CN107392926B - Remote sensing image feature selection method based on early-stage land thematic map - Google Patents

Remote sensing image feature selection method based on early-stage land thematic map Download PDF

Info

Publication number
CN107392926B
CN107392926B CN201710839939.0A CN201710839939A CN107392926B CN 107392926 B CN107392926 B CN 107392926B CN 201710839939 A CN201710839939 A CN 201710839939A CN 107392926 B CN107392926 B CN 107392926B
Authority
CN
China
Prior art keywords
image
map
land
feature
remote sensing
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.)
Active
Application number
CN201710839939.0A
Other languages
Chinese (zh)
Other versions
CN107392926A (en
Inventor
周亚男
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.)
Hohai University HHU
Original Assignee
Hohai University HHU
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 Hohai University HHU filed Critical Hohai University HHU
Priority to CN201710839939.0A priority Critical patent/CN107392926B/en
Publication of CN107392926A publication Critical patent/CN107392926A/en
Application granted granted Critical
Publication of CN107392926B publication Critical patent/CN107392926B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
    • G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20021—Dividing image into blocks, subimages or windows
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20112—Image segmentation details
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20112—Image segmentation details
    • G06T2207/20152—Watershed segmentation

Abstract

The invention discloses a remote sensing image feature selection method based on an earlier-stage land thematic map, which comprises the following steps of: selecting a remote sensing image segmentation algorithm and an image feature set of a segmented object, and setting corresponding segmentation parameters and a feature extraction algorithm; dividing the early-stage land thematic map into a corresponding number of map layers according to the land parcel type, extracting an influence intensity map and a characteristic weight matrix, and generating a characteristic weight distribution map; segmenting a remote sensing image to be analyzed, and extracting image characteristics of a segmented object; selecting an image classifier, improving the weight setting of classification features of the image classifier, and constructing a feature weighted classifier; and setting feature weights for the objects in the segmentation map according to the feature weight distribution map, inputting the image features and the weights of the objects into a feature weighting classifier, calculating the land category of the segmentation objects, and generating a land thematic map. The invention can realize the self-adaptive optimization and adjustment of the characteristic weight of the local area of the remote sensing image and can improve the analysis precision of the remote sensing image.

