CN105809205A - Classification method and system for hyperspectral images - Google Patents
Classification method and system for hyperspectral images Download PDFInfo
- Publication number
- CN105809205A CN105809205A CN201610199937.5A CN201610199937A CN105809205A CN 105809205 A CN105809205 A CN 105809205A CN 201610199937 A CN201610199937 A CN 201610199937A CN 105809205 A CN105809205 A CN 105809205A
- Authority
- CN
- China
- Prior art keywords
- interest
- spectral domain
- image
- high spectrum
- code book
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a classification method for hyperspectral images.The classification method comprises the steps that a codebook is acquired through a codebook studying mode; a support vector machine classifier is trained through a training mode; the hyperspectral images are classified through the codebook and the support vector machine classifier, and a classification result is output.The invention further provides a classification system for the hyperspectral images.According to the classification method and system, empty spectrum domain interest points of the hyperspectral images are extracted through an empty spectrum domain interest point detection algorithm, the hyperspectral images are partitioned into small segments through image segmentation to be studied, trained and classified, a novel hyperspectral image classification framework is presented, the hyperspectral images can be effectively expressed and classified, and the classification accuracy and efficiency of the hyperspectral images with the high spatial resolution are significantly improved.
Description
Technical field
The present invention relates to computer vision field, particularly relate to sorting technique and the system thereof of a kind of high spectrum image.
Background technology
Image classification is the important research direction of computer vision field, is also a typical pattern recognition problem, and it is according to the different characteristic of reflection, the image processing method that different classes of target area is separated in image information.Classification hyperspectral imagery is one of hot spot technology in EO-1 hyperion application, is also the premise realizing various practical application, in target recognition and tracking, geological prospecting and environmental monitoring etc..
Original hyperspectral image classification method is mainly for the low high-spectrum remote sensing of spatial resolution, and existing high spectrum image has significantly high spatial resolution.When the curve of spectrum adopting individual element carries out image classification, efficiency is low on the one hand, and discrimination is easily reduced by the impact of " the different spectrum of jljl " and " same object different images " on the other hand.When adopting the spectral information of pixel and the spatial information of image to carry out classification hyperspectral imagery, spatial domain and spectral domain being separated consideration, algorithm complexity is high, length consuming time, and discrimination is low, but existing algorithm is difficult to be applicable on high-resolution classification hyperspectral imagery.
Therefore, how to realize that high spatial resolution high spectrum image is carried out classification fast and accurately and be always up the target that industry urgently realizes.
Summary of the invention
In view of this, it is an object of the invention to provide sorting technique and the system thereof of a kind of high spectrum image, it is intended to solve in prior art, high spectrum image recognition efficiency is low and that accuracy is not high problem.
The present invention proposes the sorting technique of a kind of high spectrum image, and described sorting technique includes:
Code book is obtained by code book learning style；
By training method Training Support Vector Machines grader；
Described code book and described support vector machine classifier is utilized to carry out the classification of high spectrum image output category result.
Preferably, the described step by code book learning style acquisition code book specifically includes:
The high spectrum image treating study code book carries out image block；
With m thread parallel extract the empty spectral domain point of interest of all image blocks in every panel height spectrum picture, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest；
The description vectors utilizing the had sky spectral domain point of interest presetting clustering method to obtaining clusters to obtain code book.
Preferably, the described step by training method Training Support Vector Machines grader specifically includes:
The high spectrum image of the support vector machine classifier treating training carries out image block；
With m thread parallel extract the empty spectral domain point of interest of all image blocks in every panel height spectrum picture, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest；
Based on the described code book got, utilize preset model that the empty spectral domain feature set of every panel height spectrum picture is encoded, to obtain the description vector of this image；
Utilize the described description vector Training Support Vector Machines grader got.
Preferably, the described classification utilizing described code book and described support vector machine classifier to carry out high spectrum image, and the step of output category result specifically includes:
High spectrum image to be sorted is carried out image block；
With m thread parallel extract the empty spectral domain point of interest of all image blocks in high spectrum image to be sorted, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest；
Based on the described code book got, utilize preset model that the empty spectral domain interest point set in high spectrum image to be sorted is encoded, to obtain the description vector of this image；
Utilize the described support vector machine classifier trained that the described description vector got is classified, and output category result.
