, respectively by the target signature in each sub-goal region
The random forest grader built in advance is inputted, judges whether each sub-goal region is fruit-bearing forest, obtains each sub-goal area
Domain whether be fruit-bearing forest judging result.According to each sub-goal region whether be fruit-bearing forest judging result, target area can be obtained
Atural object is whole sub-goal regions of fruit-bearing forest in domain, to identify the fruit-bearing forest in target area.
Before the fruit-bearing forest in target area is identified, need to build grader in advance.Grader is used for mesh
Sub-goal region in mark region is classified, and judges whether sub-goal region is fruit-bearing forest.In the embodiment of the present invention, grader is
Random forest grader.
In machine learning, random forest is a grader for including more decision trees, and the classification of its output is
Depending on mode by the classification of every decision tree output.
Random forest grader has lot of advantages:Classification accuracy is high, can handle high-dimensional (there are many characteristic variable)
Data, energy parallelization generation decision tree, training speed are fast, realize simply, and energy balance error can assess the importance of characteristic variable
Influencing each other between detection characteristic variable, generalization ability is strong, and accuracy can be still maintained when Partial Feature is lost.
The embodiment of the present invention is used as the feature in sub-goal region by extracting spectral signature, textural characteristics and features of terrain,
And classified using random forest grader, the fruit-bearing forest in identification object region preferably overcomes " the different spectrum of jljl ", " same
The influence of spectrum foreign matter " phenomenon, improves the accuracy rate of fruit-bearing forest identification, can meet the demands such as fruit-bearing forest normality monitoring, especially in China
It is relatively previous other when southern plot is broken, type of ground objects is various, regions with complex terrain (such as Hainan Region) carries out fruit-bearing forest identification
Fruit-bearing forest recognition methods has compared with high-accuracy.
Based on above-described embodiment the tool of the high-resolution remote sensing image of target area is obtained as a preferred embodiment
Body step includes:The clear cloudless high-resolution satellite image for obtaining the target area specific period, to high-resolution satellite shadow
As being pre-processed, the high-resolution remote sensing image of target area is obtained;Pretreated side is carried out to high-resolution satellite image
Method includes radiation calibration, atmospheric correction, ortho-rectification, geometric accurate correction and projective transformation.
Specifically, in order to reduce disturbing factor, the specific period and it is clear cloudless when, obtain high-resolution satellite acquisition
The high-resolution satellite image of target area.
Since high-resolution satellite is located at outside earth atmosphere, when the high-resolution satellite image of the target area of acquisition,
It can be interfered by factors such as atmospheres, and certain image fault can be caused due to sensor, camera lens etc. itself, be adopted
After the high-resolution satellite image for collecting target area, the high-resolution satellite image of target area is pre-processed.
Carrying out pretreated method to the high-resolution satellite image of target area includes:By the high-resolution satellite of acquisition
Image carries out radiation calibration, atmospheric correction, ortho-rectification, geometric accurate correction and projective transformation, but is not limited to radiation calibration, air
Five kinds of methods such as correction, ortho-rectification, geometric accurate correction and projective transformation.
When radiation calibration is that user needs to calculate the spectral reflectivity or spectral radiance of atural object, or need to difference
When the image that time, different sensors obtain is compared, the luminance grayscale values of image are converted to absolute radiance
Process.
The global radiation brightness for the ground target that sensor finally measures not is the reflection of earth's surface real reflectance, wherein wrapping
The amount of radiation error caused by Atmospheric Absorption, especially scattering process is contained.Atmospheric correction is exactly to eliminate these by atmospheric effect
Caused radiation error reflects the process of the true surface reflectivity of atural object.
Ortho-rectification utilizes the shadow obtained originally generally by choosing some ground control points on image
As digital elevation model (DEM) data in range, slope correction is carried out at the same time to image and height displacement corrects, image is adopted again
Sample is at orthography.By ortho-rectification, the image point displacement because of caused by hypsography and sensor error can be corrected.
Geometric accurate correction is also known as geometrical registration, refers to the geometry deformation eliminated in image, generates a width and meets certain map
The new images that projection or avatars require.Geometric accurate correction be it is a kind of eliminate image geometry deformation, realize original image with
The process that the geometry of standard picture or map is integrated.
Projective transformation (projection transformation) be a kind of coordinate of map projection's point is transformed to it is another
The process of the coordinate of kind map projection point.
By above-mentioned preprocess method, can correct deformation or low-quality target area high-resolution satellite figure
Picture, to more truly reflect target area.
After being pre-processed to the high-resolution satellite image of target area, by the high-resolution of pretreated target area
High-resolution remote sensing image of the rate satellite image as target area.
The embodiment of the present invention obtains high-resolution remote sensing image by carrying out pretreatment to high-resolution satellite image, can have
Effect is interfered caused by eliminating air, sensor etc., further increases the accuracy rate of fruit-bearing forest identification.
Based on above-described embodiment, each son is obtained according to the high-resolution remote sensing image of target area and digital elevation model
The specific steps of the target signature of target area include:It is distant according to the high-resolution of target area for each sub-goal region
Sense image obtains several spectral signatures and several textural characteristics in the sub-goal region, according to the digital elevation of target area
Model obtains several featuress of terrain in the sub-goal region.
Specifically, for each sub-goal region, several spectral signatures in the sub-goal region and several textures are special
Sign is obtained by the high-resolution remote sensing image of target area.According to the specific item in the high-resolution remote sensing image of target area
The corresponding part in region is marked, several spectral signatures and several textural characteristics in the sub-goal region can be obtained.
For each sub-goal region, several featuress of terrain in the sub-goal region are high by the number of target area
Journey model obtains.According to the corresponding part in sub-goal region in the digital elevation model of target area, the specific item can be obtained
Mark several featuress of terrain in region.
Based on above-described embodiment the specific step of the random forest grader built is obtained as a preferred embodiment
Suddenly include:Step S01, the training sample that extraction obtains in advance concentrates the feature of the first quantity of each training sample;First number
The feature of amount includes several spectral signatures, several textural characteristics and several featuress of terrain;Step S02, from training sample
It is concentrated with randomly select preset first ratio training sample with putting back to;Step S03, it is taken out at random from the feature of the first quantity
The feature for taking the second quantity obtains decision tree according to splitting rule using the feature of the second quantity of the training sample of extraction
Split point generates decision tree according to split point;Second quantity is less than the first quantity;Step S04, repeat step S02 and
S03, until the number extracted reaches preset times;Step S05, it according to whole decision trees of generation, generates to be verified random
Forest classified device;Step S06, using the verification sample set obtained in advance, the classification of random forest grader to be verified is obtained
Accuracy rate, it is when classification accuracy is more than accuracy rate threshold value, random forest grader to be verified is random as what is built
Forest classified device, using the feature of current first quantity as target signature.
Specifically, since random forest grader includes more decision trees, utilize the training sample obtained in advance
Collection generates the random forest grader built on the basis of generating the decision tree of preset quantity.
Step S01 extracts the first of the training sample for each training sample that the training sample obtained in advance is concentrated
The feature of quantity.
The feature of first quantity includes several spectral signatures, several textural characteristics and several featuress of terrain.First
In the feature of quantity, spectral signature, textural characteristics and features of terrain can be preset, therefore, spectral signature, textural characteristics
After being set with features of terrain, it may be determined that the first quantity.
For example, when setting 7 spectral signatures, 7 textural characteristics and 2 featuress of terrain, the first quantity is 16.
After the feature of the first quantity for extracting each training sample, one can be generated certainly by step S02 and step S03
Plan tree.
Step S02 has randomly select preset first ratio training sample with putting back to from training sample concentration.
Preferably, the method randomly selected with putting back to uses the Bootstrap methods of samplings.
Preferably, preset first ratio is 2/3rds.Can also according to actual conditions, select suitable ratio as
Preset first ratio.
Step S03 establishes decision tree according to the training sample of extraction.Decision tree is Taxonomy and distribution
(Classification and regression tree, abbreviation CART).
The feature of the second quantity is randomly selected from the feature of the first quantity, the second quantity is less than the first quantity.
Suitable second quantity can be determined according to actual conditions.
For example, when the first quantity is 16, can select to randomly select 10 features from above-mentioned 16 features, it is above-mentioned
The 10 features i.e. feature of the second quantity.
According to splitting rule, select optimal feature as decision tree successively from the feature for randomly selecting the second quantity
Fork attribute, and determine node and the branch of decision tree, to obtain the split point of decision tree.
The division threshold value of split point is determined according to Gini coefficient minimum principles.It calculates and extracts preset first ratio
In training sample, each training sample the split point Gini coefficients, using the minimum value of wherein Gini coefficients as the division
The division threshold value of point.
By step S02 and step S03, a decision tree can be generated.
Step S04 repeats step S02 and S03.The number for repeating step S02 and S03 is preset times, i.e.,
There is the number of randomly select preset first ratio training sample with putting back to reach preset times from training sample concentration.
Such as when preset times are N, have from training sample concentration randomly select preset first ratio of n times with putting back to
Training sample can obtain the subset of N number of different training sample set;Decision is generated according to the subset of each training sample set
Tree, can generate N decision tree.
Whole decision trees after the whole decision trees for obtaining generation, are formed random forest grader by step S05.At this time
Random forest grader is random forest grader to be verified.
The classification of random forest grader output is used according to the classification of every decision tree output of composition random forest
The mode of ballot determines.
Step S06, after obtaining random forest grader to be verified, using the verification sample set obtained in advance, to be tested
The classification accuracy rate of the random forest grader of card is verified.
The feature for verifying the first quantity of each verification sample in sample set is inputted into random forest to be verified point respectively
Class device judges random forest grader to be verified to the verification according to the classification that random forest grader to be verified exports
Whether the classification of sample is correct.
When classification accuracy rate is more than accuracy rate threshold value, illustrate the classification accuracy rate symbol of random forest grader to be verified
It closes and requires, using random forest grader to be verified as the random forest grader built, and by current first quantity
Feature is as target signature.
Based on above-described embodiment, as a preferred embodiment, obtains training sample set in advance and verify the tool of sample set
Body step includes:Obtain the high-resolution remote sensing image and digital elevation model of several sample areas;By each sample areas
It is divided into several equal-sized increment one's respective areas, the size of increment one's respective area is equal in magnitude with sub-goal region;It will be complete
Portion increment one's respective area forms sample set as sample;The sample of preset second ratio is extracted from sample set as training sample
This, forms training sample set;Using the sample not being extracted in sample set as verification sample, composition verification sample set.
Specifically, training sample set and verification sample set can obtain in advance.
Using the region where typical land-use style as sample areas.Typical land-use style includes fruit-bearing forest, arable land, people
Work surface, forest, water body etc..
Obtain the high-resolution remote sensing image and digital elevation model of several sample areas.
The high-resolution remote sensing image of sample areas, is obtained by High Resolution Remote Sensing Satellites.
Preferably, the specific period and it is clear cloudless when, obtain the high-resolution of the sample areas of high-resolution satellite acquisition
Rate satellite image, and the high-resolution satellite image of sample areas is passed through into radiation calibration, atmospheric correction, ortho-rectification, geometry
The methods of fine correction and projective transformation are pre-processed, and the high-resolution remote sensing image of sample areas is obtained.
The digital elevation model of sample areas, can by photogrammetric, ground survey, have map digital or
The methods of extraction obtains in some libraries DEM.
Each sample areas is divided into several equal-sized increment one's respective areas.In order to ensure the accurate of fruit-bearing forest identification
Rate, the size of increment one's respective area are equal in magnitude with sub-goal region.
