A crop detection method and system based on flow feature selection

Abstract

The application discloses a crop detection method and system based on flow feature selection, which firstly solves the Euclidean distance of the crop samples on the earth surface by part of the crop feature information obtained, and obtains the adaptive neighborhood radius of various crops; secondly, the neighborhood relationship between the crop samples is obtained according to the adaptive radius neighborhood; finally, the flow feature selection method based on dependence and importance is used to evaluate the obtained crop feature information in turn, and the optimal feature subset for subsequent crop detection is obtained; the application can adaptively solve various crop samples in different regions and different geographical environments, improve the crop detection accuracy, and fully consider the decision of various crop samples, so that the interference of bad weather on crop detection is reduced.

Description

Technical Field

[0001] This invention belongs to the fields of geography and information science, and in particular relates to a crop detection method and system based on flow feature selection. Background Technology

[0002] To ensure that the geographical environment is suitable for the growth of certain crops, we need to detect the types of crops on the Earth's surface. Existing technologies analyze the identification ability of observed crops using raw reflectance spectra, common indices, different forms of transformed spectra, and hyperspectral feature parameters, while simultaneously constructing a BP neural network model using the most efficient data transformation method. Alternatively, they analyze the classification accuracy of common crops using rule-based decision tree (RBDT) classifiers, object-based nearest neighbor (OBNN) classifiers, and object-based image analysis (OBIA) classifiers, employing separation distance matrices and confusion matrices. However, all existing detection methods require prior acquisition of all information on crops in a given region. Acquiring information on all crops on the Earth's surface requires remote sensing satellites to take millions or even tens of millions of high-resolution images, a process that is extremely costly. Summary of the Invention

[0003] Purpose of the invention: The purpose of this invention is to provide a crop detection method that is highly accurate, less affected by weather, and fully considers the decision-making process of various crop samples.

[0004] Technical solution: The crop detection method based on flow feature selection described in this invention includes the following steps:

[0005] (1) Represent the information on crops on the Earth's surface obtained by remote sensing satellites as a flow characteristic decision system DT =<U,C,D,t> Where U is the set of crop samples, C is the set of crop features, D is the set of crop categories, and t is the time point at which the information was obtained;

[0006] (2) Calculate the remote sensing satellite's position at t using adaptive radius neighborhood. i Crop features f acquired at any time i Dependence The adaptive radius neighborhood relationship is as follows:

[0007]

[0008] Where x ij Let x be the j-th crop sample in the i-th crop sample set. k Let X be the k-th crop sample in U. i Let r(X) be the sample set of the i-th crop. i ) is the neighborhood radius, For sample x ij With Xi Mid-sample center point The Euclidean distance of the set ·, where |·| denotes the cardinality of the set ·, and c is an empirical constant;

[0009] (3) If Greater than the average dependency of all sample features, and Then f i Add B, otherwise discard f. i ; where B is the retained f i The feature subset that makes up the composition;

[0010] (4) If no new features enter the flow feature decision system, stop the flow feature selection process and output B; otherwise, jump to step (2) for iteration.

[0011] Furthermore, the method for calculating the dependency in step (2) is as follows:

[0012] (21) Define the positive region POS of D with respect to *. * (D),

[0013] in, For X i The approximate set under the adaptive radius neighborhood, IND(D) is an equivalence relation induced by D on U, where two samples are taken from U and have the same label;

[0014] (22) Calculate X for each crop i The corresponding neighborhood radius;

[0015] (23) Calculate the Euclidean distance between any two crop samples in U;

[0016] (24) Calculate x ij adaptive radius neighborhood

[0017] (25) will Each crop sample x m Labels and crop samples x ij Compare the tags, if x m The label D(x) m ) and x ij The label D(x) ij If they are the same, then x will be... m Integrate into POS * (D) in;

[0018] (26) Calculate the dependency of D on *.

[0019] Furthermore, step (1) also includes initialization. Dependence of D on B γ B (D) = 0, Mean of the average dependency of features in B γ =0.

[0020] Furthermore, the method for calculating the average dependence in step (3) is as follows:

[0021] Mean γ =Σ f′∈B γ f′ (D) / |B|

[0022] Where f' represents the crop characteristics in B.

[0023] Furthermore, step (3) also includes the following step: If Then, a redundancy elimination process is performed, and the importance Sig(f′,B,D) is calculated for each feature f′ in B. If Sig(f′,B,D)=0, remove f′ from B.

[0024] Furthermore, the representation method of the flow feature decision system described in step (1) is as follows:

[0025] (11) Knowing in advance the crop samples and their categories in a certain region, we can obtain the crop set U and the crop category set D;

[0026] (12) Remote sensing satellites take images of crops and record the time point t when the images are taken;

[0027] (13) Perform grayscale conversion and resolution extraction on the crop image;

[0028] (14) Convert the latitude and longitude information corresponding to the samples in U and the resolution extracted in step (13) into ASCII code to obtain the crop feature set C;

[0029] (15) Represent t, U, C, and D as a flow characteristic decision system DT =<U,C,D,t> .

[0030] The crop detection system based on flow feature selection according to the present invention includes:

[0031] The flow feature decision system generation unit is used to represent the crop feature information acquired by remote sensing satellites as a flow feature decision system;

[0032] The adaptive radius neighborhood calculation unit is used to calculate t based on the adaptive radius neighborhood relationship. i Crop features f acquired at any time i Dependence

[0033] The adaptive radius neighborhood relationship is as follows:

[0034]

[0035] Where x ij Let x be the j-th crop sample in the i-th crop sample set. k Let X be the k-th crop sample in U. i Let r(X) be the sample set of the i-th crop. i ) is the neighborhood radius, For sample x ij With X i Mid-sample center point The Euclidean distance of the set ·, where |·| denotes the cardinality of the set ·, and c is an empirical constant;

[0036] The redundant feature detection module is used to obtain a feature subset B by detecting redundant features based on dependency. Greater than the average dependency of all sample features, and Then f i Add B, otherwise discard f. i If no new features enter the flow feature decision system, the flow feature selection process stops and outputs B; otherwise, it jumps to the adaptive radius neighborhood calculation unit for iteration.

[0037] The electronic device of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it implements the crop detection method based on flow feature selection.

[0038] The computer-readable storage medium of the present invention stores a computer program, which, when executed by a processor, implements the crop detection method based on stream feature selection.

[0039] Beneficial effects: Compared with the prior art, the advantages of this invention are: (1) It can acquire the characteristic information of crops on the Earth's surface hourly and detect the types of crops in real time, which greatly reduces the storage requirements and prepares for the subsequent rational planting and standardized management of crops; (2) It defines a new neighborhood relationship. This method can adaptively calculate the neighborhood radius and select suitable neighbors for all crop samples; (3) It does not require any domain knowledge. The construction of the neighborhood does not require sorting operations. By using the newly defined neighborhood relationship and fully considering decision information, it can effectively resist the interference caused by weather and other factors on crop detection and significantly improve the accuracy of crop detection. Attached Figure Description

[0040] Figure 1 This is a flowchart of the crop detection method based on flow feature selection according to the present invention.

[0041] Figure 2 This is a flowchart illustrating the computational dependency in an embodiment of the present invention.

[0042] Figure 3 This is a flowchart of a flow feature decision system for representing crop feature information, according to an embodiment of the present invention. Detailed Implementation

[0043] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0044] The process of acquiring information via remote sensing satellites can be viewed as a situation where crop samples remain constant, while regional crop characteristic information dynamically increases with the images captured by remote sensing satellites. This is a typical flowing feature problem. Flowing features are defined as features that are added to the data gradually over time, while the samples remain fixed. Flowing feature selection can reduce storage requirements in the data analysis process and improve the generalization ability of downstream learning models. It can also effectively handle the high-dimensional and scalable characteristics of crop data, and can be viewed as a dynamic process performed online, which is very suitable for the dynamic and high-dimensional nature of information acquired by remote sensing satellites. However, it should be noted that not all flowing feature selection methods are suitable for crop detection, because some methods require pre-specifying parameters or frequent sorting operations during the neighborhood construction process. Regarding the specified parameters, the performance of the selected features will depend heavily on the parameter settings. Using the same parameter value when detecting crops in different regions and geographical environments is clearly unreasonable. Regarding the sorting operation, the number of samples acquired by remote sensing satellites is undoubtedly enormous. The sorting operation will inevitably lead to a significant increase in time consumption, which is not conducive to real-time detection. In addition, if the images taken by remote sensing satellites are interfered with, the sorting operation will exacerbate the impact on the data information, thereby interfering with the classification accuracy.

[0045] This invention first uses acquired partial crop feature information to calculate the Euclidean distance of crop samples on the Earth's surface and determine the adaptive neighborhood radius for each type of crop. Second, it introduces the Adapted-R neighborhood method to obtain the neighborhood relationships between crop samples. Finally, it uses a redundancy feature detection method based on dependency and importance to sequentially evaluate the acquired crop feature information, obtaining the optimal feature subset for subsequent crop detection. This method can effectively adapt to various crop samples in different regions and geographical environments, achieving a high accuracy rate in crop detection. Furthermore, this invention fully considers the decision-making processes of various crop samples, which can reduce the interference caused by severe weather on crop detection to a certain extent.

[0046] like Figure 1The diagram illustrates the flow chart of the crop detection method based on flow feature selection described in this invention. This embodiment uses the UCI dataset Statlog (Landsat Satellite) as an example to illustrate the method. The Statlog (Landsat Satellite) dataset consists of multispectral values ​​of pixels in a 3×3 neighborhood of satellite imagery and the classification associated with the center pixel in each neighborhood. It contains 6435 surface samples, 36 features, and 7 different categories. The method of this invention includes the following steps:

[0047] Step 101, represent the Statlog dataset as a streaming feature decision system, denoted as DT =<U,C,D,t> In this context, U is the universe of discourse, representing the set of surface samples; C is the feature set, representing the feature information of the surface samples; and D is the decision, representing the category to which the surface sample belongs. The surface samples are divided into 10 parts according to their quantity. Nine of these parts are used for flow feature selection, and the remaining part is used as the test sample.

[0048] Step 102, Initialization, Dependence of D on B γ B (D) = 0, Mean of the average dependency of features in B γ =0, B represents the selected feature set, which is the set of features retained by f i The feature subset that makes up the composition.

[0049] Step 103, at time point t i A new data entry is obtained from the Statlog dataset, denoted as feature f. i .

[0050] Step 104: Calculate f using the adapted-R neighborhood. i The degree of dependence is denoted as This represents the quality of the information obtained. The Adapted-R neighborhood relation on B is represented as:

[0051]

[0052] Where, x ij Let x be the j-th sample in the i-th type of land surface sample set. k Let X be the k-th sample in the entire surface sample set U. i Let P be the set of the i-th type of land surface samples (1≤i≤P, where P is the number of land surface sample types). Here, c = 1.5, which is an empirical value, |X i |For X i The basis of X represents X i The number of samples in For sample xij With X i The Euclidean distance between the center points of the sample.

[0053] Step 105, recalculate Mean γ =∑ f′∈B γ f′ (D) / |B|, if If the newly acquired surface information indicates that it has a positive effect on improving classification accuracy, then proceed to step 106; otherwise, the information is invalid, discard the surface information, and proceed to step 108.

[0054] Step 106, if Then add this information to B and proceed to step 108; if If one or more pieces of information among the selected information do not improve the detection accuracy, then step 107, the process of eliminating redundancy, will be performed.

[0055] Step 107: Calculate the importance of each feature f′ in B, denoted as Sig(f′,B,D). If Sig(f′,B,D)=0, it indicates that f′ is a redundant feature, so we delete f′ from B. The importance of feature f′ can be defined as:

[0056]

[0057] Step 108: If no new features enter the flow feature selection system, stop the flow feature selection process, output B, and use it for test samples; otherwise, jump to step 103 for the iterative process.

[0058] Step 109, End.

[0059] Figure 2 The diagram shown is a flowchart illustrating the computational dependency of an embodiment of the present invention. This embodiment uses the UCI dataset Statlog (Landsat Satellite) Data Set as an example, and the specific steps are as follows:

[0060] Step 201, Initialization, Decision D with respect to the positive region of * Defined as:

[0061]

[0062] Where * can be B, {f i}∪B、{f′}∪B、f′、f i ; For X i The approximate set under the Adapted-R neighborhood is defined as:

[0063]

[0064] IND(D) is an equivalence relation induced by decision D on the universe of discourse U, where two samples are taken from U and have the same label;

[0065] In addition, we can let U / IND(D) = {X1,X2,…,X} P} indicates the division of all surface samples based on the decision. X i The labels in the Statlog(Image Segmentation) dataset are vegetation, gray soil, etc.

[0066] Step 202, calculate each label X i The corresponding neighborhood radius r(X) i ).

[0067] Step 203: Calculate the Euclidean distance between any two surface samples in the domain of discussion U.

[0068] Step 204, obtain the surface sample x. ij Adapted-R neighborhood

[0069] Step 205, will Each surface sample x m Labels and surface samples x ij Compare the tags, if x m The label D(x) m ) and x ij The label D(x) ij If they are the same, then x will be... m It is incorporated into POS*(D).

[0070] Step 206, calculate the dependency of D on *.

[0071] Step 207, the dependency calculation process ends.

[0072] Figure 3 The diagram shows a flowchart of an embodiment of the present invention that represents the feature information of crops on the Earth's surface acquired by remote sensing satellites as a flow feature decision system. The specific steps are as follows:

[0073] Step 301: A remote sensing satellite captures an image of crops and records the capture time point t.

[0074] Step 302: Convert the image to grayscale.

[0075] Step 303: Extract the resolution of the image.

[0076] Step 304: Obtain the crop sample set of the region in advance and record it as U of the flow feature decision system.

[0077] Step 305: Convert the latitude and longitude information corresponding to each sample obtained in step 304, the resolution information obtained in step 303, and other relevant information into ASCII codes and record them as C of the flow feature decision system.

[0078] Step 306: Record the category to which each sample belongs as D, obtained in step 304;

[0079] Step 307: Represent the t, U, C, and D obtained in steps 301, 304, 305, and 306 as a flow feature decision system DT =<U,C,D,t> .

[0080] Step 308, processing ends.

Claims

1. A crop detection method based on flow feature selection, characterized in that, Includes the following steps: (1) Represent the information on crops on the Earth's surface obtained by remote sensing satellites as a flow characteristic decision system DT =<U,C,D,t> Where U is the set of crop samples, C is the set of crop features, D is the set of crop categories, and t is the time point at which the information was obtained; (2) the dependence of the crop feature f i acquired by the remote sensing satellite at time t i is calculated in combination with an adaptive radius neighborhood The adaptive radius neighborhood relationship is: Where x ij Let x be the j-th crop sample in the i-th crop sample set. k Let X be the k-th crop sample in U. i Let r(X) be the sample set of the i-th crop. i ) is the neighborhood radius, For sample x ij With X i Mid-sample center point The Euclidean distance of the set ·, where |·| denotes the cardinality of the set ·, and c is an empirical constant; (3) if greater than the average dependency of the features of the entire sample, and then f i is added to B, otherwise f i is discarded; wherein B is a subset of features consisting of the retained f i s. (4) If no new features enter the flow feature decision system, stop the flow feature selection process and output B as the optimal feature subset for subsequent crop detection; otherwise, jump to step (2) for iteration. The method for calculating the dependency in step (2) is as follows: (21) define the positive domain of D with respect to * as POS*(D), wherein is X i the adaptive radius neighborhood approximation set, IND(D) is the equivalence relation induced by D on U, two samples taken from U and with the same label; (22) Calculate the number of each crop X i Corresponding neighborhood radius; (23) Calculate the Euclidean distance between any two crop samples in U; (24) Compute x ij Adaptive radius neighborhood of (25) For each crop sample x , compare the label of x m to the label of x ij , and if the label of x m D(x m ) is the same as the label of x ij D(x ij ), then incorporate x m into POS*(D). (26) Compute the dependence of D on * 2.The crop detection method based on flow feature selection according to claim 1, characterized in that, Also included in step (1) is initialization, D dependence of B γ B (D) = 0, Mean dependence of features in B γ = 0. 3.The crop detection method based on flow feature selection according to claim 1, characterized in that, The method for calculating the average dependence in step (3) is as follows: Mean γ =∑ f′∈B γ f′ (D) / |B| Where f' represents the crop characteristics in B. 4.The crop detection method based on flow feature selection according to claim 1, characterized in that, In step (3) further comprises the following steps: if then a redundancy elimination procedure is performed, computing an importance Sig(f',B,D) for each feature f' in B, Sig(f',B,D) = γ B∪{f′} (D) - γ B (D), and if Sig(f',B,D) = 0, f' is removed from B. 5.The crop detection method based on flow feature selection according to claim 1, characterized in that, The representation method of the flow feature decision system in step (1) is as follows: (11) Knowing in advance the crop samples and their categories in a certain region, we can obtain the crop set U and the crop category set D; (12) Remote sensing satellites take images of crops and record the time point t when the images are taken; (13) Perform grayscale conversion and resolution extraction on the crop image; (14) Convert the latitude and longitude information corresponding to the samples in U and the resolution extracted in step (13) into ASCII code to obtain the crop feature set C; (15) Represent t, U, C, and D as a flow characteristic decision system DT =<U,C,D,t> .

6. A crop detection system based on flow feature selection, characterized by, include: The flow feature decision system generation unit is used to represent the crop feature information acquired by remote sensing satellites as a flow feature decision system; an adaptive radius neighborhood computing unit for computing t according to an adaptive radius neighborhood relation i the crop feature f acquired at the moment i the degree of dependence The adaptive radius neighborhood relationship is as follows: Where x ij Let x be the j-th crop sample in the i-th crop sample set. k Let X be the k-th crop sample in U. i Let r(X) be the sample set of the i-th crop. i ) is the neighborhood radius, For sample x ij With X i Mid-sample center point The Euclidean distance of the set ·, where |·| denotes the cardinality of the set ·, and c is an empirical constant; a redundancy feature detection module, configured to perform redundancy feature detection according to the dependence to obtain a feature subset B, if greater than the average dependence of all sample features, and f i is added to B, otherwise f i is discarded; if no new feature enters the flow feature decision system, the flow feature selection process is stopped, and B is output as the optimal feature subset for subsequent crop detection; otherwise, the process jumps to the adaptive radius neighborhood calculation unit for iteration; The method for calculating the dependency in the adaptive radius neighborhood calculation unit is as follows: (21) defining the positive domain POS of D with respect to * * (D), wherein is X i the adaptive radius neighborhood approximation set, IND(D) is the equivalence relation induced by D on U, two samples taken from U and with the same label; (22) Calculate the number of each crop X i corresponding neighborhood radius; (23) Calculate the Euclidean distance between any two crop samples in U; (24) Compute x ij Adaptive radius neighborhood (25) For each crop sample x , compare the label of x m to the label of x ij . If the label of x m D(x m ) is the same as the label of x ij D(x ij ), then x m is incorporated into POS*(D). (26) Compute the dependence of D on * 7. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the computer program is loaded into the processor, it implements the crop detection method based on flow feature selection according to any one of claims 1-5.

8. A computer-readable storage medium storing a computer program, the computer-readable storage medium comprising: When the computer program is executed by the processor, it implements the crop detection method based on flow feature selection according to any one of claims 1-5.

Images