On the other hand, the present invention also provides for the categorizing system of a kind of high spectrum image, and described categorizing system includes:
Study module, for obtaining code book by code book learning style；
Training module, for by training method Training Support Vector Machines grader；
Sort module, for utilizing described code book and described support vector machine classifier to carry out the classification of high spectrum image output category result.
Preferably, described study module includes:
First piecemeal submodule, the high spectrum image for treating study code book carries out image block；
First describes submodule, for extracting the empty spectral domain point of interest of all image blocks in every panel height spectrum picture with m thread parallel ground, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest；
First cluster submodule, for utilizing the description vectors of the default clustering method the had sky spectral domain point of interest to obtaining to cluster to obtain code book.
Preferably, described training module includes:
Second piecemeal submodule, the high spectrum image of the support vector machine classifier for treating training carries out image block；
Second describes submodule, for extracting the empty spectral domain point of interest of all image blocks in every panel height spectrum picture with m thread parallel ground, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest；
First encoding submodule, for based on the described code book got, utilizing preset model that the empty spectral domain feature set of every panel height spectrum picture is encoded, to obtain the description vector of this image；
First training submodule, for utilizing the described description vector Training Support Vector Machines grader got.
Preferably, described sort module includes:
3rd piecemeal submodule, for carrying out image block by high spectrum image to be sorted；
3rd describes submodule, for extracting the empty spectral domain point of interest of all image blocks in high spectrum image to be sorted with m thread parallel ground, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest；
Second encoding submodule, for based on the described code book got, utilizing preset model that the empty spectral domain interest point set in high spectrum image to be sorted is encoded, to obtain the description vector of this image；
First classification submodule, for utilizing the described support vector machine classifier trained that the described description vector got is classified, and output category result.
Technical scheme provided by the invention proposes a kind of high spatial resolution hyperspectral image classification method based on empty spectral domain point of interest, utilizes empty spectral domain point of interest can effectively utilize the information of spatial domain and spectral domain to carry out image classification.The present invention extracts the empty spectral domain point of interest of high spectrum image by empty spectral domain point of interest detection algorithm, it is divided into fritter to carry out learning, train and classifying high spectrum image by image block, propose a kind of new classification hyperspectral imagery framework, effectively high spectrum image can be indicated and classify, significantly improve the accuracy and efficiency of high spatial resolution classification hyperspectral imagery.
Accompanying drawing explanation
Fig. 1 is the sorting technique flow chart of high spectrum image in an embodiment of the present invention；
Fig. 2 is the detail flowchart of the step S11 shown in Fig. 1 in an embodiment of the present invention；
Fig. 3 is the detail flowchart of the step S12 shown in Fig. 1 in an embodiment of the present invention；
Fig. 4 is the detail flowchart of the step S13 shown in Fig. 1 in an embodiment of the present invention；
Fig. 5 is the chart of resolution classification hyperspectral imagery result between an embodiment of the present invention high and medium；
Fig. 6 is the chart of the pseudo color image classification results of high spectrum image in an embodiment of the present invention；
Fig. 7 is the general frame figure of the sorting technique of high spectrum image in an embodiment of the present invention；
Fig. 8 is the internal structure schematic diagram of the categorizing system 10 of high spectrum image in an embodiment of the present invention；
Fig. 9 is the internal structure schematic diagram of the study module 11 shown in Fig. 8 in an embodiment of the present invention；
Figure 10 is the internal structure schematic diagram of the training module 12 shown in Fig. 8 in an embodiment of the present invention；
Figure 11 is the internal structure schematic diagram of the sort module 13 shown in Fig. 8 in an embodiment of the present invention.
Detailed description of the invention
In order to make the purpose of the present invention, technical scheme and advantage clearly understand, below in conjunction with drawings and Examples, the present invention is further elaborated.Should be appreciated that specific embodiment described herein is only in order to explain the present invention, is not intended to limit the present invention.
The specific embodiment of the invention provides the sorting technique of a kind of high spectrum image, and described method mainly comprises the steps:
S11, by code book learning style obtain code book；
S12, by training method Training Support Vector Machines grader；
S13, described code book and described support vector machine classifier is utilized to carry out the classification of high spectrum image output category result.
