WO2022227486A1 - 一种基于人工智能的遥感田埂界线检测方法、系统、计算机设备及存储介质 - Google Patents

一种基于人工智能的遥感田埂界线检测方法、系统、计算机设备及存储介质 Download PDF

Info

Publication number
WO2022227486A1
WO2022227486A1 PCT/CN2021/129839 CN2021129839W WO2022227486A1 WO 2022227486 A1 WO2022227486 A1 WO 2022227486A1 CN 2021129839 W CN2021129839 W CN 2021129839W WO 2022227486 A1 WO2022227486 A1 WO 2022227486A1
Authority
WO
WIPO (PCT)
Prior art keywords
heterogeneity
remote sensing
boundary detection
ridge
artificial intelligence
Prior art date
Application number
PCT/CN2021/129839
Other languages
English (en)
French (fr)
Inventor
张翔
唐磊
钟升
匡乃亮
范建平
罗迒哉
Original Assignee
西安微电子技术研究所
西北大学
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 西安微电子技术研究所, 西北大学 filed Critical 西安微电子技术研究所
Publication of WO2022227486A1 publication Critical patent/WO2022227486A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30181—Earth observation
    • G06T2207/30188—Vegetation; Agriculture

Definitions

  • the invention belongs to the field of information technology, and in particular relates to an artificial intelligence-based remote sensing ridge boundary detection method, system, computer equipment and storage medium.
  • the ridge is the main boundary that constitutes the farmland.
  • the accurate extraction of the ridge can facilitate the quick and convenient statistics of the use of the farmland.
  • the current linear feature extraction algorithms are mainly aimed at the extraction of strong feature information such as roads and rivers in the image, and the extraction of plot information mainly relies on the detection of ridges. At present, there is very little research work on the detection of ridge boundaries.
  • the main detection technologies for ridge boundaries are as follows:
  • Extract ridge boundaries from images through edge information and shape features For example, the Canny operator is used to extract edge information from high-resolution remote sensing images, and then the ridge shape features are used to filter the edges that do not belong to the ridge features, and finally the ridge boundaries are extracted statistically. Although this method can detect the ridge boundary, it cannot detect the irregular boundary and the image resolution is not too high.
  • Another method is to use the idea of segmentation to merge pixels of the same nature into one area, so as to divide the image into several areas, and finally detect the ridge boundary. Although this type of method has achieved good results, due to the similar features (such as color, texture) in different ridge areas, there are small differences in the same ridge area. Directly adopting the idea of segmentation to achieve ridge segmentation will cause small, excessive The segmented area, which affects the detection of ridge boundaries.
  • the purpose of the present invention is to provide a remote sensing ridge boundary detection method, system, computer equipment and storage medium based on artificial intelligence, which solves the problem that the existing technology cannot detect ridge boundaries well.
  • An artificial intelligence-based remote sensing ridge boundary detection method comprising the following steps:
  • Step 1 Multi-scale segmentation of remote sensing ridges:
  • Step 3 Field ridge boundary detection:
  • the boundary of the final segmented area is detected, and a closed area is formed to obtain the ridge boundary.
  • the multi-scale segmentation algorithm adopts the region merging algorithm with the least heterogeneity.
  • two factors that affect the least heterogeneity are determined: the spectral factor and the shape factor.
  • the spectral factor includes the spectral heterogeneity
  • the shape factor includes the smoothness difference. quality and firmness heterogeneity;
  • f is the degree of heterogeneity
  • w is a user-specified color weight
  • h color represents spectral heterogeneity
  • h s represents shape heterogeneity
  • w c is the weight of the image layer
  • n merge is the number of pixels of the merged object
  • n obj1 and n obj2 represent the number of pixels of object 1 before merging and object 2 before merging, respectively
  • h s w cmpct h cmpct +(1-w cmpct )h smpct
  • w cmpct represents the compactness weight
  • h cmpct represents the compactness weight
  • h smpct represents the smoothness
  • the bad area is: in the process of using multi-scale segmentation, some areas will be divided into multiple area blocks, or some small parts will be divided into multiple blocks, these areas are called bad areas.
  • step 3 the minimum enclosing polygon method is used to complete the detection of the ridge boundary.
  • the invention also discloses a remote sensing ridge boundary detection system based on artificial intelligence, comprising a multi-scale segmentation module, an area merging module and a ridge boundary detection module, and the multi-scale segmentation module, the area merging module and the ridge boundary detection module are connected in sequence;
  • the multi-scale segmentation module adopts a multi-scale segmentation algorithm, which is used to segment the remote sensing ridge image to obtain a segmented image composed of several segmented regions;
  • the region merging module is used to merge the bad regions in the segmented image, filter out the bad regions, and obtain the final segmented region;
  • the ridge boundary detection module is used to detect the boundary of the final segmented area, and form a closed area to obtain the ridge boundary.
  • the present invention also discloses a computer device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, when the processor executes the computer program, the artificial Steps of an intelligent remote sensing ridge boundary detection method.
  • the invention also discloses a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the artificial intelligence-based remote sensing ridge boundary detection method are realized.
  • the present invention has the following beneficial technical effects:
  • the invention discloses a remote sensing ridge boundary detection method based on artificial intelligence.
  • a multi-scale segmentation algorithm is used to generate a segmented area, then small and over-segmented areas are merged, and finally a minimum external polygon is made for the merged multiple areas. Form a closed area, so as to realize the detection of remote sensing ridge boundaries.
  • the method proposes a set of solutions for the detection of remote sensing ridge boundaries, and provides a new idea for remote sensing ridge boundary detection.
  • the experimental results show that the invention can accurately and completely extract the land ridges in the remote sensing images, and the method can It alleviates the time-consuming and labor-intensive problem of artificial ridge statistics, which brings great convenience to ridge statistics.
  • the multi-scale image segmentation adopts the region merging algorithm with the least heterogeneity to minimize the weight heterogeneity of the image objects after segmentation. Only considering the minimum spectral heterogeneity will lead to the fragmentation of the polygonal boundaries of the image objects after segmentation. Therefore, The present invention uses a combination of spectral heterogeneity criteria and shape heterogeneity criteria. Before segmentation, it is necessary to determine two factors that have the least impact on heterogeneity: spectral factor and shape factor.
  • the shape factor includes smoothness heterogeneity and compactness heterogeneity. Only by ensuring that spectral heterogeneity, smoothness heterogeneity, and compactness heterogeneity are minimized can the average heterogeneity of all objects in the entire image be minimized.
  • Figure 1 shows the images before and after multi-scale segmentation
  • Figure (a) represents the original farmland image
  • Figure (b) represents the image after multi-scale segmentation
  • Fig. 2 is the image after region merge
  • Fig. 3 is the ridge boundary diagram that adopts minimum enclosing polygon method to obtain
  • Fig. 4 is the ridge boundary diagram that adopts the minimum enclosing rectangle method to obtain
  • FIG. 5 is a schematic diagram of a method for finding the minimum enclosing polygon
  • FIGS. 5( a ) to 5 ( d ) are schematic diagrams of a specific solving process.
  • the invention discloses an artificial intelligence-based remote sensing ridge boundary detection method, comprising the following steps:
  • Step 1 Multi-scale segmentation of remote sensing ridges
  • the principle of multi-scale segmentation is to minimize the heterogeneity.
  • the segmentation result is obtained by merging adjacent small objects, that is, the method with the least heterogeneity within the object.
  • segmentation methods such as the Mean-Shift segmentation algorithm, the chessboard segmentation method, and the quadtree segmentation method, none of the multi-scale segmentation methods are effective. Therefore, the multi-scale segmentation algorithm is finally used to realize the segmentation of remote sensing fields.
  • the multi-scale image segmentation adopts the region merging algorithm with the least heterogeneity.
  • the purpose is to minimize the weight heterogeneity of the image objects after segmentation. Only considering the minimum spectral heterogeneity will lead to the fragmentation of the polygonal boundaries of the image objects after segmentation.
  • the shape factor includes smoothness heterogeneity and compactness heterogeneity. Only by ensuring that spectral heterogeneity, smoothness heterogeneity, and compactness heterogeneity are minimized can the average heterogeneity of all objects in the entire image be minimized.
  • the present invention directly uses the existing tool (eCognition) to perform multi-scale segmentation on the farmland image shown in Fig. 1(a), and the experimental effect diagram is shown in Fig. 1(b).
  • the multi-scale segmentation algorithm is as follows:
  • f is the degree of heterogeneity
  • w is a user-specified color weight
  • h color and h s represent spectral heterogeneity and shape heterogeneity, respectively.
  • w c is the weight of the image layer
  • n merge is the number of pixels of the merged object
  • n obj1 and n obj2 represent the number of pixels of object 1 before merging and object 2 before merging, respectively
  • h s w cmpct h cmpct +(1-w cmpct )h smpct
  • w cmpct represents the compactness weight
  • h cmpct represents the compactness weight
  • h smpct represents the smoothness
  • the segmentation process is to segment the image according to several tunable homogeneity or heterogeneity criteria of shape and spectrum, and the size of the scale parameter indirectly affects the size of the generated object.
  • Step 3 ridge boundary detection
  • the purpose of the present invention is to detect the ridge boundary, and to detect the boundary in each farmland.
  • step 2 a very good segmentation area has been obtained. This step needs to detect the boundaries of these segmentation areas and form a closed area.
  • the present invention compares the minimum enclosing polygon and the smallest enclosing rectangle to complete the detection of the ridge boundary, and the experimental effect diagrams are shown in Figs. It can be seen from the experimental renderings that the method of using the minimum bounding polygon can detect the ridge boundary better than the method of using the minimum bounding rectangle. The main reason is that there are irregular areas in the ridge, the ridge boundary obtained by the minimum enclosing rectangle method is a rectangle, and the minimum enclosing polygon can better fit the irregular ridge boundary.
  • the minimum enclosing rectangle method specifically includes the following processes:
  • Step 3.1 Find the polygon convex hull
  • Step 3.2 Connect two adjacent points of the convex hull as a side of the rectangle;
  • Step 3.3 Find the point on the convex hull that is farthest from the obtained edge, and make a parallel line through this point to obtain the second edge of the rectangle;
  • Step 3.4 Project the point on the convex hull to the obtained side, find the two points with the farthest projection points apart, and draw a straight line through the two points as the other two sides of the rectangle.
  • Step 3.5 Traverse all two adjacent points of the convex hull and re-run steps 3.2 to 3.4, taking the rectangle with the smallest area as the result.
  • the minimum enclosing polygon method specifically includes the following processes:
  • Step 3.1 Find the starting point:
  • the selection criteria of the starting point are: select the point with the smallest x coordinate, and take the point with the largest y as the starting point A.
  • Step 3.2 Take point A as the origin, scan the positive and negative rays of the x-axis clockwise, find the point scanned when the rotation angle is the smallest, and record it as point B; take point B as the origin, scan the ray in the AB direction clockwise to find the rotation angle The point scanned when the rotation angle is the smallest is recorded as point C; with point C as the origin, the ray in the BC direction scans clockwise, and the point scanned when the rotation angle is the smallest is recorded as point D; and so on, until the starting point A is found.
  • Step 3.3 When point A is found again, the minimum enclosing polygon of the convex hull can be obtained, as shown in Figure 5.
  • the invention also discloses a remote sensing ridge boundary detection system based on artificial intelligence.
  • the segmentation module adopts a multi-scale segmentation algorithm, which is used to segment the remote sensing ridge image to obtain a segmented image composed of several segmented areas; the area merging module is used to merge the bad areas in the segmented image, filter out the bad areas, and get The final segmented area; the ridge boundary detection module is used to detect the boundary of the final segmented area and form a closed area to obtain the ridge boundary.
  • the artificial intelligence-based remote sensing ridge boundary detection method of the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects.
  • the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
  • the artificial intelligence-based remote sensing ridge boundary detection method of the present invention is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.
  • the present invention can implement all or part of the processes in the methods of the above embodiments, and can also be completed by instructing relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium.
  • the computer program includes computer program code
  • the computer program code may be in the form of source code, object code, executable file or some intermediate form, and the like.
  • Computer-readable storage media including persistent and non-permanent, removable and non-removable media, can be implemented by any method or technology for storage of information.
  • Information may be computer readable instructions, data structures, modules of programs, or other data. It should be noted that the content contained in the computer-readable media may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer-readable media Electric carrier signals and telecommunication signals are not included.
  • the computer storage medium can be any available medium or data storage device that can be accessed by a computer, including but not limited to magnetic memory (such as floppy disk, hard disk, magnetic tape, magneto-optical disk (MO), etc.), optical memory (such as CD, DVD, BD, HVD, etc.), and semiconductor memory (eg, ROM, EPROM, EEPROM, non-volatile memory (NANDFLASH), solid-state disk (SSD)), and the like.
  • magnetic memory such as floppy disk, hard disk, magnetic tape, magneto-optical disk (MO), etc.
  • optical memory such as CD, DVD, BD, HVD, etc.
  • semiconductor memory eg, ROM, EPROM, EEPROM, non-volatile memory (NANDFLASH), solid-state disk (SSD)
  • a computer apparatus comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor implementing the computer program when executed The steps of artificial intelligence-based remote sensing ridge boundary detection method.
  • the processor may be a central processing unit (CentralProcessingUnit, CPU), or other general-purpose processors, digital signal processors (DigitalSignalProcessor, DSP), application specific integrated circuit (ApplicationSpecificIntegratedCircuit, ASIC), ready-made programmable gate array (Field-ProgrammableGateArray, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
  • CPU central processing unit
  • DSP digital signal processors
  • ASIC ApplicationSpecificIntegratedCircuit
  • FPGA ready-made programmable gate array
  • FPGA field-ProgrammableGateArray

Abstract

一种基于人工智能的遥感田埂界线检测方法、系统、计算机设备及存储介质，其中方法包括以下步骤：步骤一、遥感田埂多尺度分割：采用多尺度分割算法对遥感田埂图像进行分割，得到由若干个分割区域构成的分割图像；步骤二、区域合并：对分割图像中的不良区域进行合并，过滤掉不良区域，得到最终的分割区域；步骤三、田埂界线检测：将最终的分割区域的边界检测出来，并且形成封闭区域，得到田埂界线。

Description

一种基于人工智能的遥感田埂界线检测方法、系统、计算机设备及存储介质 技术领域
本发明属于信息技术领域，具体涉及一种基于人工智能的遥感田埂界线检测方法、系统、计算机设备及存储介质。
背景技术
耕地的四周一般都具有非常明显的边界，从特征结构来看，田埂是构成农田的主要边界，田埂的准确提取能够有利于快速方便的统计农田的使用情况。目前的线状地物提取算法主要针对图像中的道路、河流等强特征信息的提取，对地块信息的提取主要依靠对田埂的检测。目前对田埂地界的检测研究工作非常少，主要的田埂界线检测技术如下：
通过边缘信息和形状特征从图像中提取田埂界线。比如使用Canny算子从高分辨率遥感图像中提取边缘信息，然后利用田埂形状特征，过滤不属于田埂特征的边缘，最终通过统计学提取田埂界线。尽管这类方法能检测出田埂界线，但是对于界线不规则、图像分辨率不太高的情况不能很好的检测出来。另外一种方法是采用分割的思路，将相同性质的像素归并为一个区域，从而将图像分为若干个区域，最终检测出田埂界线。尽管这类方法取得了较好的结果，但由于不同田埂区域存在相似的特征(如，颜色，纹理)，同一田埂区域存在细小的不同，直接采用分割的思路来实现田埂分割会造成细小、过分割的区域，从而影响田埂界线检测。
发明内容
本发明的目的在于提供一种基于人工智能的遥感田埂界线检测方法、系统、计算机设备及存储介质，解决了现有技术不能很好的检测出田埂界线的问题。
本发明是通过以下技术方案来实现：
一种基于人工智能的遥感田埂界线检测方法，包括以下步骤：
步骤一、遥感田埂多尺度分割：
采用多尺度分割算法对遥感田埂图像进行分割，得到由若干个分割区域构成的分割图像；
步骤二、区域合并：
对分割图像中的不良区域进行合并，过滤掉不良区域，得到最终的分割区域；
步骤三、田埂界线检测：
将最终的分割区域的边界检测出来，并且形成封闭区域，得到田埂界线。
进一步，多尺度分割算法采用异质性最小的区域合并算法，分割前先确定影响异质性最小的两种因子：光谱因子与形状因子，光谱因子包括光谱异质性，形状因子包括光滑度异质性和紧致度异质性；
当光谱异质性、光滑度异质性及紧致度异质性最小时，才能使整个图像所有对象的平均异质性最小。
进一步，多尺度分割算法具体如下：
f＝w*h color+(1-w)*h s
其中，f为异质度，w是用户规定的颜色权重，h color表示光谱异质性，h s表示形状异质性。
进一步，光谱异质性的算法如下：
Figure PCTCN2021129839-appb-000001
其中，w c是影像层权重；n merge是合并后对象的像元数；
Figure PCTCN2021129839-appb-000002
是合并后光谱标准差；n obj1和n obj2分别表示合并前对象1和合并前对象2的像元数；
Figure PCTCN2021129839-appb-000003
和
Figure PCTCN2021129839-appb-000004
分别表示合并前对象1的光谱标准差和合并前对象2的光谱标准差。
进一步，形状异质性的算法如下：
h s＝w cmpcth cmpct+(1-w cmpct)h smpct
其中，w cmpct表示紧致度权重；h cmpct表示紧致度权重；h smpct表示光滑度。
进一步，步骤二中，不良区域为：在用多尺度分割的过程中会出现将某些区域分割为多个区域块，或者将一些细小的部分分割多块，这些区域称为不良区域。
进一步，步骤三中，采用最小包围多边形法来完成田埂界线检测。
本发明还公开了一种基于人工智能的遥感田埂界线检测系统，包括多尺度分割模块、区域合并模块和田埂界线检测模块，多尺度分割模块、区域合并模块和田埂界线检测模块依次连接；
多尺度分割模块采用多尺度分割算法，用于对遥感田埂图像进行分割，得到由若干个分割区域构成的分割图像；
区域合并模块，用于对分割图像中的不良区域进行合并，过滤掉不良区域，得到最终的分割区域；
田埂界线检测模块，用于将最终的分割区域的边界检测出来，并且形成封闭区域，得到田埂界线。
本发明还公开了一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现所述基于人工智能的遥感田埂界线检测方法的步骤。
本发明还公开了一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，所述计算机程序被处理器执行时实现所述基于人工智能的遥感田埂界线检测方法的步骤。
与现有技术相比，本发明具有以下有益的技术效果：
本发明公开了一种基于人工智能的遥感田埂界线检测方法，首先利用多尺度分割算法来产生分割区域，然后对细小、过分割的区域进行合并，最后对合并的多个区域做最小外界多边形，形成封闭区域，从而实现对遥感田埂界线的 检测。该方法提出了一套解决遥感田埂界线检测的方案，为遥感田埂界线检测提供了新的思路，实验结果表明，本发明可以将遥感图像中的地块田埂较为准确完整的提取出来，该方法能够缓解人工田埂统计的耗时耗力的问题，为田埂统计带了极大的方便。
进一步，多尺度影像分割采用异质性最小的区域合并算法，实现了分割后影像对象的权重异质性最小化，仅仅考虑光谱异质性最小会导致分割后影像对象的多边形边界比较破碎，因此本发明把光谱异质性的标准和形状异质性的标准结合使用。分割前需要确定影响异质性最小的两种因子：光谱因子与形状因子，其中形状因子包括光滑度异质性、紧致度异质性。只有保证光谱异质性、光滑度异质性、紧致度异质性最小，才能使整个图像所有对象的平均异质胜最小。
附图说明
图1为多尺度分割前后的图像；图(a)代表原始农田图像，图(b)代表多尺度分割后的图像；
图2为区域合并后的图像；
图3为采用最小包围多边形法求出的田埂界线图；
图4为采用最小包围矩形法求出的田埂界线图；
图5为最小包围多边形求法的示意图，图5(a)～图5(d)为具体的求解过程示意图。
具体实施方式
下面结合具体的实施例对本发明做进一步的详细说明，所述是对本发明的解释而不是限定。
本发明公开了一种基于人工智能的遥感田埂界线检测方法，包括以下步骤：
步骤1、遥感田埂多尺度分割
多尺度分割的原理是异质性最小，通过合并相邻小的对象，即对象内部异 质性最小的方法得到分割结果。在实验过程中通过对比多种分割方法，比如Mean-Shift分割算法、棋盘分割方法、四叉树分割方法等都没有多尺度分割效果好，因此最终采用多尺度分割算法来实现遥感田埂的分割。
多尺度影像分割采用异质性最小的区域合并算法，目的是实现分割后影像对象的权重异质性最小化，仅仅考虑光谱异质性最小会导致分割后影像对象的多边形边界比较破碎，因此常把光谱异质性的标准和形状异质性的标准结合使用。分割前需要确定影响异质性最小的两种因子：光谱因子与形状因子，其中形状因子包括光滑度异质性、紧致度异质性。只有保证光谱异质性、光滑度异质性、紧致度异质性最小，才能使整个图像所有对象的平均异质胜最小。
本发明直接使用现有的工具(eCognition)对图1(a)所示的农田图像进行多尺度分割，实验效果图如图1(b)所示。
多尺度分割算法具体如下：
f＝w*h color+(1-w)*h s
其中，f为异质度，w是用户规定的颜色权重，h color和h s分别表示光谱异质性和形状异质性。
光谱异质性h color的算法如下：
Figure PCTCN2021129839-appb-000005
其中，w c是影像层权重；n merge是合并后对象的像元数；
Figure PCTCN2021129839-appb-000006
是合并后光谱标准差；n obj1和n obj2分别表示合并前对象1和合并前对象2的像元数；
Figure PCTCN2021129839-appb-000007
和
Figure PCTCN2021129839-appb-000008
分别表示合并前对象1光谱标准差和合并前对象2光谱标准差。
形状异质性h s的算法如下：
h s＝w cmpcth cmpct+(1-w cmpct)h smpct
其中，w cmpct表示紧致度权重；h cmpct表示紧致度权重；h smpct表示光滑度。
分割过程是按照形状和光谱的几个可调的均质性或异质性标准对图像进行分割，尺度参数的大小间接影影响生成的对象大小。
步骤2、区域合并
由于田埂的复杂程度不同，在用多尺度分割的过程中可能会出现将一些某些区域分割为多个区域块，或者将一些细小的部分分割多块，因此我们直接使用eCognition软件对这些区域进行合并，从而过滤到这些不良区域。区域合并的效果图如图2所示。
步骤3、田埂界线检测
本发明的目的是将田埂界线检测出来，在每块农田中检测出界线。在步骤2中已经得到非常好的分割区域，这一步需要将这些分割区域的边界检测出来，并且形成封闭区域。
本发明对比了最小包围多边形和最小包围矩形来完成田埂界线检测，实验效果图如图4和图5所示。从实验效果图可以看出使用最小包围多边形方法比使用最小包围矩形方法能更好的检测田埂界线。主要原因是田埂存在不规则的区域，最小包围矩形方法得到的田埂界线是矩形，而最小包围多边形能更好的拟合不规则的田埂界线。
(1)以最终的分割区域为作用对象，最小包围矩形求法具体包括以下过程：
步骤3.1：求多边形凸包；
步骤3.2：将凸包两个相邻的点连接作为矩形一条边；
步骤3.3：寻找凸包上距离已得到的边最远的点，过该点做平行线，从而得到矩形的第二条边；
步骤3.4：将凸包上点向已求得的边投影，求得投影点相距最远的两个点，过该两点做直线，作为矩形另外两条边。
步骤3.5：遍历凸包所有相邻两点重新运行步骤3.2～3.4，将面积最小的矩形作为求得结果。
(2)如图5(a)～图5(d)所示，最小包围多边形求法具体包括以下过程：
步骤3.1、求起始点：
在分割区域中首先选择起始点，起始点的选取标准是：选取x坐标最小的，并且取y大的点作为起始点A。
步骤3.2、以A点为原点，x轴正反向射线顺时针扫描，找到旋转角最小时扫描到的点，记做B点；以B点为原点，AB方向射线顺时针扫描，找到旋转角最小时扫描到的点，记做C点；以C点为原点，BC方向射线顺时针扫描，找到旋转角最小时扫描到的点，记做D点；以此类推，直到找到起始点A。
步骤3.3、当再次找到A点后就能得到凸包的最小包围多边形，如图5所示。
本发明还公开了一种基于人工智能的遥感田埂界线检测系统，包括多尺度分割模块、区域合并模块和田埂界线检测模块，多尺度分割模块、区域合并模块和田埂界线检测模块依次连接；多尺度分割模块采用多尺度分割算法，用于对遥感田埂图像进行分割，得到由若干个分割区域构成的分割图像；区域合并模块，用于对分割图像中的不良区域进行合并，过滤掉不良区域，得到最终的分割区域；田埂界线检测模块，用于将最终的分割区域的边界检测出来，并且形成封闭区域，得到田埂界线。
本发明基于人工智能的遥感田埂界线检测方法可采用完全硬件实施例、完全软件实施例或结合软件和硬件方面的实施例的形式。而且，本发明可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本发明基于人工智能的遥感田埂界线检测方法如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本发明实现上述实施例方法中的全部或部分流程，也可以通过计算机程序来指令相关的硬件来完成，所述的计算机程序可存储于一计算机可读存储介质中，该计算机程序在被处理器执行时，可实现上述各个方法实施例的步骤。其中，所述计算机程序包括计算机程序代码，所述计算机程序代码可以为源代码形式、对象代码形式、可执行文件或某些中间形式等。计算 机可读存储介质包括永久性和非永久性、可移动和非可移动媒体可以由任何方法或技术来实现信息存储。信息可以是计算机可读指令、数据结构、程序的模块或其他数据。需要说明的是，所述计算机可读介质包含的内容可以根据司法管辖区内立法和专利实践的要求进行适当的增减，例如在某些司法管辖区，根据立法和专利实践，计算机可读介质不包括电载波信号和电信信号。其中，所述计算机存储介质可以是计算机能够存取的任何可用介质或数据存储设备，包括但不限于磁性存储器(例如软盘、硬盘、磁带、磁光盘(MO)等)、光学存储器(例如CD、DVD、BD、HVD等)、以及半导体存储器(例如ROM、EPROM、EEPROM、非易失性存储器(NANDFLASH)、固态硬盘(SSD))等。
在示例性实施例中，还提供计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现所述基于人工智能的遥感田埂界线检测方法的步骤。处理器可能是中央处理单元(CentralProcessingUnit，CPU)，还可以是其他通用处理器、数字信号处理器(DigitalSignalProcessor，DSP)、专用集成电路(ApplicationSpecificIntegratedCircuit，ASIC)、现成可编程门阵列(Field-ProgrammableGateArray，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。
本发明的内容不限于上述实施例所列举，本领域技术人员不付出创造性劳动即可做出的各种修改或变形，均为本发明的权利要求所涵盖。

Claims (10)

  1. 一种基于人工智能的遥感田埂界线检测方法，其特征在于，包括以下步骤：
    步骤一、遥感田埂多尺度分割：
    采用多尺度分割算法对遥感田埂图像进行分割，得到由若干个分割区域构成的分割图像；
    步骤二、区域合并：
    对分割图像中的不良区域进行合并，过滤掉不良区域，得到最终的分割区域；
    步骤三、田埂界线检测：
    将最终的分割区域的边界检测出来，并且形成封闭区域，得到田埂界线。
  2. 根据权利要求1所述的一种基于人工智能的遥感田埂界线检测方法，其特征在于，多尺度分割算法采用异质性最小的区域合并算法，分割前先确定影响异质性最小的两种因子：光谱因子与形状因子，光谱因子包括光谱异质性，形状因子包括光滑度异质性和紧致度异质性；
    当光谱异质性、光滑度异质性及紧致度异质性最小时，才能使整个图像所有对象的平均异质性最小。
  3. 根据权利要求2所述的一种基于人工智能的遥感田埂界线检测方法，其特征在于，多尺度分割算法具体如下：
    f＝w*h color+(1-w)*h s
    其中，f为异质度，w是用户规定的颜色权重，h color表示光谱异质性，h s表示形状异质性。
  4. 根据权利要求3所述的一种基于人工智能的遥感田埂界线检测方法，其特征在于，光谱异质性的算法如下：
    Figure PCTCN2021129839-appb-100001
    其中，w c是影像层权重；n merge是合并后对象的像元数；
    Figure PCTCN2021129839-appb-100002
    是合并后光谱 标准差；n obj1和n obj2分别表示合并前对象1和合并前对象2的像元数；
    Figure PCTCN2021129839-appb-100003
    和
    Figure PCTCN2021129839-appb-100004
    分别表示合并前对象1的光谱标准差和合并前对象2的光谱标准差。
  5. 根据权利要求3所述的一种基于人工智能的遥感田埂界线检测方法，其特征在于，形状异质性的算法如下：
    h s＝w cmpcth cmpct+(1-w cmpct)h smpct
    其中，w cmpct表示紧致度权重；h cmpct表示紧致度权重；h smpct表示光滑度。
  6. 根据权利要求1所述的一种基于人工智能的遥感田埂界线检测方法，其特征在于，步骤二中，不良区域为：在用多尺度分割的过程中会出现将某些区域分割为多个区域块，或者将一些细小的部分分割多块，这些区域称为不良区域。
  7. 根据权利要求1所述的一种基于人工智能的遥感田埂界线检测方法，其特征在于，步骤三中，采用最小包围多边形法来完成田埂界线检测。
  8. 一种基于人工智能的遥感田埂界线检测系统，其特征在于，包括多尺度分割模块、区域合并模块和田埂界线检测模块，多尺度分割模块、区域合并模块和田埂界线检测模块依次连接；
    多尺度分割模块采用多尺度分割算法，用于对遥感田埂图像进行分割，得到由若干个分割区域构成的分割图像；
    区域合并模块，用于对分割图像中的不良区域进行合并，过滤掉不良区域，得到最终的分割区域；
    田埂界线检测模块，用于将最终的分割区域的边界检测出来，并且形成封闭区域，得到田埂界线。
  9. 一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序，其特征在于，所述处理器执行所述计算机程序时实现如权利要求1至7任意一项所述基于人工智能的遥感田埂界线检测方法的步骤。
  10. 一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现如权利要求1至7任意一项所述基于人工智能的遥感田埂界线检测方法的步骤。
PCT/CN2021/129839 2021-04-27 2021-11-10 一种基于人工智能的遥感田埂界线检测方法、系统、计算机设备及存储介质 WO2022227486A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110462416.5A CN113129323A (zh) 2021-04-27 2021-04-27 一种基于人工智能的遥感田埂界线检测方法、系统、计算机设备及存储介质
CN202110462416.5 2021-04-27

Publications (1)

Publication Number Publication Date
WO2022227486A1 true WO2022227486A1 (zh) 2022-11-03

Family

ID=76780410

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/129839 WO2022227486A1 (zh) 2021-04-27 2021-11-10 一种基于人工智能的遥感田埂界线检测方法、系统、计算机设备及存储介质

Country Status (2)

Country Link
CN (1) CN113129323A (zh)
WO (1) WO2022227486A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116030363A (zh) * 2023-02-20 2023-04-28 北京数慧时空信息技术有限公司 遥感图像类激活映射图优化方法

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113129323A (zh) * 2021-04-27 2021-07-16 西安微电子技术研究所 一种基于人工智能的遥感田埂界线检测方法、系统、计算机设备及存储介质
CN114067158B (zh) * 2021-11-17 2022-09-23 江苏天汇空间信息研究院有限公司 应用多源遥感数据的农田使用状态的监测系统及方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109146889A (zh) * 2018-07-13 2019-01-04 洛阳中科龙网创新科技有限公司 一种基于高分辨率遥感图像的农田边界提取方法
US20190164322A1 (en) * 2017-11-29 2019-05-30 Adobe Inc. Presenting multiple image segmentations
CN109903304A (zh) * 2019-02-25 2019-06-18 武汉大学 一种基于卷积神经元网络和多边形规则化的建筑物轮廓自动提取算法
CN113129323A (zh) * 2021-04-27 2021-07-16 西安微电子技术研究所 一种基于人工智能的遥感田埂界线检测方法、系统、计算机设备及存储介质

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103969656A (zh) * 2014-05-08 2014-08-06 北京数字绿土科技有限公司 基于机载激光雷达的建筑物建模方法和装置
CN107085708B (zh) * 2017-04-20 2020-06-09 哈尔滨工业大学 基于多尺度分割和融合的高分辨率遥感图像变化检测方法
CN107358197B (zh) * 2017-07-12 2022-01-11 安阳师范学院 一种面向耕地地块提取的影像分类及语义处理方法
CN110443817B (zh) * 2019-06-20 2021-02-02 浙江工业大学 一种提高图像分割精度的方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190164322A1 (en) * 2017-11-29 2019-05-30 Adobe Inc. Presenting multiple image segmentations
CN109146889A (zh) * 2018-07-13 2019-01-04 洛阳中科龙网创新科技有限公司 一种基于高分辨率遥感图像的农田边界提取方法
CN109903304A (zh) * 2019-02-25 2019-06-18 武汉大学 一种基于卷积神经元网络和多边形规则化的建筑物轮廓自动提取算法
CN113129323A (zh) * 2021-04-27 2021-07-16 西安微电子技术研究所 一种基于人工智能的遥感田埂界线检测方法、系统、计算机设备及存储介质

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
MENGHUA LI, YUN SHI, YONGQIANG MA, SHANSHAN HAO: "Terrace Information Extraction in Loess Hilly－gully Ｒegion Landscape Based on Object－oriented Classification Method", GEOMATICS & SPATIAL INFORMATION TECHNOLOGY, vol. 42, no. 5, 1 May 2019 (2019-05-01), pages 1 - 5, XP055982715 *
TANG, LEI: "Research on Automatic Extraction of Terraced Ridges in Loess Hilly Region Based on GF-2", CHINESE DOCTORAL DISSERTATIONS & MASTER'S THESES FULL-TEXT DATABASE (MASTER), AGRICULTURE, 1 June 2020 (2020-06-01), XP055982708, [retrieved on 20221117] *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116030363A (zh) * 2023-02-20 2023-04-28 北京数慧时空信息技术有限公司 遥感图像类激活映射图优化方法

Also Published As

Publication number Publication date
CN113129323A (zh) 2021-07-16

Similar Documents

Publication Publication Date Title
WO2022227486A1 (zh) 一种基于人工智能的遥感田埂界线检测方法、系统、计算机设备及存储介质
CN109978839B (zh) 晶圆低纹理缺陷的检测方法
CN109785285B (zh) 一种基于椭圆特征拟合的绝缘子破损检测方法
WO2019205290A1 (zh) 一种图像检测方法、装置、计算机设备及存储介质
CN107092871B (zh) 基于多尺度多特征融合的遥感影像建筑物检测方法
Ofir et al. On detection of faint edges in noisy images
CN105844625A (zh) 一种融合边缘和区域的活动轮廓图像分割方法
WO2019071976A1 (zh) 基于区域增长和眼动模型的全景图像显著性检测方法
CN109712143B (zh) 一种基于超像素多特征融合的快速图像分割方法
US20230334627A1 (en) Method for segmenting and denoising triangle mesh
CN107180436A (zh) 一种改进的kaze图像匹配算法
CN111868783B (zh) 基于边界提取的区域合并图像分割算法
CN113409332B (zh) 一种基于三维点云的建筑物平面分割方法
CN112258536B (zh) 一种胼胝体及小脑蚓部的一体化定位及分割方法
WO2019041447A1 (zh) 一种3d视频帧特征点提取的方法及系统
CN108875825B (zh) 一种基于图像分块的粮仓害虫检测方法
CN110298816B (zh) 一种基于图像重生成的桥梁裂缝检测方法
CN112560740A (zh) 一种基于PCA-Kmeans的可见光遥感影像变化检测方法
CN114155236A (zh) 一种适用于动态测量环境的激光条纹中心线提取方法
CN104156696B (zh) 基于双方向图的快速局部不变特征描述子的构造方法
CN114764810A (zh) 一种医学图像分割方法
CN105513044B (zh) 一种基于统计度量直线特征的数字直线段识别方法
Li et al. Depth image restoration method based on improved FMM algorithm
CN110648341B (zh) 一种基于尺度空间和子图的目标边界检测方法
Drechsler et al. Automatic ROI identification for fast liver tumor segmentation using graph-cuts

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: 21938948

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE