CN113269037A - Single-phase image earth surface coverage change detection automation method - Google Patents
Single-phase image earth surface coverage change detection automation method Download PDFInfo
- Publication number
- CN113269037A CN113269037A CN202110402357.2A CN202110402357A CN113269037A CN 113269037 A CN113269037 A CN 113269037A CN 202110402357 A CN202110402357 A CN 202110402357A CN 113269037 A CN113269037 A CN 113269037A
- Authority
- CN
- China
- Prior art keywords
- image
- surface coverage
- texture
- samples
- sample
- 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.)
- Withdrawn
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Abstract
The invention discloses an automatic method for detecting single-phase image earth surface coverage change, which comprises the following steps: A) segmenting the remote sensing image by utilizing the existing vector data to obtain an initial image object with a prior earth surface coverage type; B) extracting image samples of various earth surface coverage types in the research area according to the earth surface coverage type and the spatial distribution in the initial image object; C) calculating texture feature values of various image samples, and obtaining optimal texture features covered by various earth surfaces based on texture feature contribution degrees so as to retain correct samples to form an image object to be detected; D) calculating the optimal texture characteristic value of each image object to be detected, and comparing the optimal texture characteristic value with the texture characteristic values of correct samples in the image samples of the same category to obtain a changed image object; E) and re-segmenting and classifying the changed image object to obtain a change detection result. The invention can realize the automation of the detection of the earth surface coverage change target of the existing first-stage earth surface coverage vector data base and the single-stage new image condition.
Description
Technical Field
The invention relates to a method for identifying earth surface coverage change, in particular to an automatic method for detecting earth surface coverage change of a single-phase image.
Background
Remote sensing image change detection is an important content of global change research, and is applied to various fields, such as emergency and evaluation after disasters, environment change detection and spatial data updating.
The overall detection of changes can be divided into two broad categories: one is a change detection method based on a two-stage remote sensing image, and the other is a change detection method using combination of vector data and the remote sensing image. Because factors influencing the spectral characteristic change of the ground object are complex, the change detection method based on the two-stage remote sensing image has the defects of large workload, serious classification error accumulation phenomenon, strict requirement on data conditions and the like.
In the prior art, change detection is carried out through differences of texture features among image objects, so that the precision and the efficiency of a change detection result are effectively improved. However, in the change detection, a certain number of samples with the same category attribute need to be extracted from the segmented image object in a manner of manual visual identification, the workload is still huge, the time and the labor are wasted, the extraction result of the samples has certain subjectivity, and the accuracy of the remote-sensing image change detection is serious.
In view of the above, it is desirable to provide an automated method for detecting surface coverage change of a single-phase image.
Disclosure of Invention
The invention provides an automatic method for detecting the surface coverage change of a single-phase image, which can realize the automation of the detection of a surface coverage change target by combining the current vector data of a first phase with a new single-phase remote sensing image and has high detection precision.
In order to achieve the above object, the present invention provides an automatic method for detecting surface coverage change of a single-phase image, comprising the following steps: A) segmenting the remote sensing image by utilizing the existing vector data to obtain an initial image object with a prior earth surface coverage type; B) extracting image samples of various earth surface coverage types in the research area according to the earth surface coverage types contained in the initial image objects and the spatial distribution of the initial image objects; C) calculating texture feature values of the image samples of various earth surface coverage types, obtaining preferred texture features of various earth surface coverage types based on texture feature contribution degrees, eliminating abnormal samples, and reserving correct samples to form an image object to be detected; D) calculating the optimal texture characteristic value of each image object to be detected, and comparing the optimal texture characteristic value of the image object to be detected with the texture characteristic value of the correct sample in the image samples of the same category to obtain a changed image object; E) and re-segmenting and classifying the changed image object to obtain a change detection result.
Specifically, in the step B), the spatial arrangement of the initial image objects is obtained according to a study region range, distribution characteristics of the image objects in the study region, and a topographic disparity in the study region, and the method includes the following steps:
1) dividing said region of interest into r x c sampling grids;
2) dividing the study area into t levels according to the terrain difference;
3) and calculating the number of image samples distributed in the sampling grid according to the number of the initial image objects in the sampling grid and the terrain difference.
More specifically, in the step C), the preferred texture features of each type of the surface coverage type are obtained from the texture feature contribution degree of the image sample of the type of the surface coverage type; the abnormal samples of the image samples of the earth surface coverage types are obtained by the preferred texture characteristics and texture abnormal degree indexes of the image samples of the earth surface coverage types.
More specifically, the texture feature contribution of each type of the image sample is derived from an information gain rate of the texture feature.
More specifically, the texture abnormality index of each type of the image sample is obtained from the local reachable density of the sample object in the image sample.
Further specifically, the local reachable density is derived from a kth neighborhood of the sample object and a kth reachable distance between the sample object and another sample object.
Further specifically, the kth reachable distance is a maximum of the kth distance of the sample object and a euclidean distance of the sample object and the other sample object within a preferred feature space vector.
Further specifically, the preferred feature space vector is constructed by weighting the preferred texture features of the same type of surface coverage according to the texture feature contribution degree.
Further specifically, in the step D), the changed image object is obtained by screening a texture abnormality degree index between the preferred texture feature value of the image object to be detected and the texture feature value of the correct sample in the image samples of the same type of surface coverage according to a set abnormality degree threshold.
More specifically, in the step E), the variable image object is segmented by using a multi-scale segmentation algorithm, a texture abnormality degree index of the segmented variable image object in each type of the image samples of the surface coverage type is calculated, and the classification category of the variable image object is divided according to the texture abnormality degree index.
The invention discloses an automatic method for detecting single-stage image ground surface coverage change, which comprises the steps of sampling an initial image object with a prior ground surface coverage type according to the ground surface coverage type and the spatial distribution of each ground surface coverage type to obtain image samples of various ground surface coverage types, weighting the image samples of each ground surface coverage type according to texture characteristic contribution degrees to construct an optimal characteristic space vector, judging whether the image sample object is an abnormal sample according to the texture abnormality degree index of the image sample, if so, rejecting the abnormal sample, forming a reserved correct sample into a sample object of an image object to be detected, wherein the abnormal sample is a sample which is inconsistent with the prior category attribute, detecting and rejecting the sample can ensure the correctness of automatically extracting the sample, and thus, the precision of a change detection result can be improved. And then calculating texture abnormality degree indexes between the optimal texture characteristic value of the image object to be detected and texture characteristic values of correct samples in the image samples of the same type, screening according to a set abnormality degree threshold value to obtain a changed image object, finally segmenting the changed image object by adopting a multi-scale segmentation algorithm, calculating the texture abnormality degree indexes of the segmented changed image object in the image samples of the different types, and dividing the ground object type of the changed image object according to the texture abnormality degree indexes to obtain the change conditions of the ground surface coverage of the different types. The method can realize automatic detection of the earth surface coverage change of the existing first-stage earth surface coverage vector data base and the single-stage new image condition, and not only reduces the rigorous requirements of the two-stage (multi-stage) remote sensing image change detection on the data; the automatic extraction of the samples is realized, the subjectivity caused by manual visual discrimination is avoided, and the abnormal samples (including misclassified or changed samples) in the automatically extracted samples are removed, and the characteristics with high contribution degree are selected for abnormal detection, so that the precision of the change detection result can be improved.
Additional features and advantages of embodiments of the invention will be set forth in the detailed description which follows.
Drawings
FIG. 1 is a schematic diagram of an automated method for single-session image earth surface coverage change detection according to the present invention;
FIG. 2 is a flow chart of automatic sample extraction for the automated method for single-phase image earth surface coverage change detection of the present invention;
FIG. 3 is a schematic diagram of the calculation of the reachable distance of a sample object in a three-dimensional feature space;
FIG. 4 is a flowchart of the earth's surface coverage change detection of the automated method for single-session image earth's surface coverage change detection of the present invention;
FIG. 5 is a flowchart of the classification of the change image objects in the automated method for single-session image terrain coverage change detection of the present invention;
FIG. 6 is a spatial distribution characteristic diagram of cultivated land in the automated method for single-phase image land surface coverage change detection of the present invention, wherein FIG. 6-1 is an image object diagram of cultivated land type image; FIG. 6-2 is a photographic image map of cultivated land with an elevation of 650-660 m; FIG. 6-3 is a view of cultivated land images with elevation of 660-670 m; FIG. 6-4 is a view of cultivated land images with an altitude of 670-680 m; FIG. 6-5 is a view of cultivated land images with an altitude of 680-690 m; FIG. 6-6 is a view of cultivated land images with an altitude of 690-700 m; FIG. 6-7 is a view of cultivated land at an altitude of 700-710 m; FIG. 6-8 is a view of cultivated land at an altitude of 710-720 m;