The present invention proposes a kind of high spatial resolution hyperspectral image classification method based on empty spectral domain point of interest, utilizes empty spectral domain point of interest can effectively utilize the information of spatial domain and spectral domain to carry out image classification.The present invention extracts the empty spectral domain point of interest of high spectrum image by empty spectral domain point of interest detection algorithm, it is divided into fritter to carry out learning, train and classifying high spectrum image by image block, propose a kind of new classification hyperspectral imagery framework, effectively high spectrum image can be indicated and classify, significantly improve the accuracy and efficiency of high spatial resolution classification hyperspectral imagery.
The sorting technique of a kind of high spectrum image provided by the present invention will be described in detail below.
Refer to Fig. 1, for the sorting technique flow chart of high spectrum image in an embodiment of the present invention.
In step s 11, code book is obtained by code book learning style.
In the present embodiment, the described step S12 by code book learning style acquisition code book specifically includes S111 S113, as shown in Figure 2.
Refer to Fig. 2, for the detail flowchart of the step S11 shown in Fig. 1 in an embodiment of the present invention.
In step S111, the high spectrum image treating study code book carries out image block.
In the present embodiment, the high spatial resolution high-spectrum image set I={I of study code book is treated1,I2,I3,…,InIn image IiCarry out image block, it is possible to be expressed asWhereinRepresent image block respectively, wherein niRepresent the quantity of image block.
In step S112, with m thread parallel extract the empty spectral domain point of interest of all image blocks in every panel height spectrum picture, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest.
In the present embodiment, empty spectral domain point of interest is the characteristic point extracted on the empty spectral domain that the spatial domain of high spectrum image and spectral domain form, reflect the characteristic on spatial domain and spectral domain, utilize empty spectral domain point of interest can effectively utilize the information of spatial domain and spectral domain to carry out image classification.
In the present embodiment, with the empty spectral domain point of interest of all image blocks in every panel height spectrum picture in m thread parallel ground extraction step S111, and select 3DSIFT description that had sky spectral domain point of interest is carried out feature description, each empty spectral domain point of interest is described as the description vectors of 640 dimensions.
In step S113, the description vectors of the had sky spectral domain point of interest presetting clustering method to obtaining is utilized to cluster to obtain code book.
In the present embodiment, default clustering method is K-means clustering method, utilizes K-means clustering method that all of description vectors obtained in step S112 is clustered, and k the cluster centre obtained after cluster is called code word or vision word, uses viRepresent.And the combination of all vision word is code book V={v1, v2, v3..., vk}。
Please continue to refer to Fig. 1, in step s 12, by training method Training Support Vector Machines grader.
In the present embodiment, the described step S12 by training method Training Support Vector Machines (SupportVectorMachine, SVM) grader specifically includes S121 S124, as shown in Figure 3.
Refer to Fig. 3, for the detail flowchart of the step S12 shown in Fig. 1 in an embodiment of the present invention.
In step S121, the high spectrum image of the support vector machine classifier treating training carries out image block.
In the present embodiment, the high spatial resolution high-spectrum image set T={T of the SVM classifier of training is treated1,T2,T3,…,TLIn image TjCarry out image block, it is possible to be expressed asWhereinRepresent image block respectively, wherein njRepresent the quantity of image block.
In step S122, with m thread parallel extract the empty spectral domain point of interest of all image blocks in every panel height spectrum picture, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest.
In the present embodiment, with the empty spectral domain point of interest of all image blocks in every panel height spectrum picture in m thread parallel ground extraction step S121, and with 3DSIFT description, had sky spectral domain point of interest is carried out feature description, each empty spectral domain point of interest is described as the description vectors of 640 dimensions.
In step S123, based on the described code book got, utilize preset model that the empty spectral domain feature set of every panel height spectrum picture is encoded, to obtain the description vector of this image.
In the present embodiment, preset model is BoW model, based on the code book V obtained in the learning procedure shown in Fig. 2, utilizes the BoW model high spatial resolution high spectrum image T to every width SVM classifier to be trainedjEmpty spectral domain feature set be encoded, obtain the description vector of all images, be designated as: S={S1,S2,S3,…,SL, wherein Sj∈ S is TjThe iamge description vector of ∈ T.
In step S124, utilize the described description vector Training Support Vector Machines grader got.
In the present embodiment, the description vector set S utilizing the step S123 T obtained trains SVM classifier.
