WO2021248599A1

WO2021248599A1 - Method and system for automatically identifying spot of which category is abnormal

Info

Publication number: WO2021248599A1
Application number: PCT/CN2020/100451
Authority: WO
Inventors: 李冲; 李昊霖
Original assignee: 自然资源部四川测绘产品质量监督检验站(四川省测绘产品质量监督检验站)
Priority date: 2020-06-12
Filing date: 2020-07-06
Publication date: 2021-12-16
Also published as: AU2021102441A4; CN111709927A; CN111709927B

Abstract

A method and a system for automatically identifying a spot of which the category is abnormal. Said method comprises: acquiring multi-spectral remote sensing images to be identified (101); cropping, on the basis of the geometric range of spots, the multi-spectral remote sensing image to obtain images of the spots of which the categories are error-prone or easily confused (102); calculating the comprehensive spectral index of the images of the spots (103); sorting, according to the size of the comprehensive spectral index of the images of the spots, the images of the spots per element class (104); determining whether there is a spot of which the category is abnormal in the set number of images of the spots preceding and following a spot image sequence (105); if so, adding a set number of images of spots preceding and/or following the spot image sequence, and performing category abnormality identification on the added images of spots, if there is a spot of which the category is abnormal in the added images of spots, skipping to the step of "adding a set number of images of spots preceding and/or following the spot image sequence for identification" until there is no spot of which the category is abnormal in the added images of spots (106). Said method can implement automatic identification of a spot of which the category is abnormal.

Description

Method and system for automatically identifying type abnormal patterns

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on June 12, 2020, the application number is 202010534638.9, and the invention title is "A method and system for automatic identification of abnormal patterns of a type", the entire content of which is incorporated by reference In this application.

Technical field

The present invention relates to the technical field of image processing, and in particular to a method and system for automatically identifying types of abnormal patterns.

Background technique

With the development of society, in the field of geographic and national conditions monitoring, higher requirements are put forward for the accuracy and fineness of surface coverage data. However, due to the complex surface conditions in my country, the uneven quality of remote sensing images, and the large subjective influence of personnel on interpretation, the problem of map classification errors often occurs in the surface coverage data, which seriously affects the quality of the monitoring results of geographic conditions. Each production unit every year , The quality inspection unit needs to spend a lot of manpower, material resources, and financial resources to investigate the large pattern and same-spectrum foreign body errors in the surface coverage data.

Summary of the invention

The purpose of the present invention is to provide a method and system for automatically identifying types of abnormal patterns.

In order to achieve the above objectives, the present invention provides the following solutions:

A method for automatically identifying types of abnormal patterns, including:

Acquire multi-spectral remote sensing images to be identified;

Cropping the multi-spectral remote sensing image based on the geometric range of the pattern to obtain an image of pattern prone to error or confusion;

Calculate the comprehensive spectral index of each patch image, where the comprehensive spectral index is composed of a weighted combination of vegetation index, water index, and differential building index to distinguish the types of patch elements;

According to the size of the comprehensive spectral index of each patch image, sort each patch image by element type;

Determine whether there are abnormal patterns in the set number of pattern images before and after the pattern image sequence;

If there are abnormal categorical patterns, add a set number of pattern images before and/or after the pattern image sequence, and perform category anomaly recognition on the added pattern images. If there are categorical abnormalities in the added pattern image For spots, skip to the step of "adding a set number of spot images before and/or after the spot image sequence for identification" until there are no abnormal spots in the added spot images;

Output abnormal pattern image.

Optionally, before the cropping of the multispectral remote sensing image based on the geometric range of the pattern, the method further includes:

According to historical data, determine the element types that are prone to error or confusion and the weight coefficients of the vegetation index, water index, and differential building index in the comprehensive spectral index used to distinguish the element types that are prone to error or confusion.

Optionally, the sorting of each map spot image according to the size of the comprehensive spectral index of each map spot image by element category specifically includes:

Determine the median or average value of the comprehensive spectral index of each of the pattern images;

According to the size of the median or average value of the comprehensive spectral index of the patch image, each patch image is sorted by feature class.

Optionally, the method for detecting category abnormal patterns includes:

Determine whether the difference of the comprehensive spectral index of adjacent spot images is greater than the set threshold;

If it is, it means that there is an abnormal pattern, and the abnormal pattern is all the patterns between the adjacent pattern image and the first end of the pattern sequence, and the first end is the distance of the pattern sequence. The closer end of the adjacent spot image.

Optionally, before calculating the comprehensive spectral index of each spot image, the method further includes:

Normalizing the vegetation index and the water body index;

Stretching processing is performed on the differential building index and the normalized vegetation index and water body index.

Optionally, the method further includes:

The abnormal pattern image of the category recognized by the computer is output to the human-computer interaction terminal for manual secondary recognition.

Optionally, the cropping process of the multi-spectral remote sensing image is performed in a computer memory.

The present invention also provides an automatic identification system for abnormal patterns, including:

The to-be-identified multi-spectral remote sensing image acquisition module is used to acquire the to-be-identified multi-spectral remote sensing image;

The patch image cropping module is used for cropping the multi-spectral remote sensing image based on the geometric range of the patch to obtain the image of the patch that is prone to error or confusion;

The comprehensive spectral index calculation module is used to calculate the comprehensive spectral index of each patch image, and the comprehensive spectral index is composed of a weighted combination of vegetation index, water index, and differential building index to distinguish the types of the patch elements;

The sorting module is used to sort the image of each patch according to the size of the comprehensive spectral index of the image of each patch;

The category abnormality determination module is used to determine whether there are category abnormal spots in the set number of spot images before and after the spot image sequence, and to determine whether there are category abnormal spots in the added spot images.

Optionally, the system further includes:

The initial parameter determination module is used to determine the error-prone or confusing element types and the weight coefficients of the vegetation index, the water body index, and the differential building index in the comprehensive spectral index used to distinguish the error-prone or confusing element types based on historical data.

Optionally, the sorting module specifically includes:

A numerical value determining unit for determining the median or average value of the comprehensive spectral index of each of the pattern images;

The sorting unit is used for sorting each spot image according to the size of the middle value or average value of the comprehensive spectral index of the spot image.

According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects: the method and system for automatically identifying abnormal patterns provided by the present invention first extract the error-prone or confusing patterns in the multi-spectral remote sensing image to be identified, and then correct Calculate the spectral index of each pattern that can distinguish error-prone or confusing patterns, and sort the available patterns according to the size of the spectral index, and finally classify abnormalities by setting the number of patterns at both ends of the sequence Detection to determine the category of abnormal patterns. The invention realizes the automatic recognition of the abnormal pattern category and improves the detection efficiency.

Attached drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some of the present invention. Embodiments, for those of ordinary skill in the art, without creative work, other drawings can be obtained based on these drawings.

FIG. 1 is a schematic flowchart of a method for automatically identifying abnormal patterns according to Embodiment 1 of the present invention;

Fig. 2 is a schematic structural diagram of an automatic identification system for type abnormal patterns provided by Embodiment 2 of the present invention.

detailed description

The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

In order to make the above objectives, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

Example 1

Fig. 1 is a schematic flow chart of the method for automatic identification of abnormal patterns according to Embodiment 1 of the present invention. Referring to Fig. 1, the method for automatic identification of abnormal patterns provided by this embodiment includes the following steps:

Step 101: Obtain a multi-spectral remote sensing image to be identified;

Step 102: Based on the geometric range of the pattern, crop the multispectral remote sensing image to obtain an image of pattern prone to error or confusion;

Step 103: Calculate the comprehensive spectral index of each map spot image, where the comprehensive spectral index is composed of a weighted combination of vegetation index, water index, and differential building index to distinguish the types of map spot elements;

Step 104: According to the size of the comprehensive spectral index of each map spot image, sort the map spot images according to the feature class;

Step 105: Determine whether there are abnormal patterns in the set number of pattern images before and after the pattern image sequence;

Step 106: If there are abnormal categorical patterns, add a set number of pattern images before and/or after the pattern image sequence, and perform category abnormality recognition on the added pattern images. If the added pattern images are If there are abnormal patterns in the category, skip to the step of "adding a set number of pattern images before and/or after the pattern image sequence for identification" until there are no abnormal patterns in the added pattern image;

Step 107: Output the abnormal pattern image.

