CN107368813A - A kind of forest hat width recognition methods applied to airborne field hyperspectrum image - Google Patents
A kind of forest hat width recognition methods applied to airborne field hyperspectrum image Download PDFInfo
- Publication number
- CN107368813A CN107368813A CN201710603764.3A CN201710603764A CN107368813A CN 107368813 A CN107368813 A CN 107368813A CN 201710603764 A CN201710603764 A CN 201710603764A CN 107368813 A CN107368813 A CN 107368813A
- Authority
- CN
- China
- Prior art keywords
- classification
- hat width
- forest
- probability
- image
- 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
- G06V20/188—Vegetation
-
- 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
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
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)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The present invention discloses a kind of forest hat width recognition methods applied to airborne field hyperspectrum image, belongs to forestry remote sensing data processing and information extraction category.Its technical characterstic is based on support vector cassification of the tradition based on spectrum, complete support vector cassification, formed on the basis of all kinds of atural object probability figures, introduce guiding filtering, texture is carried out to classification probability figure using guiding filtering and marginal information optimizes, pass through maximum probability criterion, each pixel generic after optimization is divided, after most total classification is completed, extraction trees correspond to classification, forest edge vectors are formed, complete the accurate identification of forest hat width.
Description
First, technical field
The present invention relates to a kind of forest hat width recognition methods based on empty spectrum classification, particularly one kind is applied to airborne near-earth
The forest hat width recognition methods of airborne-remote sensing, suitable for the standing forest that forest Crown structure is complicated, belong to forestry EO-1 hyperion
Remote Sensing Data Processing technical field.
2nd, technical background
Airborne near-earth Imaging Hyperspectral Data is that one kind utilizes UAV flight's EO-1 hyperion camera to obtain atural object EO-1 hyperion
Image is remote sensing technology means.Airborne near-earth Imaging Hyperspectral Data has had both high spatial resolution, high spectral resolution simultaneously
Characteristic, and data obtaining time is flexible, can obtain high time resolution data in time as desired, is current remotely-sensed data
Processing and the study hotspot of application field.In addition, forest hat width information is obtained also with remote sensing technology by remote sensing technology means
Continuous development and turn into the focus on research direction of forestry industry instantly.With reference to the feature of near-earth Imaging Hyperspectral Data, utilize
The thought of sky spectrum classification carries out image classification segmentation, completes standing forest, Dan Mu information identification, can be based on spectral classification original
Or nicety of grading is increased substantially on the basis of image segmentation, realize the accurate identification of hat width information.
Initial stage is applied in high-spectral data, conventional Hyperspectral data classification method concentrates on k nearest neighbor classification, maximum likelihood
The sorting techniques based on machine learning such as classification, hereafter, according to the own characteristic of airborne-remote sensing, support vector cassification
Progressively risen Deng the sorting technique based on Data Dimensionality Reduction.It is and equal for near-earth Imaging Hyperspectral Data, its spectrum and textural characteristics
Extremely fine, in numerous identification sorting algorithms, the classification of sky spectrum effectively can be believed the information such as space, texture and spectrum because of it
Manner of breathing with reference to and obtain extremely wide application.However, in forestry industry, particularly in single wooden hat width complex shape, standing forest
Region complicated, with a varied topography, by the image of the factors such as data acquisition and data volume, EO-1 hyperion number is imaged using near-earth
It is still less according to the correlation technique and algorithm that enter row information identification.Therefore, the near-earth imaging bloom suitable for forestry industry is explored
Modal data analyzes and processes algorithm, realizes that the standing forest parameter information based on field hyperspectrum data accurately identifies to turn into and currently chooses very much
The research theme of war property.
It is currently applied to the identification of the forest hat width based on the classification institute facing challenges of airborne near-earth Imaging Hyperspectral Data
Mainly：
(1) under forest structure complex situations, it is having for data that existing sorting technique, which can not make full use of field hyperspectrum,
Imitate information, when forest hat width scope is extracted the mistake that easily occurs point and it is excessively coarse phenomena such as.
(2) data volume of airborne field hyperspectrum image is big, and existing sorting algorithm is complicated, the speed of service in assorting process
Slowly, time-consuming, requires high to infrastructure device, small range data processing needs to take a significant amount of time.
3rd, the content of the invention
For existing field hyperspectrum data classification algorithm in forest hat width recognition effect and the deficiency of operational efficiency, this hair
The bright forest hat width recognition methods for proposing a kind of empty spectrum classification of Support Vector Machines Optimized based on airborne field hyperspectrum image, is fitted
For various stand types forest hat width identify, particularly have in terms of the forest hat width identification of Crown structure complexity standing forest compared with
Big advantage.
The present invention solves its technical problem and adopted the technical scheme that：All kinds of atural objects are formed in original support vector cassification
On the basis of probability figure, the initialization probability figure of all kinds of atural objects is entered by the simple guiding filtering of non-iterative, principle
Row edge optimization, realize the accurate forest hat width scope identification of airborne field hyperspectrum image.Methods described is included in detail below
Step：
(1) support vector cassification：Using traditional support vector machine (SVM) sorting technique, training sample and checking are chosen
Sample, the field hyperspectrum image by pretreatment is classified pixel-by-pixel, and by analyzing under SVM different kernel functions and right
The nicety of grading in the case of parameter setting is answered, it is determined that the kernel function and its parameter setting of the optimal svm classifier suitable for studying area,
Complete preliminary classification；
(2) all kinds of atural object probability figures are formed：According to classification results, single band image corresponding to all kinds of atural objects is extracted,
Form the classification probability figure of corresponding atural object；
(3) guiding filtering optimizes：With field hyperspectrum data carry out principal component analysis (PCA), after being analyzed using PCA before
Three principal components form pseudocolour picture and scheme (or the CCD images synchronously to obtain is guiding figures) as guiding, to all kinds of atural objects
Probability figure guides filtering optimization, and default parameters controls gradient to control the windows radius r=1 of local window size
ε=0.1 of change2, with average structure similarity determination effect of optimization, realize the marginal information optimization of all kinds of atural objects；
(4) forest hat width identifies：According to maximum probability criterion, according to the probability of each pixel after optimization, final point is formed
Class result, in classification results, " trees " classification is extracted, forms forest hat width edge vectors result, completes the essence of forest hat width
Really identification.
Beneficial effects of the present invention are as follows：
(1) present invention proposes a kind of forest hat width recognition methods for airborne field hyperspectrum image data, is applied to
The hat width identification of the complicated standing forest region of forest Crown structure.
(2) present invention proposes the empty spectrum taxonomy model of Support Vector Machines Optimized, utilizes the texture and edge of guiding filtering
Holding capacity, optimize the support vector cassification result based on spectral classification, form empty spectrum taxonomy model.Algorithm realizes speed
It hurry up, forest hat width accuracy of identification improves.
4th, illustrate
The present invention is further described with reference to the accompanying drawings and examples.
Fig. 1 is flow chart of the method for the present invention；
Fig. 2 is the field hyperspectrum striograph of experimental data 1.；
Fig. 3 is the high definition CCD striographs of experimental data 1.；
Fig. 4 is the field hyperspectrum striograph of experimental data 2.；
Fig. 5 is the high definition CCD striographs of experimental data 2.；
Fig. 6 is the probability figure of experimental data 1. all kinds of atural objects；
Fig. 7 is the probability figure of experimental data 2. all kinds of atural objects；
Fig. 8 is the filter effect figure of experimental data 1. all kinds of atural objects；
Fig. 9 is the filter effect figure of experimental data 2. all kinds of atural objects；
Effect of optimization detail view under different parameters are set when Figure 10 is using high definition CCD images as guiding figure；
Figure 11 is experimental data 1. forest hat width recognition effect figure；
Figure 12 is experimental data 2. forest hat width recognition effect figure.
5th, embodiment：
For the technical characterstic for illustrating this programme can be understood, below by embodiment, and its accompanying drawing is combined, to this hair
It is bright to be described in detail.The present invention：A kind of forest hat width recognition methods applied to airborne field hyperspectrum data, methods described
Specific implementation step is as follows：
(1) support vector cassification：Using traditional support vector machine (SVM) sorting technique, training sample and checking are chosen
Sample, the field hyperspectrum image by pretreatment is classified pixel-by-pixel, and by analyzing under SVM different kernel functions and right
The nicety of grading in the case of parameter setting is answered, it is determined that the kernel function and its parameter setting of the optimal svm classifier suitable for studying area,
Complete preliminary classification；
(2) all kinds of atural object probability figures are formed：According to classification results, single band image corresponding to all kinds of atural objects is extracted,
Form the classification probability figure of corresponding atural object；
(3) guiding filtering optimizes：With field hyperspectrum data carry out principal component analysis (PCA), after being analyzed using PCA before
Three principal components form pseudocolour picture and scheme (or the CCD images synchronously to obtain is guiding figures) as guiding, to all kinds of atural objects
Probability figure guides filtering optimization, and default parameters controls gradient to control the windows radius r=1 of local window size
ε=0.1 of change2, with average structure similarity determination effect of optimization, realize the marginal information optimization of all kinds of atural objects；
(4) forest hat width identifies：According to maximum probability criterion, according to the probability of each pixel after optimization, final point is formed
Class result, in classification results, " trees " classification is extracted, forms forest hat width edge vectors result, completes the essence of forest hat width
Really identification.
For checking effectiveness of the invention and universality, using two groups of difference canopy density sample data verified, base
This situation and result are as follows：
(1) experimental data is summarized
The Pinus tabulaeformis forest that area is the different extents of injury is studied, understory species enrich, and major surface features type includes in the range of sample ground
Bare area, understory species (including short vegetation, meadow etc.), shade and Chinese pine be not (right to verify the method for the invention universality
The Chinese pine extent of injury finely divide) four classes.Enter while 30m × 30m samples of two different canopy density are determined in gamut
Row experiment.
The sample of two groups of difference canopy density data mainly include：1. low canopy density dendrolimus tabulaeformis severe endangers Chinese pine sample ground
Airborne field hyperspectrum data (such as Fig. 2) and its high definition CCD image datas (such as Fig. 3) synchronously obtained；2. medium canopy density oil
The high definition CCD image numbers that pine moth moderate endangers the airborne field hyperspectrum data (such as Fig. 4) on Chinese pine sample ground and its synchronously obtained
According to (such as Fig. 5).
Chinese pine and correspondingly endanger Chinese pine Crown structure complexity, select two kinds of different canopy density under the conditions of, include different harm
Tested the sample of degree, can fully verify the validity and universality of the method for the invention.
(2) support vector cassification
Respectively in two pieces of sample ground select 40 training samples and 20 checking samples, with the airborne near-earth on each sample ground into
Data source based on image height spectroscopic data, four kinds of conventional kernel functions and difference are joined in Matlab2015a using Libsvm
Number facilities carry out class test, by comparative analysis overall classification accuracy, Kappa coefficients and Chinese pine nicety of grading, most
Determine eventually under gaussian radial basis function (RBF) kernel function, penalty factor 20, when nuclear parameter is 0.5, comprehensive nicety of grading is most
It is excellent.Classify pixel-by-pixel it is thus determined that carrying out the traditional support vector machine based on spectrum using RBF (20,0.5).
(3) probability figure into
All kinds of atural objects in classification results are corresponded into single band image zooming-out to come out, form the probability figure of all kinds of atural objects.
The probability figure of two groups of experimental datas is as shown in Figure 6 and Figure 7.
(4) guiding filtering optimizes
First three principal component after being analyzed respectively with field hyperspectrum data by PCA forms pseudocolour picture and synchronous acquisition
CCD images for guiding figure as guiding figure, filtering optimization is guided to the probability figure of all kinds of atural objects.
Drawn with average structure similitude (Mean Structure Similarity Index, MSSIM) to evaluate in difference
Lead image, the edge in the case of different parameters setting keeps effect.If navigational figure and filtered defeated is represented with I and Q respectively
Go out image, then：
Wherein, M be image in number of pixels, μIAnd μQThe average of guiding figure and output image is represented respectively,WithFor
The variance of guiding figure and output image, σIQRepresent the covariance of guiding figure and output image.And c1=(K1L)2,c2=(k2L)2,L
For the dynamic range of pixel value, and K1=0.01, K2=0.03.MSSIM evaluation results are as shown in table 1, and two groups of data of the inside are each
Class atural object filter effect figure is as shown in Figure 8 and Figure 9.
The guiding filtering MSSIM evaluation results of table 1
From upper table analysis, when being schemed using high definition CCD images as guiding, guiding filtering parameter be arranged to (r=1, ε=
0.12) in the case of, filtering optimization effect is optimal.Precision evaluation is as shown in table 2 after optimization, and the optimization under different parameters are set is imitated
Fruit detail view is as shown in Figure 10.