Please continue to refer to Fig. 1, in step s 13, described code book and described support vector machine classifier is utilized to carry out the classification of high spectrum image output category result.
In the present embodiment, the described classification utilizing described code book and described support vector machine classifier to carry out high spectrum image, and the step S13 of output category result specifically includes S131 S134, as shown in Figure 4.
Refer to Fig. 4, for the detail flowchart of the step S13 shown in Fig. 1 in an embodiment of the present invention.
In step S131, high spectrum image to be sorted is carried out image block.
In the present embodiment, by high spatial resolution high spectrum image I to be sortedtestCarry out image block, it is possible to be expressed as Itest={ b1, b2, b3..., bN, wherein b1, b2, b3..., bNRepresenting image block respectively, N represents the quantity of image block.
In step S132, with m thread parallel extract the empty spectral domain point of interest of all image blocks in high spectrum image to be sorted, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest.
In the present embodiment, same with all image I in high spectrum image to be sorted in m thread parallel ground extraction step S131testThe empty spectral domain point of interest of corresponding image block, and with 3DSIFT description, each empty spectral domain point of interest is described, each empty spectral domain point of interest is described as the description vectors of 640 dimensions, the empty spectral domain point of interest extracted is merged form high spectrum image I in N block high spectrum image blocktestEmpty spectral domain feature set, be designated as Ftest={ f1,f2,…,fM}。
In step S133, based on the described code book got, utilize preset model that the empty spectral domain interest point set in high spectrum image to be sorted is encoded, to obtain the description vector of this image.
In the present embodiment, preset model is BoW model, based on the code book V obtained in the learning procedure shown in Fig. 2, utilizes BoW model to image I in high spatial resolution high spectrum image to be sortedtestEmpty spectral domain interest point set FtestIt is encoded, obtains image ItestDescription vector D.
In step S134, utilize the described support vector machine classifier trained that the described description vector got is classified, and output category result.
In the present embodiment, utilize the SVM classifier that the training step shown in Fig. 3 trains that the description vector D obtained in step S133 is classified, output category result.
In the present embodiment, no matter it is step S11, step S12, or step S13, it is both for great amount of images, including continuous print multiple image, and not only just for piece image.
In the present embodiment, apply the above-mentioned sorting technique based on the high spatial resolution high spectrum image of empty spectral domain point of interest and one group of high spatial resolution high spectrum image is carried out classification experiments, when learning code book, the each image of code book to be learned is extracted 3500 empty spectral domain points of interest and is used for clustering, and the vocabulary number obtained after cluster is set to 2500.Repeat 10 subseries to test us and obtain classification results as shown in Figure 5.Calculate average and the standard deviation of the classification results of these 10 times experiments, it is known that, the accuracy rate of this group classification hyperspectral imagery is 93.22 ± 0.51 (%).
In order to the effectiveness of the sorting technique of the high spatial resolution high spectrum image based on empty spectral domain point of interest that the present invention proposes is described, the pseudocolour picture being also extracted above-mentioned all high spectrum images to be sorted has carried out the classification experiments of two dimensional image, and calculates its accuracy rate.Repeat 10 experiments, obtain classification results as shown in Figure 6.Calculating average and the standard deviation of the classification results of these 10 times experiments, the accuracy rate that can obtain classification is 84.42 ± 0.70 (%).
Compared with Fig. 5, in Fig. 6, the accuracy rate of the pseudo color image classification of the high spectrum image of display is far below the accuracy rate of classification hyperspectral imagery, and this has absolutely proved the sorting technique of high spatial resolution high spectrum image based on empty spectral domain point of interest and the effectiveness of system and accuracy that the present invention proposes.
Refer to Fig. 7, for the general frame figure of the sorting technique of high spectrum image in an embodiment of the present invention.
The sorting technique of a kind of high spectrum image provided by the invention utilizes empty spectral domain point of interest can effectively utilize the information of spatial domain and spectral domain to carry out image classification, effectively high spectrum image can be indicated and classify, significantly improve the accuracy and efficiency of high spatial resolution classification hyperspectral imagery.