In this embodiment, the geometric range of each map spot is sequentially obtained from the ground cover vector data, and the image cropping function is defined. The cropped range is the geometric range of the map spot, and the cropped image is the multi-spectral remote sensing image to be identified , The number of bands of the multi-spectral remote sensing image to be identified should be no less than 4, including at least red, green, blue, near-infrared and other bands. The image cropping function defines the image data processing method for the speckle spectral index. Among them, the preferred Yes, the processing is carried out in the computer memory, and the cutting results are not output to the computer hard disk. Calculate the comprehensive spectral index of each patch image. The comprehensive spectral index is composed of a weighted combination of vegetation index, water index, and differential building index to distinguish the types of patch elements. Of course, in other embodiments, other spectral feature data that can distinguish the types of pattern elements can also be used. After calculating the comprehensive spectral index of each pattern image, the middle value of the comprehensive spectral index of the pattern image can be selected to represent that the patterns are sorted according to their size. Of course, in other embodiments, the average value of the integrated spectral index of the pattern image can also be selected to represent the pattern for pattern sorting. When sorting, it is necessary to sort by feature class, that is, instead of mixing the patches of all feature classes for sorting, the patches belonging to the same feature class are sorted separately, for example, all feature category attributes are paddy fields. The patches of the class are sorted as a group, and the patches of all the feature category attributes are lakes are sorted as a group. After the spot sequence is obtained, first determine whether there are abnormal spots in the set number (such as 2) of the spot images before and after the spot sequence. If they do not exist, it is considered that the spots in the entire spot sequence are normal. , There are no abnormal categorical patterns, if there are abnormal categorical patterns in the set number of pattern images before and after the pattern sequence, for example, there are categorical abnormalities in the set number of pattern images at the front (rear end) of the pattern sequence For spots, add a set number of spots (such as 2 to 3) at the front end (back end) of the spot sequence for category abnormality detection until the increased set number of spots are all normal spots. At the time, it is considered that the last added pattern is a normal pattern, and the patterns identified before this are all abnormal patterns.

Among them, the method for determining the intermediate value of the comprehensive spectral index of the pattern image is as follows: when the number of pixels in the pattern is an odd number, the middle value in the pixel comprehensive spectral index sequence is the middle of the comprehensive spectral index of the pattern. When the number of pixels in the pattern is an even number, the two middle values in the pixel comprehensive spectral index sequence are taken out, and the average of the two values is calculated, which is the middle value of the pattern comprehensive spectral index of the pattern.

In this embodiment, after determining the abnormal pattern, the category abnormal pattern can be directly output, or the category abnormal pattern can be output to the human-computer interaction terminal, and the secondary identification is manually performed.

In this embodiment, before step 102, it may further include: determining the element types that are prone to error or confusion based on historical data and the vegetation index, water index, and the comprehensive spectral index used to distinguish the element types that are prone to error or confusion. The weight coefficient of the difference building index. For example, the statistics on the quality of large map spots and foreign matter with the same spectrum found in the national basic geographic and national conditions monitoring data over the years, focusing on the error-prone and confusing element categories, such as paddy field and dry land, orchard and woodland, etc., research and analyze the image The spectral information of the non-visible light band and the visible light band can obtain the weight coefficients of the vegetation index, the water index and the difference building index in the comprehensive spectral index of the pattern that can distinguish the error-prone and confusing element types.

As an implementation manner, the method for detecting category abnormal patterns may specifically be:

If it is, it means that there is an abnormal pattern, and the abnormal pattern is all the patterns between the adjacent pattern image and the first end of the pattern sequence, and the first end is the distance from the phase in the pattern sequence. The nearest end of the adjacent spot image.

For example, to determine whether there are abnormal patterns in the first 5 patterns in the pattern sequence, the specific operation method is: to judge the comprehensive spectral index of two adjacent patterns among the 5 patterns (such as the middle value of the comprehensive spectral index) Whether the difference between is greater than the set threshold, the set threshold can generally be set to 0.04, if the difference between the integrated spectral index of the 4th and 5th spots (such as the middle value of the integrated spectral index) is greater than the set threshold, then It is considered that the first 4 spots are all abnormal spots.

In this embodiment, when calculating the comprehensive spectral index of each pattern, each index needs to be preprocessed. The preprocessing process may include normalization processing and stretching processing. The specific process is as follows:

First, use the following formula to calculate the normalized vegetation index NDVI of the map spots to highlight the vegetation information in the image.

Among them, B _nir is the near-infrared band _{of remote sensing images, and B red} is the red band of remote sensing images.

Then, use the following formula to calculate the normalized water index NDWI of the patch to highlight the water information in the image.

Among them, B _nir is the near-infrared band _{of remote sensing images, and B green} is the green band of remote sensing images.

After that, use the following formula to calculate the differential building index DSBI of the pattern.

DSBI=k*(B _blue -B _red )+(1-k)*(B _blue -B _green )

Among them, k is the calculation coefficient, which can generally be set to 0.5, B _blue is the blue band of the remote sensing image _{, B red} _{is the red band of the remote sensing image, and B green} is the green band of the remote sensing image.

Finally, the normalized vegetation index, normalized water index, and differential building index are stretched to stretch the value range to [0,1], and the normalized vegetation index and normalized water index are processed according to the following formula Stretching treatment.

Among them, Pixel _v is the pixel value on the original normalized water body or vegetation index, and Pixel' _v is the pixel value after stretching.

The differential building index is stretched according to the following formula.

Among them, Pixel _v is the pixel value on the original differential building index, Pixel' _v is the pixel value after stretching, Pixel _{v_min} is the minimum pixel value of the original differential building index _{for the entire scene, and Pixel v_max} is the original differential building for the entire scene The maximum pixel value of the object index.

The vegetation index NDVI, the water index NDWI, and the building index NSBI are comprehensively weighted according to the following formula to calculate the comprehensive spectral index NCI of the pattern.

Among them, k ₁ is the coefficient of NDVI, k ₂ is the coefficient of NDWI, and k ₃ is the coefficient of NSBI. The values of k ₁ , k ₂ , and k ₃ can be determined based on the statistical analysis results of the spectral information of each type of element. When the value of NDVI is easier to distinguish a certain element type from other element types, the NCI of this type of element is calculated When the _{value of k 1} is set to 1, the value of k ₂ and k ₃ is set to 0; when the value of NDWI is easier to distinguish a certain feature category from other feature categories, when calculating the NCI of this type of feature, k _The value of 2 is set to 1, the value of k ₁ and k ₃ is set to 0; when the value of NSBI is easier to distinguish a certain feature category from other feature categories, when calculating the NCI of this type of feature, the value of _{k 3 is set} The values of 1, k ₁ and k ₂ are set to zero.

Example 2

FIG. 2 is a schematic structural diagram of the automatic identification system for abnormal patterns according to Embodiment 2 of the present invention. Referring to FIG. 2, the automatic identification system for abnormal patterns provided by this embodiment includes:

The to-be-identified multi-spectral remote sensing image acquisition module 201 is used to acquire the to-be-identified multi-spectral remote sensing image;

The patch image cropping module 202 is configured to crop the multi-spectral remote sensing image based on the geometric range of the patch to obtain an image with a pattern that is prone to error or confusion;

The comprehensive spectral index calculation module 203 is used to calculate the comprehensive spectral index of each pattern image, the comprehensive spectral index is composed of a weighted combination of vegetation index, water index, and differential building index to distinguish the types of pattern elements;

The sorting module 204 is used for sorting each spot image according to the size of the comprehensive spectral index of each spot image according to the feature class;

The category abnormality determining module 205 is used to determine whether there are abnormal categorical patterns in the set number of pattern images before and after the pattern image sequence, and to determine whether there are abnormal categorical patterns in the added pattern image.

As an implementation manner, the system further includes:

The initial parameter determination module is used to determine the error-prone or confusing element types based on historical data and the weight coefficients of the vegetation index, water index, and differential building index in the comprehensive spectral index used to distinguish error-prone or confusing element types.

As an implementation manner, the sorting module 204 specifically includes:

As an implementation manner, the category abnormality determining module 205 specifically includes:

The judging unit is used for judging whether the difference of the comprehensive spectral index of adjacent spot images is greater than a set threshold;

The category abnormal pattern determination unit is used to determine the category of all the patterns between the adjacent pattern image and the first end of the pattern sequence when the difference of the comprehensive spectral index of the adjacent pattern image is greater than the set threshold value For an abnormal pattern, the first end is an end of the pattern sequence that is closer to the image of the adjacent pattern.

As an implementation manner, the system further includes:

The normalization processing module is used to perform normalization processing on the vegetation index and the water body index;

The stretching processing module is used to perform stretching processing on the differential building index and the normalized vegetation index and water body index.

The method and system for automatic identification of type abnormal patterns provided by the present invention realizes automatic extraction of large pattern and same-spectrum foreign matter errors, improves the efficiency and reliability of surface coverage data quality detection, and effectively improves the quality of results of geographic and national conditions.

The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. As for the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the related parts can be referred to the description of the method part.

Specific examples are used in this article to illustrate the principles and implementation of the present invention. The description of the above examples is only used to help understand the method and core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the present invention There will be some changes in the specific implementation and application scope of the idea. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

A method for automatically identifying types of abnormal patterns, which is characterized in that it includes:

Acquire multi-spectral remote sensing images to be identified;

Cropping the multi-spectral remote sensing image based on the geometric range of the pattern to obtain an image of pattern prone to error or confusion;

Calculate the comprehensive spectral index of each patch image, where the comprehensive spectral index is composed of a weighted combination of vegetation index, water index, and differential building index to distinguish the types of patch elements;

According to the size of the comprehensive spectral index of each patch image, sort each patch image by element type;

Determine whether there are abnormal patterns in the set number of pattern images before and after the pattern image sequence;

If there are abnormal categorical patterns, add a set number of pattern images before and/or after the pattern image sequence, and perform category anomaly recognition on the added pattern images. If there are categorical abnormalities in the added pattern image For spots, skip to the step of "adding a set number of spot images before and/or after the spot image sequence for identification" until there are no abnormal spots in the added spot images,

Output category abnormal pattern image.
The method for automatically identifying abnormal patterns of types according to claim 1, wherein before the cropping of the multispectral remote sensing image based on the geometric range of the patterns, the method further comprises:

According to historical data, determine the element types that are prone to error or confusion and the weight coefficients of the vegetation index, water index, and differential building index in the comprehensive spectral index used to distinguish the element types that are prone to error or confusion.
The method for automatically identifying types of abnormal patterns according to claim 1, wherein the sorting of each pattern image according to the comprehensive spectral index of each pattern image according to the feature class specifically includes:

Determine the median or average value of the comprehensive spectral index of each of the pattern images;

According to the size of the median or average value of the comprehensive spectral index of the patch image, each patch image is sorted by feature class.
The method for automatically identifying types of abnormal patterns according to claim 1, wherein the method for detecting abnormal patterns of categories comprises:

Determine whether the difference of the comprehensive spectral index of adjacent spot images is greater than the set threshold;

If it is, it means that there is an abnormal pattern, and the abnormal pattern is all the patterns between the adjacent pattern image and the first end of the pattern sequence, and the first end is the distance of the pattern sequence. The closer end of the adjacent spot image.
The method for automatically identifying abnormal patterns of types according to claim 1, wherein before said calculating the comprehensive spectral index of each pattern image, the method further comprises:

Normalizing the vegetation index and the water body index;

Stretching processing is performed on the differential building index and the normalized vegetation index and water body index.
The method for automatically identifying type abnormal patterns according to claim 1, wherein the method further comprises:

The abnormal pattern image of the category recognized by the computer is output to the human-computer interaction terminal for manual secondary recognition.
The method for automatically identifying type abnormal patterns according to claim 1, wherein the cutting process of the multi-spectral remote sensing image is performed in a computer memory.
A type of automatic identification system for abnormal patterns, which is characterized in that it includes:

The to-be-identified multi-spectral remote sensing image acquisition module is used to acquire the to-be-identified multi-spectral remote sensing image;

The patch image cropping module is used for cropping the multi-spectral remote sensing image based on the geometric range of the patch to obtain the image of the patch that is prone to error or confusion;

The comprehensive spectral index calculation module is used to calculate the comprehensive spectral index of each pattern image, the comprehensive spectral index is composed of a weighted combination of vegetation index, water index, and differential building index to distinguish the types of pattern elements;

The sorting module is used to sort the image of each map spot according to the size of the comprehensive spectral index of each map spot image according to the feature class;

The category anomaly determination module is used to determine whether there are categorical abnormal patterns in the set number of pattern images before and after the pattern image sequence, and to determine whether there are categorical abnormal patterns in the added pattern image.
The automatic identification system for type abnormal patterns according to claim 8, wherein the system further comprises:

The initial parameter determination module is used to determine the error-prone or confusing element types and the weight coefficients of the vegetation index, water index, and differential building index in the comprehensive spectral index used to distinguish error-prone or confusing element types based on historical data.
The automatic identification system for type abnormal patterns according to claim 8, wherein the sorting module specifically comprises:

A numerical value determining unit for determining the median or average value of the comprehensive spectral index of each of the pattern images;

The sorting unit is used for sorting each spot image according to the size of the middle value or average value of the comprehensive spectral index of the spot image.