FIG. 7 is a diagram of spatial distribution characteristics of forest land in the automated method for single-phase image surface coverage change detection of the present invention, wherein FIG. 7-1 is a diagram of image objects of forest land type; FIG. 7-2 is a diagram of a forest land type image at an altitude of 650-660 m; FIG. 7-3 is a forest land image with an altitude of 660 to 670 m; FIG. 7-4 is a forest land image with an altitude of 670-680 m; FIG. 7-5 is a forest land image with an altitude of 680-690 m; FIG. 7-6 is a forest land image with an altitude of 690-700 m; FIG. 7-7 is a forest land image with an altitude of 700-710 m; FIG. 7-8 is a forest land image with an altitude of 710-720 m;
FIG. 8 is a spatial distribution characteristic diagram of a residential area in the single-phase image ground surface coverage change detection automation method of the present invention, wherein FIG. 8-1 is an image map of the residential area class; FIG. 8-2 is a photographic image of a residential area at an altitude of 650-660 m; FIG. 8-3 is an image of a residential area at an altitude of 660 to 670 m; FIG. 8-4 is an image of a residential area with an altitude of 670-680 m; FIG. 8-5 is an image of a residential area with an altitude of 680-690 m; FIG. 8-6 is an image of a residential area at an altitude of 690-700 m; FIG. 8-7 is an image of a residential area at an altitude of 700 to 710 m; FIG. 8-8 is an image of a residential area with an altitude of 710-720 m;
FIG. 9 is a diagram showing the result of laying out image samples in the automated method for detecting surface coverage changes of single-phase images of the present invention, wherein FIG. 9-1 is a diagram showing the result of laying out images of cultivated land type; FIG. 9-2 is a diagram of the layout result of forest land type image samples; FIG. 9-3 is a diagram showing the layout result of the image samples of the residential areas;
FIG. 10 is a plot of frequency distribution of index detection of abnormality of intertillage land type according to the automated method for single-phase image surface coverage change detection of the present invention;
FIG. 11 is a histogram of forest anomaly index detection in the single-phase image earth surface coverage change detection automation method of the present invention;
FIG. 12 is a frequency distribution diagram of index detection of abnormality degree of residential land type in the single-stage image ground surface coverage change detection automation method of the present invention;
FIG. 13 is a diagram of the extraction result and abnormal samples of various image samples in the single-phase image ground surface coverage change detection automation method of the present invention, wherein FIG. 13-1 is the extraction result of cultivated land image samples; FIG. 13-2 shows the result of the forest image sample extraction; FIG. 13-3 shows the result of sampling the image of the residential area; FIG. 13-4 is a sample of abnormal farmland; FIGS. 13-5 are samples of forest land anomalies; fig. 13-6 are residential area abnormality samples.
Detailed Description
The following detailed description of embodiments of the invention refers to the accompanying drawings. It should be understood that the detailed description and specific examples, while indicating embodiments of the invention, are given by way of illustration and explanation only, not limitation.
As shown in fig. 1 and 4, in an example of the automated method for detecting surface coverage change of single-phase images provided by the present invention, the method includes the following steps:
A) the method comprises the following steps of segmenting a remote sensing image by utilizing the existing vector data to obtain an initial image object with a prior earth surface coverage type:
specifically, the latest remote sensing image data is segmented by utilizing the existing first-stage historical vector data to obtain an initial image object with a prior earth surface coverage type, the method can fully utilize the size, shape and type information of the image spots in the historical vector data, the problems of unsatisfactory segmentation effect, low classification precision and the like of the existing remote sensing image based on image characteristics are solved, and the segmentation and classification precision of the new image is improved.
B) Extracting image samples of various earth surface coverage types in the research area according to the earth surface coverage types contained in the initial image object and the spatial distribution of the initial image object;
specifically, the spatial arrangement of the initial image objects is obtained according to the range of the research region, the distribution characteristics of the initial image objects in the research region, and the terrain difference in the research region, and the specific obtaining steps are as follows:
firstly, dividing a research area into r multiplied by c sampling grids; secondly, dividing a sampling region into t levels according to terrain difference, then calculating the number of image samples arranged in the sampling grid according to the number of initial image objects in the sampling grid and the terrain fluctuation condition, and then arranging the number SN of the image samples in the r-th row and c-th column sampling gridsr×cThe calculation method comprises the following steps:
where Ion is the total number of initial image objects in the sampling region, STN is the total number of image sample layouts,
the number of the initial image objects with the shape characteristics in the grid of the r-th row and the c-th column as the j-th level is shown. The spatial arrangement of the initial image object can ensure that various sampled image samples of the ground coverage type can show the integral digital characteristics of the initial image object on one hand, and can show the correctness and uniqueness of a typical normal area data set in the initial image object on the other hand, so that the detection precision of the ground coverage change target can be improved.
C) Calculating texture feature values of the image samples of various earth surface coverage types, obtaining optimal texture features of various earth surface coverage types based on texture feature contribution degrees, eliminating abnormal samples, and reserving correct samples to form an image object to be detected:
first, the preferred texture feature acquisition process is as follows: calculating the information Gain of the texture feature of the image sample, and setting Gain (c)i,fj) Representing the jth texture feature parameter fjThe information gain of (1) is:
Gain(ci,fj)＝h(ci)-h(ci/fi)
namely Gain (c)i,fj) The jth texture feature f is shownjWhen the feature value of (2) is known, the feature class ciThe degree of increase and decrease of the information amount indicates the texture feature f as the information gain of the texture feature increasesjTo ground object class ciSince the larger the influence of the classification result is, when texture feature selection is performed, a texture feature having a large information gain is generally selected to construct a preferred feature space vector. However, when the information gain is used for selecting the texture features, the selection result is often biased to the texture features with more value intervals, so that the texture feature evaluation result is inaccurate, and therefore, when the optimal texture features of various earth surface coverage types are obtained, the information gain rate can be further preferably adopted as the basis for selecting the texture features, so that the adverse effects are effectively eliminated. The specific method is to set GainRat (c)i,fj) Representing the jth texture feature parameter fjThe information gain ratio of (1) is as follows:
GainRat(ci,fj)＝Gain(ci,fj)/h(fj)
wherein the content of the first and second substances,
h(fj) Information entropy of j texture features to measure fjThe information gain rate can quantitatively describe the contribution degree of the textural features to the ground object types, the value of the information gain rate is larger than 0 and smaller than 1, and the larger the information gain rate value of a textural feature is, the larger the influence effect of the textural feature on the ground object in the distinguishing process is. Furthermore, according to the relative size of the information gain rate, an index for evaluating the size of the recognition capability of the textural features to the ground object types, namely the contribution degree of the textural features is defined, and
for the contribution degree of the texture feature, there are:
therein, maxf(. to) express texture feature fjMaximum value of information gain ratio, max, in different surface feature classesc{. represents that the recognized feature type is ciOf the different texture features. By adopting the texture feature contribution degree, the relative contribution of the same texture feature to different types of ground objects and the relative contribution of different texture features to the same type of ground objects can be quantified.
The larger the value of (b) is, the larger the contribution of the texture feature to the identification of the ground object class is, the smaller the value of (b) is, the smaller the contribution of the texture feature is. Furthermore, when the ground object type is identified, the contribution can be made according to the texture characteristicsThe magnitude of the value effectively determines which class of texture features to select and to what extent to select such class of texture features. And finally, carrying out weighted fusion processing on different texture features according to the relative contribution of the different texture features to a certain type of ground features (namely ground surface coverage type) so as to obtain the preferred texture features of the ground surface coverage.
Then, rejecting abnormal samples which are inconsistent with the earth surface coverage type to which the image sample belongs in the image samples of certain earth surface coverage types, forming sample objects of the image objects to be detected by the retained normal samples, and checking whether errors occur in the judgment of the earth surface coverage types of the image objects when the latest remote sensing image data is segmented by using the existing historical vector data, so that the accuracy of segmentation and classification of new images can be further improved, as shown in fig. 2, the specific step of screening the abnormal samples is as follows:
first, a preferred feature space vector is established, which is constructed by the preferred texture features of the same type of surface coverage according to the texture feature contribution weighting, for example, the sample object ob in fig. 3jFeature space FSP (ob)j) Can be expressed as:
wherein
I.e. contribution degree of texture feature
n is the number of texture parameters in the sample object, when the contribution of texture of a texture parameter
Then not selecting the texture feature parameter, corresponding to
In addition, f is the difference of the value intervals of the parameter values of the texture features of various texture featuresiAll need to be normalized.
Secondly, calculating the local reachable density of the sample object, and the specific method comprises the following steps: first, the sample object ob is calculatedjTo another sample object obiDistance k between (Rdis)k(obj,obi))：
Rdisk(obj,obi)＝max(k-dis(obi),d(obj,obi))
Wherein d (ob)j,obi) Representing sample objects ob within a preferred feature space vectoriTo the sample object objEuclidean distance of, e.g., d 'in FIG. 3'2I.e. representing the sample object objTo the sample object ob2The euclidean distance between; k-dis (ob)i) Representing a sample object obiOf (a) a kth distance, i.e. the sample object obiAnd a sample object ob in a neighborhood containing k sample objectsiThe distance between the most distant sample objects, e.g. d in fig. 31I.e. representing the sample object ob1And a neighborhood of k sample objects and a sample object ob1The distance between the most distant sample objects. Then, based on the sample object obiTo the sample object objCalculates the sample object ob by the kth reachable distance therebetweenjLocal reachable density (LRD (ob)j))：
Wherein k is a neighborhood parameter indicating the number of sample objects that should be contained in a neighborhood of a sample object, and N is the number of sample objectsk(obj) Representing objects objThe k-th neighborhood of (c).
Finally according to the sample object objCalculating the sample object ob by the local reachable density ofjTexture abnormality index (Fsoi (ob) of (c)j))：
Where D represents the set of sample layout objects, maxi∈D{LRD(obj) Denotes LRD (ob) in the sample object setj) Is measured.
The calculated sample object objTexture anomaly index Fsoi (ob) ofj) And comparing the image sample with a set threshold, if the image sample is larger than the set threshold, the sample object is an abnormal sample, and the sample object is removed so that the image sample is a sample object only with correct retention. In the abnormal sample screening, "sample object obiThe value of k in the kth distance influences the texture abnormality index of the sample object, and generally, the texture abnormality index tends to be relatively stable when the value is determined according to 1/5-1/3 of the total number of samples. The different setting of the texture abnormality index thresholds can cause the difference of the abnormal sample screening results, a lower threshold can obtain a lower missing detection rate and a higher false detection rate of the abnormal sample, and a higher threshold can obtain a higher missing detection rate and a lower false detection rate of the abnormal sample. In the sample extraction result, the missing rate should be emphasized, and the missing rate of 0 may be obtained at the expense of a certain false detection rate, and generally, the missing rate of 0 and higher detection accuracy can be obtained when the abnormality threshold is set at 80% or 70%.
D) Calculating the optimal texture characteristic value of each image object to be detected, and comparing the optimal texture characteristic value of the image object to be detected with the texture characteristic value of the correct sample in the image sample of the same earth surface coverage type to obtain a changed image object; namely, the texture abnormality degree index of the image to be detected is calculated, and an abnormality degree threshold value is set so as to distinguish a changed image object from an unchanged image object.
E) The method comprises the steps of re-segmenting and classifying a changed image object to obtain a change detection result, specifically, re-segmenting the changed image object by adopting a multi-scale segmentation method according to the characteristics of the changed image, and then classifying the re-segmented changed image object. Since the prior class of the changed image object after re-segmentation is unknown, texture abnormality values of the changed image object in all classes of image samples need to be calculated, and the class with the minimum abnormality value smaller than a certain threshold (for example, 20%) is the classification class of the changed image object, as shown in fig. 5, taking the image samples including cultivated land class image samples, forest land class image samples, water body class image samples, and residential area class image samples as an example, the texture abnormality values in the four classes of image samples need to be calculated for the changed image object to be classified, so as to determine which type of surface coverage the changed image object belongs to.
According to the technical scheme, automatic sampling of samples required by the change detection of the remote sensing image in the research area can be effectively realized, so that complicated manual sampling is avoided, the sampling workload is reduced, and the efficiency of the change detection of the remote sensing image is improved; secondly, the method carries out hierarchical sample space layout by extracting prior information of a sampling layer through vector data, can automatically extract samples aiming at specific change detection targets, and can extract specific house and road samples if the damage degree of houses and roads is generally concerned more in post-disaster emergency rescue; finally, the method realizes the automation of the remote sensing image change detection, particularly the automation of the single-phase image change detection and classification, improves the change detection precision, and reduces the harsh conditions of the remote sensing image change detection on the acquisition sensor, time, resolution ratio and the like of the remote sensing image.
The process of detecting changes in the surface coverage according to the present invention will be further described with reference to a specific embodiment:
taking three types of surface coverage of cultivated land, forest land and residential area as examples, as shown in fig. 2, firstly, sampling layers of the cultivated land, the forest land and the residential area are respectively extracted from a reference period vector map, and a sampling area is divided into a regular sampling grid of 10cm by 10cm according to a mapping scale, and if the mapping scale is 1:1000, the ground horizontal distance of the regular grid of 10cm by 10cm is 100m multiplied by 100 m. And then, dividing the image data by using the sampling layer to obtain an image object. According to the topographic features represented by the DEM data (elevation data), the sampling area is divided into 10 or less elevation sampling levels according to the principle of equal altitude distance according to the height difference of the study area, and the distribution characteristics of the image objects of the cultivated land, the forest land and the residential area in different sampling levels are shown in fig. 6 to 8.
The experimental data sampling area comprises 1028 cultivated land image objects, 625 forest land image objects and 159 residential area image objects, and in the experiment, 224 cultivated land image samples, 167 forest land image samples and 92 residential area image samples are planned to be arranged in the sampling area. According to the distribution characteristics of the image objects of the cultivated land, the forest land and the residential area in each layer, the number of samples distributed in each layer in each sampling grid is calculated, then the random sampling method is adopted to distribute the samples of each grid, and the distribution result of the samples in the sampling area is shown in figure 9.
In the screening process of the abnormal samples of the image objects, an optimal characteristic space vector of the image objects is firstly constructed, according to the prior category attribute of the initial image objects, the farmland forms the optimal characteristic space vector by 7 characteristic parameters of angular second moment, contrast, inverse difference moment, entropy, square sum, difference entropy and difference variance, the forest land forms the optimal characteristic space vector by 6 characteristic parameters of angular second moment, inverse difference moment, entropy, mean value, total variance and total average, and the resident land forms the optimal characteristic space vector by 5 characteristic parameters of inverse difference moment, entropy, contrast, difference entropy and angular second moment. Secondly, calculating the kth distance and the kth reachable distance of the sample object, then calculating the local reachable density, and calculating the abnormality index according to the local reachable density. The magnitude of the k value affects the magnitude of the k-th distance and thus the abnormality index of the image object. Fig. 10 to 12 show frequency distribution histograms of sample anomaly size for different surface coverage classes at different k values.
As can be seen from fig. 10 to 12, as the k value increases, the number of low abnormality samples increases and the number of high abnormality samples decreases. When the k value reaches a certain value interval (e.g., the k value is 35 to 90 in fig. 10), the number of samples with high abnormality tends to be stable, and when the k value is too large (close to the total number of samples), almost all the samples with low abnormality (e.g., the k value is 120 in fig. 12), the abnormality detection result fails. Since the video object abnormality detection is to remove a video object having a high degree of abnormality in the sample, it is clear from the test data in fig. 10 to 12 that the k value selected at about 1/5 of the total number of samples does not affect the accuracy of the abnormality detection result.
In order to quantitatively analyze the overall effect of the image object abnormality detection, false detection, missed detection and overall accuracy in the abnormality detection are analyzed under different k values and different abnormality degree thresholds, wherein the false detection means that samples which are originally consistent with the prior category are judged as abnormal samples by mistake, and the missed detection means that the samples are originally abnormal samples and are not detected. The accuracy analysis of the abnormal detection results of the cultivated land, the forest land and the residential area is shown in a table 1.
TABLE 1 analysis of abnormal detection accuracy of cultivated land, forest land and residential area
As can be seen from Table 1, when the abnormality threshold is 80% and the k value is less than 90, the missed detection rate of cultivated land is 0, the missed detection rate of forest land is 21.3, and the missed detection rate of residential area is 32.8. When the abnormality degree threshold value is 80%, and the k value is lower than 65, the missed detection rate of the forest land is 0, and the missed detection rate of the residential area is 0 only when the k value is lower than 50, because the total number of samples of the forest land is less than the cultivated land, and the total number of samples of the residential area is least, the proportion of the k value to the total number of samples has a direct influence on the accuracy of the abnormality detection result. And when the arable land k value is 50-90 values in the interval, the forest land k value is 20-65 values in the interval, the residential land k value is 10-50 values in the interval, and the abnormal degree threshold value is 80%, the false detection rate and the missed detection rate are both zero, and the overall detection accuracy is 100%. Under the condition that the abnormality degree threshold value is set to be 70%, the omission ratio of cultivated land, forest land and residential land is 0 no matter what value k takes. In the abnormal detection of the sample image object, successful sampling is performed as long as the posterior class and the prior class of the extracted sample are consistent in attribute, so that a certain false detection rate can be sacrificed to obtain a 0 missing detection rate in the abnormal detection. Therefore, in order to obtain a 0-missing detection rate and a high detection accuracy, in general, k can be taken as 1/5-1/3 of the total number of the distributed samples, and the abnormality threshold can be set to 80%. In this test, the results of the abnormal sample screening and the surface coverage extraction samples from which the abnormal samples were removed are shown in fig. 13, where k is 50 for cultivated land, k is 35 for forest land, k is 20 for residential land, and the abnormality threshold is set to 80%.
Although the embodiments of the present invention have been described in detail with reference to the accompanying drawings, the embodiments of the present invention are not limited to the details of the above embodiments, and various simple modifications can be made to the technical solutions of the embodiments of the present invention within the technical idea of the embodiments of the present invention, and the simple modifications all belong to the protection scope of the embodiments of the present invention.
It should be noted that the various features described in the above embodiments may be combined in any suitable manner without departing from the scope of the invention. In order to avoid unnecessary repetition, the embodiments of the present invention do not describe every possible combination.
In addition, any combination of various different implementation manners of the embodiments of the present invention is also possible, and the embodiments of the present invention should be considered as disclosed in the embodiments of the present invention as long as the combination does not depart from the spirit of the embodiments of the present invention.
Claims (10)
1. An automatic method for detecting surface coverage change of a single-phase image is characterized by comprising the following steps:
A) segmenting the remote sensing image by utilizing the existing vector data to obtain an initial image object with a prior earth surface coverage type;
B) extracting image samples of various earth surface coverage types in a research area according to the earth surface coverage types contained in the initial image object and the spatial distribution of the initial image object;
C) calculating texture feature values of the image samples of various earth surface coverage types, obtaining preferred texture features of various earth surface coverage types based on texture feature contribution degrees, eliminating abnormal samples, and reserving correct samples to form an image object to be detected;
D) calculating the optimal texture characteristic value of each image object to be detected, and comparing the optimal texture characteristic value of the image object to be detected with the texture characteristic value of the correct sample in the image samples of the same category to obtain a changed image object;
E) and re-segmenting and classifying the changed image object to obtain a change detection result.
2. The automated method for detecting surface coverage changes of a single-phase image according to claim 1, wherein in the step B), the spatial distribution of the initial image objects is obtained according to the range of the research area, the distribution characteristics of the initial image objects in the research area and the terrain difference in the research area, and the method comprises the following steps:
1) dividing said region of interest into r x c sampling grids;
2) dividing the study area into t levels according to the terrain difference;
3) and calculating the number of image samples distributed in the sampling grid according to the number of the initial image objects in the sampling grid and the terrain difference.
3. The automated method for ground surface coverage change detection of single-phase images according to claim 2, wherein in the step C), the preferred texture features of each type of ground surface coverage type are derived from the texture feature contribution degree of the image sample of the type of ground surface coverage; the abnormal samples of the image samples of the earth surface coverage types are obtained by the preferred texture characteristics and texture abnormal degree indexes of the image samples of the earth surface coverage types.
4. The automated method for detecting surface coverage change of single-phase images according to claim 3, wherein the contribution of texture features of each type of image samples is derived from the information gain rate of the texture features.
5. The automated method for single-session image earth surface coverage change detection as claimed in claim 3, wherein the texture abnormality index for each type of the image sample is derived from the local reachable density of the sample object in the image sample.
6. The automated method of single-session image earth-surface coverage change detection as claimed in claim 5, wherein the local reachable density is derived from a kth neighborhood of the sample object and a kth reachable distance between the sample object and another sample object.
7. The automated method for single-session imagery ground cover change detection according to claim 6, wherein the kth reachable distance is a maximum of the kth distance of the another sample object and Euclidean distances between the sample object and the another sample object within a preferred feature space vector space.
8. The automated method for single-phase image earth surface coverage change detection as claimed in claim 7, wherein the preferred feature space vector space is constructed by weighting the preferred texture features of the same earth surface coverage type according to the texture feature contribution degree.
9. The method according to claim 1, wherein in the step D), the changed image object is obtained by screening a texture abnormality degree index between the preferred texture feature value of the image object to be detected and the texture feature value of the correct sample in the image samples of the same type of surface coverage according to a set abnormality degree threshold.
10. The method according to claim 1, wherein in the step E), the changed image object is segmented by a multi-scale segmentation algorithm, and texture abnormality indexes of the segmented changed image object in the image samples of various types of surface coverage are calculated, and the classification of the changed image object is divided according to the texture abnormality indexes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110402357.2A CN113269037A (en) | 2021-04-14 | 2021-04-14 | Single-phase image earth surface coverage change detection automation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110402357.2A CN113269037A (en) | 2021-04-14 | 2021-04-14 | Single-phase image earth surface coverage change detection automation method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113269037A true CN113269037A (en) | 2021-08-17 |
Family
ID=77228947
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110402357.2A Withdrawn CN113269037A (en) | 2021-04-14 | 2021-04-14 | Single-phase image earth surface coverage change detection automation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113269037A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113989657A (en) * | 2021-10-11 | 2022-01-28 | 中国测绘科学研究院 | Method and device for detecting farmland range change based on invariant information sample screening |
| CN113989657B (en) * | 2021-10-11 | 2024-05-10 | 中国测绘科学研究院 | Cultivated land range change detection method and device based on invariable information sample screening |
-
2021
- 2021-04-14 CN CN202110402357.2A patent/CN113269037A/en not_active Withdrawn
Non-Patent Citations (3)
| Title |
|---|
| 吴加敏: "《基于遥感与GIS的土壤盐渍化检测、预测与治理研究》", 31 October 2018, 银川：宁夏人民教育出版社 * |
| 魏东升、周晓光: "《遥感影像变化检测样本自动抽样》", 《遥感学报》 * |
| 魏东升、周晓光: "《顾及纹理特征贡献度的变化影像对象提取算法》", 《测绘学报》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113989657A (en) * | 2021-10-11 | 2022-01-28 | 中国测绘科学研究院 | Method and device for detecting farmland range change based on invariant information sample screening |
| CN113989657B (en) * | 2021-10-11 | 2024-05-10 | 中国测绘科学研究院 | Cultivated land range change detection method and device based on invariable information sample screening |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Lv et al. | Novel land cover change detection method based on K-means clustering and adaptive majority voting using bitemporal remote sensing images | |
| CN108846832B (en) | Multi-temporal remote sensing image and GIS data based change detection method and system | |
| Chen et al. | Isolating individual trees in a savanna woodland using small footprint lidar data | |
| Huang et al. | A multidirectional and multiscale morphological index for automatic building extraction from multispectral GeoEye-1 imagery | |
| CN103268358B (en) | Multi-source control point image database builds and update method | |
| Huo et al. | Towards low vegetation identification: A new method for tree crown segmentation from LiDAR data based on a symmetrical structure detection algorithm (SSD) | |
| CN109829423B (en) | Infrared imaging detection method for frozen lake | |
| CN108896021B (en) | Method for extracting artificial forest stand structure parameters based on aerial photogrammetry point cloud | |
| CN112053371A (en) | Water body extraction method and device in remote sensing image | |
| Yao et al. | Automated detection of 3D individual trees along urban road corridors by mobile laser scanning systems | |
| CN109063577B (en) | Satellite image segmentation optimal segmentation scale determination method based on information gain rate | |
| Zhao et al. | Fusing LiDAR data and aerial imagery for building detection using a vegetation-mask-based connected filter | |
| Dos Santos et al. | Automatic building change detection using multi-temporal airborne LiDAR data | |
| Deng et al. | An automatic cloud detection algorithm for Landsat remote sensing image | |
| CN111882573B (en) | Cultivated land block extraction method and system based on high-resolution image data | |
| CN111060922B (en) | Tree point cloud extraction method based on airborne laser radar point cloud spatial distribution characteristics | |
| CN112906659A (en) | Remote sensing image change detection method based on virtual sample | |
| Zhu et al. | Research on deep learning individual tree segmentation method coupling RetinaNet and point cloud clustering | |
| Yun et al. | Dynamic stratification for vertical forest structure using aerial laser scanning over multiple spatial scales | |
| CN113269037A (en) | Single-phase image earth surface coverage change detection automation method | |
| Zeng et al. | Detecting and measuring fine roots in minirhizotron images using matched filtering and local entropy thresholding | |
| Király et al. | Modelling single trees from terrestrial laser scanning data in a forest reserve | |
| Ukhnaa et al. | Modification of urban built-up area extraction method based on the thematic index-derived bands | |
| CN115063698A (en) | Automatic identification and information extraction method and system for slope surface deformation crack | |
| CN112166696B (en) | Remote sensing image change detection method based on historical spatial information and image segmentation |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20210817 |