The Support Vector Machines Optimized classification results of table 2 are evaluated
(5) forest hat width is extracted
According to maximum probability criterion, according to the probability of each pixel after optimization, final classification result is formed, in classification results
In, " trees " classification is extracted, forms trees edge vectors result, completes the identification of forest hat width.The final hat width of two groups of data is known
Other result is as is illustrated by figs. 11 and 12.
It can be seen that the wooden hat width identification of list that patent methods described of the present invention is carried out for airborne field hyperspectrum image achieves
Good result, it is possible to achieve Different forest stands, the forest hat width identification under different canopy density are applicable.Particularly in complicated hat width knot
In the case of structure, forest hat width extraction marginal information keeps good, and overall accuracy is good.
Simply the preferred embodiment of the present invention described above, for those skilled in the art,
Without departing from the principles of the invention, some improvements and modifications can also be made, these improvements and modifications are also regarded as this hair
Bright protection domain.
Claims (1)
1. a kind of forest hat width recognition methods applied to airborne field hyperspectrum image, it is characterized in that：Keep filtering using edge
Guiding filtering in ripple algorithm optimizes to the classification results of traditional support vector cassification, formed based on optimization support to
The empty spectrum taxonomy model of amount machine, realizes that the airborne field hyperspectrum image forest hat width based on classification accurately identifies.Specific steps
It is as follows：
(1) support vector cassification：Using traditional support vector machine (SVM) sorting technique to the field hyperspectrum by pretreatment
Image is classified pixel-by-pixel, and by analyzing the setting of different kernel functions and corresponding parameter under SVM, it is determined that suitable for image
The kernel function and its parameter setting of optimal svm classifier, complete preliminary classification；
(2) all kinds of atural object probability figures are formed：According to classification results, single band image corresponding to all kinds of atural objects is extracted, is formed
The probability figure of corresponding atural object；
(3) guiding filtering optimizes：Principal component analysis (PCA) is carried out with field hyperspectrum data, first three after being analyzed using PCA
Principal component forms pseudocolour picture and schemes (or the CCD images synchronously to obtain is guiding figures) as guiding, to the initial of all kinds of atural objects
Probability graph guides filtering optimization, and default parameters controls graded to control the windows radius r=1 of local window size
ε=0.12, with average structure similarity determination effect of optimization, realize the marginal information optimization of all kinds of atural objects；
(4) forest hat width identifies：According to maximum probability criterion, according to the probability of each pixel after optimization, final classification knot is formed
Fruit, in classification results, " trees " classification is extracted, forms forest hat width edge vectors result, completes the accurate knowledge of forest hat width
Not.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710603764.3A CN107368813A (en) | 2017-07-23 | 2017-07-23 | A kind of forest hat width recognition methods applied to airborne field hyperspectrum image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710603764.3A CN107368813A (en) | 2017-07-23 | 2017-07-23 | A kind of forest hat width recognition methods applied to airborne field hyperspectrum image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107368813A true CN107368813A (en) | 2017-11-21 |
Family
ID=60307449
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710603764.3A Pending CN107368813A (en) | 2017-07-23 | 2017-07-23 | A kind of forest hat width recognition methods applied to airborne field hyperspectrum image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107368813A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109902686A (en) * | 2019-01-22 | 2019-06-18 | 中国科学院植物研究所 | A kind of single wooden parameter extracting method of forest |
| CN110207670A (en) * | 2019-06-26 | 2019-09-06 | 北京林业大学 | A method of artificial forest forest hat width parameter is obtained using two dimensional image |
| CN111368712A (en) * | 2020-03-02 | 2020-07-03 | 四川九洲电器集团有限责任公司 | Hyperspectral image disguised target detection method based on deep learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104820830A (en) * | 2015-05-08 | 2015-08-05 | 南京林业大学 | Tree species identification method based on full-waveform LiDAR canopy profile model |