Description

Remote sensing image feature selection method based on early-stage land thematic map
Technical Field
The invention relates to a remote sensing image feature selection method based on an earlier-stage land thematic map, and belongs to the field of object-oriented remote sensing image analysis.
Background
With the rapid development of high (spatial) resolution remote sensing, an Object-Based Image Analysis (OBIA) method becomes a main technical means for current remote sensing application. Compared with the traditional pixel-based remote sensing analysis technology, the object-oriented image analysis method can fully mine the rich geometric characteristics, textural characteristics, spatial pattern characteristics and the like of the high-resolution image on the basis of utilizing the spectral characteristics of the image, can further integrate high-level knowledge such as social economy, spatial models and the like, and realizes image analysis with higher precision and higher efficiency. The corresponding reference documents facing the object remote sensing analysis comprise Zhou Asia man, Luo Jian bearing, Cheng xi and the like, and the self-adaptive remote sensing image multi-scale segmentation [ J ] with the blended multiple features: high resolution satellite telemetry image geography calculations [ M ]. Beijing scientific Press, 2009, Blastke, T.,2010.Object based image analysis for Remote Sensing. ISPRS Journal of Photogrammetry and Remote Sensing,65(1), 2-16, Myint, S.W., Gober, P., Brazel, A., texture-Clarke, S.Weng, Q.,2011. P.pixel vs. Object-based classification of root based image analysis, Remote sensing.115 (5): 1145).
However, in the object-oriented remote sensing analysis, various image features (including spectral features, geometric features, texture features, spatial relations and the like) are constructed and extracted, and new challenges are provided for the object-oriented analysis method. On the one hand, extracting and utilizing more image features of the segmented object consumes more processing and analyzing time; on the other hand, the image analysis accuracy is rather reduced by too many image features compared to the limited number of training samples. Therefore, researchers have attempted to devise methods to reduce feature dimensionality to find the most important features in the feature space in an effort to obtain an analysis accuracy that is comparable to using all features with fewer object features. In the field of digital image processing and analysis, common methods for reducing feature dimensions (feature dimension reduction) can be divided into two categories, feature extraction and feature selection. In feature extraction, the original object feature space is converted into a new lower-dimensional feature space to construct and extract more important features. However, the original feature space is changed by the feature extraction method, so that the theoretical explanation of the dimension reduction result is difficult to provide; and therefore more commonly feature selection methods (which are also of interest for the present invention). The goal of feature selection is to select a smaller number of more important features from the original feature set to form a feature subset, and to enable the feature subset to approach or even exceed the analysis accuracy of the original feature set. According to the method, attribute information among features is utilized or evaluation based on classification results is utilized, and the feature selection method in remote sensing application is roughly divided into three categories: filtration (filter), wrapping (wrapper) and embedding (embedded). The filtering method ranks the importance of the features based on the statistical characteristics (such as correlation, information entropy and the like) among the features, and selects the feature combination with higher importance without evaluating the classification result at the later stage. The efficiency of the filtration method is high but it is difficult to directly optimize the classification result. Accordingly, the wrapping law utilizes some machine learning method (e.g., SVM classifier) to evaluate the quality of the classification results generated by the randomly chosen subset of features and to attempt to find the combination of features that produces the optimal results. Although this type of method can achieve better analytical prediction, it is inefficient in that it requires many feature subsets to be tried. The embedding rule is a compromise between the filtering method and the wrapping method, and the evaluation and selection of the features are completed while the feature selection model is constructed, so that the subsequent evaluation of the selected feature combination is not needed. For example, the random forest (random forest) algorithm is a commonly used embedding method, which evaluates the importance of features by calculating the average influence degree of the features on classification results. References to corresponding Feature dimension reductions include, Zhong P, Zhang P, Wang r.dynamic Learning of SMLR for Feature Selection and classification of hyperspectral data [ J ]. IEEE Geoscience and removal Learning, 2008,5(2): 280. 80. quadrature. d.high-dimensional data analysis: The currents and classifications of two dimensional Learning [ J ]. AMS knowledge Learning, 2000,1:32, sight I, chess eff a.an expression to Variable Feature Selection [ J ]. Journal of Machine Learning, 2003,3: 7. map M. 1182. fe. simulation for motion analysis [ J ] and classification of collection [ P ] 2. observation, discovery, simulation [ J ] and classification of collection [ P ] and analysis, 2. P. f. h. c. h. f. c. h. P. c. h. k. c. h. P. e.g. simulation, classification of motion analysis, 2. P. k. c. f. c. k. c. f. observation of motion analysis, classification, 2. P. f. c. f. c. prediction, simulation, Selection and classification [ P. b. c. b. c. f. c. b. c. 2. c. f. c. f. c. prediction of motion estimation, simulation, 2. c. b. c. f. c. b. 2. b. f. c. b. c, 1997,19(2):153-158, Duro D C, Franklin S E, Dub é M G. Multi-scale Object-Based image analysis and Feature Selection of multi-sensor area update requirements [ J ]. International Journal of Remote Sensing,2012,33(14):4502-4526, Ma L, Fu T, BlaschT, et al. evaluation of Feature Selection Methods for Object-Based coat Mapping of manconnected audio quality estimation and Support Vector fields [ J ]. I J. J.: Journal of Information of I.S. Journal of I.S. J. 12 and I.S. J.: Journal of I.S. J.S. 7, D.S. M.S. J.: 7, D.S. 2).
Although the scholars propose a large number of feature selection methods, numerous feature selection experiments are also carried out; however, most methods and experiments are unsupervised (without support of a priori knowledge), globally consistent (the optimal feature selection for the whole experimental area is difficult to perform adaptive optimization aiming at local characteristics of the image), and therefore, the improvement on the image analysis effect is limited.
In summary, in the currently seen patent and literature research, most feature selection methods are unsupervised and globally fixed, and there are few feature selection methods that can effectively incorporate prior knowledge and can perform local adaptive optimization, and especially there are many problems in the methods of the mechanism of incorporating prior knowledge, the adaptive optimization method and flow of feature weights, the application of feature weights in remote sensing analysis, and the like, and a feasible solution is not developed.
Disclosure of Invention
The invention aims to solve the technical problem of providing a remote sensing image feature selection method based on an earlier-stage land thematic map, which can improve the analysis precision of the remote sensing image.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows:
the remote sensing image feature selection method based on the early-stage land thematic map comprises the following steps:
(1) selecting a remote sensing image segmentation algorithm and an image feature set of a segmented object, and setting corresponding segmentation parameters and a feature extraction algorithm;
(2) dividing the early-stage land thematic map into a corresponding number of map layers according to land category, and extracting an influence intensity map of each land category;
(3) registering the early-stage land thematic map to a corresponding early-stage remote sensing image, extracting image characteristics of land blocks in the early-stage land thematic map according to the characteristic extraction algorithm in the step (1), storing the image characteristics in an attribute table of the thematic map, dividing the early-stage land thematic map into a corresponding number of map layers according to land block types, and extracting a characteristic weight matrix of each land type;
(4) combining the influence intensity graph in the step (2) and the characteristic weight matrix in the step (3) to generate a characteristic weight distribution graph of the experimental area;
(5) segmenting the remote sensing image to be analyzed according to the segmentation algorithm and the feature extraction algorithm in the step (1), extracting the image features of the segmented object, and storing the image features into an attribute table of the segmentation graph;
(6) selecting an image classifier, improving the weight setting of classification features of the image classifier, and constructing a feature weighted classifier;
(7) and (4) setting feature weights for the objects in the segmentation map based on the feature weight distribution map in the step (4), inputting the image features and the weights of the objects into a feature weighting classifier, calculating the land category of the segmentation objects, and generating a land thematic map.
The segmentation algorithm in the step (1) comprises a watershed segmentation algorithm, a mean shift segmentation algorithm or a multi-resolution segmentation algorithm.
The image feature set in the step (1) comprises spectral features, geometric features and texture features.
In the step (2), trivial region elimination and vector rasterization are carried out on a single layer to generate a binary image, then distance transformation is carried out on the binary image to generate an influence intensity map of the land type, and other layers are treated in the same way to obtain the influence intensity map of each land type.
And (3) performing target-background binary reclassification on the single image layer in the step (3), generating a secondary image by taking the ground objects in the same type as the image layer as targets and other ground objects as backgrounds, calculating a characteristic weight vector of the target type by using a random forest algorithm, and performing the same treatment on other image layers to obtain a characteristic weight matrix.
The image classifier in the step (6) comprises a minimum distance classifier, a Bayesian classifier or a K-nearest neighbor classifier.
The invention achieves the following beneficial effects:
the land thematic map (such as a land utilization/coverage map, a vegetation thematic map and the like) in the early stage records various attributes (information such as characteristics, categories, space sizes, distribution and the like) of land features in the experimental area; and the land utilization/coverage property of a local area is changed slightly in a short time, so that the land thematic map at the early stage can be used as a priori knowledge for analyzing and understanding the remote sensing image. The early-stage land thematic map is registered to a remote sensing image to be analyzed, and the prior knowledge of the thematic map is extracted to assist in selecting a better feature combination (for example, in a lake water body area, the spectral feature and the texture feature of a ground object are endowed with higher feature weight, and in an urban area, the shape feature of the ground object is endowed with higher weight). On the other hand, geospatial scenes are complex and are composed of surface features of different scales, different categories and different attributes; and the property of the ground object in the local area is different from that of the whole experimental area, and the property of the ground object in different local areas is different. Therefore, the feature selection subset applicable to the whole experimental area is not necessarily the optimal feature combination of a certain local area; more targeted and targeted feature selection needs to be performed on local areas of the image, and the effects of feature selection and image analysis are further improved. According to the method, based on the early-stage land thematic map and the corresponding early-stage remote sensing image thereof, the influence intensity map and the characteristic weight matrix of each land category can be effectively extracted, and the characteristic weight map of a test area is constructed; and then, the classifier of the characteristic weight is combined, the self-adaptive optimization of the characteristic weight in the classification is realized, and the precision of the remote sensing image analysis is improved. The self-adaptive feature selection method is not only suitable for classification of the remote sensing images, but also suitable for other remote sensing image analysis, such as segmentation of the remote sensing images, thematic map extraction of the remote sensing images, target recognition and the like.
Drawings
FIG. 1 is a schematic diagram of remote sensing image feature selection supported by an earlier-stage land thematic map;
FIG. 2 is a flow chart of implementation of remote sensing image feature selection under the support of an earlier-stage land thematic map;
FIG. 3 is a flow chart of remote sensing image segmentation (taking mean shift as an example) and feature extraction of a segmented object;
FIG. 4 is a flow chart of distance (impact) intensity map generation for land category(s);
FIG. 5 is a flow chart of land category feature weight(s) calculation;
FIG. 6 is a schematic illustration of feature weight map generation;
fig. 7 is a schematic diagram of land classification effect of a remote sensing image of a certain land.
Detailed Description
The invention is further described below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present invention is not limited thereby.
FIG. 1 is a schematic diagram of remote sensing image feature selection supported by an earlier stage land thematic map, wherein in the step of calculating a feature weight map, a distance intensity map of each land category and a feature weight matrix of the land category are extracted from the earlier stage land thematic map and a corresponding remote sensing image thereof, and a feature weight map of an experimental area is constructed; in the step of classifying the feature weighting, according to the space position of an object to be analyzed in the segmentation graph, the weight of each feature of the segmentation graph is inquired from the feature weight graph, and the weight of a classifier is set according to the weight; therefore, the self-adaptive characteristic optimization of the remote sensing image is realized;
FIG. 2 is a specific flow chart of remote sensing image feature selection under the support of an earlier-stage land thematic map, which comprises 6 implementation units, such as segmentation and feature extraction of a remote sensing image, generation of a distance (influence) intensity map of each land category, construction of a feature weight matrix of the land category, calculation of a feature weight (distribution) map, design of a feature weighting classifier, and generation of a land thematic classification map;
as shown in fig. 3, the present invention is applied to remote sensing image segmentation (the mean shift segmentation algorithm is taken as an example for the specific implementation of the present invention) and feature extraction. Image segmentation based on mean shift usually includes two steps of image filtering and region merging; the mean shift algorithm is applied to a filtering process and aims to search local extreme points in an image and generate a mean image area; the merging process is to find the connected region of the mean block and form the final segmentation object. In the filtering process, the spatial position and the spectral feature of the image are jointly considered to form a joint vector x (x) with d ═ p +2 dimensionss,xr) Wherein x issRepresenting the position coordinates, x, of the grid/pixel in the imagerRepresenting the p-dimensional vector features of the grid/pixel in the image. Mean shift algorithm basis functionAnd (4) estimating the probability density of the number, and searching a local extreme point in the vector space. Wherein the multidimensional kernel function is defined as:
Figure BDA0001410478810000061
where k (x) is a kernel function in the spatial and spectral domains, hs、hrThe nuclear bandwidths of the spatial and spectral domains, respectively, and C is a normalization constant. Then the mean shift iteration function is:
Figure BDA0001410478810000062
wherein xtFor the position of the mode point (mode) after t iterations, wiAnd the function weight of the pixel point i in the field x. The positions x passes through in the iteration, i.e. the sequence { x, m (x), m (x)), } is the trajectory of x; the mean shift always points to where there is a maximum local density, where the amount of shift approaches zero at the maximum of the density function, and the iteration ends. Mean filtering achieves drift by computing a weighted sum of local (in-neighborhood) sample points in the feature space; on one hand, the image elements inside the ground objects in the image are smoothed, and on the other hand, the boundary characteristics of the ground objects are kept. Two (more) homogeneous regions with similar modes (modes) and small boundary strengths (edge strength) that are spatially adjacent are iteratively merged first in region merging using a transitive closure (transitive closure) algorithm to generate a larger segmented region. The smaller segmented regions (with fewer pixels than S) are then merged into adjacent segmented regions to produce the final segmented object.
Based on the mean shift segmentation, various attribute features such as spectral features, texture features, and shape features of the segmented object are extracted, as shown in table 1.
TABLE 1 image characteristics of segmented objects
Figure BDA0001410478810000063
Figure BDA0001410478810000071
As shown in fig. 4, in generating a distance (influence) intensity map for the land category(s). The distance transformation is used for describing the separation degree of the binary image from the interest area, and the distance transformation image records the minimum distance from the background pixel in the binary image to the interest area. The method takes a certain type of land area (such as water) as an interest area, calculates the minimum distance from a non-interest area (land objects except water, such as cultivated land, forest land and the like) to the interest area by using distance transformation, and is used for depicting an influence (distance) intensity graph of the land type; and then an influence intensity map of the land type(s) is generated. In the concrete implementation, firstly, dividing a land thematic map into a plurality of layers according to categories (one land category corresponds to one layer); then performing trivial region elimination and vector rasterization on the single layer and generating a binary image; then, distance transformation is carried out on the binary image to generate an influence (distance) intensity map of the land type; and finally, carrying out the same treatment on other land types to obtain an influence intensity graph of each land type in the experimental area. For an object S at any spatial position in the experimental area, a corresponding class intensity value can be found from the influence intensity graph of the land class, and linear normalization is carried out to obtain a class influence intensity vector D (D) ═ at the pointiI is 1,2, …, N, where 0<di<1,∑diN is the number of land categories).
As shown in fig. 5, in calculating the feature weight of the land category. Random forest is a machine learning method integrating multiple decision trees and can evaluate the importance of variables (features) in regression and classification problems. The invention will use a random forest algorithm to calculate the importance (weight) of various features for each land type. In a specific implementation, first, a previous land thematic map is geometrically mapped to a previous satellite image, and a plurality of image features (shown in table 1) of each land block in the land thematic map are calculated; then segmenting the land thematic map into a plurality of layers according to categories (one land category corresponds to one layer);secondly, carrying out 'target-background' binary reclassification (the ground objects in the same category as the image layer are targets, and other ground objects are backgrounds) on a single image layer to generate a second-class image, and estimating a characteristic weight vector of the target category by using a random forest algorithm; and finally, performing the same processing on other layers to obtain a characteristic weight matrix W (W) of the land categoryijI 1,2, …, N, j 1,2, …, M, where 0<wij<1, N is the number of land categories, M is the number of image features).
As shown in fig. 6, in generating the feature weight (distribution) map. According to the flow shown in FIG. 4, the influence intensity maps of various land types at any position of the experimental area can be calculated, and according to the flow shown in FIG. 5, the characteristic weight matrix of the land types can be calculated. Further, by using equation (1), it is possible to estimate (WF) as the feature weight vector WF at an arbitrary position SjJ-1, 2, …, M, where M is the number of image features), where diImage intensity, w, for the ith land typeijAnd the weight of the j dimension characteristic of the ith land type.
Figure BDA0001410478810000081
In constructing a feature weighted classifier (taking the K-nearest neighbor classification algorithm as an example for specific implementation of the invention) and feature extraction. The K-nearest neighbor (K-NN) classifier only considers the spatial distribution of the prior samples and ignores the assumption of the class probability distribution, and is one of the commonly used remote sensing image classification methods. In the K-NN classification process, K prior samples with the closest distance to the test sample are searched first, and then the category of the test sample is judged according to the category of the prior samples based on the rule of a majority voting scheme. The invention constructs a classifier with characteristic weighting by transforming a K-NN classification method. In a specific implementation, the nearest neighbor distance of the test sample to the prior sample is calculated using equation (2), where Δ fjThe characteristic distance in the j-th dimension.
Figure BDA0001410478810000082
In the standard K-NN classification, wfjAll set to 1.0 (equal weight setting); in the K-NN classifier with improved feature weighting, the weights wf of different features of the sample need to be setj(ii) a Namely, the design assumption that the important features of the feature weighting classifier have large effects and the non-important features have small effects is realized, and method support is provided for the subsequent generation of the land thematic map.
In generating the land topical classification map. And according to a conventional object-oriented remote sensing classification process, selecting a training sample and a verification sample of the experimental area, performing K-NN classification of improved weighting characteristics, verifying a classification result and the like in sequence. In the improved K-NN classification, the weight of each image feature is required to be inquired from the feature weight graph according to the spatial position of a sample to be classified, and the weight is brought into the classification judgment of the improved K-NN.
Fig. 7 illustrates the land topic classification effect of the satellite image of a certain land implemented by the method of the present invention. From the overall classification effect, the method can achieve better classification effect in simple areas such as water bodies and forest lands of the images and complex areas such as construction lands, water pouring lands and roads of the images.
An example of the present invention is implemented on a PC platform. Experiments prove that the method can effectively evaluate the Feature weight of the local image area in the experimental area, and has a great degree of precision improvement in subsequent object-oriented classification compared with conventional global Feature selection methods (such as Chi-square selection and Recursive Elimination) (the land classification precision of the example image provided by the invention is improved by about 8% compared with the Chi-square selection method and is improved by about 9% compared with the Recursive Elimination method). The method can be widely applied to the object-oriented image analysis, classification, identification and other processes of the high-resolution remote sensing image, such as the third national agricultural census, the national soil resource survey and other large-scale applications.
The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several modifications and variations can be made without departing from the technical principle of the present invention, and these modifications and variations should also be regarded as the protection scope of the present invention.

Claims (5)

1. The remote sensing image feature selection method based on the early-stage land thematic map is characterized by comprising the following steps of:
(1) selecting a remote sensing image segmentation algorithm and an image feature set of a segmented object, and setting corresponding segmentation parameters and a feature extraction algorithm;
(2) dividing the early-stage land thematic map into a corresponding number of map layers according to land category, and extracting an influence intensity map of each land category; the method for obtaining the influence intensity graph comprises the following steps:
performing trivial region elimination and vector rasterization on a single layer to generate a binary image, then performing distance transformation on the binary image to generate an influence intensity map of the land type, and performing the same treatment on other layers to obtain the influence intensity map of each land type;
(3) registering the early-stage land thematic map to a corresponding early-stage remote sensing image, extracting image characteristics of land blocks in the early-stage land thematic map according to the characteristic extraction algorithm in the step (1), storing the image characteristics in an attribute table of the thematic map, dividing the early-stage land thematic map into a corresponding number of map layers according to land block types, and extracting a characteristic weight matrix of each land type;
(4) combining the influence intensity graph in the step (2) and the characteristic weight matrix in the step (3) to generate a characteristic weight distribution graph of the experimental area;
(5) segmenting the remote sensing image to be analyzed according to the segmentation algorithm and the feature extraction algorithm in the step (1), extracting the image features of the segmented object, and storing the image features into an attribute table of the segmentation graph;
(6) selecting an image classifier, improving the weight setting of classification features of the image classifier, and constructing a feature weighted classifier;
(7) and (4) setting feature weights for the objects in the segmentation map based on the feature weight distribution map in the step (4), inputting the image features and the weights of the objects into a feature weighting classifier, calculating the land category of the segmentation objects, and generating a land thematic map.
2. The method for selecting characteristics of remote sensing images based on earlier-stage land thematic maps according to claim 1, wherein the segmentation algorithm in the step (1) comprises a watershed segmentation algorithm, a mean shift segmentation algorithm or a multi-resolution segmentation algorithm.
3. The method for selecting the remote sensing image features based on the early-stage land thematic map as claimed in claim 1, wherein the image feature set in the step (1) comprises spectral features, geometric features and texture features.
4. The method for selecting the remote sensing image features based on the earlier-stage land thematic map as claimed in claim 1, wherein in the step (3), binary reclassification of a target-background is performed on a single map layer, a ground object of the same type as the map layer is used as a target, other ground objects are used as backgrounds to generate a secondary map, then a feature weight vector of the target type is calculated by using a random forest algorithm, and the same processing is performed on other map layers to obtain a feature weight matrix.
5. The method for selecting remote sensing image features based on the early-stage land thematic map as claimed in claim 1, wherein the image classifier in step (6) comprises a minimum distance classifier, a Bayesian classifier or a K-nearest neighbor classifier.
CN201710839939.0A 2017-09-18 2017-09-18 Remote sensing image feature selection method based on early-stage land thematic map Active CN107392926B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710839939.0A CN107392926B (en) 2017-09-18 2017-09-18 Remote sensing image feature selection method based on early-stage land thematic map

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710839939.0A CN107392926B (en) 2017-09-18 2017-09-18 Remote sensing image feature selection method based on early-stage land thematic map

Publications (2)

Publication Number Publication Date
CN107392926A CN107392926A (en) 2017-11-24
CN107392926B true CN107392926B (en) 2020-12-01

Family

ID=60350563

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710839939.0A Active CN107392926B (en) 2017-09-18 2017-09-18 Remote sensing image feature selection method based on early-stage land thematic map

Country Status (1)

Country Link
CN (1) CN107392926B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108376232A (en) * 2018-01-04 2018-08-07 北京星衡科技有限公司 A kind of method and apparatus of automatic interpretation for remote sensing image
CN109657616A (en) * 2018-12-19 2019-04-19 四川立维空间信息技术有限公司 A kind of remote sensing image land cover pattern automatic classification method
CN111723711A (en) * 2020-06-10 2020-09-29 内蒙古农业大学 Plianes and object-oriented mulching film information extraction method and system

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6266442B1 (en) * 1998-10-23 2001-07-24 Facet Technology Corp. Method and apparatus for identifying objects depicted in a videostream
CN101661497B (en) * 2009-09-10 2011-05-11 北京农业信息技术研究中心 Remote sensing land use change detection method and system thereof
CN101840581B (en) * 2010-02-05 2011-12-28 北京交通大学 Method for extracting profile of building from satellite remote sensing image
CN101930547A (en) * 2010-06-24 2010-12-29 北京师范大学 Method for automatically classifying remote sensing image based on object-oriented unsupervised classification
CN102024258B (en) * 2010-12-15 2012-07-25 中国科学院遥感应用研究所 Multi-scale segmentation method for remote sensing image with boundary maintenance characteristics
CN103839261B (en) * 2014-02-18 2017-01-25 西安电子科技大学 SAR image segmentation method based on decomposition evolution multi-objective optimization and FCM

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
"Object-Based Land Cover Mapping using Adaptive Scale Segmentation from";Ya Nan Zhou等;《2017 IEEE International Geoscience and Remote Sensing Symposium》;20170728;全文 *

Also Published As

Publication number Publication date
CN107392926A (en) 2017-11-24

Similar Documents

Publication Publication Date Title
Wu et al. ORSIm detector: A novel object detection framework in optical remote sensing imagery using spatial-frequency channel features
Shi et al. Road detection from remote sensing images by generative adversarial networks
Chen et al. Superpixel based land cover classification of VHR satellite image combining multi-scale CNN and scale parameter estimation
Bai et al. VHR object detection based on structural feature extraction and query expansion
CN107067405B (en) Remote sensing image segmentation method based on scale optimization
Zhu et al. Scene classification based on the sparse homogeneous–heterogeneous topic feature model
CN108830870B (en) Satellite image high-precision farmland boundary extraction method based on multi-scale structure learning
CN106897679A (en) A kind of semantic change detecting method and system based on improvement fuzzy C-means clustering
CN105761238B (en) A method of passing through gray-scale statistical data depth information extraction well-marked target
Zhong et al. Multiagent object-based classifier for high spatial resolution imagery
Zhuang et al. Normal endmember spectral unmixing method for hyperspectral imagery
CN107392926B (en) Remote sensing image feature selection method based on early-stage land thematic map
Zhong et al. Multiscale and multifeature normalized cut segmentation for high spatial resolution remote sensing imagery
Zhang et al. Polygon structure-guided hyperspectral image classification with single sample for strong geometric characteristics scenes
CN109034213B (en) Hyperspectral image classification method and system based on correlation entropy principle
Xu et al. Using linear spectral unmixing for subpixel mapping of hyperspectral imagery: A quantitative assessment
CN109117739A (en) One kind identifying projection properties extracting method based on neighborhood sample orientation
Tu et al. Feature extraction using multitask superpixel auxiliary learning for hyperspectral classification
Sutha Object based classification of high resolution remote sensing image using HRSVM-CNN classifier
Bhil et al. Recent progress in object detection in satellite imagery: A review
Nurwauziyah et al. Satellite image classification using Decision Tree, SVM and k-Nearest Neighbor
Xu et al. The image segmentation algorithm of colorimetric sensor array based on fuzzy C-means clustering
Wu et al. Instant water body variation detection via analysis on remote sensing imagery
Li et al. Standardized object-based dual CNNs for very high-resolution remote sensing image classification and standardization combination effect analysis
Wang et al. A method of spatial mapping and reclassification for high-spatial-resolution remote sensing image classification

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant