WO2019218313A1 - 一种递进式动态高光谱图像分类方法 - Google Patents

一种递进式动态高光谱图像分类方法 Download PDF

Info

Publication number
WO2019218313A1
WO2019218313A1 PCT/CN2018/087329 CN2018087329W WO2019218313A1 WO 2019218313 A1 WO2019218313 A1 WO 2019218313A1 CN 2018087329 W CN2018087329 W CN 2018087329W WO 2019218313 A1 WO2019218313 A1 WO 2019218313A1
Authority
WO
WIPO (PCT)
Prior art keywords
point
points
hyperspectral image
thr
flag
Prior art date
Application number
PCT/CN2018/087329
Other languages
English (en)
French (fr)
Inventor
杨保健
关志锋
胡志群
Original Assignee
五邑大学
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 五邑大学 filed Critical 五邑大学
Priority to PCT/CN2018/087329 priority Critical patent/WO2019218313A1/zh
Publication of WO2019218313A1 publication Critical patent/WO2019218313A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition

Definitions

  • the invention relates to the field of high-spectral image classification technology, in particular to a progressive dynamic hyperspectral image classification method.
  • HIC hyperspectral classification
  • an object of the present invention is to provide a hyperspectral image classification method with low computational complexity, high classification accuracy, and fast running speed, that is, a progressive dynamic hyperspectral image classification method.
  • a progressive dynamic hyperspectral image classification method comprising the following steps:
  • Step 1 Read hyperspectral image data
  • Step 2 Calculate the difference between each pixel in the hyperspectral image and its local neighborhood pixels, and set a suitable threshold to perform discontinuity detection to divide the continuous point and the discontinuous point;
  • Step 3 cyclically classifying unclassified consecutive points by comparing successive points with points of known categories in their local neighborhoods;
  • Step 4 cyclically classifying the discontinuous points and consecutive points that cannot be classified in step three;
  • Step 5 classify all remaining points to be classified in step four;
  • Step six output the classification result.
  • step one is specifically:
  • the hyperspectral image data is derived from the remote sensing image acquired by the imaging spectrometer, and the hyperspectral image data is removed by the water vapor absorption band and the signal to noise ratio is relatively low, and the remaining hyperspectral image data is D ⁇ R m ⁇ n ⁇ l , where m and n are the number of rows and columns of the hyperspectral image, and l is the number of remaining hyperspectral image bands; and the training sample data of each type is read at the same time.
  • step 2 is specifically:
  • step three is specifically:
  • the point P(x, y) is judged as the class; if the local neighborhood of P(x, y) is known in multiple classes For the category point, the point P(x, y) is judged as one of the most similar ones; the next step is performed;
  • the marker point P(x, y) is the classified point and returns 3.1.
  • step four is specifically:
  • the point Q(x, y) is judged as the class; if the local neighborhood of Q(x, y) is known in multiple classes For the category point, the point Q(x, y) is judged as one of the most similar ones; the next step is performed;
  • the beneficial effects of the present invention are: a progressive dynamic hyperspectral image classification method adopted by the present invention, which firstly finds the position of a known sample point, and then classifies the sample adjacent to the known sample, and then Gradually promote the expansion to achieve high-precision classification of hyperspectral images. Specifically, as shown in steps 1 to 6, the unknown point in the local neighborhood of the known sample is first compared with the current known sample point. If the difference is small, it is determined as the currently known sample. The category of the point; if the difference is large, it is difficult to decide whether it is the same class, then mark it as a point that is more difficult to classify, put it first, wait for more known category points around, get more information, then Classify it.
  • the unknown point is also in the domain of other different classes of known points, you also need to consider the possibility that it belongs to other classes.
  • the points that are easy to be separated are resolved, for the remaining points that have not been classified, first check which points have higher resolution, such as those points where there are more known samples in the neighborhood (considering the distribution of features)
  • the local continuity so that you can make more use of the information of points in its local neighborhood
  • the most difficult to classify points - such as points in its local neighborhood that have no known points.
  • the above classification process embodies a kind of point-and-spoke classification thinking that is easy to difficult and progressive.
  • This classification method of progressive propulsion makes full use of the spectral-spatial information of hyperspectral, and each step avoids the misclassification as much as possible, reflecting the dynamic and intelligent nature of the classification process.