Refer to Fig. 8, it is shown that for the structural representation of the categorizing system 10 of high spectrum image in an embodiment of the present invention.In the present embodiment, the categorizing system 10 of high spectrum image mainly includes study module 11, training module 12 and sort module 13.The categorizing system 10 of this high spectrum image, empty spectral domain point of interest is utilized can effectively to utilize the information of spatial domain and spectral domain to carry out image classification, effectively high spectrum image can be indicated and classify, significantly improve the accuracy and efficiency of high spatial resolution classification hyperspectral imagery.
Wherein, study module 11, for obtaining code book by code book learning style.
In the present embodiment, study module 11 specifically includes the first piecemeal submodule 111, first describes submodule 112 and the first cluster submodule 113, as shown in Figure 7.
Refer to Fig. 9, it is shown that for the internal structure schematic diagram of the study module 11 shown in Fig. 8 in an embodiment of the present invention.
First piecemeal submodule 111, the high spectrum image for treating study code book carries out image block.
In the present embodiment, the concrete image block method of the first piecemeal submodule 111 refers to the relevant record in abovementioned steps S111, does not do repeated description at this.
First describes submodule 112, for extracting the empty spectral domain point of interest of all image blocks in every panel height spectrum picture with m thread parallel ground, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest.
In the present embodiment, first describes the relevant record that the concrete character description method of submodule 112 refers in abovementioned steps S112, does not do repeated description at this.
First cluster submodule 113, for utilizing the description vectors of the default clustering method the had sky spectral domain point of interest to obtaining to cluster to obtain code book.
In the present embodiment, the first clustering method clustering submodule 113 concrete refers to the relevant record in abovementioned steps S113, does not do repeated description at this.
Please continue to refer to Fig. 8, training module 12, for by training method Training Support Vector Machines grader.
In the present embodiment, training module 12 specifically includes the second piecemeal submodule 121, second describes submodule the 122, first encoding submodule 123 and the first training submodule 124, as shown in Figure 10.
Refer to Figure 10, it is shown that for the internal structure schematic diagram of the training module 12 shown in Fig. 8 in an embodiment of the present invention.
Second piecemeal submodule 121, the high spectrum image of the support vector machine classifier for treating training carries out image block.
In the present embodiment, the concrete image block method of the second piecemeal submodule 121 refers to the relevant record in abovementioned steps S121, does not do repeated description at this.
Second describes submodule 122, for extracting the empty spectral domain point of interest of all image blocks in every panel height spectrum picture with m thread parallel ground, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest.
In the present embodiment, second describes the relevant record that the concrete character description method of submodule 122 refers in abovementioned steps S122, does not do repeated description at this.
First encoding submodule 123, for based on the described code book got, utilizing preset model that the empty spectral domain feature set of every panel height spectrum picture is encoded, to obtain the description vector of this image.
In the present embodiment, the concrete coded method of the first encoding submodule 123 refers to the relevant record in abovementioned steps S123, does not do repeated description at this.
First training submodule 124, for utilizing the described description vector Training Support Vector Machines grader got.
In the present embodiment, the first training method training submodule 124 concrete refers to the relevant record in abovementioned steps S124, does not do repeated description at this.
Please continue to refer to Fig. 8, sort module 13, for utilizing described code book and described support vector machine classifier to carry out the classification of high spectrum image output category result.
In the present embodiment, sort module 13 specifically includes the 3rd piecemeal submodule the 131, the 3rd and describes submodule the 132, second encoding submodule 133 and the first classification submodule 134, as shown in figure 11.
Refer to Figure 11, it is shown that for the internal structure schematic diagram of the sort module 13 shown in Fig. 8 in an embodiment of the present invention.
3rd piecemeal submodule 131, for carrying out image block by high spectrum image to be sorted.
In the present embodiment, the concrete image block method of the 3rd piecemeal submodule 131 refers to the relevant record in abovementioned steps S131, does not do repeated description at this.
3rd describes submodule 132, for extracting the empty spectral domain point of interest of all image blocks in high spectrum image to be sorted with m thread parallel ground, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest.
In the present embodiment, the 3rd describes the relevant record that the concrete character description method of submodule 132 refers in abovementioned steps S132, does not do repeated description at this.
Second encoding submodule 133, for based on the described code book got, utilizing preset model that the empty spectral domain interest point set in high spectrum image to be sorted is encoded, to obtain the description vector of this image.
In the present embodiment, the concrete coded method of the second encoding submodule 133 refers to the relevant record in abovementioned steps S133, does not do repeated description at this.
First classification submodule 134, for utilizing the described support vector machine classifier trained that the described description vector got is classified, and output category result.
In the present embodiment, the first classification concrete sorting technique of submodule 134 refers to the relevant record in abovementioned steps S134, does not do repeated description at this.
The categorizing system 10 of a kind of high spectrum image provided by the invention, empty spectral domain point of interest is utilized can effectively to utilize the information of spatial domain and spectral domain to carry out image classification, effectively high spectrum image can be indicated and classify, significantly improve the accuracy and efficiency of high spatial resolution classification hyperspectral imagery.
It should be noted that in above-described embodiment, included unit is carry out dividing according to function logic, but is not limited to above-mentioned division, as long as being capable of corresponding function；It addition, the concrete title of each functional unit is also only to facilitate mutually distinguish, it is not limited to protection scope of the present invention.
Additionally, one of ordinary skill in the art will appreciate that all or part of step realizing in the various embodiments described above method can be by the hardware that program carrys out instruction relevant and completes, corresponding program can be stored in a computer read/write memory medium, described storage medium, such as ROM/RAM, disk or CD etc..
The foregoing is only presently preferred embodiments of the present invention, not in order to limit the present invention, all any amendment, equivalent replacement and improvement etc. made within the spirit and principles in the present invention, should be included within protection scope of the present invention.
Claims (8)
1. the sorting technique of a high spectrum image, it is characterised in that described sorting technique includes:
Code book is obtained by code book learning style；
By training method Training Support Vector Machines grader；
Described code book and described support vector machine classifier is utilized to carry out the classification of high spectrum image output category result.
2. the sorting technique of high spectrum image as claimed in claim 1, it is characterised in that the described step by code book learning style acquisition code book specifically includes:
The high spectrum image treating study code book carries out image block；
With m thread parallel extract the empty spectral domain point of interest of all image blocks in every panel height spectrum picture, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest；
The description vectors utilizing the had sky spectral domain point of interest presetting clustering method to obtaining clusters to obtain code book.
3. the sorting technique of high spectrum image as claimed in claim 2, it is characterised in that the described step by training method Training Support Vector Machines grader specifically includes:
The high spectrum image of the support vector machine classifier treating training carries out image block；
With m thread parallel extract the empty spectral domain point of interest of all image blocks in every panel height spectrum picture, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest；
Based on the described code book got, utilize preset model that the empty spectral domain feature set of every panel height spectrum picture is encoded, to obtain the description vector of this image；
Utilize the described description vector Training Support Vector Machines grader got.
4. the sorting technique of high spectrum image as claimed in claim 3, it is characterised in that the described classification utilizing described code book and described support vector machine classifier to carry out high spectrum image, and the step of output category result specifically includes:
High spectrum image to be sorted is carried out image block；
With m thread parallel extract the empty spectral domain point of interest of all image blocks in high spectrum image to be sorted, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest；
Based on the described code book got, utilize preset model that the empty spectral domain interest point set in high spectrum image to be sorted is encoded, to obtain the description vector of this image；
Utilize the described support vector machine classifier trained that the described description vector got is classified, and output category result.
5. the categorizing system of a high spectrum image, it is characterised in that described categorizing system includes:
Study module, for obtaining code book by code book learning style；
Training module, for by training method Training Support Vector Machines grader；
Sort module, for utilizing described code book and described support vector machine classifier to carry out the classification of high spectrum image output category result.
6. the categorizing system of high spectrum image as claimed in claim 5, it is characterised in that described study module includes:
First piecemeal submodule, the high spectrum image for treating study code book carries out image block；
First describes submodule, for extracting the empty spectral domain point of interest of all image blocks in every panel height spectrum picture with m thread parallel ground, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest；
First cluster submodule, for utilizing the description vectors of the default clustering method the had sky spectral domain point of interest to obtaining to cluster to obtain code book.
7. the categorizing system of high spectrum image as claimed in claim 6, it is characterised in that described training module includes:
Second piecemeal submodule, the high spectrum image of the support vector machine classifier for treating training carries out image block；
Second describes submodule, for extracting the empty spectral domain point of interest of all image blocks in every panel height spectrum picture with m thread parallel ground, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest；
First encoding submodule, for based on the described code book got, utilizing preset model that the empty spectral domain feature set of every panel height spectrum picture is encoded, to obtain the description vector of this image；
First training submodule, for utilizing the described description vector Training Support Vector Machines grader got.
