CN102930269B - Method for modulating multiscale semanteme of remote-sensing image - Google Patents

Method for modulating multiscale semanteme of remote-sensing image Download PDF

Info

Publication number
CN102930269B
CN102930269B CN201210344453.7A CN201210344453A CN102930269B CN 102930269 B CN102930269 B CN 102930269B CN 201210344453 A CN201210344453 A CN 201210344453A CN 102930269 B CN102930269 B CN 102930269B
Authority
CN
China
Prior art keywords
remote sensing
sensing image
scale
pixel
handling object
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.)
Expired - Fee Related
Application number
CN201210344453.7A
Other languages
Chinese (zh)
Other versions
CN102930269A (en
Inventor
张金芳
徐帆江
赵军锁
李亚平
李邦昱
黄志坚
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Institute of Software of CAS
Original Assignee
Institute of Software of CAS
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 Institute of Software of CAS filed Critical Institute of Software of CAS
Priority to CN201210344453.7A priority Critical patent/CN102930269B/en
Publication of CN102930269A publication Critical patent/CN102930269A/en
Application granted granted Critical
Publication of CN102930269B publication Critical patent/CN102930269B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Landscapes

  • Image Analysis (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种遥感影像多尺度语义的调制方法,步骤为:选择遥感影像处理对象的局部特征类型,确定遥感影像处理对象所采用的观察尺度大小范围,划分遥感影像处理对象观察尺度级别,生成单像素特征值序列,生成像素多尺度特征值,生成尺度显化影像。利用本发明原理调制所获取的尺度显化影像效果能更好提高遥感影像处理对象的内部一致性、增强遥感影像处理对象与背景的对比性的目标。试验结果证明基于多尺度特征的尺度显化影像更能比较好的反映地物的实际情况,表明本发明具有很强的使用价值。

A method for modulating multi-scale semantics of remote sensing images, the steps of which are: selecting a local feature type of a remote sensing image processing object, determining the range of observation scales adopted by the remote sensing image processing object, dividing the observation scale level of the remote sensing image processing object, and generating single-pixel features value sequence, generate pixel multi-scale feature values, and generate scale visualization images. By using the principle of the invention to modulate the obtained scale visualization image effect, the internal consistency of the remote sensing image processing object can be better improved, and the contrast between the remote sensing image processing object and the background can be enhanced. The test results prove that the scale visualization image based on multi-scale features can better reflect the actual situation of ground objects, which shows that the present invention has strong application value.

Description

一种遥感影像多尺度语义的调制方法A Modulation Method for Multi-scale Semantics of Remote Sensing Images

技术领域 technical field

本发明涉及一种遥感影像多尺度语义的调制方法,属于面向遥感影像地物目标识别的遥感影像预处理技术,可应用于基于遥感影像的道路、建筑物等地物目标的提取。The invention relates to a modulation method for multi-scale semantics of remote sensing images, which belongs to remote sensing image preprocessing technology for remote sensing image object recognition, and can be applied to the extraction of roads, buildings and other object objects based on remote sensing images.

背景技术 Background technique

遥感影像信息提取技术领域中,尺度有不同的含义,首先指遥感影像所描述的真实地物的几何尺度,这个几何尺度在能够通过遥感影像获取地物信息之前是不知道的;其次是观察尺度,该观察尺度在原始遥感影像上表现为遥感影像的分辨率,观察尺度小(高分辨率)的遥感影像可以依据成像原理生成观察尺度大(低分辨率)的遥感影像,因而高分辨率遥感影像为多尺度观察研究提供了基础;最后是在遥感影像分割研究中的尺度,尺度一般指分割的遥感影像对象大小(或控制分割单元大小的参量),控制分割单元大小的参数一般为内部一致度参数,尽管一致度与分割的单元大小之间存在着密切的关系,但是很难明确描述,这也是eCognition软件中最难以操控的方面,必须通过反复试验来获取合适的尺度参数值。In the field of remote sensing image information extraction technology, scale has different meanings. First, it refers to the geometric scale of the real ground objects described by remote sensing images. This geometric scale is unknown before the ground feature information can be obtained through remote sensing images; secondly, it refers to the observation scale. , the observation scale is expressed as the resolution of the remote sensing image on the original remote sensing image. A remote sensing image with a small observation scale (high resolution) can generate a remote sensing image with a large observation scale (low resolution) according to the imaging principle. Therefore, high resolution remote sensing The image provides the basis for multi-scale observation research; the last is the scale in the remote sensing image segmentation research. The scale generally refers to the size of the segmented remote sensing image object (or the parameter that controls the size of the segmentation unit), and the parameter that controls the size of the segmentation unit is generally internally consistent. Although there is a close relationship between the degree of coherence and the size of the segmented unit, it is difficult to describe it clearly. This is also the most difficult aspect of eCognition software to manipulate. It must be obtained by trial and error to obtain the appropriate value of the scale parameter.

所有的遥感影像特征都是某个尺度下的特征,即尺度遥感影像特征,一般简称为遥感影像特征或特征,当研究多个尺度遥感影像时,称之为多尺度特征,但在研究中往往是分别进行单尺度研究最后再综合考虑各个尺度的特征生成最终结果。分形对象具有特征不随尺度变化的特性;更普遍的情况是某个特征只在一定尺度范围内可以保持不变,SIFT方法使用了该特征;遥感影像上地物关系复杂,特征随着尺度的变化而变化,且与其赋存环境密切相关。All remote sensing image features are features at a certain scale, that is, scale remote sensing image features, which are generally referred to as remote sensing image features or features. When studying multiple scale remote sensing images, they are called multi-scale features. It is to conduct single-scale research separately and then comprehensively consider the features of each scale to generate the final result. Fractal objects have the characteristic that their features do not change with scale; more generally, a certain feature can only remain unchanged within a certain scale range, and the SIFT method uses this feature; the relationship between ground objects on remote sensing images is complex, and features change with scale It changes and is closely related to its environment.

在涉及到的多尺度遥感影像识别专利文献中,其处理方法一般是在各个单尺度图像下进行识别,最后将不同尺度条件下的识别结果合并;在涉及到的多尺度遥感影像分割中,一般采用尺度递增的顺序依次产生不同尺度的分割结果,最后将不同尺度条件下的分割结果合并,可以参见200710304466.0,200810032390.5,201110137557.6,如图1所示。在上述专利中,使用尺度进行分析时,尺度特征是隐含在遥感影像上下文中的,而且在应用中使用的是单个的尺度特征,而没有考虑多尺度的变化特征。而作为高分辨率遥感影像,一方面精度高要反映遥感影像的细部特征,另一方面遥感影像的区域都比较大,因而遥感影像上具有多尺度的特性,多尺度特性是遥感目标识别的重要依据,如果生成的尺度显化特征影像不能显式地描述了该尺度特征,则会影响遥感影像的图像增强能力。In the related multi-scale remote sensing image recognition patent documents, the processing method is generally to recognize each single-scale image, and finally combine the recognition results under different scale conditions; in the multi-scale remote sensing image segmentation involved, generally Segmentation results of different scales are sequentially generated in the order of increasing scale, and finally the segmentation results under different scale conditions are combined, as shown in Figure 1. In the above-mentioned patents, when using scales for analysis, scale features are implicit in the context of remote sensing images, and a single scale feature is used in the application without considering multi-scale change features. As a high-resolution remote sensing image, on the one hand, high precision should reflect the detailed features of the remote sensing image; Based on this, if the generated image with scale explicit feature cannot explicitly describe the scale feature, it will affect the image enhancement ability of remote sensing image.

发明内容 Contents of the invention

本发明技术解决问题:克服现有技术的不足,提供一种遥感影像多尺度语义的调制方法,通过一种多尺度语义调制方法将遥感影像尺度特征显化到遥感影像的像素中,生成尺度显化影像,生成的尺度显化影像显式地描述了遥感影像尺度特征,能够反映遥感影像的尺度变化特征,属于高阶遥感影像特征,用于高层语义遥感影像对象的识别。The technology of the present invention solves the problem: overcomes the deficiencies of the prior art, and provides a multi-scale semantic modulation method for remote sensing images. Through a multi-scale semantic modulation method, the scale features of remote sensing images are manifested into the pixels of remote sensing images, and the scale display is generated. The generated scale explicit image explicitly describes the scale characteristics of remote sensing images, which can reflect the scale change characteristics of remote sensing images, belongs to high-level remote sensing image features, and is used for the recognition of high-level semantic remote sensing image objects.

本发明解决其技术问题所采用的技术方案如下:一种遥感影像多尺度语义的调制方法,其实现步骤如下:The technical solution adopted by the present invention to solve the technical problem is as follows: a modulation method for multi-scale semantics of remote sensing images, and its realization steps are as follows:

(1)选定遥感影像处理对象以及选定遥感影像处理对象的局部特征类型;所述遥感影像处理对象是指在遥感影像中,一个感兴趣的待处理区域;遥感影像处理对象的局部特征反映在该遥感影像处理对象的局部范围内与其他相邻遥感影像处理对象的相对特征,遥感影像处理对象的局部特征也反映在该遥感影像处理对象所包含的像素集之间以及与其他相邻遥感影像处理对象所包含的像素间的相对特征,利用遥感影像处理对象的局部特征信息作为判断遥感影像处理对象间的相似性依据;(1) Select the remote sensing image processing object and the local feature type of the selected remote sensing image processing object; the remote sensing image processing object refers to an area of interest to be processed in the remote sensing image; the local feature of the remote sensing image processing object reflects The relative characteristics of the remote sensing image processing object and other adjacent remote sensing image processing objects in the local range of the remote sensing image processing object, the local characteristics of the remote sensing image processing object are also reflected in the pixel sets contained in the remote sensing image processing object and the relative characteristics of other adjacent remote sensing image processing objects. The relative features between the pixels contained in the image processing object, using the local feature information of the remote sensing image processing object as the basis for judging the similarity between the remote sensing image processing objects;

(2)在选定遥感影像处理对象的基础上,根据选定遥感影像处理对象的几何尺度大小确定所使用的遥感影像处理对象的观察尺度的大小;(2) On the basis of selecting the remote sensing image processing object, determine the observation scale of the remote sensing image processing object used according to the geometric scale of the selected remote sensing image processing object;

(3)采用尺度特征曲线以连续的方式表达遥感影像处理对象所包含的像素在观察尺度范围内的局部特征变化情况;针对尺度特征曲线采用离散的形式进行采样和编码,将步骤2中所确定遥感影像处理对象的观察尺度大小范围按照m倍级数划分成n个观察尺度;(3) Use the scale characteristic curve to express the local characteristic changes of the pixels contained in the remote sensing image processing object in a continuous manner within the observation scale range; for the scale characteristic curve, use discrete form to sample and code, and the determined in step 2 The observation scale size range of the remote sensing image processing object is divided into n observation scales according to the m-fold series;

(4)根据步骤1的遥感影像处理对象局部特征类型,计算遥感影像中每个像素在步骤3中所划分的n个观察尺度下的局部特征,遥感影像中每个像素在n个观察尺度下所获得的局部特征成为该像素的单像素特征值序列;遥感影像中每个像素都有一个该像素的单像素特征值序列;(4) According to the local feature type of the remote sensing image processing object in step 1, calculate the local features of each pixel in the remote sensing image under the n observation scales divided in step 3, and each pixel in the remote sensing image is under n observation scales The obtained local features become the single-pixel feature value sequence of the pixel; each pixel in the remote sensing image has a single-pixel feature value sequence of the pixel;

(5)对步骤4中所获得的遥感影像中每个像素的单像素特征值序列需要进行编码概括为该像素的一个特征值,作为遥感影像中该像素的多尺度特征值,以描述遥感影像像素点上的多尺度特征,遥感影像中一个像素多尺度特征值既能反映出像素在某观察尺度的局部特征,同时也能反映该像素在观察尺度范围的每个尺度上的局部特征;(5) The single-pixel feature value sequence of each pixel in the remote sensing image obtained in step 4 needs to be encoded and summarized as a feature value of the pixel, which is used as the multi-scale feature value of the pixel in the remote sensing image to describe the remote sensing image The multi-scale feature on the pixel point, the multi-scale feature value of a pixel in the remote sensing image can not only reflect the local feature of the pixel at a certain observation scale, but also reflect the local feature of the pixel at each scale of the observation scale range;

(6)对步骤5中,遥感影像中每个像素都生成一个多尺度特征值,以该像素的多尺度特征值作为该像素的像素值,形成一幅与原始遥感影像大小相同,该像素值为该像素在原始遥感影像中所表现出的多尺度特征值的遥感影像,即尺度显化影像。(6) For step 5, a multi-scale feature value is generated for each pixel in the remote sensing image, and the multi-scale feature value of the pixel is used as the pixel value of the pixel to form a picture with the same size as the original remote sensing image, and the pixel value is the remote sensing image of the multi-scale feature values shown by the pixel in the original remote sensing image, that is, the scale explicit image.

所述步骤(1)中的采用遥感影像处理对象的观察尺度大小的范围是遥感影像处理对象几何尺度大小的范围的2倍。In the step (1), the range of the observation scale of the remote sensing image processing object is twice the range of the geometric scale of the remote sensing image processing object.

所述步骤(3)中的m=2。m=2 in the step (3).

本发明与现有技术相比的优点在于:The advantage of the present invention compared with prior art is:

(1)传统多尺度遥感影像处理中,是对多个尺度进行面向最终目标的遥感影像处理,并将各个处理结果简单组合,无法使用尺度变化特征;本发明生成的尺度显化特征影像显式地描述了该特征,可以反映遥感影像的尺度变化特征,属于高阶遥感影像特征,可以用于高层语义对象的识别。(1) In the traditional multi-scale remote sensing image processing, the remote sensing image processing is performed on multiple scales for the ultimate goal, and the processing results are simply combined, and the scale change feature cannot be used; the scale manifest feature image generated by the present invention is explicit This feature is described accurately, which can reflect the scale change characteristics of remote sensing images, belongs to high-order remote sensing image features, and can be used for the recognition of high-level semantic objects.

(2)本发明的尺度显化影像以尺度作为调制框架,在遥感影像处理对象特征类型选为亮暗特征类型下进行本发明所述方法进行调制所获得的尺度显化影像,可以保留原始影像的光谱特性。(2) The scale manifested image of the present invention uses scale as the modulation frame, and the scale manifested image obtained by modulating the method of the present invention under the feature type of the remote sensing image processing object selected as the bright and dark feature type can retain the original image spectral properties.

(3)传统多尺度遥感影像处理中,大尺度下的特征因为分辨率降低往往偏离真正的位置点;如步骤6中所述,而本发明的尺度显化影像上的特征都是基于原始遥感影像中每个像素点生成多尺度特征值,形成尺度显化影像,因而保持了原始遥感影像的位置精度。(3) In traditional multi-scale remote sensing image processing, the features at large scales often deviate from the real position points due to the reduction of resolution; as described in step 6, the features on the scale visualization images of the present invention are all based on the original remote sensing Each pixel in the image generates multi-scale eigenvalues to form a scale explicit image, thus maintaining the position accuracy of the original remote sensing image.

(4)如步骤4和步骤5中所述,本发明的尺度显化特征影像中每个像素值是该像素多尺度特征值,是由该像素的单尺度特征序列概括而成,可以在概括单尺度特征序列时很方便的选择部分或者特定的几个尺度进行概括为该像素的多尺度特征值,从而实现对尺度的加强或抑制,以突出某个尺度对象的空间对比度,这在传统方法中很难做到。(4) As mentioned in step 4 and step 5, each pixel value in the scale-manifestation feature image of the present invention is the multi-scale feature value of the pixel, which is summarized by the single-scale feature sequence of the pixel, and can be summarized in When using a single-scale feature sequence, it is very convenient to select some or specific scales to summarize as the multi-scale feature value of the pixel, so as to realize the enhancement or suppression of the scale, so as to highlight the spatial contrast of a certain scale object, which is in the traditional method. difficult to do.

(5)本发明的尺度显化影像中的像素值反映的是该像素在原始影像中的多尺度的表现,因而尺度显化影像中相邻像素间的相似性反映了在原始遥感影像中这些像素是否属于同一遥感影像处理对象,据此可以将本发明的尺度显化影像中的多尺度特征应用于分类或分割研究中,这是一种新的相似性定义,可以弥补传统方法中缺少尺度变化相似性不足,使分割或分类精度更高。(5) The pixel value in the scale visualization image of the present invention reflects the multi-scale performance of the pixel in the original image, so the similarity between adjacent pixels in the scale visualization image reflects these pixels in the original remote sensing image. Whether the pixels belong to the same remote sensing image processing object, based on this, the multi-scale features in the scale visualization image of the present invention can be applied to classification or segmentation research. This is a new definition of similarity, which can make up for the lack of scale in traditional methods. Insufficient variation similarity enables higher segmentation or classification accuracy.

附图说明 Description of drawings

图1为通常尺度处理过程示意图;Figure 1 is a schematic diagram of the usual scale processing process;

图2为使用本发明尺度处理过程示意图;Fig. 2 is a schematic diagram of the scale processing process using the present invention;

图3为本发明的尺度显化影像调制过程流程图;Fig. 3 is a flow chart of the scale visualization image modulation process of the present invention;

图4为本发明对象在邻域范围内的光谱剖面,其中W为像素点或遥感影像处理对象,A、B、C为其不同邻域大小;Fig. 4 is the spectral profile of the object of the present invention within the neighborhood range, wherein W is a pixel point or a remote sensing image processing object, and A, B, and C are their different neighborhood sizes;

图5为本发明对象的相对亮度-尺度特征变化曲线;Fig. 5 is the relative brightness-scale feature change curve of the object of the present invention;

图6为本发明中调制灯柱模型;Fig. 6 is the modulating lamp post model in the present invention;

图7为本发明的尺度显化影像效果图,左上图为原始遥感影像,右上图为左上图对应的尺度显化影像,左下图为另一幅原始遥感影像,右下图为左下图对应的尺度显化影像。从中可以看出,尺度显化影像能更好的表达遥感影像处理对象。Fig. 7 is an effect diagram of the scale visualization image of the present invention, the upper left image is the original remote sensing image, the upper right image is the scale visualization image corresponding to the upper left image, the lower left image is another original remote sensing image, and the lower right image is the corresponding image of the lower left image Scale visualization image. It can be seen that the scale explicit image can better express the object of remote sensing image processing.

具体实施方式 Detailed ways

本发明属于面向遥感影像地物目标提取的遥感影像预处理技术,用于显化遥感影像处理对象的空间尺度语意,它通过一种多尺度语义调制方法将遥感影像尺度特征显化到遥感影像的像素中,生成尺度显化影像,生成的尺度显化影像显式地描述了遥感影像尺度特征,能够反映遥感影像的尺度变化特征,属于高阶遥感影像特征,可用于高层语义遥感影像对象的识别。The present invention belongs to remote sensing image preprocessing technology oriented to remote sensing image object extraction, which is used to visualize the spatial scale semantics of remote sensing image processing objects. It manifests the remote sensing image scale features to the remote sensing image through a multi-scale semantic modulation method. In the pixel, a scale explicit image is generated. The generated scale explicit image explicitly describes the scale characteristics of the remote sensing image, which can reflect the scale change characteristics of the remote sensing image. It belongs to the high-level remote sensing image feature and can be used for the recognition of high-level semantic remote sensing image objects. .

尺度显化影像调制过程流程图如图3。其中包括选择能表达遥感影像处理对象的局部特征类型,针对具体遥感影像处理对象确定观察尺度大小范围和观察尺度等级,在不同观察尺度下提取局部特征值,形成像素的单尺度特征序列,并对单尺度特征值序列进行概括形成像素的尺度特征值,最后形成尺度显化影像。The flow chart of the image modulation process for scale visualization is shown in Figure 3. It includes selecting the local feature type that can express the remote sensing image processing object, determining the observation scale size range and observation scale level for the specific remote sensing image processing object, extracting local feature values at different observation scales, forming a single-scale feature sequence of pixels, and The sequence of single-scale eigenvalues is summarized to form the scale eigenvalues of the pixels, and finally the scale manifested image is formed.

如图2、3所示,本发明具体步骤如下。As shown in Figures 2 and 3, the specific steps of the present invention are as follows.

1.选择遥感影像处理对象的局部特征类型1. Select the local feature type of remote sensing image processing object

选定遥感影像处理对象以及选定遥感影像处理对象的局部特征类型;遥感影像处理对象的局部特征反映在该遥感影像处理对象的局部范围内与其他相邻遥感影像处理对象的相对特征,遥感影像处理对象的局部特征也反映在该遥感影像处理对象所包含的像素集之间以及与其他相邻遥感影像处理对象所包含的像素间的相对特征,利用遥感影像处理对象的局部特征信息作为判断遥感影像处理对象间的相似性依据。局部特征区分为一致性特征和对比特征,一致性特征是在遥感影像中的一定区域内遥感影像处理对象间具有的共同特征,例如均值、偏差、纹理等,对比特征是在遥感影像处理对象与周围邻域内其它遥感影像处理对象进行对比而呈现出来的特征,例如明暗、多寡、前景-背景、线性方向等。The selected remote sensing image processing object and the local feature type of the selected remote sensing image processing object; the local features of the remote sensing image processing object reflect the relative characteristics of the remote sensing image processing object and other adjacent remote sensing image processing objects within the local range of the remote sensing image processing object. The local characteristics of the processing object are also reflected in the relative characteristics between the pixel sets contained in the remote sensing image processing object and the pixels contained in other adjacent remote sensing image processing objects. Similarity basis between image processing objects. Local features are divided into consistency features and contrast features. Consistency features are common features between remote sensing image processing objects in a certain area of remote sensing images, such as mean value, deviation, texture, etc. Contrast features are between remote sensing image processing objects and remote sensing image processing objects. The characteristics presented by comparing other remote sensing image processing objects in the surrounding neighborhood, such as light and shade, amount, foreground-background, linear direction, etc.

2.确定遥感影像处理对象所采用的观察尺度大小范围2. Determine the range of observation scales used for remote sensing image processing objects

在选定观察的遥感影像处理对象的基础上,根据选定的遥感影像处理对象的几何尺度大小确定所使用的遥感影像处理对象的观察尺度的大小;本发明所采用遥感影像处理对象的观察尺度大小的范围是遥感影像处理对象几何尺度大小的范围的2倍;该步骤主要针对遥感影像处理对象几何尺度大小确定遥感影像处理对象的观察尺度大小范围。要观察一个遥感影像处理对象,那么遥感影像处理对象的观察尺度大小至少要大于该遥感影像处理对象本身的几何尺度。观察尺度范围太小,不能反映遥感影像处理对象的几何尺度大小;观察尺度大小范围太大,则不足以精确表达该遥感影像处理对象的局部范围内的局部特征,因而本发明综合考虑能精确表达遥感影像处理对象局部特征前提下,同时考虑实现效率的问题,确定遥感影像处理对象的观察尺度大小范围为其几何尺度大小范围的2倍左右,即观察尺度的最小值是遥感影像处理对象细部特征尺度的一半,最大值是整体遥感影像处理对象几何尺度的2倍,对于无限延伸遥感影像处理对象,例如道路,以其宽度作为几何尺度依据。On the basis of the remote sensing image processing object selected for observation, determine the size of the observation scale of the remote sensing image processing object used according to the geometric scale size of the selected remote sensing image processing object; the observation scale of the remote sensing image processing object adopted in the present invention The size range is twice the range of the geometric scale size of the remote sensing image processing object; this step mainly determines the observation scale size range of the remote sensing image processing object according to the geometric scale size of the remote sensing image processing object. To observe a remote sensing image processing object, the observation scale of the remote sensing image processing object must be at least larger than the geometric scale of the remote sensing image processing object itself. If the observation scale range is too small, it cannot reflect the geometric scale of the remote sensing image processing object; if the observation scale range is too large, it is not enough to accurately express the local features in the local range of the remote sensing image processing object, so the present invention can accurately express Under the premise of the local characteristics of the remote sensing image processing object, and considering the problem of realization efficiency, the observation scale size range of the remote sensing image processing object is determined to be about 2 times the geometric scale size range, that is, the minimum value of the observation scale is the detailed feature of the remote sensing image processing object half of the scale, and the maximum value is twice the geometric scale of the overall remote sensing image processing object. For infinitely extending remote sensing image processing objects, such as roads, the geometric scale is based on its width.

3.划分遥感影像处理对象观察尺度级别3. Divide the observation scale level of remote sensing image processing objects

采用尺度特征曲线以连续的方式表达遥感影像处理对象所包含的像素在观察尺度范围内的局部特征变化情况;以遥感影像处理对象所包含的像素的一个像素点来说明如图4所示,其中P、Q、W为像素点或遥感影像处理对象,A、B、C为其不同邻域大小。一点对于其邻域来说可以表现为亮或暗,而且范围不同其亮暗也不一样,图中W的位置在A邻域内表现为暗,而在B邻域内则表现为亮,更大范围的C邻域内则有表现为暗,这种“暗-亮-暗”的组合就完整地表达了该点的空间尺度变化属性,同单纯的像素值相比表达了更多的语义关系,可以理解为该点是亮区内的暗斑中的一个相对亮的点。图5显示像素点在不同尺度下其相对其邻域内亮度特征的变化曲线。在离散尺度处理中将会得到一个尺度特征序列。The scale characteristic curve is used to express the local feature changes of the pixels contained in the remote sensing image processing object in a continuous manner within the observation scale; a pixel point of the pixel contained in the remote sensing image processing object is used to illustrate as shown in Figure 4, where P, Q, W are pixels or remote sensing image processing objects, and A, B, C are different neighborhood sizes. A point can be bright or dark for its neighborhood, and its brightness and darkness are different in different ranges. The position of W in the figure is dark in the neighborhood of A, and bright in the neighborhood of B. In the C neighborhood of , it is dark. This combination of "dark-bright-dark" completely expresses the spatial scale change attribute of the point, and expresses more semantic relationships than simple pixel values. It is understood that the point is a relatively bright point in the dark spot in the bright area. Fig. 5 shows the variation curves of the pixel point relative to the brightness characteristics in its neighborhood at different scales. In discrete scale processing, a sequence of scale features will be obtained.

尺度特征曲线能够很好表达像素在观察尺度范围内的不同观察尺度下特征值的连续变化情况,但在计算机中仅仅能采用离散的形式,因而需要进行采样和编码,即在观察尺度范围内划分观察尺度级别。本发明在确定遥感影像处理对象观察尺度大小范围基础上,将步骤2中所确定遥感影像处理对象的观察尺度大小范围按照2倍级数划分成n个观察尺度。这样便于遥感影像的表达与后期处理。即将观察尺度空间划分成n个级别,高一级的观察尺度大小是相邻较低一级的观察尺度大小的2倍,这样划分在操作上便于进行实现,而在理论上,观察尺度是前一个观察尺度的2倍大小后,更能够精确而又不冗余的反映出遥感影像处理对象几何尺度信息。The scale characteristic curve can well express the continuous change of the characteristic value of the pixel at different observation scales within the observation scale range, but it can only be used in a discrete form in the computer, so sampling and encoding are required, that is, divided within the observation scale range Observe the scale level. In the present invention, on the basis of determining the observation scale range of the remote sensing image processing object, the observation scale range of the remote sensing image processing object determined in step 2 is divided into n observation scales according to the double series. This facilitates the expression and post-processing of remote sensing images. That is, the observation scale space is divided into n levels, and the observation scale of the higher level is twice the size of the observation scale of the adjacent lower level. After twice the size of an observation scale, it can reflect the geometric scale information of the remote sensing image processing object more accurately and without redundancy.

4.生成单像素特征值序列4. Generate a sequence of single-pixel eigenvalues

根据步骤1的遥感影像处理对象局部特征类型,计算遥感影像中每个像素在步骤3中所划分的n个观察尺度下的局部特征,遥感影像中每个像素在n个观察尺度下所获得的局部特征成为该像素的单像素特征值序列;遥感影像中每个像素都有一个该像素的单像素特征值序列。According to the local feature type of the remote sensing image processing object in step 1, calculate the local features of each pixel in the remote sensing image under the n observation scales divided in step 3, and obtain the n observation scales for each pixel in the remote sensing image The local feature becomes the single-pixel feature value sequence of the pixel; each pixel in the remote sensing image has a single-pixel feature value sequence of the pixel.

5.生成像素多尺度特征值5. Generate pixel multi-scale feature values

对步骤4中所获得的遥感影像中每个像素的单像素特征值序列需要进行编码概括为该像素的一个特征值,作为遥感影像中该像素的多尺度特征值,以描述遥感影像像素点上的多尺度特征,遥感影像中一个像素多尺度特征值既能反映出像素在某观察尺度的局部特征,同时也能反映该像素在观察尺度范围的每个尺度上的局部特征。本发明表达多尺度特征值的编码表示为:其中,S为形成的一个像素的多尺度特征值,即整体特征值;k为观察尺度级别k=1…n;tk为第k个观察尺度的开关;wk为第k个观察尺度权重,观察尺度权重反映了不同观察下所表现的尺度特征对整体特征值的影响;fk为第k观察尺度下的尺度特征值,反映像素在第k观察尺度下的观察特征。当相邻尺度特征的权重为2倍关系时即,wk=2k,可以使用一个bit位来描述单个观察尺度下的特征,实现了单个观察尺度下特征的显式描述,而整体特征值则描述多尺度的综合特征。The single-pixel feature value sequence of each pixel in the remote sensing image obtained in step 4 needs to be encoded and summarized as a feature value of the pixel, which is used as the multi-scale feature value of the pixel in the remote sensing image to describe the remote sensing image pixel. The multi-scale features of a pixel in a remote sensing image can not only reflect the local features of the pixel at a certain observation scale, but also reflect the local features of the pixel at each scale of the observation scale range. The present invention expresses the encoding of multi-scale eigenvalues as: Among them, S is the multi-scale eigenvalue of a pixel formed, that is, the overall eigenvalue; k is the observation scale level k=1…n; t k is the switch of the k-th observation scale; w k is the weight of the k-th observation scale , the observation scale weight reflects the influence of the scale features expressed under different observations on the overall eigenvalue; f k is the scale feature value at the kth observation scale, reflecting the observation characteristics of the pixel at the kth observation scale. When the weight of adjacent scale features is 2 times the relationship, that is, w k = 2 k , one bit can be used to describe the features at a single observation scale, realizing the explicit description of features at a single observation scale, while the overall feature value Then describe the comprehensive characteristics of multi-scale.

6.生成尺度显化影像6. Generate scale visualization images

在步骤5中,遥感影像中每个像素都生成一个多尺度特征值,以该像素的多尺度特征值作为该像素的像素值,形成一幅与原始遥感影像大小相同,但像素值为该像素在原始遥感影像中所表现出的多尺度特征值的遥感影像,即尺度显化影像。之所以称之为显化影像是因为每个像素点的像素值已经不是简单的辐射亮度值或相对亮度值,而是一种反应该点多尺度观察下的特征值。In step 5, a multi-scale feature value is generated for each pixel in the remote sensing image, and the multi-scale feature value of the pixel is used as the pixel value of the pixel to form a picture with the same size as the original remote sensing image, but the pixel value is the pixel The remote sensing image with multi-scale eigenvalues shown in the original remote sensing image is the scale explicit image. The reason why it is called a manifested image is that the pixel value of each pixel point is no longer a simple radiance value or relative brightness value, but a characteristic value reflecting the multi-scale observation of the point.

对如上6个步骤的过程,称之为调制过程,即将原始反应地面辐射强度的像素值,从灰度值域,通过不同观察尺度下的观察结果(单尺度特征)调制到多尺度特征值域。在实现上,观察尺度之间采用等比例关系,相邻观察尺度特征值的权重比选择2倍关系,这样的尺度显化特征影像的调制过程可以通过灯柱模型实现。基于亮暗特征的调制方法为:如图6所示,设想一架立在像素中心的一个柱子,将原始像素值的二进制数从下到上依次排列,在每一个二进制位上设置一个灯,每盏灯的照亮范围是其下面相邻灯得照亮范围的2倍,如果对应的二进制位为1,则该位等打开,否则该位灯关闭,则总亮度代表原始辐射灰度值。调制时,对每个二进制位上灯的打开与关闭基于如下的判断原则,即如果该点的灰度值在该位等照射范围内属于亮像素,则打开该灯,否则关闭。这样调制后的二进制位序列反映的是该点上的基于亮暗特征的多尺度特征。其它特征的调制方法相同。The process of the above six steps is called the modulation process, that is, the original pixel value reflecting the ground radiation intensity is modulated from the gray scale value range to the multi-scale feature value range through the observation results (single-scale features) at different observation scales . In terms of implementation, an equal-proportional relationship is adopted between the observation scales, and the weight ratio of the feature values of adjacent observation scales is doubled. The modulation process of such scale-manifested feature images can be realized through the lamp post model. The modulation method based on bright and dark features is as follows: As shown in Figure 6, imagine a pillar standing in the center of the pixel, arrange the binary numbers of the original pixel values from bottom to top, and set a light on each binary bit, The illumination range of each lamp is twice the illumination range of the adjacent lamps below it. If the corresponding binary bit is 1, the bit is turned on, otherwise the bit is turned off, and the total brightness represents the original radiance gray value . During modulation, the turning on and off of the light on each binary bit is based on the following judgment principle, that is, if the gray value of the point belongs to a bright pixel within the irradiation range of the bit, the light is turned on, otherwise it is turned off. The modulated binary bit sequence reflects the multi-scale features based on bright and dark features at the point. The modulation methods of other features are the same.

在灯柱模型中,尺度显化影像上的二进制像素值表示了各个尺度层次的影像特征,只打开一层灯(通过屏蔽操作只保留该位的值)则尺度显化影像显示的是在该观察尺度下观察到的遥感影像特征。当把高位的灯依次熄灭,尺度显化影像描述的遥感影像特征的观察尺度将逐渐减小,最后只看到微细纹理,而不见整体对象;反之小纹理消失,大对象凸显。In the lamppost model, the binary pixel values on the scale-displayed image represent the image characteristics of each scale level, and only one layer of light is turned on (only the value of this bit is retained through the masking operation), and the scale-displayed image shows the image at that level Remote sensing image features observed at the observation scale. When the high-level lights are turned off in turn, the observation scale of the remote sensing image features described by the scale-manifested image will gradually decrease, and finally only the fine textures can be seen, but not the overall object; otherwise, the small textures disappear and the large objects stand out.

尺度显化影像通过综合考虑多尺度特征增强了地物目标对象间的差别,可以用来更准确进行遥感影像分割,分割过程中的相似度测量可以是综合多尺度特征值,也可以是多尺度特征位描述的尺度变化形态。以综合多尺度特征值作为相似度度量标准时,分割的结果具有综合观察的明暗一致性;以多尺度变化形态作为相似度度量标准时,可以将尺度特征形态变化一致的像素点归为一类,实现不同尺度同类遥感影像处理对象的统一。Scale visualization images enhance the differences between ground objects and objects by comprehensively considering multi-scale features, which can be used to segment remote sensing images more accurately. The similarity measurement in the segmentation process can be comprehensive multi-scale feature values or multi-scale The shape of the scale change described by the feature bit. When the comprehensive multi-scale feature value is used as the similarity measure standard, the segmentation result has the light and dark consistency of comprehensive observation; when the multi-scale change form is used as the similarity measure standard, the pixels with consistent change in scale feature form can be classified into one category to realize Unification of similar remote sensing image processing objects at different scales.

利用本发明原理调制所获取的尺度显化影像效果图7所示,左上图为原始遥感影像,右上图为左上图对应的尺度显化影像,左下图为另一幅原始遥感影像,右下图为左下图对应的尺度显化影像。从中可以看出,尺度显化影像能更好的表达遥感影像处理对象,从中可以看出,尺度显化影像能更好提高遥感影像处理对象的内部一致性、增强遥感影像处理对象与背景的对比性的目标。试验结果证明基于多尺度特征的尺度显化影像更能比较好的反映地物的实际情况,表明本发明具有很强的使用价值。Figure 7 shows the effect of the scale-manifested image obtained by modulating the principle of the present invention. The upper left picture is the original remote sensing image, the upper right picture is the scale-manifested image corresponding to the upper left picture, the lower left picture is another original remote sensing image, and the lower right picture It is the scale visualization image corresponding to the lower left image. It can be seen that the scale explicit image can better express the remote sensing image processing object, and it can be seen that the scale explicit image can better improve the internal consistency of the remote sensing image processing object and enhance the contrast between the remote sensing image processing object and the background sexual goals. The test results prove that the scale visualization image based on multi-scale features can better reflect the actual situation of ground objects, which shows that the present invention has strong application value.

利用本发明原理调制所获取的尺度显化影像效果能更好提高遥感影像处理对象的内部一致性、增强遥感影像处理对象与背景的对比性的目标。试验结果证明基于多尺度特征的尺度显化影像更能比较好的反映地物的实际情况,表明本发明具有很强的使用价值。By using the principle of the invention to modulate the obtained scale visualization image effect, the internal consistency of the remote sensing image processing object can be better improved, and the contrast between the remote sensing image processing object and the background can be enhanced. The test results prove that the scale visualization image based on multi-scale features can better reflect the actual situation of ground objects, which shows that the present invention has strong application value.

本发明说明书中未作详细描述的内容属于本领域专业技术人员公知的现有技术。The contents not described in detail in the description of the present invention belong to the prior art known to those skilled in the art.

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。The above is only a preferred embodiment of the present invention, it should be pointed out that, for those of ordinary skill in the art, without departing from the principle of the present invention, some improvements and modifications can also be made, and these improvements and modifications can also be made. It should be regarded as the protection scope of the present invention.

Claims (3)

1. a modulator approach for remote sensing image multiscale semanteme, is characterized in that performing step is as follows:
(1) the local feature type of selected remote sensing image handling object and selected remote sensing image handling object; Described remote sensing image handling object refers in remote sensing image, an interested pending region; The local feature of remote sensing image handling object is reflected in the relative characteristic of the interior remote sensing image handling object adjacent with other of subrange of this remote sensing image handling object, relative characteristic between the pixel that the local feature of remote sensing image handling object is also reflected between set of pixels that this remote sensing image handling object comprises and remote sensing image handling object adjacent with other comprises, utilizes the local feature information of remote sensing image handling object as the similarity foundation judged between remote sensing image handling object;
(2) on the basis of selected remote sensing image handling object, the size of the observation yardstick of used remote sensing image handling object is determined according to the geometric scale size of selected remote sensing image handling object;
(3) employing scale feature curve expresses the local feature situation of change of pixel in observation range scale that remote sensing image handling object comprises in a continuous manner; Adopt discrete form to carry out sampling and encoding for scale feature curve, by step 2 determine the observation scale size scope of remote sensing image handling object according to m times of progression be divided into n observe yardstick;
(4) according to the remote sensing image handling object local feature type of step 1, calculate the local feature under n divided in step 3 observation yardstick of each pixel in remote sensing image, in remote sensing image, each pixel becomes single pixel characteristic value sequence of this pixel at the local feature that n obtains under observing yardstick; In remote sensing image, each pixel has single pixel characteristic value sequence of this pixel;
(5) need to encode to be summarised as an eigenwert of this pixel to single pixel characteristic value sequence of pixel each in the remote sensing image obtained in step 4, as the Analysis On Multi-scale Features value of this pixel in remote sensing image, to describe the Analysis On Multi-scale Features on remote sensing image pixel, in remote sensing image, a pixel Analysis On Multi-scale Features value can reflect that pixel observes the local feature of yardstick at certain, also can reflect the local feature of this pixel on each yardstick observing range scale simultaneously;
(6) in step 5, in remote sensing image, each pixel generates an Analysis On Multi-scale Features value, using the Analysis On Multi-scale Features value of this pixel as the pixel value of this pixel, form a width identical with original remote sensing image size, the remote sensing image of the Analysis On Multi-scale Features value that this pixel value shows in original remote sensing image for this pixel, i.e. yardstick image clear.
2. the modulator approach of remote sensing image multiscale semanteme according to claim 1, is characterized in that: the observation scale size scope of the employing remote sensing image handling object in described step (3) is 2 times of the scope of remote sensing image handling object geometric scale size.
3. the modulator approach of remote sensing image multiscale semanteme according to claim 1, is characterized in that: the m=2 in described step (3).
CN201210344453.7A 2012-09-17 2012-09-17 Method for modulating multiscale semanteme of remote-sensing image Expired - Fee Related CN102930269B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201210344453.7A CN102930269B (en) 2012-09-17 2012-09-17 Method for modulating multiscale semanteme of remote-sensing image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201210344453.7A CN102930269B (en) 2012-09-17 2012-09-17 Method for modulating multiscale semanteme of remote-sensing image

Publications (2)

Publication Number Publication Date
CN102930269A CN102930269A (en) 2013-02-13
CN102930269B true CN102930269B (en) 2015-05-13

Family

ID=47645066

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201210344453.7A Expired - Fee Related CN102930269B (en) 2012-09-17 2012-09-17 Method for modulating multiscale semanteme of remote-sensing image

Country Status (1)

Country Link
CN (1) CN102930269B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109376599A (en) * 2018-09-19 2019-02-22 中国科学院东北地理与农业生态研究所 A remote sensing image processing method and system for wetland information extraction
CN110096948B (en) * 2019-03-15 2020-11-17 中国科学院西安光学精密机械研究所 Remote sensing image identification method based on characteristic aggregation convolutional network

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102169545A (en) * 2011-04-25 2011-08-31 中国科学院自动化研究所 Detection method for changes of high-resolution remote sensing images
CN102496151A (en) * 2011-12-08 2012-06-13 南京大学 Method for multi-scale segmentation of high-resolution remote sensing images

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB2398379A (en) * 2003-02-11 2004-08-18 Qinetiq Ltd Automated digital image analysis

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102169545A (en) * 2011-04-25 2011-08-31 中国科学院自动化研究所 Detection method for changes of high-resolution remote sensing images
CN102496151A (en) * 2011-12-08 2012-06-13 南京大学 Method for multi-scale segmentation of high-resolution remote sensing images

Also Published As

Publication number Publication date
CN102930269A (en) 2013-02-13

Similar Documents

Publication Publication Date Title
CN111738064B (en) A fog density identification method for haze images
Zhou et al. Multiscale water body extraction in urban environments from satellite images
Hall et al. A multiscale object-specific approach to digital change detection
CN111310558A (en) An intelligent extraction method of pavement diseases based on deep learning and image processing
CN109934154A (en) A kind of remote sensing image change detection method and detection device
CN110084108A (en) Pedestrian re-identification system and method based on GAN neural network
CN103839267B (en) Building extracting method based on morphological building indexes
CN104361589A (en) High-resolution remote sensing image segmentation method based on inter-scale mapping
US8559714B2 (en) Post processing for improved generation of intrinsic images
CN102930576A (en) Feature flow-based method for generating abstract line drawing
CN108564062A (en) A kind of island boundary Fast Segmentation Algorithm based on remote sensing image
CN114639002A (en) Infrared and visible light image fusion method based on multi-mode characteristics
JP2014534699A (en) System and method for digital image signal compression using unique images
CN108898569A (en) Fusion method for visible light and infrared remote sensing images and fusion result evaluation method thereof
CN113609984A (en) A kind of pointer meter reading identification method, device and electronic equipment
CN105427313A (en) Deconvolutional network and adaptive inference network based SAR image segmentation method
WO2022153305A1 (en) Systems and methods for hue based encoding of a digital image
CN103927759A (en) Automatic cloud detection method of aerial images
CN106485238A (en) A kind of high-spectrum remote sensing feature extraction and sorting technique and its system
CN108492288B (en) Random forest based multi-scale layered sampling high-resolution satellite image change detection method
Abd-Alameer et al. Quality of medical microscope Image at different lighting condition
CN118506067A (en) Image processing method and system for power grid identification
CN102930269B (en) Method for modulating multiscale semanteme of remote-sensing image
CN103366368B (en) Eliminate shade and the double-truncated-concodebooker codebooker foreground detection method capable of highlight noise
CN102184403B (en) Optimization-based intrinsic image extraction method

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20150513

Termination date: 20190917

CF01 Termination of patent right due to non-payment of annual fee