  • this patent proposes a progressive dynamic hyperspectral classification method. Compared with most existing methods, this method does not require advanced mathematical knowledge, and has low computational complexity and high classification accuracy. , the speed is fast and so on.
  • FIG. 1 is a flow chart of a method for classifying a progressive dynamic hyperspectral image according to the present invention
  • Figure 2 is a false color image of the hyperspectral image AVIRIS Indian pines.
  • the hyperspectral image selected for this example is AVIRIS Indian pines.
  • the hyperspectral image has a size of 145 x 145 and has 220 spectral segments that uniformly cover the spectral range of 0.2:2.5 ⁇ m.
  • the hyperspectral image contains 16 types of labeled samples. Due to the low water absorption and low signal to noise ratio, the spectral segments 104-108, 150-163 and 220 will be removed prior to classification, leaving only a total of 200 spectral segments.
  • Figure 2 shows a false color image of the AVIRIS Indian pines.
  • a specific implementation procedure of a progressive dynamic hyperspectral image classification method is as follows:
  • Step 1 Input the hyperspectral image data D ⁇ R 145 ⁇ 145 ⁇ 200 and the corresponding feature mark matrix L ⁇ R 145 ⁇ 145 .
  • Each pixel in D is the sample represented by the hyperspectral feature vector, and the dimension of the sample is 200.
  • Step 2 Discontinuity detection: In order to calculate the difference between each pixel in D and its local neighborhood pixels, this example firstly reduces the dimension of D by retaining the feature vector of 60% of the total energy through principal component analysis; The Euclidean distance of each pixel in D after dimension reduction and its local neighborhood mean of 23 ⁇ 23, and the discontinuity of each point in D is represented by the Euclidean distance; the discontinuity of each point in the assumed D obeys the normal state Distribution, and assuming that 40% of the pixels are continuous pixel points, the threshold Th of each point in the division D is a continuous point and a discontinuous point; and the degree of discontinuity of each point in the D is obtained by using the Th binarization
  • Step 3 Classification of consecutive points: Perform the following loop:
  • the point P(x, y) is judged as the class; if the 3 ⁇ 3 local neighbor of P(x, y) There are many types of known category points in the domain. According to the minimum distance criterion, the point P(x, y) is judged as the class corresponding to the known category sample with the smallest Euclidean distance, and the next step is performed;
  • the marker point P(x, y) is the classified point and returns 3.1.
  • Step 4 Classify the discontinuous points and consecutive points that cannot be classified in step three:
  • this example uses the optimized distance-weighted linear regression classifier to classify the remaining points to be classified; for each of the remaining points to be classified b, the following steps 5.1 to 5.4 are performed:
  • Step 6 Output the classification result.
  • the comparison methods include:
  • WJCR Weighted Joint Collaborative Representation (M.Xiong, Q.Ran, W.Li, J.Zou, and Q.Du, Hyperspectral Image Classification Using Weighted Joint Collaborative Representation, IEEE Geoscience and Remote Sensing Letters, vol. 12, no. 6, pp.1209-1213, Jun.2015.);
  • JSaCR Joint Spatial-Aware Collaborative Representation (J. Jiang, C. Chen, Y. Yu, X. Jiang, and J. Ma, "Spatial-Aware Collaborative Representation for Hyperspectral Remote Sensing Image Classification,” IEEE Geoscience and Remote Sensing Letters , vol.14, no.3, pp.404-408, Mar.2017.);
  • SC-MK Superpixel-based Classification via Multiple Kernels (L. Fang, S. Li, W. Duan, J. Ren, and JA Benediktsson, "Classification of hyperspectral images by exploiting spectral-patial information of superpixel via multiple kernels," IEEE Transactions on Geoscience and Remote Sensing, vol. 53, no. 12, pp. 6663-6674, 2015.);
  • R2MK Region-based Relaxed Multiple Kernel (J. Liu, Z. Wu, Z. Xiao, and J. Yang, "Region-Based Relaxed Multiple Kernel Collaborative Representation for Hyperspectral Image Classification," IEEE Access, vol. 5, pp. 20921-20933, 2017.).
  • Table 1 Comparison of classification accuracy (%) of different methods on the AVIRIS Indian Pines dataset (15 training samples per class).
  • the present invention provides a progressive dynamic hyperspectral image classification method, which is only a preferred embodiment of the present invention, and it should be noted that those skilled in the art can not deviate from the principle of the present invention. Several modifications and retouchings are also possible, which should also be considered as protection of the present invention.
  • the components that are not clear in this embodiment can be implemented by the prior art.
  • the present invention is not limited to the above embodiments, and it should be within the scope of the present invention as long as it achieves the technical effects of the present invention by the same means.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Image Analysis (AREA)