Based on ground investigation, the latitude and longitude information of each increment one's respective area is obtained.Believed according to the longitude and latitude of increment one's respective area
Breath can determine the land-use style of the increment one's respective area by consulting the modes such as statistical yearbook, on-site inspection.
Using whole increment one's respective areas as sample, sample set is formed.Sample set is divided into two parts, a part is for instructing
Practice, obtains random forest grader to be verified;Another part is for correct to the classification of random forest grader to be verified
Rate is verified
With the sample that preset second ratio randomly drawing sample is concentrated, as training sample, by whole training sample groups
At training sample set;Using the sample not being extracted as verification sample, by all verification sample composition verification sample sets.
Preferably, the second ratio is 4/5ths.That is the ratio of number of training sample and verification sample is 4:1.
Based on above-described embodiment, as a preferred embodiment, the spectral signature in the feature of the first quantity includes:Blue light
Band spectrum information, green light band spectral information, red spectral band spectral information and near infrared band spectral information, normalization vegetation
Index, soil lightness index and soil adjust vegetation index;Textural characteristics in the feature of first quantity include:Gray scale symbiosis square
Variance, correlation, non-similarity, homogeney, contrast, second moment and the entropy of battle array;Features of terrain in the feature of first quantity
Including:Elevation and the gradient.
Specifically, as a preferred embodiment, the first quantity is characterized as 16 features.Above-mentioned 16 features include 7
A spectral signature, 7 textural characteristics and 2 featuress of terrain.
7 spectral signatures include:Blue wave band spectral information (B), green light band spectral information (G), red spectral band spectrum
Information (R) and near infrared band spectral information (NIR), normalized differential vegetation index (Normalized Difference
Vegetation Index, abbreviation NDVI), soil lightness index (SBI) and soil adjust vegetation index (MSAVI).
Wherein, the vegetative coverage information in NDVI energy repercussion studies area;Bare farmland, concrete floors of the SBI to no vegetative coverage
The extraction effect of the artificial surfaces such as face, bare rock and building is preferable;MSAVI, which is soil adjusting vegetation index, can react vegetation
Non-fully cover the soil and vegetation information of lower earth's surface.
The calculation formula of NDVI, SBI and MSAVI are as follows:
Wherein, B indicates that blue wave band spectral information, G indicate that green light band spectral information, R indicate red spectral band spectrum letter
Breath, NIR indicate near infrared band spectral information.
7 textural characteristics include:Variance (Variance), correlation (Correlation), the non-phase of gray level co-occurrence matrixes
Like property (Dissimilarity), homogeney (Homogeneity), contrast (Contrast), second moment (Second
) and entropy (Entropy) Moment.
The calculation formula of above-mentioned 7 textural characteristics is as follows:
Wherein, i and j is respectively any point (x, y) and to deviate its another point (x+ △ x, y+ △ y) in image (M × N)
Gray value, k be corresponding gray level, p (i, j) be gray level co-occurrence matrixes p (i, j) point value, Mean is gray average.
If pixel indicates the area on 1 meter × 1 meter of ground, when establishing gray level co-occurrence matrixes, then centered on the pixel,
One length of side of extension is 3 times of window of the pixel, i.e., the gray scale symbiosis square of the pixel is obtained with 3 meters × 3 meters of window size
Battle array.
2 featuress of terrain include elevation and the gradient.
Based on above-described embodiment, as an alternative embodiment, step S06 further includes:When classification accuracy rate is less than accurately
It when rate threshold value, obtains training sample and concentrates several new features of each training sample, by several new features and current the
The feature of one quantity collectively forms the feature of the first new quantity, and executes step S02, until classification accuracy rate is more than accuracy rate
Threshold value.
Specifically, in step S06, after obtaining random forest grader to be verified, the verification sample obtained in advance is utilized
Collection, verifies the classification accuracy rate of random forest grader to be verified.
When classification accuracy rate is less than accuracy rate threshold value, illustrate the classification accuracy rate of random forest grader to be verified not
It meets the requirements, needs to regenerate random forest grader to be verified using training sample.
As an alternative embodiment, training sample can be extracted and concentrate several new features of each training sample, it will
The feature of several new features and current first quantity collectively forms the feature of the first new quantity.Training sample is obtained to concentrate
After the feature of the first new quantity of each training sample, step S02 is executed, decision tree and random forest to be verified are regenerated
Grader.
It has been randomly selected with putting back to from training sample concentration when regenerating decision tree, before being based on preset
The training sample of first ratio updates each decision tree;It can also be randomly selected with putting back to from training sample concentration again
The training sample of preset first ratio generates decision tree.
The second quantity can also be changed when classification accuracy rate is less than accuracy rate threshold value as an alternative embodiment, weight
Newly-generated decision tree and random forest grader to be verified, until classification accuracy rate is more than accuracy rate threshold value.
For example, when it is 10 that the first quantity, which is the 16, second quantity, if the random forest grader classification to be verified generated
Accuracy is more than accuracy rate threshold value, can the second quantity be changed to 12 or 8, regenerate decision tree and random forest to be verified
Grader.
The embodiment of the present invention is by when the classification accuracy of random forest grader to be verified does not reach requirement, changing
Second quantity extracts new feature, regenerates random forest grader to be verified, until classification accuracy rate is more than accurately
Rate threshold value can obtain the higher random forest grader of classification accuracy rate, to improve the accuracy rate of fruit-bearing forest identification.
Based on above-described embodiment, as an alternative embodiment, new feature is including at least new spectral signature and/or newly
Textural characteristics;New spectral signature includes at least:Enhancement mode meta file, ratio vegetation index, green normalization difference are planted
By one or more of index, triangle vegetation index, difference vegetation index and normalization difference aqua index;New textural characteristics
It includes at least:The unfavourable balance of gray level co-occurrence matrixes divide square, intermediate value, covariance, contrast, symbiosis and mean value, symbiosis and variance, symbiosis and
One or more of entropy, symbiosis difference mean value, symbiosis difference variance and symbiosis difference entropy.
Specifically, as an alternative embodiment, new feature includes at least new spectral signature and/or new texture is special
Sign.
New spectral signature includes at least:Enhancement mode meta file (Enhacn Vegetable Index, abbreviation EVI),
Ratio vegetation index (Ratio Vegetable Index, abbreviation RVI), green normalized site attenuation (Green
Normalized Difference Vegetable Index, guide number DVI), triangle vegetation index (Triangle
Vegetable Index, abbreviation TVI), difference vegetation index (Difference Vegetable Index, abbreviation DVI)) and
One or more of difference aqua index (Normalized Difference Water Index, abbreviation NDWI) is normalized, but
It is without being limited thereto, other vegetation indexs can also be extracted as new spectral signature.
The calculation formula of above-mentioned vegetation index is as follows:
TVI=60 × (NIR-G) -100 × (R-G)
DVI=NIR-R
Wherein, B indicates that blue wave band spectral information, G indicate that green light band spectral information, R indicate red spectral band spectrum letter
Breath, NIR indicate near infrared band spectral information.
New textural characteristics include at least:The unfavourable balance of gray level co-occurrence matrixes divide square, intermediate value, covariance, contrast, symbiosis and
One or more of value, symbiosis and variance, symbiosis and entropy, symbiosis difference mean value, symbiosis difference variance and symbiosis difference entropy, but be not limited to
This, can also extract gray level co-occurrence matrixes others statistic as new textural characteristics.
Specifically, as a preferred embodiment, further include between step S05 and step S06:
Using the training sample not being extracted when generating every decision tree, current random forest grader to be verified is obtained
The outer data error of bag;According to data error outside bag, the importance of each feature in the feature of current first quantity is obtained;According to
The importance of each feature in the feature of current first quantity retains several features in the feature of current first quantity, makees
For the feature of the first new quantity, and step S02 is executed, until generating new random forest grader to be verified.
Specifically, as a preferred embodiment, after generating random forest grader to be verified, current the is obtained
The importance of each feature in the feature of one quantity screens the feature of current first quantity according to the importance of feature.
For every decision tree in random forest, using data (OOB) outside corresponding bag, that is, when generating the decision tree not
The training sample being extracted calculates the outer data error of bag of the decision tree.
It randomly gives each feature in the feature of current first quantity that noise is added, i.e., changes current first quantity at random
The value of each feature in feature calculates the outer data error of the new bag of the decision tree.
According to data error outside the bag of the decision tree and it is random data error outside the new bag obtained after noise is added, can be with
Obtain the importance of each feature in the feature of current first quantity.
Importance W (the x of featurej) calculation formula it is as follows:
In formula, etData error outside the bag for every decision tree being calculated according to data outside bag;To change outside bag at random
J-th of feature x of datajValue after the outer data error of the new bag that is calculated, N is the number of decision tree in random forest.
If after random noise is added to certain feature, the outer data accuracy of bag significantly declines (the outer data mistake of i.e. new bag
Difference rises), illustrate that this feature has a significant impact for the prediction result of sample, further relates to this feature significance level and compare
It is high.
Importance W (x in the feature of current first quantity can be retainedj) maximum several features, by the several of reservation
Feature of a feature as the first new quantity, executes step S02, regenerates decision tree and random forest to be verified classification
Device.After generating random forest grader to be verified again, no longer feature is screened.
The embodiment of the present invention can reduce the spy extracted when the fruit-bearing forest in identification object region by data screening feature outside bag
The calculation amount for system of seeking peace improves the speed of identification fruit-bearing forest.
Below by taking mango woods as an example, illustrate fruit-bearing forest recognition methods provided by the invention.
No. two satellite images of high score and dem data in main mango seed of forest growing area on October 17th, 2015 are obtained first,
And carry out Yunnan snub-nosed monkey.
Using 3 meters × 3 meters of size as sub-goal region, extract each sub-goal region spectral signature (including B, G,
R, 7 characteristic informations such as NIR, NDVI, SBI and MSAVI), textural characteristics it is (including the variance of gray level co-occurrence matrixes, correlation, non-
7 textural characteristics such as similitude, homogeney, contrast, second moment and entropy) and features of terrain information (including elevation, gradient etc.
2 featuress of terrain).
By whole sub-goal regions point with 4:1 ratio is divided into training sample and verification sample.
Training sample is trained using random forest sorting algorithm, is classified, random forest classification to be verified is generated
Device.
The classification results of random forest grader to be verified are verified using verification sample.Verification result shows
The cartographic accuracy of fruit-bearing forest recognition methods identification mango woods provided by the invention reaches 89.29%, and user's precision reaches 97.40%,
Identify that the cartographic accuracy of mango woods improves 17.04% compared with just with spectral signature, user's precision improves 7.08%.This hair
The accuracy rate that the fruit-bearing forest recognition methods of bright offer carries out Classification and Identification fruit-bearing forest using random forest grader is also significantly better than use
Support vector machines carries out the accuracy rate of sub-category fruit-bearing forest.
By random forest grader to be verified for identification the mango woods in the Sanya precipice of Hainan Province in 2015 city when, will
Recognition result is compared with current year statistical yearbook data, and the rate of accuracy reached of this method identification is to 91.7%.
Fig. 2 is a kind of functional block diagram of fruit-bearing forest identifying system of the embodiment of the present invention.Based on above-described embodiment, as shown in Fig. 2,
A kind of fruit-bearing forest identifying system includes:Remote sensing module 201, high-resolution remote sensing image and digital elevation for obtaining target area
Model;Characteristic extracting module 202, for target area to be divided into several equal-sized sub-goal regions, according to target
The high-resolution remote sensing image and digital elevation model in region obtain the target signature in each sub-goal region;Target signature includes
Several spectral signatures, several textural characteristics and several featuress of terrain;Identification module 203, for respectively by each specific item
The target signature in mark region inputs the random forest grader built in advance, judges whether each sub-goal region is fruit-bearing forest,
According to each sub-goal region whether be fruit-bearing forest judging result, the fruit-bearing forest in identification object region.
Specifically, remote sensing module 201 is electrically connected with characteristic extracting module 202, characteristic extracting module 202 and identification module
203。
After remote sensing module 201 obtains the high-resolution remote sensing image and digital elevation model of target area, by target area
High-resolution remote sensing image and digital elevation model be sent to characteristic extracting module 202.
Characteristic extracting module 202 obtains each according to the high-resolution remote sensing image and digital elevation model of target area
Several spectral signatures, several textural characteristics and several featuress of terrain in sub-goal region, as each sub-goal region
Target signature, and the target signature in each sub-goal region is sent to identification module 203.
Identification module 203 respectively classifies the random forest that the input of the target signature in each sub-goal region is built in advance
Device judges whether each sub-goal region is fruit-bearing forest, according to each sub-goal region whether be fruit-bearing forest judging result, identify mesh
Mark the fruit-bearing forest in region.
Fruit-bearing forest identifying system provided by the invention is for executing fruit-bearing forest recognition methods provided by the invention.Fruit-bearing forest identifying system
Including each module realize that the specific method of corresponding function and flow refer to the embodiment of above-mentioned fruit-bearing forest recognition methods, herein no longer
It repeats.
The embodiment of the present invention is used as the feature in sub-goal region by extracting spectral signature, textural characteristics and features of terrain,
And classified using random forest grader, the fruit-bearing forest in identification object region preferably overcomes " the different spectrum of jljl ", " same
The influence of spectrum foreign matter " phenomenon, improves the accuracy rate of fruit-bearing forest identification, can meet the demands such as fruit-bearing forest normality monitoring, especially in China
When southern plot is broken, type of ground objects is various, regions with complex terrain (such as Hainan Region) carries out fruit-bearing forest identification, there is higher fruit
The accuracy rate of woods identification.
Based on above-described embodiment, the present embodiment discloses a kind of computer program product, and computer program product includes storage
Computer program in non-transient computer readable storage medium, computer program include program instruction, when program instruction quilt
When computer executes, computer is able to carry out the method that above-mentioned each method embodiment is provided, such as including:Fruit-bearing forest identification side
Method, the extracting method of target signature, the method etc. for obtaining the random forest grader built.
Based on above-described embodiment, the present embodiment provides a kind of non-transient computer readable storage medium, non-transient computers
Readable storage medium storing program for executing stores computer instruction, and computer instruction makes computer execute the side that above-mentioned each method embodiment is provided
Method, such as including:The method for the random forest grader that fruit-bearing forest recognition methods, the extracting method of target signature, acquisition are built
Deng.
One of ordinary skill in the art will appreciate that:Realize that all or part of step of above method embodiment can pass through
The relevant hardware of program instruction is completed, and program above-mentioned can be stored in a computer read/write memory medium, the program
When being executed, step including the steps of the foregoing method embodiments is executed;And storage medium above-mentioned includes:ROM, RAM, magnetic disc or light
The various media that can store program code such as disk.
The embodiments such as fruit-bearing forest identifying system described above are only schematical, wherein illustrate as separating component
Unit may or may not be physically separated, and the component as unit may or may not be physics list
Member, you can be located at a place, or may be distributed over multiple network units.It can be selected according to the actual needs
In some or all of module achieve the purpose of the solution of this embodiment.Those of ordinary skill in the art are not paying creativeness
Labour in the case of, you can to understand and implement.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
It is realized by the mode of software plus required general hardware platform, naturally it is also possible to pass through hardware.Based on this understanding, on
Stating technical solution, substantially the part that contributes to existing technology can be expressed in the form of software products in other words, should
Computer software product can store in a computer-readable storage medium, such as ROM/RAM, magnetic disc, CD, including several fingers
It enables and using so that a computer equipment (can be personal computer, server or the network equipment etc.) executes each implementation
The method of certain parts of example or embodiment.
Finally, the above embodiment of the present invention is only preferable embodiment, is not intended to limit the protection model of the present invention
It encloses.All within the spirits and principles of the present invention, any modification, equivalent replacement, improvement and so on should be included in the present invention
Protection domain within.
Claims (10)
1. a kind of fruit-bearing forest recognition methods, which is characterized in that including:
S1, the high-resolution remote sensing image and digital elevation model for obtaining target area;
S2, the target area is divided into several equal-sized sub-goal regions, according to the high score of the target area
Resolution remote sensing images and digital elevation model obtain the target signature in each sub-goal region;If the target signature includes
Dry spectral signature, several textural characteristics and several featuress of terrain;
S3, the target signature in each sub-goal region is inputted to the random forest grader built in advance respectively, judged
Whether each sub-goal region is fruit-bearing forest, according to each sub-goal region whether be fruit-bearing forest judging result, identification
Fruit-bearing forest in the target area.
2. fruit-bearing forest recognition methods according to claim 1, which is characterized in that obtain the random forest grader built
Specific steps include:
The training sample that S01, extraction obtain in advance concentrates the feature of the first quantity of each training sample;First quantity
Feature includes several spectral signatures, several textural characteristics and several featuress of terrain;
S02, there is randomly select preset first ratio training sample with putting back to from training sample concentration;
S03, the feature that the second quantity is randomly selected from the feature of first quantity use extraction according to splitting rule
The feature of second quantity of training sample obtains the split point of decision tree, and decision tree is generated according to the split point;It is described
Second quantity is less than first quantity;
S04, step S02 and S03 are repeated, until the number extracted reaches preset times;
S05, the whole decision tree according to generation, generate random forest grader to be verified;
S06, using the verification sample set obtained in advance, obtain the classification accuracy of the random forest grader to be verified,
When the classification accuracy is more than accuracy rate threshold value, the random forest grader to be verified is built as described in
Random forest grader, using the feature of current first quantity as the target signature.
3. fruit-bearing forest recognition methods according to claim 2, which is characterized in that obtain the training sample set and described in advance
Verification sample set specific steps include:
Obtain the high-resolution remote sensing image and digital elevation model of several sample areas;
Each sample areas is divided into several equal-sized increment one's respective areas, the size of the increment one's respective area with
The target area it is equal in magnitude;
To sample set all be formed as sample in the increment one's respective area;
The sample of preset second ratio is extracted from the sample set as training sample, forms the training sample set;It will
The sample not being extracted in the sample set forms the verification sample set as verification sample.
4. fruit-bearing forest recognition methods according to claim 2, which is characterized in that the Spectral Properties in the feature of first quantity
Sign includes:
Blue wave band spectral information, green light band spectral information, red spectral band spectral information and near infrared band spectral information are returned
One, which changes vegetation index, soil lightness index and soil, adjusts vegetation index;
Textural characteristics in the feature of first quantity include:It is the variances of gray level co-occurrence matrixes, correlation, non-similarity, same
Matter, contrast, second moment and entropy;
Features of terrain in the feature of first quantity includes:Elevation and the gradient.
5. fruit-bearing forest recognition methods according to claim 4, which is characterized in that the step S06 further includes:
When the classification accuracy is less than accuracy rate threshold value, obtaining each training sample of the training sample concentration, several are new
Feature, the feature of several described new features and current first quantity is collectively formed to the feature of the first new quantity, and
Step S02 is executed, until the classification accuracy rate is more than accuracy rate threshold value.
6. fruit-bearing forest recognition methods according to claim 5, which is characterized in that the new feature includes at least new spectrum
Feature and/or new textural characteristics;
The new spectral signature includes at least:Enhancement mode meta file, ratio vegetation index, green normalization difference vegetation refer to
Number, triangle vegetation index, difference vegetation index and normalization one or more of difference aqua index;
The new textural characteristics include at least:The unfavourable balance of gray level co-occurrence matrixes divide square, intermediate value, covariance, contrast, symbiosis and
One or more of value, symbiosis and variance, symbiosis and entropy, symbiosis difference mean value, symbiosis difference variance and symbiosis difference entropy.
7. according to any fruit-bearing forest recognition methods of claim 2 to 6, which is characterized in that the step S05 and step S06
Between further include:
Using the training sample not being extracted when generating every decision tree, the bag of current random forest grader to be verified is obtained
Outer data error;
According to data error outside the bag, the importance of each feature in the feature of current first quantity is obtained;
According to the importance of each feature in the feature of current first quantity, retain several in the feature of current first quantity
Feature as the feature of the first new quantity, and executes step S02, until generating new random forest grader to be verified.
8. fruit-bearing forest recognition methods according to claim 1, which is characterized in that the high-resolution for obtaining target area is distant
Sense image specific steps include:
The clear cloudless high-resolution satellite image for obtaining the target area specific period, to the high-resolution satellite shadow
As being pre-processed, the high-resolution remote sensing image of the target area is obtained;
It includes radiation calibration, atmospheric correction, ortho-rectification, geometry to carry out pretreated method to the high-resolution satellite image
Fine correction and projective transformation.
9. the fruit-bearing forest recognition methods according to claim 1 or 8, which is characterized in that the height according to the target area
The specific steps that resolution remote sensing images and digital elevation model obtain the target signature in each sub-goal region include:
For each sub-goal region, which is obtained according to the high-resolution remote sensing image of the target area
Several spectral signatures and several textural characteristics, which is obtained according to the digital elevation model of the target area
Several featuress of terrain in domain.
10. a kind of fruit-bearing forest identifying system, which is characterized in that including:
Remote sensing module, high-resolution remote sensing image and digital elevation model for obtaining target area;
Characteristic extracting module, for the target area to be divided into several equal-sized sub-goal regions, according to described
The high-resolution remote sensing image and digital elevation model of target area obtain the target signature in each sub-goal region;It is described
Target signature includes several spectral signatures, several textural characteristics and several featuress of terrain;
Identification module, the random forest point for respectively building the input of the target signature in each sub-goal region in advance
Class device judges whether each sub-goal region is fruit-bearing forest, according to each sub-goal region whether be fruit-bearing forest judgement
As a result, identifying the fruit-bearing forest in the target area.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810133935.5A CN108280440A (en) | 2018-02-09 | 2018-02-09 | A kind of fruit-bearing forest recognition methods and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810133935.5A CN108280440A (en) | 2018-02-09 | 2018-02-09 | A kind of fruit-bearing forest recognition methods and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108280440A true CN108280440A (en) | 2018-07-13 |
Family
ID=62808178
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810133935.5A Pending CN108280440A (en) | 2018-02-09 | 2018-02-09 | A kind of fruit-bearing forest recognition methods and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108280440A (en) |
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109241970A (en) * | 2018-09-28 | 2019-01-18 | 深圳市飞点健康管理有限公司 | Urine examination method, mobile terminal and computer readable storage medium |
CN109272001A (en) * | 2018-09-28 | 2019-01-25 | 深圳市飞点健康管理有限公司 | Construction training method, device and the computer equipment of urine examination recognition classifier |
CN109344550A (en) * | 2018-11-26 | 2019-02-15 | 国智恒北斗科技集团股份有限公司 | A kind of forest reserves inversion method and system based on domestic high score satellite remote sensing date |
CN109447028A (en) * | 2018-11-12 | 2019-03-08 | 南京大学 | A kind of farmland ditch network extraction method |
CN109800815A (en) * | 2019-01-24 | 2019-05-24 | 北华航天工业学院 | Training method, wheat recognition methods and training system based on Random Forest model |
WO2020063461A1 (en) * | 2018-09-30 | 2020-04-02 | 广州地理研究所 | Urban extent extraction method and apparatus based on random forest classification algorithm, and electronic device |
CN111079672A (en) * | 2019-12-20 | 2020-04-28 | 中国科学院遥感与数字地球研究所 | Grape classification extraction method based on maximum entropy features |
CN111178186A (en) * | 2019-12-16 | 2020-05-19 | 广州地理研究所 | Rice extraction method, device and equipment based on sentinel remote sensing data |
CN111582035A (en) * | 2020-04-07 | 2020-08-25 | 北京农业信息技术研究中心 | Method, device and equipment for identifying tree age of fruit tree and storage medium |
CN111611921A (en) * | 2020-05-21 | 2020-09-01 | 佛山市高明曦逻科技有限公司 | Solar panel identification system based on remote sensing big data |
CN111860688A (en) * | 2020-07-31 | 2020-10-30 | 北京无线电测量研究所 | Target type identification method and system and electronic equipment |
CN112232171A (en) * | 2020-10-10 | 2021-01-15 | 中煤航测遥感集团有限公司 | Remote sensing image information extraction method and device based on random forest and storage medium |
CN112329829A (en) * | 2020-10-27 | 2021-02-05 | 宁波大学 | Hyperspectral data-based mangrove forest extraction method |
CN112513874A (en) * | 2019-12-06 | 2021-03-16 | 深圳市大疆创新科技有限公司 | Crop growth condition evaluation method and device |
CN113378785A (en) * | 2021-07-05 | 2021-09-10 | 中煤航测遥感集团有限公司 | Forest type identification method and device |
CN113640244A (en) * | 2021-07-28 | 2021-11-12 | 湖南师范大学 | Fruit tree variety identification method based on visible near infrared spectrum |
CN114397250A (en) * | 2021-12-27 | 2022-04-26 | 中国林业科学研究院木材工业研究所 | Wood identification method based on spectral features and image features |
CN114694036A (en) * | 2022-03-18 | 2022-07-01 | 南京农业大学 | High-altitude area crop classification and identification method based on high-resolution images and machine learning |
CN114882359A (en) * | 2022-05-07 | 2022-08-09 | 中国科学院空天信息创新研究院 | Soybean planting area extraction method and system based on vegetation index time series spectrum characteristics |
CN116229280A (en) * | 2023-01-09 | 2023-06-06 | 广东省科学院广州地理研究所 | Method and device for identifying collapse sentry, electronic equipment and storage medium |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101359371A (en) * | 2008-07-30 | 2009-02-04 | 上海同盛工程建设配套管理有限公司 | Buildings auto extracting method based on digital elevation model |
CN104992455A (en) * | 2015-07-29 | 2015-10-21 | 中国神华能源股份有限公司 | Collapse pit detecting method and apparatus |
-
2018
- 2018-02-09 CN CN201810133935.5A patent/CN108280440A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101359371A (en) * | 2008-07-30 | 2009-02-04 | 上海同盛工程建设配套管理有限公司 | Buildings auto extracting method based on digital elevation model |
CN104992455A (en) * | 2015-07-29 | 2015-10-21 | 中国神华能源股份有限公司 | Collapse pit detecting method and apparatus |
Non-Patent Citations (6)
Title |
---|
任传帅等: "利用高分二号数据提取香蕉林信息及精度分析", 《遥感信息》 * |
任传帅等: "基于面向对象分类的芒果林遥感提取方法研究", 《资源科学》 * |
姚明煌: "随机森林及其在遥感图像分类中的应用", 《中国优秀硕士论文全文数据库》 * |
庞小平: "《遥感制图与应用》", 30 June 2016, 测绘出版社 * |
张过等: "《高分辨率SAR卫星标准产品分级体系研究》", 30 November 2012, 测绘出版社 * |
敖祖锐等: "《星载SAR影像模拟与正射校正》", 31 January 2017, 中国环境科学出版社 * |
Cited By (31)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109241970B (en) * | 2018-09-28 | 2021-07-30 | 深圳市飞点健康管理有限公司 | Urine test method, mobile terminal and computer readable storage medium |
CN109272001A (en) * | 2018-09-28 | 2019-01-25 | 深圳市飞点健康管理有限公司 | Construction training method, device and the computer equipment of urine examination recognition classifier |
CN109241970A (en) * | 2018-09-28 | 2019-01-18 | 深圳市飞点健康管理有限公司 | Urine examination method, mobile terminal and computer readable storage medium |
CN109272001B (en) * | 2018-09-28 | 2021-09-03 | 深圳市飞点健康管理有限公司 | Structure training method and device of urine test recognition classifier and computer equipment |
WO2020063461A1 (en) * | 2018-09-30 | 2020-04-02 | 广州地理研究所 | Urban extent extraction method and apparatus based on random forest classification algorithm, and electronic device |
CN109447028A (en) * | 2018-11-12 | 2019-03-08 | 南京大学 | A kind of farmland ditch network extraction method |
CN109344550A (en) * | 2018-11-26 | 2019-02-15 | 国智恒北斗科技集团股份有限公司 | A kind of forest reserves inversion method and system based on domestic high score satellite remote sensing date |
CN109800815B (en) * | 2019-01-24 | 2023-11-24 | 北华航天工业学院 | Training method, wheat recognition method and training system based on random forest model |
CN109800815A (en) * | 2019-01-24 | 2019-05-24 | 北华航天工业学院 | Training method, wheat recognition methods and training system based on Random Forest model |
CN112513874A (en) * | 2019-12-06 | 2021-03-16 | 深圳市大疆创新科技有限公司 | Crop growth condition evaluation method and device |
CN111178186A (en) * | 2019-12-16 | 2020-05-19 | 广州地理研究所 | Rice extraction method, device and equipment based on sentinel remote sensing data |
CN111079672B (en) * | 2019-12-20 | 2023-10-13 | 中国科学院空天信息创新研究院 | Grape classification extraction method based on maximum entropy characteristics |
CN111079672A (en) * | 2019-12-20 | 2020-04-28 | 中国科学院遥感与数字地球研究所 | Grape classification extraction method based on maximum entropy features |
CN111582035A (en) * | 2020-04-07 | 2020-08-25 | 北京农业信息技术研究中心 | Method, device and equipment for identifying tree age of fruit tree and storage medium |
CN111582035B (en) * | 2020-04-07 | 2023-09-12 | 北京农业信息技术研究中心 | Fruit tree age identification method, device, equipment and storage medium |
CN111611921B (en) * | 2020-05-21 | 2021-05-14 | 苏州科知律信息科技有限公司 | Solar panel identification system based on remote sensing big data |
CN111611921A (en) * | 2020-05-21 | 2020-09-01 | 佛山市高明曦逻科技有限公司 | Solar panel identification system based on remote sensing big data |
CN111860688B (en) * | 2020-07-31 | 2023-10-31 | 北京无线电测量研究所 | Target type identification method and system and electronic equipment |
CN111860688A (en) * | 2020-07-31 | 2020-10-30 | 北京无线电测量研究所 | Target type identification method and system and electronic equipment |
CN112232171A (en) * | 2020-10-10 | 2021-01-15 | 中煤航测遥感集团有限公司 | Remote sensing image information extraction method and device based on random forest and storage medium |
CN112232171B (en) * | 2020-10-10 | 2024-08-09 | 中煤航测遥感集团有限公司 | Remote sensing image information extraction method and device based on random forest and storage medium |
CN112329829A (en) * | 2020-10-27 | 2021-02-05 | 宁波大学 | Hyperspectral data-based mangrove forest extraction method |
CN112329829B (en) * | 2020-10-27 | 2021-08-27 | 宁波大学 | Hyperspectral data-based mangrove forest extraction method |
CN113378785A (en) * | 2021-07-05 | 2021-09-10 | 中煤航测遥感集团有限公司 | Forest type identification method and device |
CN113640244A (en) * | 2021-07-28 | 2021-11-12 | 湖南师范大学 | Fruit tree variety identification method based on visible near infrared spectrum |
CN114397250B (en) * | 2021-12-27 | 2024-02-13 | 中国林业科学研究院木材工业研究所 | Wood identification method based on spectral features and image features |
CN114397250A (en) * | 2021-12-27 | 2022-04-26 | 中国林业科学研究院木材工业研究所 | Wood identification method based on spectral features and image features |
CN114694036A (en) * | 2022-03-18 | 2022-07-01 | 南京农业大学 | High-altitude area crop classification and identification method based on high-resolution images and machine learning |
CN114882359A (en) * | 2022-05-07 | 2022-08-09 | 中国科学院空天信息创新研究院 | Soybean planting area extraction method and system based on vegetation index time series spectrum characteristics |
CN116229280A (en) * | 2023-01-09 | 2023-06-06 | 广东省科学院广州地理研究所 | Method and device for identifying collapse sentry, electronic equipment and storage medium |
CN116229280B (en) * | 2023-01-09 | 2024-06-04 | 广东省科学院广州地理研究所 | Method and device for identifying collapse sentry, electronic equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108280440A (en) | A kind of fruit-bearing forest recognition methods and system | |
Lopatin et al. | How canopy shadow affects invasive plant species classification in high spatial resolution remote sensing | |
CN110363215B (en) | Method for converting SAR image into optical image based on generating type countermeasure network | |
Halme et al. | Utility of hyperspectral compared to multispectral remote sensing data in estimating forest biomass and structure variables in Finnish boreal forest | |
Laurent et al. | Bayesian object-based estimation of LAI and chlorophyll from a simulated Sentinel-2 top-of-atmosphere radiance image | |
Saranathan et al. | Uniformity-based superpixel segmentation of hyperspectral images | |
CN104484681B (en) | Hyperspectral Remote Sensing Imagery Classification method based on spatial information and integrated study | |
CN109146889A (en) | A kind of field boundary extracting method based on high-resolution remote sensing image | |
David et al. | Vegetation indices and textures in object-based weed detection from UAV imagery | |
Marconi et al. | Continental-scale hyperspectral tree species classification in the United States National Ecological Observatory Network | |
Tang et al. | A multiple-point spatially weighted k-NN method for object-based classification | |
CN108932521A (en) | A kind of Crop classification method and system based on deep learning | |
Iordache et al. | A dynamic unmixing framework for plant production system monitoring | |
CN111783555A (en) | Deep learning satellite data cloud detection algorithm using unified samples | |
CN110443139A (en) | A kind of target in hyperspectral remotely sensed image noise wave band detection method of Classification Oriented | |
Jin et al. | Combining 3D radiative transfer model and convolutional neural network to accurately estimate Forest canopy cover from very high-resolution satellite images | |
Tao et al. | Deep Learning for Hyperspectral Image Analysis and Classification | |
Asming et al. | Processing and classification of landsat and sentinel images for oil palm plantation detection | |
Mianji et al. | Enhanced self-training superresolution mapping technique for hyperspectral imagery | |
CN117115669B (en) | Object-level ground object sample self-adaptive generation method and system with double-condition quality constraint | |
Tomková et al. | Semantic classification of sandstone landscape point cloud based on neighbourhood features | |
RU2693880C1 (en) | Method of plant objects identification based on remote sensing satellite images | |
Soto Vega et al. | Deforestation detection with weak supervised convolutional neural networks in tropical biomes | |
Bochow et al. | Automatic shadow detection in hyperspectral VIS-NIR images | |
Radecka et al. | Mapping secondary succession species in agricultural landscape with the use of hyperspectral and airborne laser scanning data |
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: 20180713 |