| CN105654440A (en) * | 2015-12-30 | 2016-06-08 | 首都师范大学 | Regression model-based fast single-image defogging algorithm and system |
| CN105913017A (en) * | 2016-04-08 | 2016-08-31 | 南京林业大学 | Corresponding period double high resolution remote sensing image-based forest biomass estimation method |
| CN106951877A (en) * | 2017-03-28 | 2017-07-14 | 北京恒华伟业科技股份有限公司 | A kind of Objects extraction method and device to high resolution image |
-
2017
- 2017-07-23 CN CN201710603764.3A patent/CN107368813A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104820830A (en) * | 2015-05-08 | 2015-08-05 | 南京林业大学 | Tree species identification method based on full-waveform LiDAR canopy profile model |
| CN105654440A (en) * | 2015-12-30 | 2016-06-08 | 首都师范大学 | Regression model-based fast single-image defogging algorithm and system |
| CN105913017A (en) * | 2016-04-08 | 2016-08-31 | 南京林业大学 | Corresponding period double high resolution remote sensing image-based forest biomass estimation method |
| CN106951877A (en) * | 2017-03-28 | 2017-07-14 | 北京恒华伟业科技股份有限公司 | A kind of Objects extraction method and device to high resolution image |
Non-Patent Citations (4)
| Title |
|---|
| 孙宁 等: "基于滤波后处理的主动学习高光谱遥感图像分类", 《南昌工程学院学报》 * |
| 康旭东 等: "高光谱遥感影像空谱特征提取与分类方法研究", 《万方数据知识服务平台》 * |
| 王理 等: "《消费品质量安全信息分析：理论、方法与技术》", 30 April 2014 * |
| 黎良财 等: "基于 SVM 方法的 SPOT－5 影像植被分类", 《东北林业大学学报》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109902686A (en) * | 2019-01-22 | 2019-06-18 | 中国科学院植物研究所 | A kind of single wooden parameter extracting method of forest |
| CN109902686B (en) * | 2019-01-22 | 2021-01-15 | 中国科学院植物研究所 | Forest single tree parameter extraction method |
| CN110207670A (en) * | 2019-06-26 | 2019-09-06 | 北京林业大学 | A method of artificial forest forest hat width parameter is obtained using two dimensional image |
| CN111368712A (en) * | 2020-03-02 | 2020-07-03 | 四川九洲电器集团有限责任公司 | Hyperspectral image disguised target detection method based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Gao et al. | Discriminant saliency for visual recognition from cluttered scenes | |
| US8879855B2 (en) | Image segmentation for large-scale fine-grained recognition | |
| Huang et al. | Robust scene text detection with convolution neural network induced mser trees | |
| CN102214298B (en) | Method for detecting and identifying airport target by using remote sensing image based on selective visual attention mechanism | |
| CN107610114B (en) | optical satellite remote sensing image cloud and snow fog detection method based on support vector machine | |
| Wang et al. | Comparison of different color spaces for image segmentation using graph-cut | |
| CN109002851B (en) | Fruit classification method based on image multi-feature fusion and application | |
| CN107273832B (en) | License plate recognition method and system based on integral channel characteristics and convolutional neural network | |
| CN113591766B (en) | Multi-source remote sensing tree species identification method for unmanned aerial vehicle | |
| Almogdady et al. | A flower recognition system based on image processing and neural networks | |
| CN106610969A (en) | Multimodal information-based video content auditing system and method | |
| CN105718866B (en) | A kind of detection of sensation target and recognition methods | |
| CN104680127A (en) | Gesture identification method and gesture identification system | |
| Liu et al. | Identifying immature and mature pomelo fruits in trees by elliptical model fitting in the Cr–Cb color space | |
| CN107220664B (en) | Oil bottle boxing and counting method based on structured random forest | |
| US20130342694A1 (en) | Method and system for use of intrinsic images in an automotive driver-vehicle-assistance device | |
| CN102194108A (en) | Smiley face expression recognition method based on clustering linear discriminant analysis of feature selection | |
| CN107368813A (en) | A kind of forest hat width recognition methods applied to airborne field hyperspectrum image | |
| CN105913090B (en) | SAR image objective classification method based on SDAE-SVM | |
| CN106446925A (en) | Dolphin identity recognition method based on image processing | |
| CN113112498B (en) | Grape leaf spot identification method based on fine-grained countermeasure generation network | |
| CN113222959B (en) | Fresh jujube wormhole detection method based on hyperspectral image convolutional neural network | |
| CN110503140A (en) | Classification method based on depth migration study and neighborhood noise reduction | |
| CN110599463A (en) | Tongue image detection and positioning algorithm based on lightweight cascade neural network | |
| CN109190456A (en) | Pedestrian detection method is overlooked based on the multiple features fusion of converging channels feature and gray level co-occurrence matrixes |
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: 20171121 |