8. the categorizing system of high spectrum image as claimed in claim 7, it is characterised in that described sort module includes:
3rd piecemeal submodule, for carrying out image block by high spectrum image to be sorted；
3rd describes submodule, for extracting the empty spectral domain point of interest of all image blocks in high spectrum image to be sorted with m thread parallel ground, and had sky spectral domain point of interest is carried out feature description, to obtain the description vectors of each empty spectral domain point of interest；
Second encoding submodule, for based on the described code book got, utilizing preset model that the empty spectral domain interest point set in high spectrum image to be sorted is encoded, to obtain the description vector of this image；
First classification submodule, for utilizing the described support vector machine classifier trained that the described description vector got is classified, and output category result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610199937.5A CN105809205B (en) | 2016-03-31 | 2016-03-31 | A kind of classification method and its system of high spectrum image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610199937.5A CN105809205B (en) | 2016-03-31 | 2016-03-31 | A kind of classification method and its system of high spectrum image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105809205A true CN105809205A (en) | 2016-07-27 |
| CN105809205B CN105809205B (en) | 2019-07-02 |
Family
ID=56460028
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610199937.5A Active CN105809205B (en) | 2016-03-31 | 2016-03-31 | A kind of classification method and its system of high spectrum image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105809205B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106408009A (en) * | 2016-09-08 | 2017-02-15 | 哈尔滨工业大学 | Depth belief network-based neighborhood weighted averaging hyperspectral image classification method |
| CN106529472A (en) * | 2016-10-24 | 2017-03-22 | 深圳大学 | Target detection method and apparatus based on large-scale high-resolution and high-spectral image |
| CN106682672A (en) * | 2016-10-24 | 2017-05-17 | 深圳大学 | Method and device for acquiring feature descriptor of hyper-spectral image |
| WO2018076137A1 (en) * | 2016-10-24 | 2018-05-03 | 深圳大学 | Method and device for obtaining hyper-spectral image feature descriptor |
| WO2018076138A1 (en) * | 2016-10-24 | 2018-05-03 | 深圳大学 | Target detection method and apparatus based on large-scale high-resolution hyper-spectral image |
| WO2019090509A1 (en) * | 2017-11-08 | 2019-05-16 | 深圳大学 | Hyperspectral image classification method and system |
| CN110244381A (en) * | 2019-06-17 | 2019-09-17 | 中国地质大学（北京） | Unmanned plane geological exploration methods, devices and systems |
| WO2020042070A1 (en) * | 2018-08-30 | 2020-03-05 | 深圳大学 | Method for improving classification accuracy of hyperspectral image, device, apparatus, and storage medium |
| CN112257603A (en) * | 2020-10-23 | 2021-01-22 | 深圳大学 | Hyperspectral image classification method and related equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103400158A (en) * | 2013-07-29 | 2013-11-20 | 电子科技大学 | Dynamic shape codebook learning-based level set tracking method |
| CN105389588A (en) * | 2015-11-04 | 2016-03-09 | 上海交通大学 | Multi-semantic-codebook-based image feature representation method |
-
2016
- 2016-03-31 CN CN201610199937.5A patent/CN105809205B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103400158A (en) * | 2013-07-29 | 2013-11-20 | 电子科技大学 | Dynamic shape codebook learning-based level set tracking method |
| CN105389588A (en) * | 2015-11-04 | 2016-03-09 | 上海交通大学 | Multi-semantic-codebook-based image feature representation method |
Non-Patent Citations (4)
| Title |
|---|
| 于施淼等: "基于高光谱图像与视觉词袋模型的稻种发芽率预测研究", 《激光与光电子学进展》 * |
| 任越美等: "基于稀疏表示和词袋模型的高光谱图像分类", 《计算机科学》 * |
| 董振宇等: "基于词袋模型的图像分类研究", 《中国优秀硕士学位论文全文数据库 信息科技I辑 》 * |
| 高恒振: "一种基于谱域-空域组合特征支持向量机的高光谱图像分类算法", 《宇航学报》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106408009B (en) * | 2016-09-08 | 2019-04-12 | 哈尔滨工业大学 | Neighborhood weighted average hyperspectral image classification method based on depth confidence network |
