CN103927538B - Threshold selection method for improving spectral angle mapping precision - Google Patents

Abstract

The invention belongs to the technical field of remote sensing information science, and particularly discloses a threshold selection method for improving spectral angle mapping precision. The threshold selection method for improving the spectral angle mapping precision comprises the following steps of obtaining pure end member wave spectrum curves of a target image; classifying the wave spectrum end member curves, and building a wave spectrum curve set of each kind of ground features; calculating spectral angle radian similar values between different kinds of ground features; obtaining the maximum value of spectral angle radian similar value sets of every two target ground features; according to the maximum value of the spectral angle radian similar value sets, obtaining an upper limit value of extraction thresholds of the ground features; calculating spectral angle radian similar values of the same kind of ground features; according to the spectral angle radian similar values of the same kind of ground features, obtaining the minimum value of the radian similar value set of each kind of ground features; according to the minimum value of the radian similar value set of each kind of ground features, obtaining a lower limit value of the extraction thresholds of the ground features; comparing the upper limit value with the lower limit value, selecting the extraction thresholds, and guaranteeing accuracy and integrity of information extraction of the ground features. The method can guarantee the accuracy and the integrity of information extraction of the ground features of the target image.

Description

Threshold selection method for improving spectral angle mapping precision

Technical Field

The invention belongs to the technical field of remote sensing information science. In particular to a threshold value selection method of a spectrum angle mapping method.

Background

The spectral angle mapping method is one of the main methods for identifying the ground features in the imaging spectral image processing technology. The algorithm is an interactive classification method for directly matching an image spectrum with a reference spectrum, and is an automatic classification method for comparing the image spectrum with a ground object spectrum or ground object spectra in a spectrum library. The method fully utilizes the information of the spectral dimension, emphasizes the shape characteristic of the spectrum and greatly reduces the characteristic information.

Currently, the spectral angle mapping method has been successfully applied to surface feature identification in multiple fields, mainly because the method mainly focuses on the form of the spectrum, and reduces the influence of the spectral feature gain and drift. In the spectral corner-filling technique, the selection of the threshold is crucial. The threshold value is small, the information extraction result is missing, the threshold value is large, and the extraction accuracy is reduced. Therefore, a new method is needed to select the threshold value to improve the accuracy and completeness of the spectral corner mapping result.

Disclosure of Invention

The invention aims to provide a threshold value selection method for improving spectral angle mapping precision aiming at the setting defect of a threshold value of a spectral angle, and the method can ensure the accuracy and the integrity of the extraction of the ground object information of a target image.

In order to achieve the purpose, the technical scheme of the invention is as follows: a threshold value selection method for improving spectral angle mapping precision comprises the following steps:

step 1, acquiring a pure end member wave spectrum curve of a target image ground object;

step 2, classifying the pure end member spectrum curves of the target image ground features obtained in the step 1 according to the forms and the characteristic positions of the spectrum curves, determining the number of the ground features of the target image, and constructing a spectrum curve set of each type of target image ground features;

step 3, respectively calculating spectral angle radian similarity values rad between different types of target image ground objects A, B, C by using a spectral analysis tool, wherein rad (A, B), rad (A, C) and rad (B, C) respectively represent spectral angle radian similarity value sets between two types of ground objects of three types of target images A, B, A, C, B and C;

step 4, obtaining the maximum value of the set of similarity values of spectral angle radian between two target objects of A, B, A, C, B and C in the step 3, namely rad (A, B), rad (A, C) and rad (B, C)_AB、_AC、_BC；

Step 5, obtaining the maximum value of the set of similarity values of the spectral angle and the radian according to the step 4_AB、_AC、_BCObtaining the extraction threshold A of the target image ground object A, B, C_threshold、B_threshold、C_thresholdUpper limit value A of_thresholdmax、B_thresholdmax、C_thresholdmax；

Step 6, calculating spectral angle radian similarity values rad (A, A), rad (B, B) and rad (C, C) of the ground objects of the same type of target images by using a spectral analysis tool;

and 7: acquiring each type of target image according to the radian similarity values rad (A, A), rad (B, B) and rad (C, C) of the same type of target image ground object spectral angles obtained in the step 6Minimum β of similarity value set of image feature spectrum angle camber_AA、β_BB、β_CC；

Step 8, obtaining the minimum value β of the radian similarity value set of each type of target image surface feature obtained in the step 7_AA、β_BB、β_CCObtaining the extraction threshold A of the target image ground object A, B, C_threshold、B_threshold、C_thresholdLower limit value A of_thresholdmin、B_thresholdmin、C_thresholdmin；

And step 9: comparing the upper limit value A of the extraction threshold obtained in the step 5_thresholdmax、B_thresholdmax、C_thresholdmaxAnd the lower limit value A obtained in the above step 8_thresholdmin、B_thresholdmin、C_thresholdminThe extraction threshold value is selected, and the accuracy and the integrity of the extraction of the ground feature information of the target image are guaranteed.

In the step 1, a pure end-member wave spectrum curve of the ground object of the target image is obtained by adopting an hourglass method or a continuous maximum-angle convex cone method.

The specific steps of the step 2 are as follows: when there are three types of image target feature A, B, C, the set of spectrum curves of each type of image target feature A, B, C is represented by SA ═ a₁、a₂、a₃…a_i},SB＝{b₁、b₂、b₃…b_j},SC＝{c₁、c₂、c₃…c_k}。

The specific formulas of rad (A, B), rad (A, C) and rad (B, C) in step 3 are respectively as follows:

said step 4_AB、_AC、_BCThe specific formulas are respectively as follows:

_AB＝max(rad(A,B))＝max{rad(a₁,b₁),rad(a₁,b₂),rad(a₁,b₃)...rad(a_i,b_j)}

_AC＝max(rad(A,C))＝max{rad(a₁,c₁),rad(a₁,c₂),rad(a₁,c₃)...rad(a_i,c_k)}

_BC＝max(rad(B,C))＝max{rad(b₁,c₁),rad(b₁,c₂),rad(b₁,c₃)...rad(b_j,c_k)}。

a in the step 5_thresholdmax、B_thresholdmax、C_thresholdmaxThe specific formulas are respectively as follows:

A_thresholdmax＝1-Max(_AB,_AC)

B_thresholdmax＝1-Max(_AB,_BC)

C_thresholdmax＝1-Max(_AC,_BC)。

the specific formulas of rad (A, A), rad (B, B) and rad (C, C) in step 6 are respectively as follows:

rad(A,A)＝{rad(a_n,a_m) Where n ≠ m }is 1,2, … i-1, m ═ n +1, n +2, … i, and n ≠ m }

rad(B,B)＝{rad(b_n,b_m) Wherein n is 1,2, … j-1, m is n +1, n +2, …j and n ≠ m }

rad(C,C)＝{rad(c_n,c_m) Where n ≠ m }is 1,2, … k-1, m ═ n +1, n +2, … k and n ≠ m }

β in the step 7_AA、β_BB、β_CCThe specific formulas are respectively as follows:

β_AA＝min(rad(A,A))＝min{rad(a_n,a_m) Where n ≠ m }is 1,2, … i-1, m ═ n +1, n +2, … i, and n ≠ m }

β_BB＝min(rad(B,B))＝min{rad(b_n,b_m) Where n ≠ m }is 1,2, … j-1, m ═ n +1, n +2, … j, and n ≠ m }

β_CC＝min(rad(C,C))＝min{rad(c_n,c_m) Where n is 1,2, … k-1, m is n +1, n +2, … k, and n ≠ m }.

A in the step (8)_thresholdmin、B_thresholdmin、C_thresholdminThe specific formulas are respectively as follows:

A_thresholdmin＝1-β_AA

B_thresholdmin＝1-β_BB

C_thresholdmin＝1-β_CC。

the step 9 specifically includes the following two cases:

(9.1) when the upper limit value of the extraction threshold value of a certain type of target image surface feature is smaller than the lower limit value, the accuracy and the integrity of the extraction of the information of the target image surface feature cannot be ensured at the same time;

(9.2) when the upper limit value of a certain type of target image ground object is greater than or equal to the lower limit value, the threshold value of the spectrum angle mapping of the target image ground object is equal to the two values or takes any value in the interval between the upper limit value and the lower limit value, and the completeness and the accuracy of information extraction of the target image ground object can be ensured at the same time.

The invention has the following beneficial effects: according to the method, the upper limit value and the lower limit value of the threshold value in the spectrum angle mapping method are selected, and the sizes of the upper limit value and the lower limit value are judged, so that the situation that other ground objects are mixed in the extracted target ground object is avoided, the accuracy and the integrity of the spectrum angle mapping result of the target image ground object are improved, the phenomena of extraction omission and extraction error caused by improper threshold value selection are made up, and the accuracy of extracting the information of the target image ground object is improved.

Detailed Description

The invention is further described below with reference to the accompanying drawings and examples.

The invention provides a threshold selection method suitable for improving the accuracy of spectral angle mapping, which comprises the following steps:

step 1, obtaining a pure end member wave spectrum curve of a target image

The purpose of obtaining the pure end-member wave spectrum curve of the target image is to reduce the later-stage calculation amount to the maximum extent on the premise of not reducing the classification precision. Typically, the number of pixels of an image is thousands or more, and the number of clean end-member spectra is typically less than 100.

An hourglass method or a continuous maximum angle convex cone method can be adopted to obtain the pure end-member wave spectrum curve of the target ground object in the actual image.

And 2, classifying the end member spectrum curves of the target image ground features obtained in the step 1 according to the forms and the characteristic positions of the spectrum curves, determining the number of the ground features of the target image, and constructing a spectrum curve set of each type of target image ground features.

Assuming that the number of end elements of the clean spectrum obtained in step (1) is 50 and the number of types of actual image target land objects is A, B, C, the set of spectrum curves of each type of image target land object A, B, C is represented as SA ═ a₁、a₂、a₃…a_i},SB＝{b₁、b₂、b₃…b_j},SC＝{c₁、c₂、c₃…c_k}. Wherein a, b and c are two-dimensional spectrum scatter curves of the target image ground object A, B, C respectively, a₁The first curve representing the feature a, and so on, i, j, k are the wave curves of the target image feature A, B, C, respectively. 0<i<50,0<j<50,0<k<50 and the sum of the three is equal to 50.

Step 3, using the spectral analysis tool to calculate the spectral angle radian Similarity (rads for short) between different kinds of target image ground objects A, B, C respectively

The range of rad is (0,1), and the two digits after the decimal point are reserved. The larger the value, the higher the similarity between the two spectral curves, and the radian similarity value of the two identical spectral curves is 1.

rad (A, B), rad (A, C), rad (B, C) respectively represent the spectrum angle radian similarity value set between two types of ground objects of three types of target images A and B, A and C, B and C, and the specific formulas of rad (A, B), rad (A, C), rad (B, C) are respectively as follows:

rad(a₁,b₁) Representing A, B spectral angle curvature similarity between two pairs of first spectral curves of two types of ground objects, and rad (a)₁,b₁)＝rad(b₁,a₁). The meaning of i, j, k in the above formula is the same as that in step (2).

The smaller the spectral angle radian similarity value of the spectral curves of the two types of target ground objects is, the larger the separability of the two types of target image ground objects is, namely the larger the morphological difference of the two types of ground object spectral curves is; the larger the similarity value of the spectral angle radian is, the smaller the separability of the two types of target image ground objects is proved, namely the difference of the spectral curve forms of the two types of ground objects is smaller.

And directly calculating the similarity value of the spectral angle radian of the two spectral curves by adopting an ENVI spectral analysis tool.

Step 4, obtaining the maximum value of the set of similarity values of spectral angle radian between every two target images ground objects of A, B, A, C, B and C in the step (3), rad (A, B), rad (A, C) and rad (B, C)_AB、_AC、_BCThe concrete formulas are respectively as follows:

_AB＝max(rad(A,B))＝max{rad(a₁,b₁),rad(a₁,b₂),rad(a₁,b₃)...rad(a_i,b_j)}

_AC＝max(rad(A,C))＝max{rad(a₁,c₁),rad(a₁,c₂),rad(a₁,c₃)...rad(a_i,c_k)}

_BC＝max(rad(B,C))＝max{rad(b₁,c₁),rad(b₁,c₂),rad(b₁,c₃)...rad(b_j,c_k)}

wherein,_AB、_AC、_BCrespectively representing the spectral radian similarity between each two types of target image ground objects. For class A terrain extraction threshold, when_AB＞_ACThen, the separability between the class A feature and the class C feature is larger, and the threshold for extracting the class A feature should not be larger than 1-_ABOtherwise, part of the B-class ground objects are mixed in the extraction result, and the extraction accuracy is reduced; all the same as_AB＜_ACThen, the separability between the class A feature and the class B feature is larger, and the threshold for extracting the class A feature should not be larger than 1-_ACOtherwiseThe extraction result is mixed with part of the class C feature, and the accuracy of extraction is reduced. And (4) analogizing other ground object extraction accuracy in turn, wherein the meaning of i, j and k in the formula is the same as that in the step (2).

Step 5, according to the maximum value of the spectrum angle arc degree similarity value set obtained in the step (4)_AB、_AC、_BCObtaining the extraction threshold A of the target image ground object A, B, C_threshold、B_threshold、C_thresholdUpper limit value A of_thresholdmax、B_thresholdmax、C_thresholdmax，A_thresholdmax、B_thresholdmax、C_thresholdmaxThe specific formulas are respectively as follows:

A_thresholdmax＝1-Max(_AB,_AC)

B_thresholdmax＝1-Max(_AB,_BC)

C_thresholdmax＝1-Max(_AC,_BC)

assuming that in the above formula_AB、_AC、_BCRespectively 0.83, 0.91 and 0.88, and A is obtained by calculation_thresholdmax＝0.09、B_thresholdmax＝0.12、C_thresholdmax0.09. If the upper limit value of the extraction threshold is not calculated, and the target ground object is extracted by directly adopting a default threshold (Maximum radians)0.1 in the ENVI software, the pixels with radian similarity more than or equal to 0.9 with the spectral curve radian of the target ground object are extracted as the target ground object.

When A-type ground objects are extracted, C-type pixels with the similarity of 0.91 with A are extracted as the A-type ground objects. When the class B ground object is extracted, the condition that ground objects of other ground object types are extracted as the class B ground object does not exist; when C-class ground features are extracted, the A-class pixels with the similarity of 0.91 to the C-class ground features are extracted as the C-class ground features.

Step 6, calculating spectral angular arc similarity values rad (A, A), rad (B, B) and rad (C, C) of the same type of target image ground objects by using a spectral analysis tool, wherein the rad (A, A), rad (B, B) and rad (C, C) respectively represent the spectral angular arc similarity value set of the spectral curve among A, B, C types of target image ground objects, and the specific formulas are respectively as follows:

rad(A,A)＝{rad(a_n,a_m) Where n ≠ m }is 1,2, … i-1, m ═ n +1, n +2, … i, and n ≠ m }

rad(B,B)＝{rad(b_n,b_m) Where n ≠ m }is 1,2, … j-1, m ═ n +1, n +2, … j, and n ≠ m }

rad(C,C)＝{rad(c_n,c_m) Where n ≠ m }is 1,2, … k-1, m ═ n +1, n +2, … k and n ≠ m }

The meaning of i, j and k in the formula is the same as that in the step (2), m and n are integer variables, and the value ranges are different in different surface feature types.

Radian similarity values rad (A, A), rad (B, B) and rad (C, C) of the spectral angles represent the variation degree of the spectral curves of the ground objects between classes, and the larger the radian similarity value of the spectral angles is, the smaller the variation of the spectral curves of the ground objects of each class is, and the larger the radian similarity value of the spectral angles is, the larger the radian similarity value of the spectral angles is.

Step 7, acquiring the minimum value β of the radian similarity value set of the ground object of each type of target image according to the radian similarity values rad (A, A), rad (B, B) and rad (C, C) of the spectral angles of the ground object of the same type of target image obtained in the step (6)_AA、β_BB、β_CCThe concrete formulas are respectively as follows:

β_AA＝min(rad(A,A))＝min{rad(a_n,a_m) Where n ≠ m }is 1,2, … i-1, m ═ n +1, n +2, … i, and n ≠ m }

β_BB＝min(rad(B,B))＝min{rad(b_n,b_m) Where n ≠ m }is 1,2, … j-1, m ═ n +1, n +2, … j, and n ≠ m }

β_CC＝min(rad(C,C))＝min{rad(c_n,c_m) Where n ≠ m }is 1,2, … k-1, m ═ n +1, n +2, … k and n ≠ m }

The meaning of i, j, k, m, n in the formula is the same as that in step (6).

The three values β_AA、β_BB、β_CCRespectively representing the approximation degree of the spectral radian in each type of target image ground object, wherein the smaller the value is, the larger the spectral variation of the type of ground object is represented, and the threshold value set value for extracting each type of ground object is more than or equal to 1- β, so that the ground object extraction result is ensured not to have the phenomenon of extraction omission.

Step 8, obtaining the minimum value β of the radian similarity value set of each type of target image ground object obtained in the step (7)_AA、β_BB、β_CCObtaining the extraction threshold A of the target image ground object A, B, C_threshold、B_threshold、C_thresholdLower limit value A of_thresholdmin、B_thresholdmin、C_thresholdmin，A_thresholdmin、B_thresholdmin、C_thresholdminThe specific formulas are respectively as follows:

A_thresholdmin＝1-β_AA

B_thresholdmin＝1-β_BB

C_thresholdmin＝1-β_CC

assume β in the above formula_AA、β _BB 、β _CC Respectively 0.93, 0.88 and 0.89, and A is obtained by calculation_thresholdmin＝0.07、B_thresholdmin＝0.12、C_thresholdmin0.11. If the lower limit value of the extraction threshold is not calculated, directly according to the upper limit value A in the step (5)_thresholdmax＝0.09、B_thresholdmax＝0.12、C_thresholdmaxA threshold value satisfying the condition, i.e., a, is arbitrarily set at 0.09_threshold＝0.05、B_threshold＝0.05、C_thresholdThese three thresholds guarantee the accuracy of each type of extraction, but do not guarantee the integrity of the extraction results, 0.05.

When A-type ground objects are extracted, target pixels with similarity degree of more than or equal to 0.95 with the A-type standard curve are extracted, and target pixels with similarity degree of 0.93-0.95 are not extracted; when B-class ground objects are extracted, target pixels with similarity degree of more than or equal to 0.95 with the B-class standard curve are extracted, and target pixels with similarity degree of 0.88-0.95 are not extracted; when C-class ground objects are extracted, target pixels with similarity degree of more than or equal to 0.95 with the B-class standard curve are extracted, and target pixels with similarity degree of 0.89-0.95 are not extracted; therefore, the accuracy of the spectral angle classification can only be improved only by simply depending on the upper limit value of the threshold, and the completeness of the spectral angle classification cannot be ensured.

And step 9: comparing the upper limit value A of the extraction threshold obtained in the step 5_thresholdmax、B_thresholdmax、C_thresholdmaxAnd the lower limit value A obtained in the above step 8_thresholdmin、B_thresholdmin、C_thresholdminThe extraction threshold value is selected, and the accuracy and the integrity of the extraction of the ground feature information of the target image are guaranteed.

(9.1) when the upper limit value of the extraction threshold value of the target image surface feature is smaller than the lower limit value, the accuracy and the integrity of the extraction of the target image surface feature information can not be ensured at the same time.

For example, the above-mentioned case, namely C, exists as the C-class feature extraction threshold in step (5) and step (8)_thresholdmin＝0.11，C_thresholdmax＝0.09，C_thresholdmin＞C_thresholdmaxIn the extraction process, the completeness of the C-class ground object information extraction can be met only, namely C_threshold0.11; or the accuracy of C-type ground object information extraction is satisfied, namely C_threshold＝0.09。

(9.2) when the upper limit value of a certain type of target image ground object is greater than or equal to the lower limit value, the integrity and the accuracy of information extraction of the target image ground object can be simultaneously ensured;

when the upper limit value of a certain type of target image ground object is equal to the lower limit value, the threshold value of the spectral angle map of the target image ground object is equal to the two values. For exampleThe above condition is present as the class B feature extraction threshold in step (5) and step (8), that is, B_thresholdmin＝0.12，B_thresholdmax＝0.12，B_thresholdmin＝B_thresholdmaxThe threshold B is usually set during the extraction process_threshold0.12, the extraction result of the B-type ground object can be ensured to have no omission or false extraction phenomenon, namely the integrity of the B-type ground object information extraction can be ensured; meanwhile, the accuracy of B-type ground object information extraction can be ensured.

When the upper limit value of a certain type of target image ground object is larger than the lower limit value, the threshold value of the spectral angle mapping is any value in the interval between the upper limit value and the lower limit value. For example, the above-mentioned case exists as the class a feature extraction threshold in step (5) and step (8), that is, a_thresholdmin＝0.07，A_thresholdmax＝0.09，A_thresholdmin＜A_thresholdmaxIf the threshold A is set during the extraction process_threshold0.08, the extraction result can be ensured to have no phenomena of omission and false extraction, namely the integrity of the extraction of the A-type ground object information can be ensured; meanwhile, the completeness and the accuracy of the extraction of the A-type ground objects can be ensured.

The present invention has been described in detail with reference to the embodiments, but the present invention is not limited to the embodiments, and various changes can be made without departing from the gist of the present invention within the knowledge of those skilled in the art. The prior art can be adopted in the content which is not described in detail in the invention.

Claims (8)

1. A threshold value selection method for improving spectral angle mapping precision is characterized by comprising the following steps: the method comprises the following steps:

step 1, acquiring a pure end member wave spectrum curve of a target image ground object;

step 2, according to the form and the characteristic position of the spectrum curve, classifying the pure end member spectrum curve of the target image ground object obtained in the step 1, determining the number of the ground object types of the target image, and constructing a spectrum curve set of each type of target image ground object;

step 3, respectively calculating spectral angle radian similarity values rad between different types of target image ground objects A, B, C by using a spectral analysis tool, wherein rad (A, B), rad (A, C) and rad (B, C) respectively represent spectral angle radian similarity value sets between two types of ground objects of three types of target images A, B, A, C, B and C;

step 4, obtaining the maximum value of the set of similarity values of spectral angle radian between two target objects of A, B, A, C, B and C in the step 3, namely rad (A, B), rad (A, C) and rad (B, C)_AB、_AC、_BC；

Step 5, obtaining the maximum value of the set of similarity values of the spectral angle and the radian according to the step 4_AB、_AC、_BCObtaining the extraction threshold A of the target image ground object A, B, C_threshold、B_threshold、C_thresholdUpper limit value A of_thresholdmax、B_thresholdmax、C_thresholdmax；

A in the step 5_thresholdmax、B_thresholdmax、C_thresholdmaxThe specific formulas are respectively as follows:

A_thresholdmax＝1-Max(_AB,_AC)

B_thresholdmax＝1-Max(_AB,_BC)

C_thresholdmax＝1-Max(_AC,_BC)；

step 6, calculating spectral angle radian similarity values rad (A, A), rad (B, B) and rad (C, C) of the ground objects of the same type of target images by using a spectral analysis tool;

step 7, obtaining the minimum value β of the similarity value set of the spectral angle radian of the ground object of each type of target image according to the similarity values rad (A, A), rad (B, B) and rad (C, C) of the spectral angle radian of the ground object of the same type of target image obtained in the step 6_AA、β_BB、β_CC；

Step 8, obtaining the minimum value β of the similar value set of the angle radian of each type of object image ground object spectrum obtained in the step 7_AA、β_BB、β_CCObtaining the extraction threshold A of the target image ground object A, B, C_threshold、B_threshold、C_thresholdLower limit value A of_thresholdmin、B_thresholdmin、C_thresholdmin；

A in the step 8_thresholdmin、B_thresholdmin、C_thresholdminThe specific formulas are respectively as follows:

A_thresholdmin＝1-β_AA

B_thresholdmin＝1-β_BB

C_thresholdmin＝1-β_CC；

and step 9: comparing the upper limit value A of the extraction threshold obtained in the step 5_thresholdmax、B_thresholdmax、C_thresholdmaxAnd the lower limit value A obtained in the above step 8_thresholdmin、B_thresholdmin、C_thresholdminThe extraction threshold value is selected, and the accuracy and the integrity of the extraction of the ground feature information of the target image are guaranteed.

2. The threshold selection method for improving spectral corner mapping accuracy according to claim 1, wherein: in the step 1, a pure end-member wave spectrum curve of the ground object of the target image is obtained by adopting an hourglass method or a continuous maximum angle convex cone method.

3. The threshold selection method for improving spectral corner mapping accuracy according to claim 2, wherein: the specific steps of the step 2 are as follows: when there are three types of target image feature A, B, C, the set of spectral curves for each type of target image feature A, B, C is denoted as SA ═ a₁、a₂、a₃…a_i},SB＝{b₁、b₂、b₃…b_j},SC＝{c₁、c₂、c₃…c_k}；

Where i, j, and k are the number of the spectral curves of the target image feature A, B, C, respectively.

4. The threshold selection method for improving spectral corner mapping accuracy according to claim 3, wherein: the specific formulas of rad (A, B), rad (A, C) and rad (B, C) in step 3 are respectively as follows:

where i, j, and k are the number of the spectral curves of the target image feature A, B, C, respectively.

5. The threshold selection method for improving spectral corner mapping accuracy according to claim 4, wherein: said step 4_AB、_AC、_BCThe specific formulas are respectively as follows:

_AB＝max(rad(A,B))＝max{rad(a₁,b₁),rad(a₁,b₂),rad(a₁,b₃)...rad(a_i,b_j)}

_AC＝max(rad(A,C))＝max{rad(a₁,c₁),rad(a₁,c₂),rad(a₁,c₃)...rad(a_i,c_k)}

_BC＝max(rad(B,C))＝max{rad(b₁,c₁),rad(b₁,c₂),rad(b₁,c₃)...rad(b_j,c_k)}；

where i, j, and k are the number of the spectral curves of the target image feature A, B, C, respectively.

6. The threshold selection method for improving spectral corner mapping accuracy according to claim 5, wherein: the specific formulas of rad (A, A), rad (B, B) and rad (C, C) in step 6 are respectively as follows:

rad(A,A)＝{rad(a_n,a_m) Where n ≠ m }is 1,2, … i-1, m ═ n +1, n +2, … i, and n ≠ m }

rad(B,B)＝{rad(b_n,b_m) Where n ≠ m }is 1,2, … j-1, m ═ n +1, n +2, … j, and n ≠ m }

rad(C,C)＝{rad(c_n,c_m) Where n ≠ m }is 1,2, … k-1, m ═ n +1, n +2, … k and n ≠ m }

Where i, j, and k are the number of the spectral curves of the target image feature A, B, C, respectively.

7. The method of claim 6, wherein β in step 7 is a threshold selection method for improving spectral corner-fill accuracy_AA、β_BB、β_CCThe specific formulas are respectively as follows:

β_AA＝min(rad(A,A))＝min{rad(a_n,a_m) Where n ≠ m }is 1,2, … i-1, m ═ n +1, n +2, … i, and n ≠ m }

β_BB＝min(rad(B,B))＝min{rad(b_n,b_m) Where n ≠ m }is 1,2, … j-1, m ═ n +1, n +2, … j, and n ≠ m }

β_CC＝min(rad(C,C))＝min{rad(c_n,c_m) Wherein n is 1,2, … k-1, m is n +1, n +2, … k and n ≠ m };

where i, j, and k are the number of the spectral curves of the target image feature A, B, C, respectively.

8. The threshold selection method for improving spectral corner mapping accuracy according to claim 7, wherein: the step 9 specifically includes the following two cases:

(9.1) when the upper limit value of the extraction threshold value of a certain type of target image surface feature is smaller than the lower limit value, the accuracy and the integrity of the extraction of the information of the target image surface feature cannot be ensured at the same time;

(9.2) when the upper limit value of a certain type of target image ground object is greater than or equal to the lower limit value, the threshold value of the spectrum angle mapping of the target image ground object is equal to the two values or takes any value in the interval between the upper limit value and the lower limit value, and the completeness and the accuracy of information extraction of the target image ground object can be ensured at the same time.