Abstract

一种递进式动态高光谱图像分类方法,包括以下步骤:步骤一:读取高光谱图像数据;步骤二:计算高光谱图像中各像其局部邻域像素的差异,通过设定合适的阈值,进行非连续性检测,划分出连续点和非连续点;步骤三,通过将连续点与其局部邻域内已知类别的点进行比较,对未分类的连续点进行分类;步骤四,对非连续点以及步骤三中未能分类的连续点进行分类;步骤五,对步骤四中全部剩余待分类点进行分类;步骤六,输出分类结果。相比于现有多数方法,该方法不需要高深的数学知识,兼具计算复杂度低,分类精度高,运行速度快等优点。

Description

一种递进式动态高光谱图像分类方法 技术领域
本发明涉及高光谱图像分类技术领域,特别涉及一种递进式动态高光谱图像分类方法。
背景技术
经过最近数十年的发展,高光谱分类(HIC)的方法已经非常丰富。早期的HIC方法侧重于直接将其他领域常用的分类器的借用于HIC。如基于SVM、决策树等分类器的方法。后来,人们逐近开始针对HIC的特殊性,将空间信息融入到HIC中,提出了谱-空间HIC方法。谱-空间HIC方法是当前最主流的HIC方法。这其中,基于超像素分割的方法近年来获得了较多的关注。随着深度学习在各个领域的发展,基于深度学习的HIC也开始出现。
人类之所以区别于其他动物,在于在各种活动中,人类均体现出自己的智慧和技巧。但是,现有的HIC方法要么侧重于分类器的使用,要么侧重于特征提取方法,却少有思考,如果HIC需要人工来进行分类,该如何充分利用地物分布的局部连续性特征,借鉴人类手工HIC的过程和策略,最大限度地降低分类错误,建立高效的HIC方法。
发明内容
为解决上述问题,本发明的目的在于提供一种计算复杂度低,分类精度高,运行速度快的高光谱图像分类方法,即一种递进式动态高 光谱图像分类方法。
本发明解决其问题所采用的技术方案是:
一种递进式动态高光谱图像分类方法,包括以下步骤:
步骤一:读取高光谱图像数据;
步骤二:计算高光谱图像中各像素与其局部邻域像素的差异,通过设定合适的阈值,进行非连续性检测,划分出连续点和非连续点;
步骤三,通过将连续点与其局部邻域内已知类别的点进行比较,循环地对未分类的连续点进行分类;
步骤四,对非连续点以及步骤三中未能分类的连续点循环地进行分类;
步骤五,对步骤四中全部剩余待分类点进行分类;
步骤六,输出分类结果。
进一步,所述步骤一具体为:
所述高光谱图像数据来源于成像光谱仪采集到的遥感图像,将高光谱图像数据去除被水汽吸收的波段和信噪比较低的波段,设剩余的高光谱图像数据为D∈R m×n×l,其中m、n为高光谱图像的行数和列数,l为剩余的高光谱图像波段数;同时读取各类的训练样本数据。
进一步,所述步骤二具体为:
计算D中各像素与其局部邻域像素的差异,通过设定合适的阈值,将高光谱图像各点划分为连续点和非连续点,其结果为0和1,并建立一个矩阵S∈R m×n用于存储结果,若S(x,y)=0,则表示(x,y)点为连 续点,若S(x,y)=1,则表示(x,y)点为非连续点。
进一步,所述步骤三具体为:
3.1扫描高光谱图像的未分类连续点,若找不到未分类的连续点,则执行步骤四;若找到未分类连续点,设该点为P(x,y),执行下一步;
3.2若P(x,y)的局部邻域内存在已知类别的点,则执行下一步;否则返回3.1;
3.3若P(x,y)的局部邻域内只有一类已知类别点,则将点P(x,y)判为该类;若P(x,y)的局部邻域内存在多类已知类别点,则将点P(x,y)判为其中与其最相似的一类;执行下一步;
3.4标记点P(x,y)为已分类点,返回3.1。
进一步,所述步骤四具体为:
4.1设定鉴别度阈值Thr及其下界Thr 0,设定Thr的下降步长step,引入标记值flag和trigger,初始化flag=true:
4.2当flag=true时,执行下一步,否则执行步骤五;
4.3令flag=false,trigger=false;
4.4扫描高光谱图像,若找到未分类点,则记当前未分类点为Q(x,y),执行下一步;否则执行4.9;
4.5若Q(x,y)的局部邻域内存在已知类别的点,则执行下一步;否则返回4.4;
4.6计算Q(x,y)的鉴别度,记为ρ(x,y);若ρ(x,y)<Thr,则trigger=true,返回4.4;否则执行下一步;
4.7若Q(x,y)的局部邻域内只有一类已知类别点,则将点Q(x,y)判为该类;若Q(x,y)的局部邻域内存在多类已知类别点,则将点Q(x,y)判为其中与其最相似的一类;执行下一步;
4.8令flag=true,标记点Q(x,y)为已分类点,执行下一步;
4.9若flag=false,trigger=true,且Thr>Thr 0,则令Thr=Thr-step,flag=true;返回4.2。
本发明的有益效果是:本发明采用的一种递进式动态高光谱图像分类方法,其原理是通过先找到已知样本点的位置,然后将已知样本邻近的样本进行分类处理,然后再逐步推进扩展实现高光谱图像的高精度分类。具体地,如步骤一至步骤六所示,对已知样本局部邻域内的未知点,会先拿它跟当前的已知样本点进行比对,如果差异很小,将它判为当前已知样本点的类别;如果差异较大,难以决定它是否是同一类,则把它标记为比较难分类的点,先放一放,待周围有更多已知类别点,获得更多信息后,再对它进行分类。如果未知点同时也在其他不同类已知点的领域内,则同时也需要考虑它属于其他类的可能性。待容易分的都分辨完了之后,对留下来的尚未分类的点,先去检视哪些点可分辨性比较高的点,比如说那些邻域内已经有较多已知样本的点(考虑地物分布的局部连续性,这样就可以更充分利用其局部邻域内的点的信息),最后才是那些最难分类的点——如其局部邻域内尚无已知点的点。上面这样一个分类过程,体现了一种由易到难的,渐进式的点扩展式的分类思维。这种渐进式推进的分类方法,充分利 用了高光谱的谱-空间信息,每一步都尽可能避免发生错分,体现了分类过程的动态性和智能性。正是基于这样一种分类思维,本专利提出一种递进式动态高光谱分类方法,相比于现有多数方法,该方法不需要高深的数学知识,兼具计算复杂度低,分类精度高,运行速度快等优点。
附图说明
下面结合附图和实例对本发明作进一步说明。
图1是本发明一种递进式动态高光谱图像分类方法的方法流程图;
图2是高光谱图像AVIRIS Indian pines的一幅假彩色图像。
具体实施方式
本实例选用的高光谱图像为AVIRIS Indian pines。该高光谱图像大小为145×145,拥有220个谱段,均匀覆盖0.2:2.5μm的波谱范围。该高光谱图像包含16类已标注样本。由于被水吸收及信噪比较低的缘故,分类之前,谱段104-108、150-163及220将被去除,而只留下总共200个谱段。图2给出了AVIRIS Indian pines的一幅假彩色图像。
参照图1,一种递进式动态高光谱图像分类方法的具体实施步骤如下:
步骤一:输入高光谱图像数据D∈R 145×145×200及对应的地物标记矩阵L∈R 145×145,D中每个像素即样本用高光谱特征向量表示,样本的维数为200;L(x,y)=c表示图像位置(x,y)的像素点属于第c类(c=1,2…16); 在已经标记的数据中每类随机选取15个样本作为训练样本,其余样本作为测试样本;训练样本构成最初的已知类别点。
步骤二:非连续性检测:为计算D中各像素与其局部邻域像素的差异,本实例先经由主成分分析,通过保留占总能量60%的特征向量,对D进行降维;然后再计算降维后的D中各像素与其大小为23×23的局部邻域均值的欧式距离,并用该欧式距离表示D中各点的非连续度;在假定D中各点的非连续度服从正态分布,并假定40%的像素点为连续像素点的基础上,求出划分D中各点为连续点和非连续点的阈值Th;利用Th二值化D中各点的非连续度,获得非连续性检测结果S,S为0、1矩阵,若S(x,y)=0,则表示(x,y)点为连续点,若S(x,y)=1,则表示(x,y)点为非连续点。
步骤三:对连续点的分类:执行如下循环:
3.1扫描高光谱图像的未分类连续点,若找不到未分类的连续点,则跳转步骤四;若找到未分类连续点,设该点为P(x,y),执行下一步;
3.2若P(x,y)的3×3局部邻域内存在已知类别的点(已知类别点包括已标注的训练样本点和已分类点),则执行下一步;否则返回3.1;
3.3若P(x,y)的3×3局部邻域内只有一类已知类别点,则将点P(x,y)判为该类;若P(x,y)的3×3局部邻域内存在多类已知类别点,则依据最小距离准则,则将点P(x,y)判为其中与其欧式距离最小的已知类别样本所对应的类,执行下一步;
3.4标记点P(x,y)为已分类点,返回3.1。
步骤四:对非连续点以及步骤三未能分类的连续点进行分类:
4.1设定鉴别度阈值Thr=6及其下界Thr 0=1,设定Thr的下降步长step=1,初始化flag=true;
4.2当flag=true时,执行下一步,否则跳转步骤五;
4.3令flag=false,trigger=false;
4.4扫描高光谱图像,若找到未分类点,则记当前未分类点为Q(x,y),执行下一步;否则跳转4.9;
4.5若Q(x,y)的3×3局部邻域内存在已知类别的点(已知类别的点包括已标注的训练样本点和已分类点),记ρ为该邻域内已知类别的点的数目,执行下一步;否则返回4.4;
4.6用ρ表示Q(x,y)的鉴别度,若ρ<Thr,则令trigger=true,返回4.4;否则执行下一步;
4.7若Q(x,y)的3×3局部邻域内只有一类已知类别点,则将点Q(x,y)判为该类;若Q(x,y)的3×3局部邻域内存在多类已知类别点,则依据最小距离准则,将点Q(x,y)判为其中与其欧式距离最小的已知类别样本所对应的类,执行下一步;
4.8令flag=true,标记点Q(x,y)为已分类点,执行下一步;
4.9若flag=false,trigger=true,且Thr>Thr 0,则令Thr=Thr-step,flag=true;返回4.2。
对于高光谱图像数据,经过上述四个步骤的操作,已经可以获得一个很高的分类精度;若经过上述四个步骤仍有未分类点,则继续执 行步骤五:
具体地,对全部剩余待分类点进行分类:本实例使用优化后的距离加权线性回归分类器对剩余的待分类点进行分类;对每一个剩余待分类点b,执行如下步骤5.1至5.4:
5.1计算b的图像坐标位置与第c类(c=1,2…16)各已分类点的图像坐标位置的欧式距离,通过对该距离进行排序,选择出其中与b最近的n c个已分类点,构成第c类的训练样本矩阵,并表示为A c;本实例中取n c=40,n c的取值可以上下浮动一点,其对结果产生的差异并不会太大。
5.2计算距离加权矩阵
Figure PCTCN2018087329-appb-000001
(c=1,2…16),其中(x t,y t)表示b的图像坐标,
Figure PCTCN2018087329-appb-000002
表示A c的第i个样本的图像坐标(i=1,2…n c);
5.3计算
Figure PCTCN2018087329-appb-000003
5.4计算
Figure PCTCN2018087329-appb-000004
并以arg min c{r c|c=1,2,…16}作为b的类别。
步骤六:输出分类结果。
采用本实例方法与国际一流方法进行对比,对比方法包括:
WJCR:Weighted Joint Collaborative Representation(M.Xiong,Q.Ran,W.Li,J.Zou,and Q.Du,Hyperspectral Image Classification Using  Weighted Joint Collaborative Representation,IEEE Geoscience and Remote Sensing Letters,vol.12,no.6,pp.1209-1213,Jun.2015.);
JSaCR:Joint Spatial-Aware Collaborative Representation(J.Jiang,C.Chen,Y.Yu,X.Jiang,and J.Ma,“Spatial-Aware Collaborative Representation for Hyperspectral Remote Sensing Image Classification,”IEEE Geoscience and Remote Sensing Letters,vol.14,no.3,pp.404-408,Mar.2017.);
SC-MK:Superpixel-based Classification via Multiple Kernels(L.Fang,S.Li,W.Duan,J.Ren,and J.A.Benediktsson,“Classification of hyperspectral images by exploiting spectral-patial information of superpixel via multiple kernels,”IEEE Transactions on Geoscience and Remote Sensing,vol.53,no.12,pp.6663-6674,2015.);
R2MK:Region-based Relaxed Multiple Kernel(J.Liu,Z.Wu,Z.Xiao,and J.Yang,“Region-Based Relaxed Multiple Kernel Collaborative Representation for Hyperspectral Image Classification,”IEEE Access,vol.5,pp.20921-20933,2017.)。
对比实验结果如表1所示,表中数据均为20次随机实验的平均值,其中OA(Overall Accuracy)表示各类总体精度,AA(Average Accuracy)表示各类平均精度,KA(Kappa Coefficient of Agreement)表示Kappa一致性系数。从表1可以看出,无论从各类的单类分类精 度看,还是从各类总体精度、各类平均精度、Kappa一致性系数看,本发明的分类精度均显著优于其他方法。
表1:不同方法在AVIRIS Indian Pines数据集上的分类精度(%)对比(每类随机选15个训练样本)。
表1
Figure PCTCN2018087329-appb-000005
本发明提供了一种递进式动态高光谱图像分类方法,以上所述仅是本发明的优选实施方式,应当指出,对于本技术领域的普通技术人员来说,在不脱离本发明原理的前提下,还可以做出若干改进和润饰,这些改进和润饰也应视为本发明的保护范围。本实施例中未明确的各组成部分均可用现有技术加以实现。本发明并不局限于上述实施方式, 只要其以相同的手段达到本发明的技术效果,都应属于本发明的保护范围。

Claims (5)

  1. 一种递进式动态高光谱图像分类方法,其特征在于,包括以下步骤:
    步骤一:读取高光谱图像数据;
    步骤二:计算高光谱图像中各像素与其局部邻域像素的差异,通过设定合适的阈值,进行非连续性检测,划分出连续点和非连续点;
    步骤三,通过将连续点与其局部邻域内已知类别的点进行比较,循环地对未分类的连续点进行分类;
    步骤四,对非连续点以及步骤三中未能分类的连续点循环地进行分类;
    步骤五,对步骤四中全部剩余待分类点进行分类;
    步骤六,输出分类结果。
  2. 根据权利要求1所述的一种递进式动态高光谱图像分类方法,其特征在于,所述步骤一具体为:
    所述高光谱图像数据来源于成像光谱仪采集到的遥感图像,将高光谱图像数据去除被水汽吸收的波段和信噪比较低的波段,设剩余的高光谱图像数据为D∈R m×n×l,其中m、n为高光谱图像的行数和列数,l为剩余的高光谱图像波段数;同时读取各类的训练样本数据。
  3. 根据权利要求2所述的一种递进式动态高光谱图像分类方法,其特征在于,所述步骤二具体为:
    计算D中各像素与其局部邻域像素的差异,通过设定合适的阈值, 将高光谱图像各点划分为连续点和非连续点,其结果为0和1,并建立一个矩阵S∈R m×n用于存储结果,若S(x,y)=0,则表示(x,y)点为连续点,若S(x,y)=1,则表示(x,y)点为非连续点。
  4. 根据权利要求3所述的一种递进式动态高光谱图像分类方法,其特征在于,所述步骤三具体为:
    3.1扫描高光谱图像的未分类连续点,若找不到未分类的连续点,则执行步骤四;若找到未分类连续点,设该点为P(x,y),执行下一步;
    3.2若P(x,y)的局部邻域内存在已知类别的点,则执行下一步;否则返回3.1;
    3.3若P(x,y)的局部邻域内只有一类已知类别点,则将点P(x,y)判为该类;若P(x,y)的局部邻域内存在多类已知类别点,则将点P(x,y)判为其中与其最相似的一类;执行下一步;
    3.4标记点P(x,y)为已分类点,返回3.1。
  5. 根据权利要求4所述的一种递进式动态高光谱图像分类方法,其特征在于,所述步骤四具体为:
    4.1设定鉴别度阈值Thr及其下界Thr 0,设定Thr的下降步长step,引入标记值flag和trigger,初始化flag=true:
    4.2当flag=true时,执行下一步,否则执行步骤五;
    4.3令flag=false,trigger=false;
    4.4扫描高光谱图像,若找到未分类点,则记当前未分类点为Q(x,y),执行下一步;否则执行4.9;
    4.5若Q(x,y)的局部邻域内存在已知类别的点,则执行下一步;否则返回4.4;
    4.6计算Q(x,y)的鉴别度,记为ρ(x,y);若ρ(x,y)<Thr,则trigger=true,返回4.4;否则执行下一步;
    4.7若Q(x,y)的局部邻域内只有一类已知类别点,则将点Q(x,y)判为该类;若Q(x,y)的局部邻域内存在多类已知类别点,则将点Q(x,y)判为其中与其最相似的一类;执行下一步;
    4.8令flag=true,标记点Q(x,y)为已分类点,执行下一步;
    4.9若flag=false,trigger=true,且Thr>Thr 0,则令Thr=Thr-step,flag=true;返回4.2。
PCT/CN2018/087329 2018-05-17 2018-05-17 一种递进式动态高光谱图像分类方法 WO2019218313A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2018/087329 WO2019218313A1 (zh) 2018-05-17 2018-05-17 一种递进式动态高光谱图像分类方法

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2018/087329 WO2019218313A1 (zh) 2018-05-17 2018-05-17 一种递进式动态高光谱图像分类方法

Publications (1)

Publication Number Publication Date
WO2019218313A1 true WO2019218313A1 (zh) 2019-11-21

Family

ID=68539547

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/087329 WO2019218313A1 (zh) 2018-05-17 2018-05-17 一种递进式动态高光谱图像分类方法

Country Status (1)

Country Link
WO (1) WO2019218313A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112115795A (zh) * 2020-08-21 2020-12-22 河海大学 一种基于Triple GAN的高光谱图像分类方法
CN113537150A (zh) * 2021-08-11 2021-10-22 西安交通大学 高光谱图像目标异常检测方法、系统、终端及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101236106A (zh) * 2008-01-11 2008-08-06 北京航空航天大学 一种光谱和空间信息结合的高光谱数据分类方法
CN102096825A (zh) * 2011-03-23 2011-06-15 西安电子科技大学 基于图的半监督高光谱遥感图像分类方法
CN102708374A (zh) * 2012-01-06 2012-10-03 香港理工大学 融合边缘信息和支持向量机对遥感图像进行分类的方法及装置
CN103903010A (zh) * 2014-03-28 2014-07-02 哈尔滨工程大学 基于稀疏特征和邻域同属性的高光谱图像分类方法
CN108764309A (zh) * 2018-05-16 2018-11-06 五邑大学 一种递进式动态高光谱图像分类方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101236106A (zh) * 2008-01-11 2008-08-06 北京航空航天大学 一种光谱和空间信息结合的高光谱数据分类方法
CN102096825A (zh) * 2011-03-23 2011-06-15 西安电子科技大学 基于图的半监督高光谱遥感图像分类方法
CN102708374A (zh) * 2012-01-06 2012-10-03 香港理工大学 融合边缘信息和支持向量机对遥感图像进行分类的方法及装置
CN103903010A (zh) * 2014-03-28 2014-07-02 哈尔滨工程大学 基于稀疏特征和邻域同属性的高光谱图像分类方法
CN108764309A (zh) * 2018-05-16 2018-11-06 五邑大学 一种递进式动态高光谱图像分类方法

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112115795A (zh) * 2020-08-21 2020-12-22 河海大学 一种基于Triple GAN的高光谱图像分类方法
CN112115795B (zh) * 2020-08-21 2022-08-05 河海大学 一种基于Triple GAN的高光谱图像分类方法
CN113537150A (zh) * 2021-08-11 2021-10-22 西安交通大学 高光谱图像目标异常检测方法、系统、终端及存储介质
CN113537150B (zh) * 2021-08-11 2024-04-02 西安交通大学 高光谱图像目标异常检测方法、系统、终端及存储介质

Similar Documents

Publication Publication Date Title
Lu et al. Grid r-cnn
Wang et al. Face r-cnn
Zalpour et al. A new approach for oil tank detection using deep learning features with control false alarm rate in high-resolution satellite imagery
Li et al. Hyperspectral image classification with robust sparse representation
Khoreva et al. Classifier based graph construction for video segmentation
Cai et al. Robust deformable and occluded object tracking with dynamic graph
Tian et al. Spatiotemporal deformable part models for action detection
Zhang et al. Salient object detection based on progressively supervised learning for remote sensing images
CN107480620B (zh) 基于异构特征融合的遥感图像自动目标识别方法
Cao et al. A traffic-sign detection algorithm based on improved sparse r-cnn
CN102622607A (zh) 一种基于多特征融合的遥感图像分类方法
Lim et al. Shape classification using local and global features
Fadaeddini et al. A deep residual neural network for low altitude remote sensing image classification
Xiao et al. Salient object detection based on eye tracking data
Baeta et al. Learning deep features on multiple scales for coffee crop recognition
Ghodrati et al. DeepProposals: Hunting objects and actions by cascading deep convolutional layers
Uijlings et al. Situational object boundary detection
CN113177503A (zh) 基于yolov5的任意朝向目标十二参数检测方法
Qi et al. Exploring illumination robust descriptors for human epithelial type 2 cell classification
WO2019218313A1 (zh) 一种递进式动态高光谱图像分类方法
CN108764309B (zh) 一种递进式动态高光谱图像分类方法
Zhou et al. Etdnet: efficient transformer-based detection network for surface defect detection
Le et al. Document retrieval based on logo spotting using key-point matching
Wang et al. MashFormer: A novel multiscale aware hybrid detector for remote sensing object detection
Sultani et al. Automatic action annotation in weakly labeled videos

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18919002

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18919002

Country of ref document: EP

Kind code of ref document: A1