| CN106408009A (en) * | 2016-09-08 | 2017-02-15 | 哈尔滨工业大学 | Depth belief network-based neighborhood weighted averaging hyperspectral image classification method |
| CN106682672B (en) * | 2016-10-24 | 2020-04-24 | 深圳大学 | Method and device for acquiring hyperspectral image feature descriptor |
| CN106529472A (en) * | 2016-10-24 | 2017-03-22 | 深圳大学 | Target detection method and apparatus based on large-scale high-resolution and high-spectral image |
| CN106682672A (en) * | 2016-10-24 | 2017-05-17 | 深圳大学 | Method and device for acquiring feature descriptor of hyper-spectral image |
| WO2018076137A1 (en) * | 2016-10-24 | 2018-05-03 | 深圳大学 | Method and device for obtaining hyper-spectral image feature descriptor |
| WO2018076138A1 (en) * | 2016-10-24 | 2018-05-03 | 深圳大学 | Target detection method and apparatus based on large-scale high-resolution hyper-spectral image |
| CN106529472B (en) * | 2016-10-24 | 2019-08-02 | 深圳大学 | Object detection method and device based on large scale high-resolution high spectrum image |
| WO2019090509A1 (en) * | 2017-11-08 | 2019-05-16 | 深圳大学 | Hyperspectral image classification method and system |
| WO2020042070A1 (en) * | 2018-08-30 | 2020-03-05 | 深圳大学 | Method for improving classification accuracy of hyperspectral image, device, apparatus, and storage medium |
| CN110244381A (en) * | 2019-06-17 | 2019-09-17 | 中国地质大学（北京） | Unmanned plane geological exploration methods, devices and systems |
| CN112257603A (en) * | 2020-10-23 | 2021-01-22 | 深圳大学 | Hyperspectral image classification method and related equipment |
| CN112257603B (en) * | 2020-10-23 | 2022-06-17 | 深圳大学 | Hyperspectral image classification method and related equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105809205B (en) | 2019-07-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105809205A (en) | Classification method and system for hyperspectral images | |
| Serna et al. | Classification of traffic signs: The european dataset | |
| Long et al. | Accurate object localization in remote sensing images based on convolutional neural networks | |
| Zhou et al. | Robust vehicle detection in aerial images using bag-of-words and orientation aware scanning | |
| US7853072B2 (en) | System and method for detecting still objects in images | |
| CN101859382B (en) | License plate detection and identification method based on maximum stable extremal region | |
| CN112686812B (en) | Bank card inclination correction detection method and device, readable storage medium and terminal | |
| US20140219563A1 (en) | Label-embedding for text recognition | |
| Derpanis et al. | Classification of traffic video based on a spatiotemporal orientation analysis | |
| Feris et al. | Attribute-based vehicle search in crowded surveillance videos | |
| CN103761531A (en) | Sparse-coding license plate character recognition method based on shape and contour features | |
| CN110008899B (en) | Method for extracting and classifying candidate targets of visible light remote sensing image | |
| Sikirić et al. | Image representations on a budget: Traffic scene classification in a restricted bandwidth scenario | |
| Thubsaeng et al. | Vehicle logo detection using convolutional neural network and pyramid of histogram of oriented gradients | |
| CN103984965A (en) | Pedestrian detection method based on multi-resolution character association | |
| WO2017166156A1 (en) | Hyperspectral image classification method and system | |
| Satti et al. | R‐ICTS: Recognize the Indian cautionary traffic signs in real‐time using an optimized adaptive boosting cascade classifier and a convolutional neural network | |
| Thu et al. | Pyramidal Part‐Based Model for Partial Occlusion Handling in Pedestrian Classification | |
| Zhu et al. | Car detection based on multi-cues integration | |
| Chen et al. | Context-aware lane marking detection on urban roads | |
| Peng et al. | Binary coding-based vehicle image classification | |
| Li | Vehicle license plate detection and recognition | |
| Bellam et al. | A Practical Approach of Recognizing and Detecting Traffic Signs using Deep Neural Network Model | |
| Rasche | Computer Vision | |
| Pu et al. | A robust and real-time approach for license plate detection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |